The kernel address space of a 64 bit kernel currently uses a three level
page table and the vmemmap array has a fixed address and a fixed maximum
size. A three level page table is good enough for systems with less than
3.8TB of memory, for bigger systems four page table levels need to be
used. Each page table level costs a bit of performance, use 3 levels for
normal systems and 4 levels only for the really big systems.
To avoid bloating sparse.o too much set MAX_PHYSMEM_BITS to 46 for a
maximum of 64TB of memory.
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Linux on System z uses a ballooner based on diagnose 0x10. (aka as
collaborative memory management). This patch implements diagnose
0x10 on the guest address space.
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
gmap_fault needs to walk the guest page table. However, parts of
that may change if some other thread does munmap. In that case
gmap_unmap_notifier will also unmap the corresponding parts from
the guest page table. We need to take mmap_sem in order to serialize
these operations.
do_exception now calls __gmap_fault with mmap_sem held which does
not get exported to modules. The exported function, which is called
from KVM, now takes mmap_sem.
Reported-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Carsten Otte <cotte@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
This introduces locking via mm->page_table_lock to protect
the rmap list for guest mappings from being corrupted by concurrent
operations.
Signed-off-by: Carsten Otte <cotte@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Fix possible deadlock reported by lockdep:
qemu-system-s39/2963 is trying to acquire lock:
(&mm->mmap_sem){++++++}, at: gmap_alloc_table+0x9c/0x120
but task is already holding lock:
(&mm->mmap_sem){++++++}, at: gmap_map_segment+0xa6/0x27c
Actually gmap_alloc_table is the only called in gmap_map_segment with
mmap_sem held, thus it's safe to simply remove the inner lock.
Signed-off-by: Carsten Otte <cotte@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
The rcu page table free code uses a couple of bits in the page table
pointer passed to tlb_remove_table to discern the different page table
types. __tlb_remove_table extracts the type with an incorrect mask which
leads to memory leaks. The correct mask is ((FRAG_MASK << 4) | FRAG_MASK).
Cc: stable@kernel.org
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
If gmap_unmap_segment figures that the segment was not mapped in the
first place, it need to up mmap_sem on exit.
Cc: <stable@kernel.org>
Signed-off-by: Carsten Otte <cotte@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
598841ca99 ([S390] use gmap address
spaces for kvm guest images) changed kvm to use a separate address
space for kvm guests. This address space was switched in __vcpu_run
In some cases (preemption, page fault) there is the possibility that
this address space switch is lost.
The typical symptom was a huge amount of validity intercepts or
random guest addressing exceptions.
Fix this by doing the switch in sie_loop and sie_exit and saving the
address space in the gmap structure itself. Also use the preempt
notifier.
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
Acked-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Fix the following compile warning for !CONFIG_PGSTE:
CC arch/s390/mm/pgtable.o
arch/s390/mm/pgtable.c: In function ‘page_table_alloc_pgste’:
arch/s390/mm/pgtable.c:531:1: warning: no return statement in function returning non-void [-Wreturn-type]
Signed-off-by: Jan Glauber <jang@linux.vnet.ibm.com>
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Add code that allows KVM to control the virtual memory layout that
is seen by a guest. The guest address space uses a second page table
that shares the last level pte-tables with the process page table.
If a page is unmapped from the process page table it is automatically
unmapped from the guest page table as well.
The guest address space mapping starts out empty, KVM can map any
individual 1MB segments from the process virtual memory to any 1MB
aligned location in the guest virtual memory. If a target segment in
the process virtual memory does not exist or is unmapped while a
guest mapping exists the desired target address is stored as an
invalid segment table entry in the guest page table.
The population of the guest page table is fault driven.
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Replace the s390 specific rcu page-table freeing code with the
generic variant. This requires to duplicate the definition for the
struct mmu_table_batch as s390 does not use the generic tlb flush
code.
While we are at it remove the restriction that page table fragments
can not be reused after a single fragment has been freed with rcu
and split out allocation and freeing of page tables with pgstes.
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Quite a few functions that get called from the tlb gather code require that
preemption must be disabled. So disable preemption inside of the called
functions instead.
The only drawback is that rcu_table_freelist_finish() doesn't get necessarily
called on the cpu(s) that filled the free lists. So we may see a delay, until
we finally see an rcu callback. However over time this shouldn't matter.
So we get rid of lots of "BUG: using smp_processor_id() in preemptible"
messages.
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Fold all the mmu_gather rework patches into one for submission
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Reported-by: Hugh Dickins <hughd@google.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: David Miller <davem@davemloft.net>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: Russell King <rmk@arm.linux.org.uk>
Cc: Paul Mundt <lethal@linux-sh.org>
Cc: Jeff Dike <jdike@addtoit.com>
Cc: Richard Weinberger <richard@nod.at>
Cc: Tony Luck <tony.luck@intel.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Mel Gorman <mel@csn.ul.ie>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Nick Piggin <npiggin@kernel.dk>
Cc: Namhyung Kim <namhyung@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The noexec support on s390 does not rely on a bit in the page table
entry but utilizes the secondary space mode to distinguish between
memory accesses for instructions vs. data. The noexec code relies
on the assumption that the cpu will always use the secondary space
page table for data accesses while it is running in the secondary
space mode. Up to the z9-109 class machines this has been the case.
Unfortunately this is not true anymore with z10 and later machines.
The load-relative-long instructions lrl, lgrl and lgfrl access the
memory operand using the same addressing-space mode that has been
used to fetch the instruction.
This breaks the noexec mode for all user space binaries compiled
with march=z10 or later. The only option is to remove the current
noexec support.
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
After page_table_free_rcu removed a page from the pgtable_list
page_table_free better not add it again. Otherwise a page_table_alloc
can reuse a page table fragment that is still in the rcu process.
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
commit 050eef364a
[S390] fix tlb flushing vs. concurrent /proc accesses
broke KVM on s390x. On every schedule a
Badness at include/asm/mmu_context.h:83 appears. s390_enable_sie
replaces the mm on the __running__ task, therefore, we have to
increase the attach count of the new mm.
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
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>
Introduce user_mode to replace the two variables switch_amode and
s390_noexec. There are three valid combinations of the old values:
1) switch_amode == 0 && s390_noexec == 0
2) switch_amode == 1 && s390_noexec == 0
3) switch_amode == 1 && s390_noexec == 1
They get replaced by
1) user_mode == HOME_SPACE_MODE
2) user_mode == PRIMARY_SPACE_MODE
3) user_mode == SECONDARY_SPACE_MODE
The new kernel parameter user_mode=[primary,secondary,home] lets
you choose the address space mode the user space processes should
use. In addition the CONFIG_S390_SWITCH_AMODE config option
is removed.
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
next-20090925 randconfig build breaks on s390x, with CONFIG_AIO=n.
arch/s390/mm/pgtable.c: In function 's390_enable_sie':
arch/s390/mm/pgtable.c:282: error: 'struct mm_struct' has no member named 'ioctx_list'
arch/s390/mm/pgtable.c:298: error: 'struct mm_struct' has no member named 'ioctx_list'
make[1]: *** [arch/s390/mm/pgtable.o] Error 1
Reported-by: Kamalesh Babulal <kamalesh@linux.vnet.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Make the inline assembly look like all others.
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Suzuki Poulose reported the following recursive locking bug on s390:
Here is the stack trace : (see Appendix I for more info)
[<0000000000406ed6>] _spin_lock+0x52/0x94
[<0000000000103bde>] crst_table_free+0x14e/0x1a4
[<00000000001ba684>] __pmd_alloc+0x114/0x1ec
[<00000000001be8d0>] handle_mm_fault+0x2cc/0xb80
[<0000000000407d62>] do_dat_exception+0x2b6/0x3a0
[<0000000000114f8c>] sysc_return+0x0/0x8
[<00000200001642b2>] 0x200001642b2
The page_table_lock is already acquired in __pmd_alloc (mm/memory.c) and
it tries to populate the pud/pgd with a new pmd allocated. If another
thread populates it before we get a chance, we free the pmd using
pmd_free().
On s390x, pmd_free(even pud_free ) is #defined to crst_table_free(),
which acquires the page_table_lock to protect the crst_table index updates.
Hence this ends up in a recursive locking of the page_table_lock.
The solution suggested by Dave Hansen is to use a new spin lock in the mmu
context to protect the access to the crst_list and the pgtable_list.
Reported-by: Suzuki Poulose <suzuki@in.ibm.com>
Cc: Dave Hansen <dave@linux.vnet.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Fix the following build failure caused by make allyesconfig using
CONFIG_HIBERNATION and CONFIG_DEBUG_PAGEALLOC
kernel/built-in.o: In function `saveable_page':
kernel/power/snapshot.c:897: undefined reference to `kernel_page_present'
kernel/built-in.o: In function `safe_copy_page':
kernel/power/snapshot.c:948: undefined reference to `kernel_page_present'
make: *** [.tmp_vmlinux1] Error 1
Signed-off-by: Hans-Joachim Picht <hans@linux.vnet.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
With the kernel parameter 'vmalloc=<size>' the size of the vmalloc area
can be specified. This can be used to increase or decrease the size of
the area. Works in the same way as on some other architectures.
This can be useful for features which make excessive use of vmalloc and
wouldn't work otherwise.
The default sizes remain unchanged: 96MB for 31 bit kernels and 1GB for
64 bit kernels.
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Makes code futureproof against the impending change to mm->cpu_vm_mask.
It's also a chance to use the new cpumask_ ops which take a pointer
(the older ones are deprecated, but there's no hurry for arch code).
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
The sie instruction requires address spaces to be switched
to run proper. This patch verifies that this is the case
in s390_enable_sie, otherwise the kernel would crash badly
as soon as the process runs into sie.
Signed-off-by: Carsten Otte <cotte@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Make page table walking on s390 more robust. The current code requires
that the pgd/pud/pmd/pte loop is only done for address ranges that are
below the end address of the last vma of the address space. But this
is not always true, e.g. the generic page table walker does not guarantee
this. Change TASK_SIZE/TASK_SIZE_OF to reflect the current size of the
address space. This makes the generic page table walker happy but it
breaks the upgrade of a 3 level page table to a 4 level page table.
To make the upgrade work again another fix is required.
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
The mm->ioctx_list is currently protected by a reader-writer lock,
so we always grab that lock on the read side for doing ioctx
lookups. As the workload is extremely reader biased, turn this into
an rcu hlist so we can make lookup_ioctx() lockless. Get rid of
the rwlock and use a spinlock for providing update side exclusion.
There's usually only 1 entry on this list, so it doesn't make sense
to look into fancier data structures.
Reviewed-by: Jeff Moyer <jmoyer@redhat.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
The current enable_sie code sets the mm->context.pgstes bit to tell
dup_mm that the new mm should have extended page tables. This bit is also
used by the s390 specific page table primitives to decide about the page
table layout - which means context.pgstes has two meanings. This can cause
any kind of bugs. For example - e.g. shrink_zone can call
ptep_clear_flush_young while enable_sie is running. ptep_clear_flush_young
will test for context.pgstes. Since enable_sie changed that value of the old
struct mm without changing the page table layout ptep_clear_flush_young will
do the wrong thing.
The solution is to split pgstes into two bits
- one for the allocation
- one for the current state
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
There are potential locking problem in enable_sie. We take the task_lock
and the mmap_sem. As exit_mm uses the same locks vice versa, this triggers
a lockdep warning.
The second problem is that dup_mm and mmput might sleep, so we must not
hold the task_lock at that moment.
The solution is to dup the mm unconditional and use the task_lock before and
afterwards to check if we can use the new mm. dup_mm and mmput are called
outside the task_lock, but we run update_mm while holding the task_lock,
protection us against ptrace.
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Carsten Otte <cotte@de.ibm.com>
Acked-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: Avi Kivity <avi@qumranet.com>
The SIE instruction on s390 uses the 2nd half of the page table page to
virtualize the storage keys of a guest. This patch offers the s390_enable_sie
function, which reorganizes the page tables of a single-threaded process to
reserve space in the page table:
s390_enable_sie makes sure that the process is single threaded and then uses
dup_mm to create a new mm with reorganized page tables. The old mm is freed
and the process has now a page status extended field after every page table.
Code that wants to exploit pgstes should SELECT CONFIG_PGSTE.
This patch has a small common code hit, namely making dup_mm non-static.
Edit (Carsten): I've modified Martin's patch, following Jeremy Fitzhardinge's
review feedback. Now we do have the prototype for dup_mm in
include/linux/sched.h. Following Martin's suggestion, s390_enable_sie() does now
call task_lock() to prevent race against ptrace modification of mm_users.
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: Carsten Otte <cotte@de.ibm.com>
Acked-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Avi Kivity <avi@qumranet.com>
Add support for different number of page table levels dependent
on the highest address used for a process. This will cause a 31 bit
process to use a two level page table instead of the four level page
table that is the default after the pud has been introduced. Likewise
a normal 64 bit process will use three levels instead of four. Only
if a process runs out of the 4 tera bytes which can be addressed with
a three level page table the fourth level is dynamically added. Then
the process can use up to 8 peta byte.
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Background: I've implemented 1K/2K page tables for s390. These sub-page
page tables are required to properly support the s390 virtualization
instruction with KVM. The SIE instruction requires that the page tables
have 256 page table entries (pte) followed by 256 page status table entries
(pgste). The pgstes are only required if the process is using the SIE
instruction. The pgstes are updated by the hardware and by the hypervisor
for a number of reasons, one of them is dirty and reference bit tracking.
To avoid wasting memory the standard pte table allocation should return
1K/2K (31/64 bit) and 2K/4K if the process is using SIE.
Problem: Page size on s390 is 4K, page table size is 1K or 2K. That means
the s390 version for pte_alloc_one cannot return a pointer to a struct
page. Trouble is that with the CONFIG_HIGHPTE feature on x86 pte_alloc_one
cannot return a pointer to a pte either, since that would require more than
32 bit for the return value of pte_alloc_one (and the pte * would not be
accessible since its not kmapped).
Solution: The only solution I found to this dilemma is a new typedef: a
pgtable_t. For s390 pgtable_t will be a (pte *) - to be introduced with a
later patch. For everybody else it will be a (struct page *). The
additional problem with the initialization of the ptl lock and the
NR_PAGETABLE accounting is solved with a constructor pgtable_page_ctor and
a destructor pgtable_page_dtor. The page table allocation and free
functions need to call these two whenever a page table page is allocated or
freed. pmd_populate will get a pgtable_t instead of a struct page pointer.
To get the pgtable_t back from a pmd entry that has been installed with
pmd_populate a new function pmd_pgtable is added. It replaces the pmd_page
call in free_pte_range and apply_to_pte_range.
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: <linux-arch@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
- De-confuse the defines for the address-space-control-elements
and the segment/region table entries.
- Create out of line functions for page table allocation / freeing.
- Simplify get_shadow_xxx functions.
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Martin Schwidefsky