Xiaohui Xin and some other folks at Intel have been looking into what's
behind the performance hit of paravirt_ops when running native.
It appears that the hit is entirely due to the paravirtualized
spinlocks introduced by:
| commit 8efcbab674
| Date: Mon Jul 7 12:07:51 2008 -0700
|
| paravirt: introduce a "lock-byte" spinlock implementation
The extra call/return in the spinlock path is somehow
causing an increase in the cycles/instruction of somewhere around 2-7%
(seems to vary quite a lot from test to test). The working theory is
that the CPU's pipeline is getting upset about the
call->call->locked-op->return->return, and seems to be failing to
speculate (though I haven't seen anything definitive about the precise
reasons). This doesn't entirely make sense, because the performance
hit is also visible on unlock and other operations which don't involve
locked instructions. But spinlock operations clearly swamp all the
other pvops operations, even though I can't imagine that they're
nearly as common (there's only a .05% increase in instructions
executed).
If I disable just the pv-spinlock calls, my tests show that pvops is
identical to non-pvops performance on native (my measurements show that
it is actually about .1% faster, but Xiaohui shows a .05% slowdown).
Summary of results, averaging 10 runs of the "mmperf" test, using a
no-pvops build as baseline:
nopv Pv-nospin Pv-spin
CPU cycles 100.00% 99.89% 102.18%
instructions 100.00% 100.10% 100.15%
CPI 100.00% 99.79% 102.03%
cache ref 100.00% 100.84% 100.28%
cache miss 100.00% 90.47% 88.56%
cache miss rate 100.00% 89.72% 88.31%
branches 100.00% 99.93% 100.04%
branch miss 100.00% 103.66% 107.72%
branch miss rt 100.00% 103.73% 107.67%
wallclock 100.00% 99.90% 102.20%
The clear effect here is that the 2% increase in CPI is
directly reflected in the final wallclock time.
(The other interesting effect is that the more ops are
out of line calls via pvops, the lower the cache access
and miss rates. Not too surprising, but it suggests that
the non-pvops kernel is over-inlined. On the flipside,
the branch misses go up correspondingly...)
So, what's the fix?
Paravirt patching turns all the pvops calls into direct calls, so
_spin_lock etc do end up having direct calls. For example, the compiler
generated code for paravirtualized _spin_lock is:
<_spin_lock+0>: mov %gs:0xb4c8,%rax
<_spin_lock+9>: incl 0xffffffffffffe044(%rax)
<_spin_lock+15>: callq *0xffffffff805a5b30
<_spin_lock+22>: retq
The indirect call will get patched to:
<_spin_lock+0>: mov %gs:0xb4c8,%rax
<_spin_lock+9>: incl 0xffffffffffffe044(%rax)
<_spin_lock+15>: callq <__ticket_spin_lock>
<_spin_lock+20>: nop; nop /* or whatever 2-byte nop */
<_spin_lock+22>: retq
One possibility is to inline _spin_lock, etc, when building an
optimised kernel (ie, when there's no spinlock/preempt
instrumentation/debugging enabled). That will remove the outer
call/return pair, returning the instruction stream to a single
call/return, which will presumably execute the same as the non-pvops
case. The downsides arel 1) it will replicate the
preempt_disable/enable code at eack lock/unlock callsite; this code is
fairly small, but not nothing; and 2) the spinlock definitions are
already a very heavily tangled mass of #ifdefs and other preprocessor
magic, and making any changes will be non-trivial.
The other obvious answer is to disable pv-spinlocks. Making them a
separate config option is fairly easy, and it would be trivial to
enable them only when Xen is enabled (as the only non-default user).
But it doesn't really address the common case of a distro build which
is going to have Xen support enabled, and leaves the open question of
whether the native performance cost of pv-spinlocks is worth the
performance improvement on a loaded Xen system (10% saving of overall
system CPU when guests block rather than spin). Still it is a
reasonable short-term workaround.
[ Impact: fix pvops performance regression when running native ]
Analysed-by: "Xin Xiaohui" <xiaohui.xin@intel.com>
Analysed-by: "Li Xin" <xin.li@intel.com>
Analysed-by: "Nakajima Jun" <jun.nakajima@intel.com>
Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
Acked-by: H. Peter Anvin <hpa@zytor.com>
Cc: Nick Piggin <npiggin@suse.de>
Cc: Xen-devel <xen-devel@lists.xensource.com>
LKML-Reference: <4A0B62F7.5030802@goop.org>
[ fixed the help text ]
Signed-off-by: Ingo Molnar <mingo@elte.hu>
mmu.c needs to #include module.h to prevent these warnings:
arch/x86/xen/mmu.c:239: warning: data definition has no type or storage class
arch/x86/xen/mmu.c:239: warning: type defaults to 'int' in declaration of 'EXPORT_SYMBOL_GPL'
arch/x86/xen/mmu.c:239: warning: parameter names (without types) in function declaration
[ Impact: cleanup ]
Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com>
Acked-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
The Xen pagetables are no longer implicitly reserved as part of the other
i386_start_kernel reservations, so make sure we explicitly reserve them.
This prevents them from being released into the general kernel free page
pool and reused.
[ Impact: fix Xen guest crash ]
Also-Bisected-by: Bryan Donlan <bdonlan@gmail.com>
Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
Cc: Xen-devel <xen-devel@lists.xensource.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
LKML-Reference: <4A032EEC.30509@goop.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Pass clocksource pointer to the read() callback for clocksources. This
allows us to share the callback between multiple instances.
[hugh@veritas.com: fix powerpc build of clocksource pass clocksource mods]
[akpm@linux-foundation.org: cleanup]
Signed-off-by: Magnus Damm <damm@igel.co.jp>
Acked-by: John Stultz <johnstul@us.ibm.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
* 'for-rc1/xen/core' of git://git.kernel.org/pub/scm/linux/kernel/git/jeremy/xen:
xen: add FIX_TEXT_POKE to fixmap
xen: honour VCPU availability on boot
xen: clean up gate trap/interrupt constants
xen: set _PAGE_NX in __supported_pte_mask before pagetable construction
xen: resume interrupts before system devices.
xen/mmu: weaken flush_tlb_other test
xen/mmu: some early pagetable cleanups
Xen: Add virt_to_pfn helper function
x86-64: remove PGE from must-have feature list
xen: mask XSAVE from cpuid
NULL noise: arch/x86/xen/smp.c
xen: remove xen_load_gdt debug
xen: make xen_load_gdt simpler
xen: clean up xen_load_gdt
xen: split construction of p2m mfn tables from registration
xen: separate p2m allocation from setting
xen: disable preempt for leave_lazy_mmu
Use phys_addr_t for receiving a physical address argument instead of
unsigned long. This allows fixmap to handle pages higher than 4GB on
x86-32.
Signed-off-by: Masami Hiramatsu <mhiramat@redhat.com>
Cc: Ingo Molnar <mingo@elte.hu>
Acked-by: Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
FIX_TEXT_POKE[01] are used to map kernel addresses, so they're mapping
pfns, not mfns.
Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
Some 64-bit machines don't support the NX flag in ptes.
Check for NX before constructing the kernel pagetables.
Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
Impact: fixes crashing bug
There's no particular problem with getting an empty cpu mask,
so just shortcut-return if we get one.
Avoids crash reported by Christophe Saout <christophe@saout.de>
Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
1. make sure early-allocated ptes are pinned, so they can be later
unpinned
2. don't pin pmd+pud, just make them RO
3. scatter some __inits around
Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
Xen leaves XSAVE set in cpuid, but doesn't allow cr4.OSXSAVE
to be set. This confuses the kernel and it ends up crashing on
an xsetbv instruction.
At boot time, try to set cr4.OSXSAVE, and mask XSAVE out of
cpuid it we can't. This will produce a spurious error from Xen,
but allows us to support XSAVE if/when Xen does.
This also factors out the cpuid mask decisions to boot time.
Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
Fix this sparse warnings:
arch/x86/xen/smp.c:316:52: warning: Using plain integer as NULL pointer
arch/x86/xen/smp.c:421:60: warning: Using plain integer as NULL pointer
Signed-off-by: Hannes Eder <hannes@hanneseder.net>
Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
Remove use of multicall machinery which is unused (gdt loading
is never performance critical). This removes the implicit use
of percpu variables, which simplifies understanding how
the percpu code's use of load_gdt interacts with this code.
Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
Build the p2m_mfn_list_list early with the rest of the p2m table, but
register it later when the real shared_info structure is in place.
Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
When doing very early p2m setting, we need to separate setting
from allocation, so split things up accordingly.
Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
xen_mc_flush() requires preemption to be disabled for its own sanity,
so disable it while we're flushing.
Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
Impact: new interface
Add a brk()-like allocator which effectively extends the bss in order
to allow very early code to do dynamic allocations. This is better than
using statically allocated arrays for data in subsystems which may never
get used.
The space for brk allocations is in the bss ELF segment, so that the
space is mapped properly by the code which maps the kernel, and so
that bootloaders keep the space free rather than putting a ramdisk or
something into it.
The bss itself, delimited by __bss_stop, ends before the brk area
(__brk_base to __brk_limit). The kernel text, data and bss is reserved
up to __bss_stop.
Any brk-allocated data is reserved separately just before the kernel
pagetable is built, as that code allocates from unreserved spaces
in the e820 map, potentially allocating from any unused brk memory.
Ultimately any unused memory in the brk area is used in the general
kernel memory pool.
Initially the brk space is set to 1MB, which is probably much larger
than any user needs (the largest current user is i386 head_32.S's code
to build the pagetables to map the kernel, which can get fairly large
with a big kernel image and no PSE support). So long as the system
has sufficient memory for the bootloader to reserve the kernel+1MB brk,
there are no bad effects resulting from an over-large brk.
Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
The virtually mapped percpu space causes us two problems:
- for hypercalls which take an mfn, we need to do a full pagetable
walk to convert the percpu va into an mfn, and
- when a hypercall requires a page to be mapped RO via all its aliases,
we need to make sure its RO in both the percpu mapping and in the
linear mapping
This primarily affects the gdt and the vcpu info structure.
Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
Cc: Xen-devel <xen-devel@lists.xensource.com>
Cc: Gerd Hoffmann <kraxel@redhat.com>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Tejun Heo <htejun@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
This avoids a lockdep warning from:
if (DEBUG_LOCKS_WARN_ON(unlikely(!early_boot_irqs_enabled)))
return;
in trace_hardirqs_on_caller();
Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
Cc: Mark McLoughlin <markmc@redhat.com>
Cc: Xen-devel <xen-devel@lists.xensource.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: remove unused/broken code
The Voyager subarch last built successfully on the v2.6.26 kernel
and has been stale since then and does not build on the v2.6.27,
v2.6.28 and v2.6.29-rc5 kernels.
No actual users beyond the maintainer reported this breakage.
Patches were sent and most of the fixes were accepted but the
discussion around how to do a few remaining issues cleanly
fizzled out with no resolution and the code remained broken.
In the v2.6.30 x86 tree development cycle 32-bit subarch support
has been reworked and removed - and the Voyager code, beyond the
build problems already known, needs serious and significant
changes and probably a rewrite to support it.
CONFIG_X86_VOYAGER has been marked BROKEN then. The maintainer has
been notified but no patches have been sent so far to fix it.
While all other subarchs have been converted to the new scheme,
voyager is still broken. We'd prefer to receive patches which
clean up the current situation in a constructive way, but even in
case of removal there is no obstacle to add that support back
after the issues have been sorted out in a mutually acceptable
fashion.
So remove this inactive code for now.
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: cleanup
make it simpler, don't need have one extra struct.
v2: fix the sgi_uv build
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
We can't call the callbacks after enabling interrupts, as we may get a
nested multicall call, which would cause a great deal of havok.
Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
If one of the components of a multicall fails, WARN rather than BUG,
to help with debugging.
Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Store the caller for each multicall so we can report it on failure.
Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
In general, the only definitions that assembly files can use
are in _types.S headers (where available), so convert them.
Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
The commit
commit 4595f9620c
Author: Rusty Russell <rusty@rustcorp.com.au>
Date: Sat Jan 10 21:58:09 2009 -0800
x86: change flush_tlb_others to take a const struct cpumask
causes xen_flush_tlb_others to allocate a multicall and then issue it
without initializing it in the case where the cpumask is empty,
leading to:
[ 8.354898] 1 multicall(s) failed: cpu 1
[ 8.354921] Pid: 2213, comm: bootclean Not tainted 2.6.29-rc3-x86_32p-xenU-tip #135
[ 8.354937] Call Trace:
[ 8.354955] [<c01036e3>] xen_mc_flush+0x133/0x1b0
[ 8.354971] [<c0105d2a>] ? xen_force_evtchn_callback+0x1a/0x30
[ 8.354988] [<c0105a60>] xen_flush_tlb_others+0xb0/0xd0
[ 8.355003] [<c0126643>] flush_tlb_page+0x53/0xa0
[ 8.355018] [<c0176a80>] do_wp_page+0x2a0/0x7c0
[ 8.355034] [<c0238f0a>] ? notify_remote_via_irq+0x3a/0x70
[ 8.355049] [<c0178950>] handle_mm_fault+0x7b0/0xa50
[ 8.355065] [<c0131a3e>] ? wake_up_new_task+0x8e/0xb0
[ 8.355079] [<c01337b5>] ? do_fork+0xe5/0x320
[ 8.355095] [<c0121919>] do_page_fault+0xe9/0x240
[ 8.355109] [<c0121830>] ? do_page_fault+0x0/0x240
[ 8.355125] [<c032457a>] error_code+0x72/0x78
[ 8.355139] call 1/1: op=2863311530 arg=[aaaaaaaa] result=-38 xen_flush_tlb_others+0x41/0xd0
Since empty cpumasks are rare and undoing an xen_mc_entry() is tricky
just issue such requests normally.
Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: pt_regs changed, lazy gs handling made optional, add slight
overhead to SAVE_ALL, simplifies error_code path a bit
On x86_32, %gs hasn't been used by kernel and handled lazily. pt_regs
doesn't have place for it and gs is saved/loaded only when necessary.
In preparation for stack protector support, this patch makes lazy %gs
handling optional by doing the followings.
* Add CONFIG_X86_32_LAZY_GS and place for gs in pt_regs.
* Save and restore %gs along with other registers in entry_32.S unless
LAZY_GS. Note that this unfortunately adds "pushl $0" on SAVE_ALL
even when LAZY_GS. However, it adds no overhead to common exit path
and simplifies entry path with error code.
* Define different user_gs accessors depending on LAZY_GS and add
lazy_save_gs() and lazy_load_gs() which are noop if !LAZY_GS. The
lazy_*_gs() ops are used to save, load and clear %gs lazily.
* Define ELF_CORE_COPY_KERNEL_REGS() which always read %gs directly.
xen and lguest changes need to be verified.
Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: Jeremy Fitzhardinge <jeremy@xensource.com>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Rather than overloading vectors for event channels, take full
responsibility for mapping an event channel to irq directly. With
this patch Xen has its own irq allocator.
When the kernel gets an event channel upcall, it maps the event
channel number to an irq and injects it into the normal interrupt
path.
Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Make the following style cleanups:
* drop unnecessary //#include from xen-asm_32.S
* compulsive adding of space after comma
* reformat multiline comments
Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Enable the use of the direct vcpu-access operations on 64-bit.
Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
Now that x86-64 has directly accessible percpu variables, it can also
implement the direct versions of these operations, which operate on a
vcpu_info structure directly embedded in the percpu area.
In fact, the 64-bit versions are more or less identical, and so can be
shared. The only two differences are:
1. xen_restore_fl_direct takes its argument in eax on 32-bit, and rdi on 64-bit.
Unfortunately it isn't possible to directly refer to the 2nd lsb of rdi directly
(as you can with %ah), so the code isn't quite as dense.
2. check_events needs to variants to save different registers.
Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
We need to access percpu data fairly early, so set up the percpu
registers as soon as possible. We only need to load the appropriate
segment register. We already have a GDT, but its hard to change it
early because we need to manipulate the pagetable to do so, and that
hasn't been set up yet.
Also, set the kernel stack when bringing up secondary CPUs. If we
don't they all end up sharing the same stack...
Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
Jeremy Fitzhardinge