linux/arch/x86/kvm/svm.c

3042 lines
77 KiB
C
Raw Normal View History

[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
/*
* Kernel-based Virtual Machine driver for Linux
*
* AMD SVM support
*
* Copyright (C) 2006 Qumranet, Inc.
*
* Authors:
* Yaniv Kamay <yaniv@qumranet.com>
* Avi Kivity <avi@qumranet.com>
*
* This work is licensed under the terms of the GNU GPL, version 2. See
* the COPYING file in the top-level directory.
*
*/
#include <linux/kvm_host.h>
#include "irq.h"
#include "mmu.h"
#include "kvm_cache_regs.h"
#include "x86.h"
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
#include <linux/module.h>
#include <linux/kernel.h>
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
#include <linux/vmalloc.h>
#include <linux/highmem.h>
#include <linux/sched.h>
#include <linux/ftrace_event.h>
include cleanup: Update gfp.h and slab.h includes to prepare for breaking implicit slab.h inclusion from percpu.h percpu.h is included by sched.h and module.h and thus ends up being included when building most .c files. percpu.h includes slab.h which in turn includes gfp.h making everything defined by the two files universally available and complicating inclusion dependencies. percpu.h -> slab.h dependency is about to be removed. Prepare for this change by updating users of gfp and slab facilities include those headers directly instead of assuming availability. As this conversion needs to touch large number of source files, the following script is used as the basis of conversion. http://userweb.kernel.org/~tj/misc/slabh-sweep.py The script does the followings. * Scan files for gfp and slab usages and update includes such that only the necessary includes are there. ie. if only gfp is used, gfp.h, if slab is used, slab.h. * When the script inserts a new include, it looks at the include blocks and try to put the new include such that its order conforms to its surrounding. It's put in the include block which contains core kernel includes, in the same order that the rest are ordered - alphabetical, Christmas tree, rev-Xmas-tree or at the end if there doesn't seem to be any matching order. * If the script can't find a place to put a new include (mostly because the file doesn't have fitting include block), it prints out an error message indicating which .h file needs to be added to the file. The conversion was done in the following steps. 1. The initial automatic conversion of all .c files updated slightly over 4000 files, deleting around 700 includes and adding ~480 gfp.h and ~3000 slab.h inclusions. The script emitted errors for ~400 files. 2. Each error was manually checked. Some didn't need the inclusion, some needed manual addition while adding it to implementation .h or embedding .c file was more appropriate for others. This step added inclusions to around 150 files. 3. The script was run again and the output was compared to the edits from #2 to make sure no file was left behind. 4. Several build tests were done and a couple of problems were fixed. e.g. lib/decompress_*.c used malloc/free() wrappers around slab APIs requiring slab.h to be added manually. 5. The script was run on all .h files but without automatically editing them as sprinkling gfp.h and slab.h inclusions around .h files could easily lead to inclusion dependency hell. Most gfp.h inclusion directives were ignored as stuff from gfp.h was usually wildly available and often used in preprocessor macros. Each slab.h inclusion directive was examined and added manually as necessary. 6. percpu.h was updated not to include slab.h. 7. Build test were done on the following configurations and failures were fixed. CONFIG_GCOV_KERNEL was turned off for all tests (as my distributed build env didn't work with gcov compiles) and a few more options had to be turned off depending on archs to make things build (like ipr on powerpc/64 which failed due to missing writeq). * x86 and x86_64 UP and SMP allmodconfig and a custom test config. * powerpc and powerpc64 SMP allmodconfig * sparc and sparc64 SMP allmodconfig * ia64 SMP allmodconfig * s390 SMP allmodconfig * alpha SMP allmodconfig * um on x86_64 SMP allmodconfig 8. percpu.h modifications were reverted so that it could be applied as a separate patch and serve as bisection point. Given the fact that I had only a couple of failures from tests on step 6, I'm fairly confident about the coverage of this conversion patch. If there is a breakage, it's likely to be something in one of the arch headers which should be easily discoverable easily on most builds of the specific arch. Signed-off-by: Tejun Heo <tj@kernel.org> Guess-its-ok-by: Christoph Lameter <cl@linux-foundation.org> Cc: Ingo Molnar <mingo@redhat.com> Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
2010-03-24 08:04:11 +00:00
#include <linux/slab.h>
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
#include <asm/desc.h>
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
#include <asm/virtext.h>
#include "trace.h"
#define __ex(x) __kvm_handle_fault_on_reboot(x)
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
MODULE_AUTHOR("Qumranet");
MODULE_LICENSE("GPL");
#define IOPM_ALLOC_ORDER 2
#define MSRPM_ALLOC_ORDER 1
#define SEG_TYPE_LDT 2
#define SEG_TYPE_BUSY_TSS16 3
#define SVM_FEATURE_NPT (1 << 0)
#define SVM_FEATURE_LBRV (1 << 1)
#define SVM_FEATURE_SVML (1 << 2)
#define SVM_FEATURE_PAUSE_FILTER (1 << 10)
#define NESTED_EXIT_HOST 0 /* Exit handled on host level */
#define NESTED_EXIT_DONE 1 /* Exit caused nested vmexit */
#define NESTED_EXIT_CONTINUE 2 /* Further checks needed */
#define DEBUGCTL_RESERVED_BITS (~(0x3fULL))
static const u32 host_save_user_msrs[] = {
#ifdef CONFIG_X86_64
MSR_STAR, MSR_LSTAR, MSR_CSTAR, MSR_SYSCALL_MASK, MSR_KERNEL_GS_BASE,
MSR_FS_BASE,
#endif
MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
};
#define NR_HOST_SAVE_USER_MSRS ARRAY_SIZE(host_save_user_msrs)
struct kvm_vcpu;
struct nested_state {
struct vmcb *hsave;
u64 hsave_msr;
u64 vmcb;
/* These are the merged vectors */
u32 *msrpm;
/* gpa pointers to the real vectors */
u64 vmcb_msrpm;
/* A VMEXIT is required but not yet emulated */
bool exit_required;
/* cache for intercepts of the guest */
u16 intercept_cr_read;
u16 intercept_cr_write;
u16 intercept_dr_read;
u16 intercept_dr_write;
u32 intercept_exceptions;
u64 intercept;
};
struct vcpu_svm {
struct kvm_vcpu vcpu;
struct vmcb *vmcb;
unsigned long vmcb_pa;
struct svm_cpu_data *svm_data;
uint64_t asid_generation;
uint64_t sysenter_esp;
uint64_t sysenter_eip;
u64 next_rip;
u64 host_user_msrs[NR_HOST_SAVE_USER_MSRS];
u64 host_gs_base;
u32 *msrpm;
struct nested_state nested;
bool nmi_singlestep;
};
/* enable NPT for AMD64 and X86 with PAE */
#if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE)
static bool npt_enabled = true;
#else
static bool npt_enabled = false;
#endif
static int npt = 1;
module_param(npt, int, S_IRUGO);
static int nested = 1;
module_param(nested, int, S_IRUGO);
static void svm_flush_tlb(struct kvm_vcpu *vcpu);
static void svm_complete_interrupts(struct vcpu_svm *svm);
static int nested_svm_exit_handled(struct vcpu_svm *svm);
static int nested_svm_vmexit(struct vcpu_svm *svm);
static int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
bool has_error_code, u32 error_code);
static inline struct vcpu_svm *to_svm(struct kvm_vcpu *vcpu)
{
return container_of(vcpu, struct vcpu_svm, vcpu);
}
static inline bool is_nested(struct vcpu_svm *svm)
{
return svm->nested.vmcb;
}
static inline void enable_gif(struct vcpu_svm *svm)
{
svm->vcpu.arch.hflags |= HF_GIF_MASK;
}
static inline void disable_gif(struct vcpu_svm *svm)
{
svm->vcpu.arch.hflags &= ~HF_GIF_MASK;
}
static inline bool gif_set(struct vcpu_svm *svm)
{
return !!(svm->vcpu.arch.hflags & HF_GIF_MASK);
}
static unsigned long iopm_base;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
struct kvm_ldttss_desc {
u16 limit0;
u16 base0;
unsigned base1 : 8, type : 5, dpl : 2, p : 1;
unsigned limit1 : 4, zero0 : 3, g : 1, base2 : 8;
u32 base3;
u32 zero1;
} __attribute__((packed));
struct svm_cpu_data {
int cpu;
u64 asid_generation;
u32 max_asid;
u32 next_asid;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
struct kvm_ldttss_desc *tss_desc;
struct page *save_area;
};
static DEFINE_PER_CPU(struct svm_cpu_data *, svm_data);
static uint32_t svm_features;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
struct svm_init_data {
int cpu;
int r;
};
static u32 msrpm_ranges[] = {0, 0xc0000000, 0xc0010000};
#define NUM_MSR_MAPS ARRAY_SIZE(msrpm_ranges)
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
#define MSRS_RANGE_SIZE 2048
#define MSRS_IN_RANGE (MSRS_RANGE_SIZE * 8 / 2)
#define MAX_INST_SIZE 15
static inline u32 svm_has(u32 feat)
{
return svm_features & feat;
}
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
static inline void clgi(void)
{
asm volatile (__ex(SVM_CLGI));
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
}
static inline void stgi(void)
{
asm volatile (__ex(SVM_STGI));
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
}
static inline void invlpga(unsigned long addr, u32 asid)
{
asm volatile (__ex(SVM_INVLPGA) :: "a"(addr), "c"(asid));
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
}
static inline void force_new_asid(struct kvm_vcpu *vcpu)
{
to_svm(vcpu)->asid_generation--;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
}
static inline void flush_guest_tlb(struct kvm_vcpu *vcpu)
{
force_new_asid(vcpu);
}
static void svm_set_efer(struct kvm_vcpu *vcpu, u64 efer)
{
if (!npt_enabled && !(efer & EFER_LMA))
efer &= ~EFER_LME;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
to_svm(vcpu)->vmcb->save.efer = efer | EFER_SVME;
vcpu->arch.efer = efer;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
}
static void svm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
bool has_error_code, u32 error_code)
{
struct vcpu_svm *svm = to_svm(vcpu);
/* If we are within a nested VM we'd better #VMEXIT and let the
guest handle the exception */
if (nested_svm_check_exception(svm, nr, has_error_code, error_code))
return;
svm->vmcb->control.event_inj = nr
| SVM_EVTINJ_VALID
| (has_error_code ? SVM_EVTINJ_VALID_ERR : 0)
| SVM_EVTINJ_TYPE_EXEPT;
svm->vmcb->control.event_inj_err = error_code;
}
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
static int is_external_interrupt(u32 info)
{
info &= SVM_EVTINJ_TYPE_MASK | SVM_EVTINJ_VALID;
return info == (SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR);
}
static u32 svm_get_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
{
struct vcpu_svm *svm = to_svm(vcpu);
u32 ret = 0;
if (svm->vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK)
ret |= X86_SHADOW_INT_STI | X86_SHADOW_INT_MOV_SS;
return ret & mask;
}
static void svm_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
{
struct vcpu_svm *svm = to_svm(vcpu);
if (mask == 0)
svm->vmcb->control.int_state &= ~SVM_INTERRUPT_SHADOW_MASK;
else
svm->vmcb->control.int_state |= SVM_INTERRUPT_SHADOW_MASK;
}
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
{
struct vcpu_svm *svm = to_svm(vcpu);
if (!svm->next_rip) {
if (emulate_instruction(vcpu, 0, 0, EMULTYPE_SKIP) !=
EMULATE_DONE)
printk(KERN_DEBUG "%s: NOP\n", __func__);
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
return;
}
if (svm->next_rip - kvm_rip_read(vcpu) > MAX_INST_SIZE)
printk(KERN_ERR "%s: ip 0x%lx next 0x%llx\n",
__func__, kvm_rip_read(vcpu), svm->next_rip);
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
kvm_rip_write(vcpu, svm->next_rip);
svm_set_interrupt_shadow(vcpu, 0);
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
}
static int has_svm(void)
{
const char *msg;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
if (!cpu_has_svm(&msg)) {
printk(KERN_INFO "has_svm: %s\n", msg);
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
return 0;
}
return 1;
}
static void svm_hardware_disable(void *garbage)
{
cpu_svm_disable();
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
}
static int svm_hardware_enable(void *garbage)
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
{
struct svm_cpu_data *sd;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
uint64_t efer;
struct desc_ptr gdt_descr;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
struct desc_struct *gdt;
int me = raw_smp_processor_id();
rdmsrl(MSR_EFER, efer);
if (efer & EFER_SVME)
return -EBUSY;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
if (!has_svm()) {
printk(KERN_ERR "svm_hardware_enable: err EOPNOTSUPP on %d\n",
me);
return -EINVAL;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
}
sd = per_cpu(svm_data, me);
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
if (!sd) {
printk(KERN_ERR "svm_hardware_enable: svm_data is NULL on %d\n",
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
me);
return -EINVAL;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
}
sd->asid_generation = 1;
sd->max_asid = cpuid_ebx(SVM_CPUID_FUNC) - 1;
sd->next_asid = sd->max_asid + 1;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
kvm_get_gdt(&gdt_descr);
gdt = (struct desc_struct *)gdt_descr.address;
sd->tss_desc = (struct kvm_ldttss_desc *)(gdt + GDT_ENTRY_TSS);
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
wrmsrl(MSR_EFER, efer | EFER_SVME);
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
wrmsrl(MSR_VM_HSAVE_PA, page_to_pfn(sd->save_area) << PAGE_SHIFT);
return 0;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
}
static void svm_cpu_uninit(int cpu)
{
struct svm_cpu_data *sd = per_cpu(svm_data, raw_smp_processor_id());
if (!sd)
return;
per_cpu(svm_data, raw_smp_processor_id()) = NULL;
__free_page(sd->save_area);
kfree(sd);
}
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
static int svm_cpu_init(int cpu)
{
struct svm_cpu_data *sd;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
int r;
sd = kzalloc(sizeof(struct svm_cpu_data), GFP_KERNEL);
if (!sd)
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
return -ENOMEM;
sd->cpu = cpu;
sd->save_area = alloc_page(GFP_KERNEL);
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
r = -ENOMEM;
if (!sd->save_area)
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
goto err_1;
per_cpu(svm_data, cpu) = sd;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
return 0;
err_1:
kfree(sd);
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
return r;
}
static void set_msr_interception(u32 *msrpm, unsigned msr,
int read, int write)
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
{
int i;
for (i = 0; i < NUM_MSR_MAPS; i++) {
if (msr >= msrpm_ranges[i] &&
msr < msrpm_ranges[i] + MSRS_IN_RANGE) {
u32 msr_offset = (i * MSRS_IN_RANGE + msr -
msrpm_ranges[i]) * 2;
u32 *base = msrpm + (msr_offset / 32);
u32 msr_shift = msr_offset % 32;
u32 mask = ((write) ? 0 : 2) | ((read) ? 0 : 1);
*base = (*base & ~(0x3 << msr_shift)) |
(mask << msr_shift);
return;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
}
}
BUG();
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
}
static void svm_vcpu_init_msrpm(u32 *msrpm)
{
memset(msrpm, 0xff, PAGE_SIZE * (1 << MSRPM_ALLOC_ORDER));
#ifdef CONFIG_X86_64
set_msr_interception(msrpm, MSR_GS_BASE, 1, 1);
set_msr_interception(msrpm, MSR_FS_BASE, 1, 1);
set_msr_interception(msrpm, MSR_KERNEL_GS_BASE, 1, 1);
set_msr_interception(msrpm, MSR_LSTAR, 1, 1);
set_msr_interception(msrpm, MSR_CSTAR, 1, 1);
set_msr_interception(msrpm, MSR_SYSCALL_MASK, 1, 1);
#endif
set_msr_interception(msrpm, MSR_K6_STAR, 1, 1);
set_msr_interception(msrpm, MSR_IA32_SYSENTER_CS, 1, 1);
}
static void svm_enable_lbrv(struct vcpu_svm *svm)
{
u32 *msrpm = svm->msrpm;
svm->vmcb->control.lbr_ctl = 1;
set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 1, 1);
set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 1, 1);
set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 1, 1);
set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 1, 1);
}
static void svm_disable_lbrv(struct vcpu_svm *svm)
{
u32 *msrpm = svm->msrpm;
svm->vmcb->control.lbr_ctl = 0;
set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 0, 0);
set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 0, 0);
set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 0, 0);
set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 0, 0);
}
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
static __init int svm_hardware_setup(void)
{
int cpu;
struct page *iopm_pages;
void *iopm_va;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
int r;
iopm_pages = alloc_pages(GFP_KERNEL, IOPM_ALLOC_ORDER);
if (!iopm_pages)
return -ENOMEM;
iopm_va = page_address(iopm_pages);
memset(iopm_va, 0xff, PAGE_SIZE * (1 << IOPM_ALLOC_ORDER));
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
iopm_base = page_to_pfn(iopm_pages) << PAGE_SHIFT;
if (boot_cpu_has(X86_FEATURE_NX))
kvm_enable_efer_bits(EFER_NX);
if (boot_cpu_has(X86_FEATURE_FXSR_OPT))
kvm_enable_efer_bits(EFER_FFXSR);
if (nested) {
printk(KERN_INFO "kvm: Nested Virtualization enabled\n");
kvm_enable_efer_bits(EFER_SVME);
}
for_each_possible_cpu(cpu) {
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
r = svm_cpu_init(cpu);
if (r)
goto err;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
}
svm_features = cpuid_edx(SVM_CPUID_FUNC);
if (!svm_has(SVM_FEATURE_NPT))
npt_enabled = false;
if (npt_enabled && !npt) {
printk(KERN_INFO "kvm: Nested Paging disabled\n");
npt_enabled = false;
}
if (npt_enabled) {
printk(KERN_INFO "kvm: Nested Paging enabled\n");
kvm_enable_tdp();
} else
kvm_disable_tdp();
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
return 0;
err:
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
__free_pages(iopm_pages, IOPM_ALLOC_ORDER);
iopm_base = 0;
return r;
}
static __exit void svm_hardware_unsetup(void)
{
int cpu;
for_each_possible_cpu(cpu)
svm_cpu_uninit(cpu);
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
__free_pages(pfn_to_page(iopm_base >> PAGE_SHIFT), IOPM_ALLOC_ORDER);
iopm_base = 0;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
}
static void init_seg(struct vmcb_seg *seg)
{
seg->selector = 0;
seg->attrib = SVM_SELECTOR_P_MASK | SVM_SELECTOR_S_MASK |
SVM_SELECTOR_WRITE_MASK; /* Read/Write Data Segment */
seg->limit = 0xffff;
seg->base = 0;
}
static void init_sys_seg(struct vmcb_seg *seg, uint32_t type)
{
seg->selector = 0;
seg->attrib = SVM_SELECTOR_P_MASK | type;
seg->limit = 0xffff;
seg->base = 0;
}
static void init_vmcb(struct vcpu_svm *svm)
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
{
struct vmcb_control_area *control = &svm->vmcb->control;
struct vmcb_save_area *save = &svm->vmcb->save;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
svm->vcpu.fpu_active = 1;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
control->intercept_cr_read = INTERCEPT_CR0_MASK |
INTERCEPT_CR3_MASK |
INTERCEPT_CR4_MASK;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
control->intercept_cr_write = INTERCEPT_CR0_MASK |
INTERCEPT_CR3_MASK |
INTERCEPT_CR4_MASK |
INTERCEPT_CR8_MASK;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
control->intercept_dr_read = INTERCEPT_DR0_MASK |
INTERCEPT_DR1_MASK |
INTERCEPT_DR2_MASK |
INTERCEPT_DR3_MASK |
INTERCEPT_DR4_MASK |
INTERCEPT_DR5_MASK |
INTERCEPT_DR6_MASK |
INTERCEPT_DR7_MASK;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
control->intercept_dr_write = INTERCEPT_DR0_MASK |
INTERCEPT_DR1_MASK |
INTERCEPT_DR2_MASK |
INTERCEPT_DR3_MASK |
INTERCEPT_DR4_MASK |
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
INTERCEPT_DR5_MASK |
INTERCEPT_DR6_MASK |
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
INTERCEPT_DR7_MASK;
control->intercept_exceptions = (1 << PF_VECTOR) |
(1 << UD_VECTOR) |
(1 << MC_VECTOR);
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
control->intercept = (1ULL << INTERCEPT_INTR) |
(1ULL << INTERCEPT_NMI) |
(1ULL << INTERCEPT_SMI) |
(1ULL << INTERCEPT_SELECTIVE_CR0) |
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
(1ULL << INTERCEPT_CPUID) |
(1ULL << INTERCEPT_INVD) |
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
(1ULL << INTERCEPT_HLT) |
(1ULL << INTERCEPT_INVLPG) |
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
(1ULL << INTERCEPT_INVLPGA) |
(1ULL << INTERCEPT_IOIO_PROT) |
(1ULL << INTERCEPT_MSR_PROT) |
(1ULL << INTERCEPT_TASK_SWITCH) |
(1ULL << INTERCEPT_SHUTDOWN) |
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
(1ULL << INTERCEPT_VMRUN) |
(1ULL << INTERCEPT_VMMCALL) |
(1ULL << INTERCEPT_VMLOAD) |
(1ULL << INTERCEPT_VMSAVE) |
(1ULL << INTERCEPT_STGI) |
(1ULL << INTERCEPT_CLGI) |
(1ULL << INTERCEPT_SKINIT) |
(1ULL << INTERCEPT_WBINVD) |
(1ULL << INTERCEPT_MONITOR) |
(1ULL << INTERCEPT_MWAIT);
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
control->iopm_base_pa = iopm_base;
control->msrpm_base_pa = __pa(svm->msrpm);
control->tsc_offset = 0;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
control->int_ctl = V_INTR_MASKING_MASK;
init_seg(&save->es);
init_seg(&save->ss);
init_seg(&save->ds);
init_seg(&save->fs);
init_seg(&save->gs);
save->cs.selector = 0xf000;
/* Executable/Readable Code Segment */
save->cs.attrib = SVM_SELECTOR_READ_MASK | SVM_SELECTOR_P_MASK |
SVM_SELECTOR_S_MASK | SVM_SELECTOR_CODE_MASK;
save->cs.limit = 0xffff;
/*
* cs.base should really be 0xffff0000, but vmx can't handle that, so
* be consistent with it.
*
* Replace when we have real mode working for vmx.
*/
save->cs.base = 0xf0000;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
save->gdtr.limit = 0xffff;
save->idtr.limit = 0xffff;
init_sys_seg(&save->ldtr, SEG_TYPE_LDT);
init_sys_seg(&save->tr, SEG_TYPE_BUSY_TSS16);
save->efer = EFER_SVME;
save->dr6 = 0xffff0ff0;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
save->dr7 = 0x400;
save->rflags = 2;
save->rip = 0x0000fff0;
svm->vcpu.arch.regs[VCPU_REGS_RIP] = save->rip;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
/* This is the guest-visible cr0 value.
* svm_set_cr0() sets PG and WP and clears NW and CD on save->cr0.
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
*/
svm->vcpu.arch.cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
kvm_set_cr0(&svm->vcpu, svm->vcpu.arch.cr0);
save->cr4 = X86_CR4_PAE;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
/* rdx = ?? */
if (npt_enabled) {
/* Setup VMCB for Nested Paging */
control->nested_ctl = 1;
control->intercept &= ~((1ULL << INTERCEPT_TASK_SWITCH) |
(1ULL << INTERCEPT_INVLPG));
control->intercept_exceptions &= ~(1 << PF_VECTOR);
control->intercept_cr_read &= ~INTERCEPT_CR3_MASK;
control->intercept_cr_write &= ~INTERCEPT_CR3_MASK;
save->g_pat = 0x0007040600070406ULL;
save->cr3 = 0;
save->cr4 = 0;
}
force_new_asid(&svm->vcpu);
svm->nested.vmcb = 0;
svm->vcpu.arch.hflags = 0;
if (svm_has(SVM_FEATURE_PAUSE_FILTER)) {
control->pause_filter_count = 3000;
control->intercept |= (1ULL << INTERCEPT_PAUSE);
}
enable_gif(svm);
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
}
static int svm_vcpu_reset(struct kvm_vcpu *vcpu)
{
struct vcpu_svm *svm = to_svm(vcpu);
init_vmcb(svm);
if (!kvm_vcpu_is_bsp(vcpu)) {
kvm_rip_write(vcpu, 0);
svm->vmcb->save.cs.base = svm->vcpu.arch.sipi_vector << 12;
svm->vmcb->save.cs.selector = svm->vcpu.arch.sipi_vector << 8;
}
vcpu->arch.regs_avail = ~0;
vcpu->arch.regs_dirty = ~0;
return 0;
}
static struct kvm_vcpu *svm_create_vcpu(struct kvm *kvm, unsigned int id)
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
{
struct vcpu_svm *svm;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
struct page *page;
struct page *msrpm_pages;
struct page *hsave_page;
struct page *nested_msrpm_pages;
int err;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
svm = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
if (!svm) {
err = -ENOMEM;
goto out;
}
err = kvm_vcpu_init(&svm->vcpu, kvm, id);
if (err)
goto free_svm;
err = -ENOMEM;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
page = alloc_page(GFP_KERNEL);
if (!page)
goto uninit;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
if (!msrpm_pages)
goto free_page1;
nested_msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
if (!nested_msrpm_pages)
goto free_page2;
hsave_page = alloc_page(GFP_KERNEL);
if (!hsave_page)
goto free_page3;
svm->nested.hsave = page_address(hsave_page);
svm->msrpm = page_address(msrpm_pages);
svm_vcpu_init_msrpm(svm->msrpm);
svm->nested.msrpm = page_address(nested_msrpm_pages);
svm->vmcb = page_address(page);
clear_page(svm->vmcb);
svm->vmcb_pa = page_to_pfn(page) << PAGE_SHIFT;
svm->asid_generation = 0;
init_vmcb(svm);
fx_init(&svm->vcpu);
svm->vcpu.arch.apic_base = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
if (kvm_vcpu_is_bsp(&svm->vcpu))
svm->vcpu.arch.apic_base |= MSR_IA32_APICBASE_BSP;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
return &svm->vcpu;
free_page3:
__free_pages(nested_msrpm_pages, MSRPM_ALLOC_ORDER);
free_page2:
__free_pages(msrpm_pages, MSRPM_ALLOC_ORDER);
free_page1:
__free_page(page);
uninit:
kvm_vcpu_uninit(&svm->vcpu);
free_svm:
kmem_cache_free(kvm_vcpu_cache, svm);
out:
return ERR_PTR(err);
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
}
static void svm_free_vcpu(struct kvm_vcpu *vcpu)
{
struct vcpu_svm *svm = to_svm(vcpu);
__free_page(pfn_to_page(svm->vmcb_pa >> PAGE_SHIFT));
__free_pages(virt_to_page(svm->msrpm), MSRPM_ALLOC_ORDER);
__free_page(virt_to_page(svm->nested.hsave));
__free_pages(virt_to_page(svm->nested.msrpm), MSRPM_ALLOC_ORDER);
kvm_vcpu_uninit(vcpu);
kmem_cache_free(kvm_vcpu_cache, svm);
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
}
static void svm_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
{
struct vcpu_svm *svm = to_svm(vcpu);
int i;
if (unlikely(cpu != vcpu->cpu)) {
u64 delta;
if (check_tsc_unstable()) {
/*
* Make sure that the guest sees a monotonically
* increasing TSC.
*/
delta = vcpu->arch.host_tsc - native_read_tsc();
svm->vmcb->control.tsc_offset += delta;
if (is_nested(svm))
svm->nested.hsave->control.tsc_offset += delta;
}
vcpu->cpu = cpu;
kvm_migrate_timers(vcpu);
svm->asid_generation = 0;
}
for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
rdmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
}
static void svm_vcpu_put(struct kvm_vcpu *vcpu)
{
struct vcpu_svm *svm = to_svm(vcpu);
int i;
++vcpu->stat.host_state_reload;
for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
wrmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
vcpu->arch.host_tsc = native_read_tsc();
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
}
static unsigned long svm_get_rflags(struct kvm_vcpu *vcpu)
{
return to_svm(vcpu)->vmcb->save.rflags;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
}
static void svm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
{
to_svm(vcpu)->vmcb->save.rflags = rflags;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
}
static void svm_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
{
switch (reg) {
case VCPU_EXREG_PDPTR:
BUG_ON(!npt_enabled);
load_pdptrs(vcpu, vcpu->arch.cr3);
break;
default:
BUG();
}
}
static void svm_set_vintr(struct vcpu_svm *svm)
{
svm->vmcb->control.intercept |= 1ULL << INTERCEPT_VINTR;
}
static void svm_clear_vintr(struct vcpu_svm *svm)
{
svm->vmcb->control.intercept &= ~(1ULL << INTERCEPT_VINTR);
}
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
static struct vmcb_seg *svm_seg(struct kvm_vcpu *vcpu, int seg)
{
struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
switch (seg) {
case VCPU_SREG_CS: return &save->cs;
case VCPU_SREG_DS: return &save->ds;
case VCPU_SREG_ES: return &save->es;
case VCPU_SREG_FS: return &save->fs;
case VCPU_SREG_GS: return &save->gs;
case VCPU_SREG_SS: return &save->ss;
case VCPU_SREG_TR: return &save->tr;
case VCPU_SREG_LDTR: return &save->ldtr;
}
BUG();
return NULL;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
}
static u64 svm_get_segment_base(struct kvm_vcpu *vcpu, int seg)
{
struct vmcb_seg *s = svm_seg(vcpu, seg);
return s->base;
}
static void svm_get_segment(struct kvm_vcpu *vcpu,
struct kvm_segment *var, int seg)
{
struct vmcb_seg *s = svm_seg(vcpu, seg);
var->base = s->base;
var->limit = s->limit;
var->selector = s->selector;
var->type = s->attrib & SVM_SELECTOR_TYPE_MASK;
var->s = (s->attrib >> SVM_SELECTOR_S_SHIFT) & 1;
var->dpl = (s->attrib >> SVM_SELECTOR_DPL_SHIFT) & 3;
var->present = (s->attrib >> SVM_SELECTOR_P_SHIFT) & 1;
var->avl = (s->attrib >> SVM_SELECTOR_AVL_SHIFT) & 1;
var->l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1;
var->db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1;
var->g = (s->attrib >> SVM_SELECTOR_G_SHIFT) & 1;
/* AMD's VMCB does not have an explicit unusable field, so emulate it
* for cross vendor migration purposes by "not present"
*/
var->unusable = !var->present || (var->type == 0);
switch (seg) {
case VCPU_SREG_CS:
/*
* SVM always stores 0 for the 'G' bit in the CS selector in
* the VMCB on a VMEXIT. This hurts cross-vendor migration:
* Intel's VMENTRY has a check on the 'G' bit.
*/
var->g = s->limit > 0xfffff;
break;
case VCPU_SREG_TR:
/*
* Work around a bug where the busy flag in the tr selector
* isn't exposed
*/
var->type |= 0x2;
break;
case VCPU_SREG_DS:
case VCPU_SREG_ES:
case VCPU_SREG_FS:
case VCPU_SREG_GS:
/*
* The accessed bit must always be set in the segment
* descriptor cache, although it can be cleared in the
* descriptor, the cached bit always remains at 1. Since
* Intel has a check on this, set it here to support
* cross-vendor migration.
*/
if (!var->unusable)
var->type |= 0x1;
break;
case VCPU_SREG_SS:
/* On AMD CPUs sometimes the DB bit in the segment
* descriptor is left as 1, although the whole segment has
* been made unusable. Clear it here to pass an Intel VMX
* entry check when cross vendor migrating.
*/
if (var->unusable)
var->db = 0;
break;
}
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
}
static int svm_get_cpl(struct kvm_vcpu *vcpu)
{
struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
return save->cpl;
}
static void svm_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
{
struct vcpu_svm *svm = to_svm(vcpu);
dt->size = svm->vmcb->save.idtr.limit;
dt->address = svm->vmcb->save.idtr.base;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
}
static void svm_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
{
struct vcpu_svm *svm = to_svm(vcpu);
svm->vmcb->save.idtr.limit = dt->size;
svm->vmcb->save.idtr.base = dt->address ;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
}
static void svm_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
{
struct vcpu_svm *svm = to_svm(vcpu);
dt->size = svm->vmcb->save.gdtr.limit;
dt->address = svm->vmcb->save.gdtr.base;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
}
static void svm_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
{
struct vcpu_svm *svm = to_svm(vcpu);
svm->vmcb->save.gdtr.limit = dt->size;
svm->vmcb->save.gdtr.base = dt->address ;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
}
static void svm_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
{
}
static void svm_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
{
}
static void update_cr0_intercept(struct vcpu_svm *svm)
{
ulong gcr0 = svm->vcpu.arch.cr0;
u64 *hcr0 = &svm->vmcb->save.cr0;
if (!svm->vcpu.fpu_active)
*hcr0 |= SVM_CR0_SELECTIVE_MASK;
else
*hcr0 = (*hcr0 & ~SVM_CR0_SELECTIVE_MASK)
| (gcr0 & SVM_CR0_SELECTIVE_MASK);
if (gcr0 == *hcr0 && svm->vcpu.fpu_active) {
svm->vmcb->control.intercept_cr_read &= ~INTERCEPT_CR0_MASK;
svm->vmcb->control.intercept_cr_write &= ~INTERCEPT_CR0_MASK;
} else {
svm->vmcb->control.intercept_cr_read |= INTERCEPT_CR0_MASK;
svm->vmcb->control.intercept_cr_write |= INTERCEPT_CR0_MASK;
}
}
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
static void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
{
struct vcpu_svm *svm = to_svm(vcpu);
#ifdef CONFIG_X86_64
if (vcpu->arch.efer & EFER_LME) {
if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
vcpu->arch.efer |= EFER_LMA;
svm->vmcb->save.efer |= EFER_LMA | EFER_LME;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
}
if (is_paging(vcpu) && !(cr0 & X86_CR0_PG)) {
vcpu->arch.efer &= ~EFER_LMA;
svm->vmcb->save.efer &= ~(EFER_LMA | EFER_LME);
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
}
}
#endif
vcpu->arch.cr0 = cr0;
if (!npt_enabled)
cr0 |= X86_CR0_PG | X86_CR0_WP;
if (!vcpu->fpu_active)
cr0 |= X86_CR0_TS;
/*
* re-enable caching here because the QEMU bios
* does not do it - this results in some delay at
* reboot
*/
cr0 &= ~(X86_CR0_CD | X86_CR0_NW);
svm->vmcb->save.cr0 = cr0;
update_cr0_intercept(svm);
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
}
static void svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
{
unsigned long host_cr4_mce = read_cr4() & X86_CR4_MCE;
unsigned long old_cr4 = to_svm(vcpu)->vmcb->save.cr4;
if (npt_enabled && ((old_cr4 ^ cr4) & X86_CR4_PGE))
force_new_asid(vcpu);
vcpu->arch.cr4 = cr4;
if (!npt_enabled)
cr4 |= X86_CR4_PAE;
cr4 |= host_cr4_mce;
to_svm(vcpu)->vmcb->save.cr4 = cr4;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
}
static void svm_set_segment(struct kvm_vcpu *vcpu,
struct kvm_segment *var, int seg)
{
struct vcpu_svm *svm = to_svm(vcpu);
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
struct vmcb_seg *s = svm_seg(vcpu, seg);
s->base = var->base;
s->limit = var->limit;
s->selector = var->selector;
if (var->unusable)
s->attrib = 0;
else {
s->attrib = (var->type & SVM_SELECTOR_TYPE_MASK);
s->attrib |= (var->s & 1) << SVM_SELECTOR_S_SHIFT;
s->attrib |= (var->dpl & 3) << SVM_SELECTOR_DPL_SHIFT;
s->attrib |= (var->present & 1) << SVM_SELECTOR_P_SHIFT;
s->attrib |= (var->avl & 1) << SVM_SELECTOR_AVL_SHIFT;
s->attrib |= (var->l & 1) << SVM_SELECTOR_L_SHIFT;
s->attrib |= (var->db & 1) << SVM_SELECTOR_DB_SHIFT;
s->attrib |= (var->g & 1) << SVM_SELECTOR_G_SHIFT;
}
if (seg == VCPU_SREG_CS)
svm->vmcb->save.cpl
= (svm->vmcb->save.cs.attrib
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
>> SVM_SELECTOR_DPL_SHIFT) & 3;
}
static void update_db_intercept(struct kvm_vcpu *vcpu)
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
{
struct vcpu_svm *svm = to_svm(vcpu);
svm->vmcb->control.intercept_exceptions &=
~((1 << DB_VECTOR) | (1 << BP_VECTOR));
if (svm->nmi_singlestep)
svm->vmcb->control.intercept_exceptions |= (1 << DB_VECTOR);
if (vcpu->guest_debug & KVM_GUESTDBG_ENABLE) {
if (vcpu->guest_debug &
(KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
svm->vmcb->control.intercept_exceptions |=
1 << DB_VECTOR;
if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
svm->vmcb->control.intercept_exceptions |=
1 << BP_VECTOR;
} else
vcpu->guest_debug = 0;
}
static void svm_guest_debug(struct kvm_vcpu *vcpu, struct kvm_guest_debug *dbg)
{
struct vcpu_svm *svm = to_svm(vcpu);
if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
svm->vmcb->save.dr7 = dbg->arch.debugreg[7];
else
svm->vmcb->save.dr7 = vcpu->arch.dr7;
update_db_intercept(vcpu);
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
}
static void load_host_msrs(struct kvm_vcpu *vcpu)
{
#ifdef CONFIG_X86_64
wrmsrl(MSR_GS_BASE, to_svm(vcpu)->host_gs_base);
#endif
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
}
static void save_host_msrs(struct kvm_vcpu *vcpu)
{
#ifdef CONFIG_X86_64
rdmsrl(MSR_GS_BASE, to_svm(vcpu)->host_gs_base);
#endif
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
}
static void new_asid(struct vcpu_svm *svm, struct svm_cpu_data *sd)
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
{
if (sd->next_asid > sd->max_asid) {
++sd->asid_generation;
sd->next_asid = 1;
svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ALL_ASID;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
}
svm->asid_generation = sd->asid_generation;
svm->vmcb->control.asid = sd->next_asid++;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
}
static int svm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *dest)
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
{
struct vcpu_svm *svm = to_svm(vcpu);
switch (dr) {
case 0 ... 3:
*dest = vcpu->arch.db[dr];
break;
case 4:
if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
return EMULATE_FAIL; /* will re-inject UD */
/* fall through */
case 6:
if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
*dest = vcpu->arch.dr6;
else
*dest = svm->vmcb->save.dr6;
break;
case 5:
if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
return EMULATE_FAIL; /* will re-inject UD */
/* fall through */
case 7:
if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
*dest = vcpu->arch.dr7;
else
*dest = svm->vmcb->save.dr7;
break;
}
return EMULATE_DONE;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
}
static int svm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long value)
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
{
struct vcpu_svm *svm = to_svm(vcpu);
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
switch (dr) {
case 0 ... 3:
vcpu->arch.db[dr] = value;
if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
vcpu->arch.eff_db[dr] = value;
break;
case 4:
if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
return EMULATE_FAIL; /* will re-inject UD */
/* fall through */
case 6:
vcpu->arch.dr6 = (value & DR6_VOLATILE) | DR6_FIXED_1;
break;
case 5:
if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
return EMULATE_FAIL; /* will re-inject UD */
/* fall through */
case 7:
vcpu->arch.dr7 = (value & DR7_VOLATILE) | DR7_FIXED_1;
if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
svm->vmcb->save.dr7 = vcpu->arch.dr7;
vcpu->arch.switch_db_regs = (value & DR7_BP_EN_MASK);
}
break;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
}
return EMULATE_DONE;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
}
static int pf_interception(struct vcpu_svm *svm)
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
{
u64 fault_address;
u32 error_code;
fault_address = svm->vmcb->control.exit_info_2;
error_code = svm->vmcb->control.exit_info_1;
trace_kvm_page_fault(fault_address, error_code);
if (!npt_enabled && kvm_event_needs_reinjection(&svm->vcpu))
kvm_mmu_unprotect_page_virt(&svm->vcpu, fault_address);
return kvm_mmu_page_fault(&svm->vcpu, fault_address, error_code);
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
}
static int db_interception(struct vcpu_svm *svm)
{
struct kvm_run *kvm_run = svm->vcpu.run;
if (!(svm->vcpu.guest_debug &
(KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) &&
!svm->nmi_singlestep) {
kvm_queue_exception(&svm->vcpu, DB_VECTOR);
return 1;
}
if (svm->nmi_singlestep) {
svm->nmi_singlestep = false;
if (!(svm->vcpu.guest_debug & KVM_GUESTDBG_SINGLESTEP))
svm->vmcb->save.rflags &=
~(X86_EFLAGS_TF | X86_EFLAGS_RF);
update_db_intercept(&svm->vcpu);
}
if (svm->vcpu.guest_debug &
(KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)){
kvm_run->exit_reason = KVM_EXIT_DEBUG;
kvm_run->debug.arch.pc =
svm->vmcb->save.cs.base + svm->vmcb->save.rip;
kvm_run->debug.arch.exception = DB_VECTOR;
return 0;
}
return 1;
}
static int bp_interception(struct vcpu_svm *svm)
{
struct kvm_run *kvm_run = svm->vcpu.run;
kvm_run->exit_reason = KVM_EXIT_DEBUG;
kvm_run->debug.arch.pc = svm->vmcb->save.cs.base + svm->vmcb->save.rip;
kvm_run->debug.arch.exception = BP_VECTOR;
return 0;
}
static int ud_interception(struct vcpu_svm *svm)
{
int er;
er = emulate_instruction(&svm->vcpu, 0, 0, EMULTYPE_TRAP_UD);
if (er != EMULATE_DONE)
kvm_queue_exception(&svm->vcpu, UD_VECTOR);
return 1;
}
static void svm_fpu_activate(struct kvm_vcpu *vcpu)
{
struct vcpu_svm *svm = to_svm(vcpu);
svm->vmcb->control.intercept_exceptions &= ~(1 << NM_VECTOR);
svm->vcpu.fpu_active = 1;
update_cr0_intercept(svm);
}
static int nm_interception(struct vcpu_svm *svm)
{
svm_fpu_activate(&svm->vcpu);
return 1;
}
static int mc_interception(struct vcpu_svm *svm)
{
/*
* On an #MC intercept the MCE handler is not called automatically in
* the host. So do it by hand here.
*/
asm volatile (
"int $0x12\n");
/* not sure if we ever come back to this point */
return 1;
}
static int shutdown_interception(struct vcpu_svm *svm)
{
struct kvm_run *kvm_run = svm->vcpu.run;
/*
* VMCB is undefined after a SHUTDOWN intercept
* so reinitialize it.
*/
clear_page(svm->vmcb);
init_vmcb(svm);
kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
return 0;
}
static int io_interception(struct vcpu_svm *svm)
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
{
u32 io_info = svm->vmcb->control.exit_info_1; /* address size bug? */
int size, in, string;
unsigned port;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
++svm->vcpu.stat.io_exits;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
svm->next_rip = svm->vmcb->control.exit_info_2;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
string = (io_info & SVM_IOIO_STR_MASK) != 0;
if (string) {
if (emulate_instruction(&svm->vcpu,
0, 0, 0) == EMULATE_DO_MMIO)
return 0;
return 1;
}
in = (io_info & SVM_IOIO_TYPE_MASK) != 0;
port = io_info >> 16;
size = (io_info & SVM_IOIO_SIZE_MASK) >> SVM_IOIO_SIZE_SHIFT;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
skip_emulated_instruction(&svm->vcpu);
return kvm_emulate_pio(&svm->vcpu, in, size, port);
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
}
static int nmi_interception(struct vcpu_svm *svm)
{
return 1;
}
static int intr_interception(struct vcpu_svm *svm)
{
++svm->vcpu.stat.irq_exits;
return 1;
}
static int nop_on_interception(struct vcpu_svm *svm)
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
{
return 1;
}
static int halt_interception(struct vcpu_svm *svm)
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
{
svm->next_rip = kvm_rip_read(&svm->vcpu) + 1;
skip_emulated_instruction(&svm->vcpu);
return kvm_emulate_halt(&svm->vcpu);
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
}
static int vmmcall_interception(struct vcpu_svm *svm)
{
svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
skip_emulated_instruction(&svm->vcpu);
kvm_emulate_hypercall(&svm->vcpu);
return 1;
}
static int nested_svm_check_permissions(struct vcpu_svm *svm)
{
if (!(svm->vcpu.arch.efer & EFER_SVME)
|| !is_paging(&svm->vcpu)) {
kvm_queue_exception(&svm->vcpu, UD_VECTOR);
return 1;
}
if (svm->vmcb->save.cpl) {
kvm_inject_gp(&svm->vcpu, 0);
return 1;
}
return 0;
}
static int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
bool has_error_code, u32 error_code)
{
if (!is_nested(svm))
return 0;
svm->vmcb->control.exit_code = SVM_EXIT_EXCP_BASE + nr;
svm->vmcb->control.exit_code_hi = 0;
svm->vmcb->control.exit_info_1 = error_code;
svm->vmcb->control.exit_info_2 = svm->vcpu.arch.cr2;
return nested_svm_exit_handled(svm);
}
static inline int nested_svm_intr(struct vcpu_svm *svm)
{
if (!is_nested(svm))
return 0;
if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK))
return 0;
if (!(svm->vcpu.arch.hflags & HF_HIF_MASK))
return 0;
svm->vmcb->control.exit_code = SVM_EXIT_INTR;
if (svm->nested.intercept & 1ULL) {
/*
* The #vmexit can't be emulated here directly because this
* code path runs with irqs and preemtion disabled. A
* #vmexit emulation might sleep. Only signal request for
* the #vmexit here.
*/
svm->nested.exit_required = true;
trace_kvm_nested_intr_vmexit(svm->vmcb->save.rip);
return 1;
}
return 0;
}
static void *nested_svm_map(struct vcpu_svm *svm, u64 gpa, struct page **_page)
{
struct page *page;
page = gfn_to_page(svm->vcpu.kvm, gpa >> PAGE_SHIFT);
if (is_error_page(page))
goto error;
*_page = page;
return kmap(page);
error:
kvm_release_page_clean(page);
kvm_inject_gp(&svm->vcpu, 0);
return NULL;
}
static void nested_svm_unmap(struct page *page)
{
kunmap(page);
kvm_release_page_dirty(page);
}
static bool nested_svm_exit_handled_msr(struct vcpu_svm *svm)
{
u32 param = svm->vmcb->control.exit_info_1 & 1;
u32 msr = svm->vcpu.arch.regs[VCPU_REGS_RCX];
struct page *page;
bool ret = false;
u32 t0, t1;
u8 *msrpm;
if (!(svm->nested.intercept & (1ULL << INTERCEPT_MSR_PROT)))
return false;
msrpm = nested_svm_map(svm, svm->nested.vmcb_msrpm, &page);
if (!msrpm)
goto out;
switch (msr) {
case 0 ... 0x1fff:
t0 = (msr * 2) % 8;
t1 = msr / 8;
break;
case 0xc0000000 ... 0xc0001fff:
t0 = (8192 + msr - 0xc0000000) * 2;
t1 = (t0 / 8);
t0 %= 8;
break;
case 0xc0010000 ... 0xc0011fff:
t0 = (16384 + msr - 0xc0010000) * 2;
t1 = (t0 / 8);
t0 %= 8;
break;
default:
ret = true;
goto out;
}
ret = msrpm[t1] & ((1 << param) << t0);
out:
nested_svm_unmap(page);
return ret;
}
static int nested_svm_exit_special(struct vcpu_svm *svm)
{
u32 exit_code = svm->vmcb->control.exit_code;
switch (exit_code) {
case SVM_EXIT_INTR:
case SVM_EXIT_NMI:
return NESTED_EXIT_HOST;
/* For now we are always handling NPFs when using them */
case SVM_EXIT_NPF:
if (npt_enabled)
return NESTED_EXIT_HOST;
break;
/* When we're shadowing, trap PFs */
case SVM_EXIT_EXCP_BASE + PF_VECTOR:
if (!npt_enabled)
return NESTED_EXIT_HOST;
break;
default:
break;
}
return NESTED_EXIT_CONTINUE;
}
/*
* If this function returns true, this #vmexit was already handled
*/
static int nested_svm_exit_handled(struct vcpu_svm *svm)
{
u32 exit_code = svm->vmcb->control.exit_code;
int vmexit = NESTED_EXIT_HOST;
switch (exit_code) {
case SVM_EXIT_MSR:
vmexit = nested_svm_exit_handled_msr(svm);
break;
case SVM_EXIT_READ_CR0 ... SVM_EXIT_READ_CR8: {
u32 cr_bits = 1 << (exit_code - SVM_EXIT_READ_CR0);
if (svm->nested.intercept_cr_read & cr_bits)
vmexit = NESTED_EXIT_DONE;
break;
}
case SVM_EXIT_WRITE_CR0 ... SVM_EXIT_WRITE_CR8: {
u32 cr_bits = 1 << (exit_code - SVM_EXIT_WRITE_CR0);
if (svm->nested.intercept_cr_write & cr_bits)
vmexit = NESTED_EXIT_DONE;
break;
}
case SVM_EXIT_READ_DR0 ... SVM_EXIT_READ_DR7: {
u32 dr_bits = 1 << (exit_code - SVM_EXIT_READ_DR0);
if (svm->nested.intercept_dr_read & dr_bits)
vmexit = NESTED_EXIT_DONE;
break;
}
case SVM_EXIT_WRITE_DR0 ... SVM_EXIT_WRITE_DR7: {
u32 dr_bits = 1 << (exit_code - SVM_EXIT_WRITE_DR0);
if (svm->nested.intercept_dr_write & dr_bits)
vmexit = NESTED_EXIT_DONE;
break;
}
case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
u32 excp_bits = 1 << (exit_code - SVM_EXIT_EXCP_BASE);
if (svm->nested.intercept_exceptions & excp_bits)
vmexit = NESTED_EXIT_DONE;
break;
}
default: {
u64 exit_bits = 1ULL << (exit_code - SVM_EXIT_INTR);
if (svm->nested.intercept & exit_bits)
vmexit = NESTED_EXIT_DONE;
}
}
if (vmexit == NESTED_EXIT_DONE) {
nested_svm_vmexit(svm);
}
return vmexit;
}
static inline void copy_vmcb_control_area(struct vmcb *dst_vmcb, struct vmcb *from_vmcb)
{
struct vmcb_control_area *dst = &dst_vmcb->control;
struct vmcb_control_area *from = &from_vmcb->control;
dst->intercept_cr_read = from->intercept_cr_read;
dst->intercept_cr_write = from->intercept_cr_write;
dst->intercept_dr_read = from->intercept_dr_read;
dst->intercept_dr_write = from->intercept_dr_write;
dst->intercept_exceptions = from->intercept_exceptions;
dst->intercept = from->intercept;
dst->iopm_base_pa = from->iopm_base_pa;
dst->msrpm_base_pa = from->msrpm_base_pa;
dst->tsc_offset = from->tsc_offset;
dst->asid = from->asid;
dst->tlb_ctl = from->tlb_ctl;
dst->int_ctl = from->int_ctl;
dst->int_vector = from->int_vector;
dst->int_state = from->int_state;
dst->exit_code = from->exit_code;
dst->exit_code_hi = from->exit_code_hi;
dst->exit_info_1 = from->exit_info_1;
dst->exit_info_2 = from->exit_info_2;
dst->exit_int_info = from->exit_int_info;
dst->exit_int_info_err = from->exit_int_info_err;
dst->nested_ctl = from->nested_ctl;
dst->event_inj = from->event_inj;
dst->event_inj_err = from->event_inj_err;
dst->nested_cr3 = from->nested_cr3;
dst->lbr_ctl = from->lbr_ctl;
}
static int nested_svm_vmexit(struct vcpu_svm *svm)
{
struct vmcb *nested_vmcb;
struct vmcb *hsave = svm->nested.hsave;
struct vmcb *vmcb = svm->vmcb;
struct page *page;
trace_kvm_nested_vmexit_inject(vmcb->control.exit_code,
vmcb->control.exit_info_1,
vmcb->control.exit_info_2,
vmcb->control.exit_int_info,
vmcb->control.exit_int_info_err);
nested_vmcb = nested_svm_map(svm, svm->nested.vmcb, &page);
if (!nested_vmcb)
return 1;
/* Give the current vmcb to the guest */
disable_gif(svm);
nested_vmcb->save.es = vmcb->save.es;
nested_vmcb->save.cs = vmcb->save.cs;
nested_vmcb->save.ss = vmcb->save.ss;
nested_vmcb->save.ds = vmcb->save.ds;
nested_vmcb->save.gdtr = vmcb->save.gdtr;
nested_vmcb->save.idtr = vmcb->save.idtr;
if (npt_enabled)
nested_vmcb->save.cr3 = vmcb->save.cr3;
nested_vmcb->save.cr2 = vmcb->save.cr2;
nested_vmcb->save.rflags = vmcb->save.rflags;
nested_vmcb->save.rip = vmcb->save.rip;
nested_vmcb->save.rsp = vmcb->save.rsp;
nested_vmcb->save.rax = vmcb->save.rax;
nested_vmcb->save.dr7 = vmcb->save.dr7;
nested_vmcb->save.dr6 = vmcb->save.dr6;
nested_vmcb->save.cpl = vmcb->save.cpl;
nested_vmcb->control.int_ctl = vmcb->control.int_ctl;
nested_vmcb->control.int_vector = vmcb->control.int_vector;
nested_vmcb->control.int_state = vmcb->control.int_state;
nested_vmcb->control.exit_code = vmcb->control.exit_code;
nested_vmcb->control.exit_code_hi = vmcb->control.exit_code_hi;
nested_vmcb->control.exit_info_1 = vmcb->control.exit_info_1;
nested_vmcb->control.exit_info_2 = vmcb->control.exit_info_2;
nested_vmcb->control.exit_int_info = vmcb->control.exit_int_info;
nested_vmcb->control.exit_int_info_err = vmcb->control.exit_int_info_err;
/*
* If we emulate a VMRUN/#VMEXIT in the same host #vmexit cycle we have
* to make sure that we do not lose injected events. So check event_inj
* here and copy it to exit_int_info if it is valid.
* Exit_int_info and event_inj can't be both valid because the case
* below only happens on a VMRUN instruction intercept which has
* no valid exit_int_info set.
*/
if (vmcb->control.event_inj & SVM_EVTINJ_VALID) {
struct vmcb_control_area *nc = &nested_vmcb->control;
nc->exit_int_info = vmcb->control.event_inj;
nc->exit_int_info_err = vmcb->control.event_inj_err;
}
nested_vmcb->control.tlb_ctl = 0;
nested_vmcb->control.event_inj = 0;
nested_vmcb->control.event_inj_err = 0;
/* We always set V_INTR_MASKING and remember the old value in hflags */
if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK))
nested_vmcb->control.int_ctl &= ~V_INTR_MASKING_MASK;
/* Restore the original control entries */
copy_vmcb_control_area(vmcb, hsave);
kvm_clear_exception_queue(&svm->vcpu);
kvm_clear_interrupt_queue(&svm->vcpu);
/* Restore selected save entries */
svm->vmcb->save.es = hsave->save.es;
svm->vmcb->save.cs = hsave->save.cs;
svm->vmcb->save.ss = hsave->save.ss;
svm->vmcb->save.ds = hsave->save.ds;
svm->vmcb->save.gdtr = hsave->save.gdtr;
svm->vmcb->save.idtr = hsave->save.idtr;
svm->vmcb->save.rflags = hsave->save.rflags;
svm_set_efer(&svm->vcpu, hsave->save.efer);
svm_set_cr0(&svm->vcpu, hsave->save.cr0 | X86_CR0_PE);
svm_set_cr4(&svm->vcpu, hsave->save.cr4);
if (npt_enabled) {
svm->vmcb->save.cr3 = hsave->save.cr3;
svm->vcpu.arch.cr3 = hsave->save.cr3;
} else {
kvm_set_cr3(&svm->vcpu, hsave->save.cr3);
}
kvm_register_write(&svm->vcpu, VCPU_REGS_RAX, hsave->save.rax);
kvm_register_write(&svm->vcpu, VCPU_REGS_RSP, hsave->save.rsp);
kvm_register_write(&svm->vcpu, VCPU_REGS_RIP, hsave->save.rip);
svm->vmcb->save.dr7 = 0;
svm->vmcb->save.cpl = 0;
svm->vmcb->control.exit_int_info = 0;
/* Exit nested SVM mode */
svm->nested.vmcb = 0;
nested_svm_unmap(page);
kvm_mmu_reset_context(&svm->vcpu);
kvm_mmu_load(&svm->vcpu);
return 0;
}
static bool nested_svm_vmrun_msrpm(struct vcpu_svm *svm)
{
u32 *nested_msrpm;
struct page *page;
int i;
nested_msrpm = nested_svm_map(svm, svm->nested.vmcb_msrpm, &page);
if (!nested_msrpm)
return false;
for (i=0; i< PAGE_SIZE * (1 << MSRPM_ALLOC_ORDER) / 4; i++)
svm->nested.msrpm[i] = svm->msrpm[i] | nested_msrpm[i];
svm->vmcb->control.msrpm_base_pa = __pa(svm->nested.msrpm);
nested_svm_unmap(page);
return true;
}
static bool nested_svm_vmrun(struct vcpu_svm *svm)
{
struct vmcb *nested_vmcb;
struct vmcb *hsave = svm->nested.hsave;
struct vmcb *vmcb = svm->vmcb;
struct page *page;
nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
if (!nested_vmcb)
return false;
/* nested_vmcb is our indicator if nested SVM is activated */
svm->nested.vmcb = svm->vmcb->save.rax;
trace_kvm_nested_vmrun(svm->vmcb->save.rip - 3, svm->nested.vmcb,
nested_vmcb->save.rip,
nested_vmcb->control.int_ctl,
nested_vmcb->control.event_inj,
nested_vmcb->control.nested_ctl);
/* Clear internal status */
kvm_clear_exception_queue(&svm->vcpu);
kvm_clear_interrupt_queue(&svm->vcpu);
/* Save the old vmcb, so we don't need to pick what we save, but
can restore everything when a VMEXIT occurs */
hsave->save.es = vmcb->save.es;
hsave->save.cs = vmcb->save.cs;
hsave->save.ss = vmcb->save.ss;
hsave->save.ds = vmcb->save.ds;
hsave->save.gdtr = vmcb->save.gdtr;
hsave->save.idtr = vmcb->save.idtr;
hsave->save.efer = svm->vcpu.arch.efer;
hsave->save.cr0 = kvm_read_cr0(&svm->vcpu);
hsave->save.cr4 = svm->vcpu.arch.cr4;
hsave->save.rflags = vmcb->save.rflags;
hsave->save.rip = svm->next_rip;
hsave->save.rsp = vmcb->save.rsp;
hsave->save.rax = vmcb->save.rax;
if (npt_enabled)
hsave->save.cr3 = vmcb->save.cr3;
else
hsave->save.cr3 = svm->vcpu.arch.cr3;
copy_vmcb_control_area(hsave, vmcb);
if (svm->vmcb->save.rflags & X86_EFLAGS_IF)
svm->vcpu.arch.hflags |= HF_HIF_MASK;
else
svm->vcpu.arch.hflags &= ~HF_HIF_MASK;
/* Load the nested guest state */
svm->vmcb->save.es = nested_vmcb->save.es;
svm->vmcb->save.cs = nested_vmcb->save.cs;
svm->vmcb->save.ss = nested_vmcb->save.ss;
svm->vmcb->save.ds = nested_vmcb->save.ds;
svm->vmcb->save.gdtr = nested_vmcb->save.gdtr;
svm->vmcb->save.idtr = nested_vmcb->save.idtr;
svm->vmcb->save.rflags = nested_vmcb->save.rflags;
svm_set_efer(&svm->vcpu, nested_vmcb->save.efer);
svm_set_cr0(&svm->vcpu, nested_vmcb->save.cr0);
svm_set_cr4(&svm->vcpu, nested_vmcb->save.cr4);
if (npt_enabled) {
svm->vmcb->save.cr3 = nested_vmcb->save.cr3;
svm->vcpu.arch.cr3 = nested_vmcb->save.cr3;
} else {
kvm_set_cr3(&svm->vcpu, nested_vmcb->save.cr3);
kvm_mmu_reset_context(&svm->vcpu);
}
svm->vmcb->save.cr2 = svm->vcpu.arch.cr2 = nested_vmcb->save.cr2;
kvm_register_write(&svm->vcpu, VCPU_REGS_RAX, nested_vmcb->save.rax);
kvm_register_write(&svm->vcpu, VCPU_REGS_RSP, nested_vmcb->save.rsp);
kvm_register_write(&svm->vcpu, VCPU_REGS_RIP, nested_vmcb->save.rip);
/* In case we don't even reach vcpu_run, the fields are not updated */
svm->vmcb->save.rax = nested_vmcb->save.rax;
svm->vmcb->save.rsp = nested_vmcb->save.rsp;
svm->vmcb->save.rip = nested_vmcb->save.rip;
svm->vmcb->save.dr7 = nested_vmcb->save.dr7;
svm->vmcb->save.dr6 = nested_vmcb->save.dr6;
svm->vmcb->save.cpl = nested_vmcb->save.cpl;
/* We don't want a nested guest to be more powerful than the guest,
so all intercepts are ORed */
svm->vmcb->control.intercept_cr_read |=
nested_vmcb->control.intercept_cr_read;
svm->vmcb->control.intercept_cr_write |=
nested_vmcb->control.intercept_cr_write;
svm->vmcb->control.intercept_dr_read |=
nested_vmcb->control.intercept_dr_read;
svm->vmcb->control.intercept_dr_write |=
nested_vmcb->control.intercept_dr_write;
svm->vmcb->control.intercept_exceptions |=
nested_vmcb->control.intercept_exceptions;
svm->vmcb->control.intercept |= nested_vmcb->control.intercept;
svm->nested.vmcb_msrpm = nested_vmcb->control.msrpm_base_pa;
/* cache intercepts */
svm->nested.intercept_cr_read = nested_vmcb->control.intercept_cr_read;
svm->nested.intercept_cr_write = nested_vmcb->control.intercept_cr_write;
svm->nested.intercept_dr_read = nested_vmcb->control.intercept_dr_read;
svm->nested.intercept_dr_write = nested_vmcb->control.intercept_dr_write;
svm->nested.intercept_exceptions = nested_vmcb->control.intercept_exceptions;
svm->nested.intercept = nested_vmcb->control.intercept;
force_new_asid(&svm->vcpu);
svm->vmcb->control.int_ctl = nested_vmcb->control.int_ctl | V_INTR_MASKING_MASK;
if (nested_vmcb->control.int_ctl & V_INTR_MASKING_MASK)
svm->vcpu.arch.hflags |= HF_VINTR_MASK;
else
svm->vcpu.arch.hflags &= ~HF_VINTR_MASK;
svm->vmcb->control.int_vector = nested_vmcb->control.int_vector;
svm->vmcb->control.int_state = nested_vmcb->control.int_state;
svm->vmcb->control.tsc_offset += nested_vmcb->control.tsc_offset;
svm->vmcb->control.event_inj = nested_vmcb->control.event_inj;
svm->vmcb->control.event_inj_err = nested_vmcb->control.event_inj_err;
nested_svm_unmap(page);
enable_gif(svm);
return true;
}
static void nested_svm_vmloadsave(struct vmcb *from_vmcb, struct vmcb *to_vmcb)
{
to_vmcb->save.fs = from_vmcb->save.fs;
to_vmcb->save.gs = from_vmcb->save.gs;
to_vmcb->save.tr = from_vmcb->save.tr;
to_vmcb->save.ldtr = from_vmcb->save.ldtr;
to_vmcb->save.kernel_gs_base = from_vmcb->save.kernel_gs_base;
to_vmcb->save.star = from_vmcb->save.star;
to_vmcb->save.lstar = from_vmcb->save.lstar;
to_vmcb->save.cstar = from_vmcb->save.cstar;
to_vmcb->save.sfmask = from_vmcb->save.sfmask;
to_vmcb->save.sysenter_cs = from_vmcb->save.sysenter_cs;
to_vmcb->save.sysenter_esp = from_vmcb->save.sysenter_esp;
to_vmcb->save.sysenter_eip = from_vmcb->save.sysenter_eip;
}
static int vmload_interception(struct vcpu_svm *svm)
{
struct vmcb *nested_vmcb;
struct page *page;
if (nested_svm_check_permissions(svm))
return 1;
svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
skip_emulated_instruction(&svm->vcpu);
nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
if (!nested_vmcb)
return 1;
nested_svm_vmloadsave(nested_vmcb, svm->vmcb);
nested_svm_unmap(page);
return 1;
}
static int vmsave_interception(struct vcpu_svm *svm)
{
struct vmcb *nested_vmcb;
struct page *page;
if (nested_svm_check_permissions(svm))
return 1;
svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
skip_emulated_instruction(&svm->vcpu);
nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
if (!nested_vmcb)
return 1;
nested_svm_vmloadsave(svm->vmcb, nested_vmcb);
nested_svm_unmap(page);
return 1;
}
static int vmrun_interception(struct vcpu_svm *svm)
{
if (nested_svm_check_permissions(svm))
return 1;
svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
skip_emulated_instruction(&svm->vcpu);
if (!nested_svm_vmrun(svm))
return 1;
if (!nested_svm_vmrun_msrpm(svm))
goto failed;
return 1;
failed:
svm->vmcb->control.exit_code = SVM_EXIT_ERR;
svm->vmcb->control.exit_code_hi = 0;
svm->vmcb->control.exit_info_1 = 0;
svm->vmcb->control.exit_info_2 = 0;
nested_svm_vmexit(svm);
return 1;
}
static int stgi_interception(struct vcpu_svm *svm)
{
if (nested_svm_check_permissions(svm))
return 1;
svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
skip_emulated_instruction(&svm->vcpu);
enable_gif(svm);
return 1;
}
static int clgi_interception(struct vcpu_svm *svm)
{
if (nested_svm_check_permissions(svm))
return 1;
svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
skip_emulated_instruction(&svm->vcpu);
disable_gif(svm);
/* After a CLGI no interrupts should come */
svm_clear_vintr(svm);
svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
return 1;
}
static int invlpga_interception(struct vcpu_svm *svm)
{
struct kvm_vcpu *vcpu = &svm->vcpu;
trace_kvm_invlpga(svm->vmcb->save.rip, vcpu->arch.regs[VCPU_REGS_RCX],
vcpu->arch.regs[VCPU_REGS_RAX]);
/* Let's treat INVLPGA the same as INVLPG (can be optimized!) */
kvm_mmu_invlpg(vcpu, vcpu->arch.regs[VCPU_REGS_RAX]);
svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
skip_emulated_instruction(&svm->vcpu);
return 1;
}
static int skinit_interception(struct vcpu_svm *svm)
{
trace_kvm_skinit(svm->vmcb->save.rip, svm->vcpu.arch.regs[VCPU_REGS_RAX]);
kvm_queue_exception(&svm->vcpu, UD_VECTOR);
return 1;
}
static int invalid_op_interception(struct vcpu_svm *svm)
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
{
kvm_queue_exception(&svm->vcpu, UD_VECTOR);
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
return 1;
}
static int task_switch_interception(struct vcpu_svm *svm)
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
{
u16 tss_selector;
int reason;
int int_type = svm->vmcb->control.exit_int_info &
SVM_EXITINTINFO_TYPE_MASK;
int int_vec = svm->vmcb->control.exit_int_info & SVM_EVTINJ_VEC_MASK;
uint32_t type =
svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_TYPE_MASK;
uint32_t idt_v =
svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_VALID;
tss_selector = (u16)svm->vmcb->control.exit_info_1;
if (svm->vmcb->control.exit_info_2 &
(1ULL << SVM_EXITINFOSHIFT_TS_REASON_IRET))
reason = TASK_SWITCH_IRET;
else if (svm->vmcb->control.exit_info_2 &
(1ULL << SVM_EXITINFOSHIFT_TS_REASON_JMP))
reason = TASK_SWITCH_JMP;
else if (idt_v)
reason = TASK_SWITCH_GATE;
else
reason = TASK_SWITCH_CALL;
if (reason == TASK_SWITCH_GATE) {
switch (type) {
case SVM_EXITINTINFO_TYPE_NMI:
svm->vcpu.arch.nmi_injected = false;
break;
case SVM_EXITINTINFO_TYPE_EXEPT:
kvm_clear_exception_queue(&svm->vcpu);
break;
case SVM_EXITINTINFO_TYPE_INTR:
kvm_clear_interrupt_queue(&svm->vcpu);
break;
default:
break;
}
}
if (reason != TASK_SWITCH_GATE ||
int_type == SVM_EXITINTINFO_TYPE_SOFT ||
(int_type == SVM_EXITINTINFO_TYPE_EXEPT &&
(int_vec == OF_VECTOR || int_vec == BP_VECTOR)))
skip_emulated_instruction(&svm->vcpu);
return kvm_task_switch(&svm->vcpu, tss_selector, reason);
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
}
static int cpuid_interception(struct vcpu_svm *svm)
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
{
svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
kvm_emulate_cpuid(&svm->vcpu);
return 1;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
}
static int iret_interception(struct vcpu_svm *svm)
{
++svm->vcpu.stat.nmi_window_exits;
svm->vmcb->control.intercept &= ~(1UL << INTERCEPT_IRET);
svm->vcpu.arch.hflags |= HF_IRET_MASK;
return 1;
}
static int invlpg_interception(struct vcpu_svm *svm)
{
if (emulate_instruction(&svm->vcpu, 0, 0, 0) != EMULATE_DONE)
pr_unimpl(&svm->vcpu, "%s: failed\n", __func__);
return 1;
}
static int emulate_on_interception(struct vcpu_svm *svm)
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
{
if (emulate_instruction(&svm->vcpu, 0, 0, 0) != EMULATE_DONE)
pr_unimpl(&svm->vcpu, "%s: failed\n", __func__);
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
return 1;
}
static int cr8_write_interception(struct vcpu_svm *svm)
{
struct kvm_run *kvm_run = svm->vcpu.run;
u8 cr8_prev = kvm_get_cr8(&svm->vcpu);
/* instruction emulation calls kvm_set_cr8() */
emulate_instruction(&svm->vcpu, 0, 0, 0);
if (irqchip_in_kernel(svm->vcpu.kvm)) {
svm->vmcb->control.intercept_cr_write &= ~INTERCEPT_CR8_MASK;
return 1;
}
if (cr8_prev <= kvm_get_cr8(&svm->vcpu))
return 1;
kvm_run->exit_reason = KVM_EXIT_SET_TPR;
return 0;
}
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
static int svm_get_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 *data)
{
struct vcpu_svm *svm = to_svm(vcpu);
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
switch (ecx) {
case MSR_IA32_TSC: {
u64 tsc_offset;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
if (is_nested(svm))
tsc_offset = svm->nested.hsave->control.tsc_offset;
else
tsc_offset = svm->vmcb->control.tsc_offset;
*data = tsc_offset + native_read_tsc();
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
break;
}
case MSR_K6_STAR:
*data = svm->vmcb->save.star;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
break;
#ifdef CONFIG_X86_64
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
case MSR_LSTAR:
*data = svm->vmcb->save.lstar;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
break;
case MSR_CSTAR:
*data = svm->vmcb->save.cstar;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
break;
case MSR_KERNEL_GS_BASE:
*data = svm->vmcb->save.kernel_gs_base;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
break;
case MSR_SYSCALL_MASK:
*data = svm->vmcb->save.sfmask;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
break;
#endif
case MSR_IA32_SYSENTER_CS:
*data = svm->vmcb->save.sysenter_cs;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
break;
case MSR_IA32_SYSENTER_EIP:
*data = svm->sysenter_eip;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
break;
case MSR_IA32_SYSENTER_ESP:
*data = svm->sysenter_esp;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
break;
/* Nobody will change the following 5 values in the VMCB so
we can safely return them on rdmsr. They will always be 0
until LBRV is implemented. */
case MSR_IA32_DEBUGCTLMSR:
*data = svm->vmcb->save.dbgctl;
break;
case MSR_IA32_LASTBRANCHFROMIP:
*data = svm->vmcb->save.br_from;
break;
case MSR_IA32_LASTBRANCHTOIP:
*data = svm->vmcb->save.br_to;
break;
case MSR_IA32_LASTINTFROMIP:
*data = svm->vmcb->save.last_excp_from;
break;
case MSR_IA32_LASTINTTOIP:
*data = svm->vmcb->save.last_excp_to;
break;
case MSR_VM_HSAVE_PA:
*data = svm->nested.hsave_msr;
break;
case MSR_VM_CR:
*data = 0;
break;
case MSR_IA32_UCODE_REV:
*data = 0x01000065;
break;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
default:
return kvm_get_msr_common(vcpu, ecx, data);
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
}
return 0;
}
static int rdmsr_interception(struct vcpu_svm *svm)
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
{
u32 ecx = svm->vcpu.arch.regs[VCPU_REGS_RCX];
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
u64 data;
if (svm_get_msr(&svm->vcpu, ecx, &data)) {
trace_kvm_msr_read_ex(ecx);
kvm_inject_gp(&svm->vcpu, 0);
} else {
trace_kvm_msr_read(ecx, data);
svm->vcpu.arch.regs[VCPU_REGS_RAX] = data & 0xffffffff;
svm->vcpu.arch.regs[VCPU_REGS_RDX] = data >> 32;
svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
skip_emulated_instruction(&svm->vcpu);
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
}
return 1;
}
static int svm_set_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 data)
{
struct vcpu_svm *svm = to_svm(vcpu);
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
switch (ecx) {
case MSR_IA32_TSC: {
u64 tsc_offset = data - native_read_tsc();
u64 g_tsc_offset = 0;
if (is_nested(svm)) {
g_tsc_offset = svm->vmcb->control.tsc_offset -
svm->nested.hsave->control.tsc_offset;
svm->nested.hsave->control.tsc_offset = tsc_offset;
}
svm->vmcb->control.tsc_offset = tsc_offset + g_tsc_offset;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
break;
}
case MSR_K6_STAR:
svm->vmcb->save.star = data;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
break;
#ifdef CONFIG_X86_64
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
case MSR_LSTAR:
svm->vmcb->save.lstar = data;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
break;
case MSR_CSTAR:
svm->vmcb->save.cstar = data;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
break;
case MSR_KERNEL_GS_BASE:
svm->vmcb->save.kernel_gs_base = data;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
break;
case MSR_SYSCALL_MASK:
svm->vmcb->save.sfmask = data;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
break;
#endif
case MSR_IA32_SYSENTER_CS:
svm->vmcb->save.sysenter_cs = data;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
break;
case MSR_IA32_SYSENTER_EIP:
svm->sysenter_eip = data;
svm->vmcb->save.sysenter_eip = data;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
break;
case MSR_IA32_SYSENTER_ESP:
svm->sysenter_esp = data;
svm->vmcb->save.sysenter_esp = data;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
break;
case MSR_IA32_DEBUGCTLMSR:
if (!svm_has(SVM_FEATURE_LBRV)) {
pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTL 0x%llx, nop\n",
__func__, data);
break;
}
if (data & DEBUGCTL_RESERVED_BITS)
return 1;
svm->vmcb->save.dbgctl = data;
if (data & (1ULL<<0))
svm_enable_lbrv(svm);
else
svm_disable_lbrv(svm);
break;
case MSR_VM_HSAVE_PA:
svm->nested.hsave_msr = data;
break;
case MSR_VM_CR:
case MSR_VM_IGNNE:
pr_unimpl(vcpu, "unimplemented wrmsr: 0x%x data 0x%llx\n", ecx, data);
break;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
default:
return kvm_set_msr_common(vcpu, ecx, data);
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
}
return 0;
}
static int wrmsr_interception(struct vcpu_svm *svm)
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
{
u32 ecx = svm->vcpu.arch.regs[VCPU_REGS_RCX];
u64 data = (svm->vcpu.arch.regs[VCPU_REGS_RAX] & -1u)
| ((u64)(svm->vcpu.arch.regs[VCPU_REGS_RDX] & -1u) << 32);
svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
if (svm_set_msr(&svm->vcpu, ecx, data)) {
trace_kvm_msr_write_ex(ecx, data);
kvm_inject_gp(&svm->vcpu, 0);
} else {
trace_kvm_msr_write(ecx, data);
skip_emulated_instruction(&svm->vcpu);
}
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
return 1;
}
static int msr_interception(struct vcpu_svm *svm)
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
{
if (svm->vmcb->control.exit_info_1)
return wrmsr_interception(svm);
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
else
return rdmsr_interception(svm);
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
}
static int interrupt_window_interception(struct vcpu_svm *svm)
{
struct kvm_run *kvm_run = svm->vcpu.run;
svm_clear_vintr(svm);
svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
/*
* If the user space waits to inject interrupts, exit as soon as
* possible
*/
if (!irqchip_in_kernel(svm->vcpu.kvm) &&
kvm_run->request_interrupt_window &&
!kvm_cpu_has_interrupt(&svm->vcpu)) {
++svm->vcpu.stat.irq_window_exits;
kvm_run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
return 0;
}
return 1;
}
static int pause_interception(struct vcpu_svm *svm)
{
kvm_vcpu_on_spin(&(svm->vcpu));
return 1;
}
static int (*svm_exit_handlers[])(struct vcpu_svm *svm) = {
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
[SVM_EXIT_READ_CR0] = emulate_on_interception,
[SVM_EXIT_READ_CR3] = emulate_on_interception,
[SVM_EXIT_READ_CR4] = emulate_on_interception,
[SVM_EXIT_READ_CR8] = emulate_on_interception,
[SVM_EXIT_CR0_SEL_WRITE] = emulate_on_interception,
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
[SVM_EXIT_WRITE_CR0] = emulate_on_interception,
[SVM_EXIT_WRITE_CR3] = emulate_on_interception,
[SVM_EXIT_WRITE_CR4] = emulate_on_interception,
[SVM_EXIT_WRITE_CR8] = cr8_write_interception,
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
[SVM_EXIT_READ_DR0] = emulate_on_interception,
[SVM_EXIT_READ_DR1] = emulate_on_interception,
[SVM_EXIT_READ_DR2] = emulate_on_interception,
[SVM_EXIT_READ_DR3] = emulate_on_interception,
[SVM_EXIT_READ_DR4] = emulate_on_interception,
[SVM_EXIT_READ_DR5] = emulate_on_interception,
[SVM_EXIT_READ_DR6] = emulate_on_interception,
[SVM_EXIT_READ_DR7] = emulate_on_interception,
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
[SVM_EXIT_WRITE_DR0] = emulate_on_interception,
[SVM_EXIT_WRITE_DR1] = emulate_on_interception,
[SVM_EXIT_WRITE_DR2] = emulate_on_interception,
[SVM_EXIT_WRITE_DR3] = emulate_on_interception,
[SVM_EXIT_WRITE_DR4] = emulate_on_interception,
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
[SVM_EXIT_WRITE_DR5] = emulate_on_interception,
[SVM_EXIT_WRITE_DR6] = emulate_on_interception,
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
[SVM_EXIT_WRITE_DR7] = emulate_on_interception,
[SVM_EXIT_EXCP_BASE + DB_VECTOR] = db_interception,
[SVM_EXIT_EXCP_BASE + BP_VECTOR] = bp_interception,
[SVM_EXIT_EXCP_BASE + UD_VECTOR] = ud_interception,
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
[SVM_EXIT_EXCP_BASE + PF_VECTOR] = pf_interception,
[SVM_EXIT_EXCP_BASE + NM_VECTOR] = nm_interception,
[SVM_EXIT_EXCP_BASE + MC_VECTOR] = mc_interception,
[SVM_EXIT_INTR] = intr_interception,
[SVM_EXIT_NMI] = nmi_interception,
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
[SVM_EXIT_SMI] = nop_on_interception,
[SVM_EXIT_INIT] = nop_on_interception,
[SVM_EXIT_VINTR] = interrupt_window_interception,
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
/* [SVM_EXIT_CR0_SEL_WRITE] = emulate_on_interception, */
[SVM_EXIT_CPUID] = cpuid_interception,
[SVM_EXIT_IRET] = iret_interception,
[SVM_EXIT_INVD] = emulate_on_interception,
[SVM_EXIT_PAUSE] = pause_interception,
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
[SVM_EXIT_HLT] = halt_interception,
[SVM_EXIT_INVLPG] = invlpg_interception,
[SVM_EXIT_INVLPGA] = invlpga_interception,
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
[SVM_EXIT_IOIO] = io_interception,
[SVM_EXIT_MSR] = msr_interception,
[SVM_EXIT_TASK_SWITCH] = task_switch_interception,
[SVM_EXIT_SHUTDOWN] = shutdown_interception,
[SVM_EXIT_VMRUN] = vmrun_interception,
[SVM_EXIT_VMMCALL] = vmmcall_interception,
[SVM_EXIT_VMLOAD] = vmload_interception,
[SVM_EXIT_VMSAVE] = vmsave_interception,
[SVM_EXIT_STGI] = stgi_interception,
[SVM_EXIT_CLGI] = clgi_interception,
[SVM_EXIT_SKINIT] = skinit_interception,
[SVM_EXIT_WBINVD] = emulate_on_interception,
[SVM_EXIT_MONITOR] = invalid_op_interception,
[SVM_EXIT_MWAIT] = invalid_op_interception,
[SVM_EXIT_NPF] = pf_interception,
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
};
static int handle_exit(struct kvm_vcpu *vcpu)
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
{
struct vcpu_svm *svm = to_svm(vcpu);
struct kvm_run *kvm_run = vcpu->run;
u32 exit_code = svm->vmcb->control.exit_code;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
trace_kvm_exit(exit_code, svm->vmcb->save.rip);
if (unlikely(svm->nested.exit_required)) {
nested_svm_vmexit(svm);
svm->nested.exit_required = false;
return 1;
}
if (is_nested(svm)) {
int vmexit;
trace_kvm_nested_vmexit(svm->vmcb->save.rip, exit_code,
svm->vmcb->control.exit_info_1,
svm->vmcb->control.exit_info_2,
svm->vmcb->control.exit_int_info,
svm->vmcb->control.exit_int_info_err);
vmexit = nested_svm_exit_special(svm);
if (vmexit == NESTED_EXIT_CONTINUE)
vmexit = nested_svm_exit_handled(svm);
if (vmexit == NESTED_EXIT_DONE)
return 1;
}
svm_complete_interrupts(svm);
if (!(svm->vmcb->control.intercept_cr_write & INTERCEPT_CR0_MASK))
vcpu->arch.cr0 = svm->vmcb->save.cr0;
if (npt_enabled)
vcpu->arch.cr3 = svm->vmcb->save.cr3;
if (svm->vmcb->control.exit_code == SVM_EXIT_ERR) {
kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
kvm_run->fail_entry.hardware_entry_failure_reason
= svm->vmcb->control.exit_code;
return 0;
}
if (is_external_interrupt(svm->vmcb->control.exit_int_info) &&
exit_code != SVM_EXIT_EXCP_BASE + PF_VECTOR &&
exit_code != SVM_EXIT_NPF && exit_code != SVM_EXIT_TASK_SWITCH)
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
printk(KERN_ERR "%s: unexpected exit_ini_info 0x%x "
"exit_code 0x%x\n",
__func__, svm->vmcb->control.exit_int_info,
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
exit_code);
if (exit_code >= ARRAY_SIZE(svm_exit_handlers)
|| !svm_exit_handlers[exit_code]) {
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
kvm_run->hw.hardware_exit_reason = exit_code;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
return 0;
}
return svm_exit_handlers[exit_code](svm);
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
}
static void reload_tss(struct kvm_vcpu *vcpu)
{
int cpu = raw_smp_processor_id();
struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
sd->tss_desc->type = 9; /* available 32/64-bit TSS */
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
load_TR_desc();
}
static void pre_svm_run(struct vcpu_svm *svm)
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
{
int cpu = raw_smp_processor_id();
struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
svm->vmcb->control.tlb_ctl = TLB_CONTROL_DO_NOTHING;
/* FIXME: handle wraparound of asid_generation */
if (svm->asid_generation != sd->asid_generation)
new_asid(svm, sd);
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
}
static void svm_inject_nmi(struct kvm_vcpu *vcpu)
{
struct vcpu_svm *svm = to_svm(vcpu);
svm->vmcb->control.event_inj = SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_NMI;
vcpu->arch.hflags |= HF_NMI_MASK;
svm->vmcb->control.intercept |= (1UL << INTERCEPT_IRET);
++vcpu->stat.nmi_injections;
}
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
static inline void svm_inject_irq(struct vcpu_svm *svm, int irq)
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
{
struct vmcb_control_area *control;
trace_kvm_inj_virq(irq);
++svm->vcpu.stat.irq_injections;
control = &svm->vmcb->control;
control->int_vector = irq;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
control->int_ctl &= ~V_INTR_PRIO_MASK;
control->int_ctl |= V_IRQ_MASK |
((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT);
}
static void svm_set_irq(struct kvm_vcpu *vcpu)
{
struct vcpu_svm *svm = to_svm(vcpu);
BUG_ON(!(gif_set(svm)));
svm->vmcb->control.event_inj = vcpu->arch.interrupt.nr |
SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR;
}
static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
{
struct vcpu_svm *svm = to_svm(vcpu);
if (irr == -1)
return;
if (tpr >= irr)
svm->vmcb->control.intercept_cr_write |= INTERCEPT_CR8_MASK;
}
static int svm_nmi_allowed(struct kvm_vcpu *vcpu)
{
struct vcpu_svm *svm = to_svm(vcpu);
struct vmcb *vmcb = svm->vmcb;
return !(vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK) &&
!(svm->vcpu.arch.hflags & HF_NMI_MASK);
}
static bool svm_get_nmi_mask(struct kvm_vcpu *vcpu)
{
struct vcpu_svm *svm = to_svm(vcpu);
return !!(svm->vcpu.arch.hflags & HF_NMI_MASK);
}
static void svm_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
{
struct vcpu_svm *svm = to_svm(vcpu);
if (masked) {
svm->vcpu.arch.hflags |= HF_NMI_MASK;
svm->vmcb->control.intercept |= (1UL << INTERCEPT_IRET);
} else {
svm->vcpu.arch.hflags &= ~HF_NMI_MASK;
svm->vmcb->control.intercept &= ~(1UL << INTERCEPT_IRET);
}
}
static int svm_interrupt_allowed(struct kvm_vcpu *vcpu)
{
struct vcpu_svm *svm = to_svm(vcpu);
struct vmcb *vmcb = svm->vmcb;
int ret;
if (!gif_set(svm) ||
(vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK))
return 0;
ret = !!(vmcb->save.rflags & X86_EFLAGS_IF);
if (is_nested(svm))
return ret && !(svm->vcpu.arch.hflags & HF_VINTR_MASK);
return ret;
}
static void enable_irq_window(struct kvm_vcpu *vcpu)
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
{
struct vcpu_svm *svm = to_svm(vcpu);
nested_svm_intr(svm);
/* In case GIF=0 we can't rely on the CPU to tell us when
* GIF becomes 1, because that's a separate STGI/VMRUN intercept.
* The next time we get that intercept, this function will be
* called again though and we'll get the vintr intercept. */
if (gif_set(svm)) {
svm_set_vintr(svm);
svm_inject_irq(svm, 0x0);
}
}
static void enable_nmi_window(struct kvm_vcpu *vcpu)
{
struct vcpu_svm *svm = to_svm(vcpu);
if ((svm->vcpu.arch.hflags & (HF_NMI_MASK | HF_IRET_MASK))
== HF_NMI_MASK)
return; /* IRET will cause a vm exit */
/* Something prevents NMI from been injected. Single step over
possible problem (IRET or exception injection or interrupt
shadow) */
svm->nmi_singlestep = true;
svm->vmcb->save.rflags |= (X86_EFLAGS_TF | X86_EFLAGS_RF);
update_db_intercept(vcpu);
}
static int svm_set_tss_addr(struct kvm *kvm, unsigned int addr)
{
return 0;
}
static void svm_flush_tlb(struct kvm_vcpu *vcpu)
{
force_new_asid(vcpu);
}
static void svm_prepare_guest_switch(struct kvm_vcpu *vcpu)
{
}
static inline void sync_cr8_to_lapic(struct kvm_vcpu *vcpu)
{
struct vcpu_svm *svm = to_svm(vcpu);
if (!(svm->vmcb->control.intercept_cr_write & INTERCEPT_CR8_MASK)) {
int cr8 = svm->vmcb->control.int_ctl & V_TPR_MASK;
kvm_set_cr8(vcpu, cr8);
}
}
static inline void sync_lapic_to_cr8(struct kvm_vcpu *vcpu)
{
struct vcpu_svm *svm = to_svm(vcpu);
u64 cr8;
cr8 = kvm_get_cr8(vcpu);
svm->vmcb->control.int_ctl &= ~V_TPR_MASK;
svm->vmcb->control.int_ctl |= cr8 & V_TPR_MASK;
}
static void svm_complete_interrupts(struct vcpu_svm *svm)
{
u8 vector;
int type;
u32 exitintinfo = svm->vmcb->control.exit_int_info;
if (svm->vcpu.arch.hflags & HF_IRET_MASK)
svm->vcpu.arch.hflags &= ~(HF_NMI_MASK | HF_IRET_MASK);
svm->vcpu.arch.nmi_injected = false;
kvm_clear_exception_queue(&svm->vcpu);
kvm_clear_interrupt_queue(&svm->vcpu);
if (!(exitintinfo & SVM_EXITINTINFO_VALID))
return;
vector = exitintinfo & SVM_EXITINTINFO_VEC_MASK;
type = exitintinfo & SVM_EXITINTINFO_TYPE_MASK;
switch (type) {
case SVM_EXITINTINFO_TYPE_NMI:
svm->vcpu.arch.nmi_injected = true;
break;
case SVM_EXITINTINFO_TYPE_EXEPT:
/* In case of software exception do not reinject an exception
vector, but re-execute and instruction instead */
if (is_nested(svm))
break;
if (kvm_exception_is_soft(vector))
break;
if (exitintinfo & SVM_EXITINTINFO_VALID_ERR) {
u32 err = svm->vmcb->control.exit_int_info_err;
kvm_queue_exception_e(&svm->vcpu, vector, err);
} else
kvm_queue_exception(&svm->vcpu, vector);
break;
case SVM_EXITINTINFO_TYPE_INTR:
kvm_queue_interrupt(&svm->vcpu, vector, false);
break;
default:
break;
}
}
#ifdef CONFIG_X86_64
#define R "r"
#else
#define R "e"
#endif
static void svm_vcpu_run(struct kvm_vcpu *vcpu)
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
{
struct vcpu_svm *svm = to_svm(vcpu);
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
u16 fs_selector;
u16 gs_selector;
u16 ldt_selector;
/*
* A vmexit emulation is required before the vcpu can be executed
* again.
*/
if (unlikely(svm->nested.exit_required))
return;
svm->vmcb->save.rax = vcpu->arch.regs[VCPU_REGS_RAX];
svm->vmcb->save.rsp = vcpu->arch.regs[VCPU_REGS_RSP];
svm->vmcb->save.rip = vcpu->arch.regs[VCPU_REGS_RIP];
pre_svm_run(svm);
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
sync_lapic_to_cr8(vcpu);
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
save_host_msrs(vcpu);
fs_selector = kvm_read_fs();
gs_selector = kvm_read_gs();
ldt_selector = kvm_read_ldt();
svm->vmcb->save.cr2 = vcpu->arch.cr2;
/* required for live migration with NPT */
if (npt_enabled)
svm->vmcb->save.cr3 = vcpu->arch.cr3;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
clgi();
local_irq_enable();
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
asm volatile (
"push %%"R"bp; \n\t"
"mov %c[rbx](%[svm]), %%"R"bx \n\t"
"mov %c[rcx](%[svm]), %%"R"cx \n\t"
"mov %c[rdx](%[svm]), %%"R"dx \n\t"
"mov %c[rsi](%[svm]), %%"R"si \n\t"
"mov %c[rdi](%[svm]), %%"R"di \n\t"
"mov %c[rbp](%[svm]), %%"R"bp \n\t"
#ifdef CONFIG_X86_64
"mov %c[r8](%[svm]), %%r8 \n\t"
"mov %c[r9](%[svm]), %%r9 \n\t"
"mov %c[r10](%[svm]), %%r10 \n\t"
"mov %c[r11](%[svm]), %%r11 \n\t"
"mov %c[r12](%[svm]), %%r12 \n\t"
"mov %c[r13](%[svm]), %%r13 \n\t"
"mov %c[r14](%[svm]), %%r14 \n\t"
"mov %c[r15](%[svm]), %%r15 \n\t"
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
#endif
/* Enter guest mode */
"push %%"R"ax \n\t"
"mov %c[vmcb](%[svm]), %%"R"ax \n\t"
__ex(SVM_VMLOAD) "\n\t"
__ex(SVM_VMRUN) "\n\t"
__ex(SVM_VMSAVE) "\n\t"
"pop %%"R"ax \n\t"
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
/* Save guest registers, load host registers */
"mov %%"R"bx, %c[rbx](%[svm]) \n\t"
"mov %%"R"cx, %c[rcx](%[svm]) \n\t"
"mov %%"R"dx, %c[rdx](%[svm]) \n\t"
"mov %%"R"si, %c[rsi](%[svm]) \n\t"
"mov %%"R"di, %c[rdi](%[svm]) \n\t"
"mov %%"R"bp, %c[rbp](%[svm]) \n\t"
#ifdef CONFIG_X86_64
"mov %%r8, %c[r8](%[svm]) \n\t"
"mov %%r9, %c[r9](%[svm]) \n\t"
"mov %%r10, %c[r10](%[svm]) \n\t"
"mov %%r11, %c[r11](%[svm]) \n\t"
"mov %%r12, %c[r12](%[svm]) \n\t"
"mov %%r13, %c[r13](%[svm]) \n\t"
"mov %%r14, %c[r14](%[svm]) \n\t"
"mov %%r15, %c[r15](%[svm]) \n\t"
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
#endif
"pop %%"R"bp"
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
:
: [svm]"a"(svm),
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
[vmcb]"i"(offsetof(struct vcpu_svm, vmcb_pa)),
[rbx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBX])),
[rcx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RCX])),
[rdx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDX])),
[rsi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RSI])),
[rdi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDI])),
[rbp]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBP]))
#ifdef CONFIG_X86_64
, [r8]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R8])),
[r9]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R9])),
[r10]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R10])),
[r11]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R11])),
[r12]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R12])),
[r13]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R13])),
[r14]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R14])),
[r15]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R15]))
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
#endif
: "cc", "memory"
, R"bx", R"cx", R"dx", R"si", R"di"
#ifdef CONFIG_X86_64
, "r8", "r9", "r10", "r11" , "r12", "r13", "r14", "r15"
#endif
);
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
vcpu->arch.cr2 = svm->vmcb->save.cr2;
vcpu->arch.regs[VCPU_REGS_RAX] = svm->vmcb->save.rax;
vcpu->arch.regs[VCPU_REGS_RSP] = svm->vmcb->save.rsp;
vcpu->arch.regs[VCPU_REGS_RIP] = svm->vmcb->save.rip;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
kvm_load_fs(fs_selector);
kvm_load_gs(gs_selector);
kvm_load_ldt(ldt_selector);
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
load_host_msrs(vcpu);
reload_tss(vcpu);
local_irq_disable();
stgi();
sync_cr8_to_lapic(vcpu);
svm->next_rip = 0;
if (npt_enabled) {
vcpu->arch.regs_avail &= ~(1 << VCPU_EXREG_PDPTR);
vcpu->arch.regs_dirty &= ~(1 << VCPU_EXREG_PDPTR);
}
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
}
#undef R
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
static void svm_set_cr3(struct kvm_vcpu *vcpu, unsigned long root)
{
struct vcpu_svm *svm = to_svm(vcpu);
if (npt_enabled) {
svm->vmcb->control.nested_cr3 = root;
force_new_asid(vcpu);
return;
}
svm->vmcb->save.cr3 = root;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
force_new_asid(vcpu);
}
static int is_disabled(void)
{
u64 vm_cr;
rdmsrl(MSR_VM_CR, vm_cr);
if (vm_cr & (1 << SVM_VM_CR_SVM_DISABLE))
return 1;
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
return 0;
}
static void
svm_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
{
/*
* Patch in the VMMCALL instruction:
*/
hypercall[0] = 0x0f;
hypercall[1] = 0x01;
hypercall[2] = 0xd9;
}
static void svm_check_processor_compat(void *rtn)
{
*(int *)rtn = 0;
}
static bool svm_cpu_has_accelerated_tpr(void)
{
return false;
}
static int get_npt_level(void)
{
#ifdef CONFIG_X86_64
return PT64_ROOT_LEVEL;
#else
return PT32E_ROOT_LEVEL;
#endif
}
static u64 svm_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
{
return 0;
}
static void svm_cpuid_update(struct kvm_vcpu *vcpu)
{
}
static const struct trace_print_flags svm_exit_reasons_str[] = {
{ SVM_EXIT_READ_CR0, "read_cr0" },
{ SVM_EXIT_READ_CR3, "read_cr3" },
{ SVM_EXIT_READ_CR4, "read_cr4" },
{ SVM_EXIT_READ_CR8, "read_cr8" },
{ SVM_EXIT_WRITE_CR0, "write_cr0" },
{ SVM_EXIT_WRITE_CR3, "write_cr3" },
{ SVM_EXIT_WRITE_CR4, "write_cr4" },
{ SVM_EXIT_WRITE_CR8, "write_cr8" },
{ SVM_EXIT_READ_DR0, "read_dr0" },
{ SVM_EXIT_READ_DR1, "read_dr1" },
{ SVM_EXIT_READ_DR2, "read_dr2" },
{ SVM_EXIT_READ_DR3, "read_dr3" },
{ SVM_EXIT_WRITE_DR0, "write_dr0" },
{ SVM_EXIT_WRITE_DR1, "write_dr1" },
{ SVM_EXIT_WRITE_DR2, "write_dr2" },
{ SVM_EXIT_WRITE_DR3, "write_dr3" },
{ SVM_EXIT_WRITE_DR5, "write_dr5" },
{ SVM_EXIT_WRITE_DR7, "write_dr7" },
{ SVM_EXIT_EXCP_BASE + DB_VECTOR, "DB excp" },
{ SVM_EXIT_EXCP_BASE + BP_VECTOR, "BP excp" },
{ SVM_EXIT_EXCP_BASE + UD_VECTOR, "UD excp" },
{ SVM_EXIT_EXCP_BASE + PF_VECTOR, "PF excp" },
{ SVM_EXIT_EXCP_BASE + NM_VECTOR, "NM excp" },
{ SVM_EXIT_EXCP_BASE + MC_VECTOR, "MC excp" },
{ SVM_EXIT_INTR, "interrupt" },
{ SVM_EXIT_NMI, "nmi" },
{ SVM_EXIT_SMI, "smi" },
{ SVM_EXIT_INIT, "init" },
{ SVM_EXIT_VINTR, "vintr" },
{ SVM_EXIT_CPUID, "cpuid" },
{ SVM_EXIT_INVD, "invd" },
{ SVM_EXIT_HLT, "hlt" },
{ SVM_EXIT_INVLPG, "invlpg" },
{ SVM_EXIT_INVLPGA, "invlpga" },
{ SVM_EXIT_IOIO, "io" },
{ SVM_EXIT_MSR, "msr" },
{ SVM_EXIT_TASK_SWITCH, "task_switch" },
{ SVM_EXIT_SHUTDOWN, "shutdown" },
{ SVM_EXIT_VMRUN, "vmrun" },
{ SVM_EXIT_VMMCALL, "hypercall" },
{ SVM_EXIT_VMLOAD, "vmload" },
{ SVM_EXIT_VMSAVE, "vmsave" },
{ SVM_EXIT_STGI, "stgi" },
{ SVM_EXIT_CLGI, "clgi" },
{ SVM_EXIT_SKINIT, "skinit" },
{ SVM_EXIT_WBINVD, "wbinvd" },
{ SVM_EXIT_MONITOR, "monitor" },
{ SVM_EXIT_MWAIT, "mwait" },
{ SVM_EXIT_NPF, "npf" },
{ -1, NULL }
};
static int svm_get_lpage_level(void)
{
return PT_PDPE_LEVEL;
}
static bool svm_rdtscp_supported(void)
{
return false;
}
static void svm_fpu_deactivate(struct kvm_vcpu *vcpu)
{
struct vcpu_svm *svm = to_svm(vcpu);
update_cr0_intercept(svm);
svm->vmcb->control.intercept_exceptions |= 1 << NM_VECTOR;
}
static struct kvm_x86_ops svm_x86_ops = {
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
.cpu_has_kvm_support = has_svm,
.disabled_by_bios = is_disabled,
.hardware_setup = svm_hardware_setup,
.hardware_unsetup = svm_hardware_unsetup,
.check_processor_compatibility = svm_check_processor_compat,
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
.hardware_enable = svm_hardware_enable,
.hardware_disable = svm_hardware_disable,
.cpu_has_accelerated_tpr = svm_cpu_has_accelerated_tpr,
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
.vcpu_create = svm_create_vcpu,
.vcpu_free = svm_free_vcpu,
.vcpu_reset = svm_vcpu_reset,
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
.prepare_guest_switch = svm_prepare_guest_switch,
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
.vcpu_load = svm_vcpu_load,
.vcpu_put = svm_vcpu_put,
.set_guest_debug = svm_guest_debug,
.get_msr = svm_get_msr,
.set_msr = svm_set_msr,
.get_segment_base = svm_get_segment_base,
.get_segment = svm_get_segment,
.set_segment = svm_set_segment,
.get_cpl = svm_get_cpl,
.get_cs_db_l_bits = kvm_get_cs_db_l_bits,
.decache_cr0_guest_bits = svm_decache_cr0_guest_bits,
.decache_cr4_guest_bits = svm_decache_cr4_guest_bits,
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
.set_cr0 = svm_set_cr0,
.set_cr3 = svm_set_cr3,
.set_cr4 = svm_set_cr4,
.set_efer = svm_set_efer,
.get_idt = svm_get_idt,
.set_idt = svm_set_idt,
.get_gdt = svm_get_gdt,
.set_gdt = svm_set_gdt,
.get_dr = svm_get_dr,
.set_dr = svm_set_dr,
.cache_reg = svm_cache_reg,
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
.get_rflags = svm_get_rflags,
.set_rflags = svm_set_rflags,
.fpu_activate = svm_fpu_activate,
.fpu_deactivate = svm_fpu_deactivate,
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
.tlb_flush = svm_flush_tlb,
.run = svm_vcpu_run,
.handle_exit = handle_exit,
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
.skip_emulated_instruction = skip_emulated_instruction,
.set_interrupt_shadow = svm_set_interrupt_shadow,
.get_interrupt_shadow = svm_get_interrupt_shadow,
.patch_hypercall = svm_patch_hypercall,
.set_irq = svm_set_irq,
.set_nmi = svm_inject_nmi,
.queue_exception = svm_queue_exception,
.interrupt_allowed = svm_interrupt_allowed,
.nmi_allowed = svm_nmi_allowed,
.get_nmi_mask = svm_get_nmi_mask,
.set_nmi_mask = svm_set_nmi_mask,
.enable_nmi_window = enable_nmi_window,
.enable_irq_window = enable_irq_window,
.update_cr8_intercept = update_cr8_intercept,
.set_tss_addr = svm_set_tss_addr,
.get_tdp_level = get_npt_level,
.get_mt_mask = svm_get_mt_mask,
.exit_reasons_str = svm_exit_reasons_str,
.get_lpage_level = svm_get_lpage_level,
.cpuid_update = svm_cpuid_update,
.rdtscp_supported = svm_rdtscp_supported,
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
};
static int __init svm_init(void)
{
return kvm_init(&svm_x86_ops, sizeof(struct vcpu_svm),
THIS_MODULE);
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
}
static void __exit svm_exit(void)
{
kvm_exit();
[PATCH] kvm: userspace interface web site: http://kvm.sourceforge.net mailing list: kvm-devel@lists.sourceforge.net (http://lists.sourceforge.net/lists/listinfo/kvm-devel) The following patchset adds a driver for Intel's hardware virtualization extensions to the x86 architecture. The driver adds a character device (/dev/kvm) that exposes the virtualization capabilities to userspace. Using this driver, a process can run a virtual machine (a "guest") in a fully virtualized PC containing its own virtual hard disks, network adapters, and display. Using this driver, one can start multiple virtual machines on a host. Each virtual machine is a process on the host; a virtual cpu is a thread in that process. kill(1), nice(1), top(1) work as expected. In effect, the driver adds a third execution mode to the existing two: we now have kernel mode, user mode, and guest mode. Guest mode has its own address space mapping guest physical memory (which is accessible to user mode by mmap()ing /dev/kvm). Guest mode has no access to any I/O devices; any such access is intercepted and directed to user mode for emulation. The driver supports i386 and x86_64 hosts and guests. All combinations are allowed except x86_64 guest on i386 host. For i386 guests and hosts, both pae and non-pae paging modes are supported. SMP hosts and UP guests are supported. At the moment only Intel hardware is supported, but AMD virtualization support is being worked on. Performance currently is non-stellar due to the naive implementation of the mmu virtualization, which throws away most of the shadow page table entries every context switch. We plan to address this in two ways: - cache shadow page tables across tlb flushes - wait until AMD and Intel release processors with nested page tables Currently a virtual desktop is responsive but consumes a lot of CPU. Under Windows I tried playing pinball and watching a few flash movies; with a recent CPU one can hardly feel the virtualization. Linux/X is slower, probably due to X being in a separate process. In addition to the driver, you need a slightly modified qemu to provide I/O device emulation and the BIOS. Caveats (akpm: might no longer be true): - The Windows install currently bluescreens due to a problem with the virtual APIC. We are working on a fix. A temporary workaround is to use an existing image or install through qemu - Windows 64-bit does not work. That's also true for qemu, so it's probably a problem with the device model. [bero@arklinux.org: build fix] [simon.kagstrom@bth.se: build fix, other fixes] [uril@qumranet.com: KVM: Expose interrupt bitmap] [akpm@osdl.org: i386 build fix] [mingo@elte.hu: i386 fixes] [rdreier@cisco.com: add log levels to all printks] [randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings] [anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support] Signed-off-by: Yaniv Kamay <yaniv@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com> Cc: Simon Kagstrom <simon.kagstrom@bth.se> Cc: Bernhard Rosenkraenzer <bero@arklinux.org> Signed-off-by: Uri Lublin <uril@qumranet.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by: Anthony Liguori <anthony@codemonkey.ws> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 10:21:36 +00:00
}
module_init(svm_init)
module_exit(svm_exit)