Skip to content

R
raspberrypi-kernel

openeuler / raspberrypi-kernel

通知 13
Star 1
Fork 0
R
raspberrypi-kernel
切换分支/标签
  1. 21 5月, 2014 1 次提交
  3. 08 4月, 2014 1 次提交
    • D
      mm: per-thread vma caching · 615d6e87
      由 Davidlohr Bueso 提交于 
      This patch is a continuation of efforts trying to optimize find_vma(),
avoiding potentially expensive rbtree walks to locate a vma upon faults.
The original approach (https://lkml.org/lkml/2013/11/1/410), where the
largest vma was also cached, ended up being too specific and random,
thus further comparison with other approaches were needed.  There are
two things to consider when dealing with this, the cache hit rate and
the latency of find_vma().  Improving the hit-rate does not necessarily
translate in finding the vma any faster, as the overhead of any fancy
caching schemes can be too high to consider.

We currently cache the last used vma for the whole address space, which
provides a nice optimization, reducing the total cycles in find_vma() by
up to 250%, for workloads with good locality.  On the other hand, this
simple scheme is pretty much useless for workloads with poor locality.
Analyzing ebizzy runs shows that, no matter how many threads are
running, the mmap_cache hit rate is less than 2%, and in many situations
below 1%.

The proposed approach is to replace this scheme with a small per-thread
cache, maximizing hit rates at a very low maintenance cost.
Invalidations are performed by simply bumping up a 32-bit sequence
number.  The only expensive operation is in the rare case of a seq
number overflow, where all caches that share the same address space are
flushed.  Upon a miss, the proposed replacement policy is based on the
page number that contains the virtual address in question.  Concretely,
the following results are seen on an 80 core, 8 socket x86-64 box:

1) System bootup: Most programs are single threaded, so the per-thread
   scheme does improve ~50% hit rate by just adding a few more slots to
   the cache.

+----------------+----------+------------------+
| caching scheme | hit-rate | cycles (billion) |
+----------------+----------+------------------+
| baseline       | 50.61%   | 19.90            |
| patched        | 73.45%   | 13.58            |
+----------------+----------+------------------+

2) Kernel build: This one is already pretty good with the current
   approach as we're dealing with good locality.

+----------------+----------+------------------+
| caching scheme | hit-rate | cycles (billion) |
+----------------+----------+------------------+
| baseline       | 75.28%   | 11.03            |
| patched        | 88.09%   | 9.31             |
+----------------+----------+------------------+

3) Oracle 11g Data Mining (4k pages): Similar to the kernel build workload.

+----------------+----------+------------------+
| caching scheme | hit-rate | cycles (billion) |
+----------------+----------+------------------+
| baseline       | 70.66%   | 17.14            |
| patched        | 91.15%   | 12.57            |
+----------------+----------+------------------+

4) Ebizzy: There's a fair amount of variation from run to run, but this
   approach always shows nearly perfect hit rates, while baseline is just
   about non-existent.  The amounts of cycles can fluctuate between
   anywhere from ~60 to ~116 for the baseline scheme, but this approach
   reduces it considerably.  For instance, with 80 threads:

+----------------+----------+------------------+
| caching scheme | hit-rate | cycles (billion) |
+----------------+----------+------------------+
| baseline       | 1.06%    | 91.54            |
| patched        | 99.97%   | 14.18            |
+----------------+----------+------------------+

[akpm@linux-foundation.org: fix nommu build, per Davidlohr]
[akpm@linux-foundation.org: document vmacache_valid() logic]
[akpm@linux-foundation.org: attempt to untangle header files]
[akpm@linux-foundation.org: add vmacache_find() BUG_ON]
[hughd@google.com: add vmacache_valid_mm() (from Oleg)]
[akpm@linux-foundation.org: coding-style fixes]
[akpm@linux-foundation.org: adjust and enhance comments]
Signed-off-by: NDavidlohr Bueso <davidlohr@hp.com>
Reviewed-by: NRik van Riel <riel@redhat.com>
Acked-by: NLinus Torvalds <torvalds@linux-foundation.org>
Reviewed-by: NMichel Lespinasse <walken@google.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Tested-by: NHugh Dickins <hughd@google.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>
      615d6e87
  5. 04 4月, 2014 1 次提交
  7. 31 3月, 2014 1 次提交
  9. 19 3月, 2014 1 次提交
  11. 24 1月, 2014 2 次提交
    • C
      mm: ignore VM_SOFTDIRTY on VMA merging · 34228d47
      由 Cyrill Gorcunov 提交于 
      The VM_SOFTDIRTY bit affects vma merge routine: if two VMAs has all bits
in vm_flags matched except dirty bit the kernel can't longer merge them
and this forces the kernel to generate new VMAs instead.

It finally may lead to the situation when userspace application reaches
vm.max_map_count limit and get crashed in worse case

 | (gimp:11768): GLib-ERROR **: gmem.c:110: failed to allocate 4096 bytes
 |
 | (file-tiff-load:12038): LibGimpBase-WARNING **: file-tiff-load: gimp_wire_read(): error
 | xinit: connection to X server lost
 |
 | waiting for X server to shut down
 | /usr/lib64/gimp/2.0/plug-ins/file-tiff-load terminated: Hangup
 | /usr/lib64/gimp/2.0/plug-ins/script-fu terminated: Hangup
 | /usr/lib64/gimp/2.0/plug-ins/script-fu terminated: Hangup

  https://bugzilla.kernel.org/show_bug.cgi?id=67651
  https://bugzilla.gnome.org/show_bug.cgi?id=719619#c0

Initial problem came from missed VM_SOFTDIRTY in do_brk() routine but
even if we would set up VM_SOFTDIRTY here, there is still a way to
prevent VMAs from merging: one can call

 | echo 4 > /proc/$PID/clear_refs

and clear all VM_SOFTDIRTY over all VMAs presented in memory map, then
new do_brk() will try to extend old VMA and finds that dirty bit doesn't
match thus new VMA will be generated.

As discussed with Pavel, the right approach should be to ignore
VM_SOFTDIRTY bit when we're trying to merge VMAs and if merge successed
we mark extended VMA with dirty bit where needed.
Signed-off-by: NCyrill Gorcunov <gorcunov@openvz.org>
Reported-by: NBastian Hougaard <gnome@rvzt.net>
Reported-by: NMel Gorman <mgorman@suse.de>
Cc: Pavel Emelyanov <xemul@parallels.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>
      34228d47
    • P
      mm: audit/fix non-modular users of module_init in core code · a64fb3cd
      由 Paul Gortmaker 提交于 
      Code that is obj-y (always built-in) or dependent on a bool Kconfig
(built-in or absent) can never be modular.  So using module_init as an
alias for __initcall can be somewhat misleading.

Fix these up now, so that we can relocate module_init from init.h into
module.h in the future.  If we don't do this, we'd have to add module.h
to obviously non-modular code, and that would be a worse thing.

The audit targets the following module_init users for change:
 mm/ksm.c                       bool KSM
 mm/mmap.c                      bool MMU
 mm/huge_memory.c               bool TRANSPARENT_HUGEPAGE
 mm/mmu_notifier.c              bool MMU_NOTIFIER

Note that direct use of __initcall is discouraged, vs.  one of the
priority categorized subgroups.  As __initcall gets mapped onto
device_initcall, our use of subsys_initcall (which makes sense for these
files) will thus change this registration from level 6-device to level
4-subsys (i.e.  slightly earlier).

However no observable impact of that difference has been observed during
testing.

One might think that core_initcall (l2) or postcore_initcall (l3) would
be more appropriate for anything in mm/ but if we look at some actual
init functions themselves, we see things like:

mm/huge_memory.c --> hugepage_init     --> hugepage_init_sysfs
mm/mmap.c        --> init_user_reserve --> sysctl_user_reserve_kbytes
mm/ksm.c         --> ksm_init          --> sysfs_create_group

and hence the choice of subsys_initcall (l4) seems reasonable, and at
the same time minimizes the risk of changing the priority too
drastically all at once.  We can adjust further in the future.

Also, several instances of missing ";" at EOL are fixed.
Signed-off-by: NPaul Gortmaker <paul.gortmaker@windriver.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>
      a64fb3cd
  13. 22 1月, 2014 2 次提交
  15. 15 11月, 2013 1 次提交
    • K
      mm: convert mm->nr_ptes to atomic_long_t · e1f56c89
      由 Kirill A. Shutemov 提交于 
      With split page table lock for PMD level we can't hold mm->page_table_lock
while updating nr_ptes.

Let's convert it to atomic_long_t to avoid races.
Signed-off-by: NKirill A. Shutemov <kirill.shutemov@linux.intel.com>
Tested-by: NAlex Thorlton <athorlton@sgi.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Cc: "Eric W . Biederman" <ebiederm@xmission.com>
Cc: "Paul E . McKenney" <paulmck@linux.vnet.ibm.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: Dave Jones <davej@redhat.com>
Cc: David Howells <dhowells@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Kees Cook <keescook@chromium.org>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Michael Kerrisk <mtk.manpages@gmail.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Rik van Riel <riel@redhat.com>
Cc: Robin Holt <robinmholt@gmail.com>
Cc: Sedat Dilek <sedat.dilek@gmail.com>
Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Hugh Dickins <hughd@google.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>
      e1f56c89
  17. 13 11月, 2013 3 次提交
    • J
      mm: factor commit limit calculation · 00619bcc
      由 Jerome Marchand 提交于 
      The same calculation is currently done in three differents places.
Factor that code so future changes has to be made at only one place.

[akpm@linux-foundation.org: uninline vm_commit_limit()]
Signed-off-by: NJerome Marchand <jmarchan@redhat.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Michal Hocko <mhocko@suse.cz>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>
      00619bcc
    • A
      mm: ensure get_unmapped_area() returns higher address than mmap_min_addr · 2afc745f
      由 Akira Takeuchi 提交于 
      This patch fixes the problem that get_unmapped_area() can return illegal
address and result in failing mmap(2) etc.

In case that the address higher than PAGE_SIZE is set to
/proc/sys/vm/mmap_min_addr, the address lower than mmap_min_addr can be
returned by get_unmapped_area(), even if you do not pass any virtual
address hint (i.e.  the second argument).

This is because the current get_unmapped_area() code does not take into
account mmap_min_addr.

This leads to two actual problems as follows:

1. mmap(2) can fail with EPERM on the process without CAP_SYS_RAWIO,
   although any illegal parameter is not passed.

2. The bottom-up search path after the top-down search might not work in
   arch_get_unmapped_area_topdown().

Note: The first and third chunk of my patch, which changes "len" check,
are for more precise check using mmap_min_addr, and not for solving the
above problem.

[How to reproduce]

	--- test.c -------------------------------------------------
	#include <stdio.h>
	#include <unistd.h>
	#include <sys/mman.h>
	#include <sys/errno.h>

	int main(int argc, char *argv[])
	{
		void *ret = NULL, *last_map;
		size_t pagesize = sysconf(_SC_PAGESIZE);

		do {
			last_map = ret;
			ret = mmap(0, pagesize, PROT_NONE,
				MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
	//		printf("ret=%p\n", ret);
		} while (ret != MAP_FAILED);

		if (errno != ENOMEM) {
			printf("ERR: unexpected errno: %d (last map=%p)\n",
			errno, last_map);
		}

		return 0;
	}
	---------------------------------------------------------------

	$ gcc -m32 -o test test.c
	$ sudo sysctl -w vm.mmap_min_addr=65536
	vm.mmap_min_addr = 65536
	$ ./test  (run as non-priviledge user)
	ERR: unexpected errno: 1 (last map=0x10000)
Signed-off-by: NAkira Takeuchi <takeuchi.akr@jp.panasonic.com>
Signed-off-by: NKiyoshi Owada <owada.kiyoshi@jp.panasonic.com>
Reviewed-by: NNaoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>
      2afc745f
    • H
      mmap: arch_get_unmapped_area(): use proper mmap base for bottom up direction · 4e99b021
      由 Heiko Carstens 提交于 
      This is more or less the generic variant of commit 41aacc1e ("x86
get_unmapped_area: Access mmap_legacy_base through mm_struct member").

So effectively architectures which use an own arch_pick_mmap_layout()
implementation but call the generic arch_get_unmapped_area() now can
also randomize their mmap_base.

All architectures which have an own arch_pick_mmap_layout() and call the
generic arch_get_unmapped_area() (arm64, s390, tile) currently set
mmap_base to TASK_UNMAPPED_BASE.  This is also true for the generic
arch_pick_mmap_layout() function.  So this change is a no-op currently.
Signed-off-by: NHeiko Carstens <heiko.carstens@de.ibm.com>
Cc: Radu Caragea <sinaelgl@gmail.com>
Cc: Michel Lespinasse <walken@google.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Cc: Chris Metcalf <cmetcalf@tilera.com>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>
      4e99b021
  19. 25 10月, 2013 1 次提交
  21. 12 9月, 2013 6 次提交
  23. 16 8月, 2013 1 次提交
    • L
      Fix TLB gather virtual address range invalidation corner cases · 2b047252
      由 Linus Torvalds 提交于 
      Ben Tebulin reported:

 "Since v3.7.2 on two independent machines a very specific Git
  repository fails in 9/10 cases on git-fsck due to an SHA1/memory
  failures.  This only occurs on a very specific repository and can be
  reproduced stably on two independent laptops.  Git mailing list ran
  out of ideas and for me this looks like some very exotic kernel issue"

and bisected the failure to the backport of commit 53a59fc6 ("mm:
limit mmu_gather batching to fix soft lockups on !CONFIG_PREEMPT").

That commit itself is not actually buggy, but what it does is to make it
much more likely to hit the partial TLB invalidation case, since it
introduces a new case in tlb_next_batch() that previously only ever
happened when running out of memory.

The real bug is that the TLB gather virtual memory range setup is subtly
buggered.  It was introduced in commit 597e1c35 ("mm/mmu_gather:
enable tlb flush range in generic mmu_gather"), and the range handling
was already fixed at least once in commit e6c495a9 ("mm: fix the TLB
range flushed when __tlb_remove_page() runs out of slots"), but that fix
was not complete.

The problem with the TLB gather virtual address range is that it isn't
set up by the initial tlb_gather_mmu() initialization (which didn't get
the TLB range information), but it is set up ad-hoc later by the
functions that actually flush the TLB.  And so any such case that forgot
to update the TLB range entries would potentially miss TLB invalidates.

Rather than try to figure out exactly which particular ad-hoc range
setup was missing (I personally suspect it's the hugetlb case in
zap_huge_pmd(), which didn't have the same logic as zap_pte_range()
did), this patch just gets rid of the problem at the source: make the
TLB range information available to tlb_gather_mmu(), and initialize it
when initializing all the other tlb gather fields.

This makes the patch larger, but conceptually much simpler.  And the end
result is much more understandable; even if you want to play games with
partial ranges when invalidating the TLB contents in chunks, now the
range information is always there, and anybody who doesn't want to
bother with it won't introduce subtle bugs.

Ben verified that this fixes his problem.
Reported-bisected-and-tested-by: NBen Tebulin <tebulin@googlemail.com>
Build-testing-by: NStephen Rothwell <sfr@canb.auug.org.au>
Build-testing-by: NRichard Weinberger <richard.weinberger@gmail.com>
Reviewed-by: NMichal Hocko <mhocko@suse.cz>
Acked-by: NPeter Zijlstra <peterz@infradead.org>
Cc: stable@vger.kernel.org
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>
      2b047252
  25. 01 8月, 2013 1 次提交
  27. 11 7月, 2013 1 次提交
  29. 10 7月, 2013 2 次提交
  31. 04 7月, 2013 1 次提交
  33. 10 5月, 2013 1 次提交
  35. 08 5月, 2013 1 次提交
  37. 30 4月, 2013 7 次提交
    • C
      mm/mmap: check for RLIMIT_AS before unmapping · e8420a8e
      由 Cyril Hrubis 提交于 
      Fix a corner case for MAP_FIXED when requested mapping length is larger
than rlimit for virtual memory.  In such case any overlapping mappings
are unmapped before we check for the limit and return ENOMEM.

The check is moved before the loop that unmaps overlapping parts of
existing mappings.  When we are about to hit the limit (currently mapped
pages + len > limit) we scan for overlapping pages and check again
accounting for them.

This fixes situation when userspace program expects that the previous
mappings are preserved after the mmap() syscall has returned with error.
(POSIX clearly states that successfull mapping shall replace any
previous mappings.)

This corner case was found and can be tested with LTP testcase:

testcases/open_posix_testsuite/conformance/interfaces/mmap/24-2.c

In this case the mmap, which is clearly over current limit, unmaps
dynamic libraries and the testcase segfaults right after returning into
userspace.

I've also looked at the second instance of the unmapping loop in the
do_brk().  The do_brk() is called from brk() syscall and from vm_brk().
The brk() syscall checks for overlapping mappings and bails out when
there are any (so it can't be triggered from the brk syscall).  The
vm_brk() is called only from binmft handlers so it shouldn't be
triggered unless binmft handler created overlapping mappings.
Signed-off-by: NCyril Hrubis <chrubis@suse.cz>
Reviewed-by: NMel Gorman <mgorman@suse.de>
Reviewed-by: NWanpeng Li <liwanp@linux.vnet.ibm.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>
      e8420a8e
    • A
      mm: reinititalise user and admin reserves if memory is added or removed · 1640879a
      由 Andrew Shewmaker 提交于 
      Alter the admin and user reserves of the previous patches in this series
when memory is added or removed.

If memory is added and the reserves have been eliminated or increased
above the default max, then we'll trust the admin.

If memory is removed and there isn't enough free memory, then we need to
reset the reserves.

Otherwise keep the reserve set by the admin.

The reserve reset code is the same as the reserve initialization code.

I tested hot addition and removal by triggering it via sysfs.  The
reserves shrunk when they were set high and memory was removed.  They
were reset higher when memory was added again.

[akpm@linux-foundation.org: use register_hotmemory_notifier()]
[akpm@linux-foundation.org: init_user_reserve() and init_admin_reserve can no longer be __meminit]
[fengguang.wu@intel.com: make init_reserve_notifier() static]
[akpm@linux-foundation.org: coding-style fixes]
Signed-off-by: NAndrew Shewmaker <agshew@gmail.com>
Signed-off-by: NFengguang Wu <fengguang.wu@intel.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>
      1640879a
    • A
      mm: replace hardcoded 3% with admin_reserve_pages knob · 4eeab4f5
      由 Andrew Shewmaker 提交于 
      Add an admin_reserve_kbytes knob to allow admins to change the hardcoded
memory reserve to something other than 3%, which may be multiple
gigabytes on large memory systems.  Only about 8MB is necessary to
enable recovery in the default mode, and only a few hundred MB are
required even when overcommit is disabled.

This affects OVERCOMMIT_GUESS and OVERCOMMIT_NEVER.

admin_reserve_kbytes is initialized to min(3% free pages, 8MB)

I arrived at 8MB by summing the RSS of sshd or login, bash, and top.

Please see first patch in this series for full background, motivation,
testing, and full changelog.

[akpm@linux-foundation.org: coding-style fixes]
[akpm@linux-foundation.org: make init_admin_reserve() static]
Signed-off-by: NAndrew Shewmaker <agshew@gmail.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>
      4eeab4f5
    • A
      mm: limit growth of 3% hardcoded other user reserve · c9b1d098
      由 Andrew Shewmaker 提交于 
      Add user_reserve_kbytes knob.

Limit the growth of the memory reserved for other user processes to
min(3% current process size, user_reserve_pages).  Only about 8MB is
necessary to enable recovery in the default mode, and only a few hundred
MB are required even when overcommit is disabled.

user_reserve_pages defaults to min(3% free pages, 128MB)

I arrived at 128MB by taking the max VSZ of sshd, login, bash, and top ...
then adding the RSS of each.

This only affects OVERCOMMIT_NEVER mode.

Background

1. user reserve

__vm_enough_memory reserves a hardcoded 3% of the current process size for
other applications when overcommit is disabled.  This was done so that a
user could recover if they launched a memory hogging process.  Without the
reserve, a user would easily run into a message such as:

bash: fork: Cannot allocate memory

2. admin reserve

Additionally, a hardcoded 3% of free memory is reserved for root in both
overcommit 'guess' and 'never' modes.  This was intended to prevent a
scenario where root-cant-log-in and perform recovery operations.

Note that this reserve shrinks, and doesn't guarantee a useful reserve.

Motivation

The two hardcoded memory reserves should be updated to account for current
memory sizes.

Also, the admin reserve would be more useful if it didn't shrink too much.

When the current code was originally written, 1GB was considered
"enterprise".  Now the 3% reserve can grow to multiple GB on large memory
systems, and it only needs to be a few hundred MB at most to enable a user
or admin to recover a system with an unwanted memory hogging process.

I've found that reducing these reserves is especially beneficial for a
specific type of application load:

 * single application system
 * one or few processes (e.g. one per core)
 * allocating all available memory
 * not initializing every page immediately
 * long running

I've run scientific clusters with this sort of load.  A long running job
sometimes failed many hours (weeks of CPU time) into a calculation.  They
weren't initializing all of their memory immediately, and they weren't
using calloc, so I put systems into overcommit 'never' mode.  These
clusters run diskless and have no swap.

However, with the current reserves, a user wishing to allocate as much
memory as possible to one process may be prevented from using, for
example, almost 2GB out of 32GB.

The effect is less, but still significant when a user starts a job with
one process per core.  I have repeatedly seen a set of processes
requesting the same amount of memory fail because one of them could not
allocate the amount of memory a user would expect to be able to allocate.
For example, Message Passing Interfce (MPI) processes, one per core.  And
it is similar for other parallel programming frameworks.

Changing this reserve code will make the overcommit never mode more useful
by allowing applications to allocate nearly all of the available memory.

Also, the new admin_reserve_kbytes will be safer than the current behavior
since the hardcoded 3% of available memory reserve can shrink to something
useless in the case where applications have grabbed all available memory.

Risks

* "bash: fork: Cannot allocate memory"

  The downside of the first patch-- which creates a tunable user reserve
  that is only used in overcommit 'never' mode--is that an admin can set
  it so low that a user may not be able to kill their process, even if
  they already have a shell prompt.

  Of course, a user can get in the same predicament with the current 3%
  reserve--they just have to launch processes until 3% becomes negligible.

* root-cant-log-in problem

  The second patch, adding the tunable rootuser_reserve_pages, allows
  the admin to shoot themselves in the foot by setting it too small.  They
  can easily get the system into a state where root-can't-log-in.

  However, the new admin_reserve_kbytes will be safer than the current
  behavior since the hardcoded 3% of available memory reserve can shrink
  to something useless in the case where applications have grabbed all
  available memory.

Alternatives

 * Memory cgroups provide a more flexible way to limit application memory.

   Not everyone wants to set up cgroups or deal with their overhead.

 * We could create a fourth overcommit mode which provides smaller reserves.

   The size of useful reserves may be drastically different depending
   on the whether the system is embedded or enterprise.

 * Force users to initialize all of their memory or use calloc.

   Some users don't want/expect the system to overcommit when they malloc.
   Overcommit 'never' mode is for this scenario, and it should work well.

The new user and admin reserve tunables are simple to use, with low
overhead compared to cgroups.  The patches preserve current behavior where
3% of memory is less than 128MB, except that the admin reserve doesn't
shrink to an unusable size under pressure.  The code allows admins to tune
for embedded and enterprise usage.

FAQ

 * How is the root-cant-login problem addressed?
   What happens if admin_reserve_pages is set to 0?

   Root is free to shoot themselves in the foot by setting
   admin_reserve_kbytes too low.

   On x86_64, the minimum useful reserve is:
     8MB for overcommit 'guess'
   128MB for overcommit 'never'

   admin_reserve_pages defaults to min(3% free memory, 8MB)

   So, anyone switching to 'never' mode needs to adjust
   admin_reserve_pages.

 * How do you calculate a minimum useful reserve?

   A user or the admin needs enough memory to login and perform
   recovery operations, which includes, at a minimum:

   sshd or login + bash (or some other shell) + top (or ps, kill, etc.)

   For overcommit 'guess', we can sum resident set sizes (RSS)
   because we only need enough memory to handle what the recovery
   programs will typically use. On x86_64 this is about 8MB.

   For overcommit 'never', we can take the max of their virtual sizes (VSZ)
   and add the sum of their RSS. We use VSZ instead of RSS because mode
   forces us to ensure we can fulfill all of the requested memory allocations--
   even if the programs only use a fraction of what they ask for.
   On x86_64 this is about 128MB.

   When swap is enabled, reserves are useful even when they are as
   small as 10MB, regardless of overcommit mode.

   When both swap and overcommit are disabled, then the admin should
   tune the reserves higher to be absolutley safe. Over 230MB each
   was safest in my testing.

 * What happens if user_reserve_pages is set to 0?

   Note, this only affects overcomitt 'never' mode.

   Then a user will be able to allocate all available memory minus
   admin_reserve_kbytes.

   However, they will easily see a message such as:

   "bash: fork: Cannot allocate memory"

   And they won't be able to recover/kill their application.
   The admin should be able to recover the system if
   admin_reserve_kbytes is set appropriately.

 * What's the difference between overcommit 'guess' and 'never'?

   "Guess" allows an allocation if there are enough free + reclaimable
   pages. It has a hardcoded 3% of free pages reserved for root.

   "Never" allows an allocation if there is enough swap + a configurable
   percentage (default is 50) of physical RAM. It has a hardcoded 3% of
   free pages reserved for root, like "Guess" mode. It also has a
   hardcoded 3% of the current process size reserved for additional
   applications.

 * Why is overcommit 'guess' not suitable even when an app eventually
   writes to every page? It takes free pages, file pages, available
   swap pages, reclaimable slab pages into consideration. In other words,
   these are all pages available, then why isn't overcommit suitable?

   Because it only looks at the present state of the system. It
   does not take into account the memory that other applications have
   malloced, but haven't initialized yet. It overcommits the system.

Test Summary

There was little change in behavior in the default overcommit 'guess'
mode with swap enabled before and after the patch. This was expected.

Systems run most predictably (i.e. no oom kills) in overcommit 'never'
mode with swap enabled. This also allowed the most memory to be allocated
to a user application.

Overcommit 'guess' mode without swap is a bad idea. It is easy to
crash the system. None of the other tested combinations crashed.
This matches my experience on the Roadrunner supercomputer.

Without the tunable user reserve, a system in overcommit 'never' mode
and without swap does not allow the admin to recover, although the
admin can.

With the new tunable reserves, a system in overcommit 'never' mode
and without swap can be configured to:

1. maximize user-allocatable memory, running close to the edge of
recoverability

2. maximize recoverability, sacrificing allocatable memory to
ensure that a user cannot take down a system

Test Description

Fedora 18 VM - 4 x86_64 cores, 5725MB RAM, 4GB Swap

System is booted into multiuser console mode, with unnecessary services
turned off. Caches were dropped before each test.

Hogs are user memtester processes that attempt to allocate all free memory
as reported by /proc/meminfo

In overcommit 'never' mode, memory_ratio=100

Test Results

3.9.0-rc1-mm1

Overcommit | Swap | Hogs | MB Got/Wanted | OOMs | User Recovery | Admin Recovery
----------   ----   ----   -------------   ----   -------------   --------------
guess        yes    1      5432/5432       no     yes             yes
guess        yes    4      5444/5444       1      yes             yes
guess        no     1      5302/5449       no     yes             yes
guess        no     4      -               crash  no              no

never        yes    1      5460/5460       1      yes             yes
never        yes    4      5460/5460       1      yes             yes
never        no     1      5218/5432       no     no              yes
never        no     4      5203/5448       no     no              yes

3.9.0-rc1-mm1-tunablereserves

User and Admin Recovery show their respective reserves, if applicable.

Overcommit | Swap | Hogs | MB Got/Wanted | OOMs | User Recovery | Admin Recovery
----------   ----   ----   -------------   ----   -------------   --------------
guess        yes    1      5419/5419       no     - yes           8MB yes
guess        yes    4      5436/5436       1      - yes           8MB yes
guess        no     1      5440/5440       *      - yes           8MB yes
guess        no     4      -               crash  - no            8MB no

* process would successfully mlock, then the oom killer would pick it

never        yes    1      5446/5446       no     10MB yes        20MB yes
never        yes    4      5456/5456       no     10MB yes        20MB yes
never        no     1      5387/5429       no     128MB no        8MB barely
never        no     1      5323/5428       no     226MB barely    8MB barely
never        no     1      5323/5428       no     226MB barely    8MB barely

never        no     1      5359/5448       no     10MB no         10MB barely

never        no     1      5323/5428       no     0MB no          10MB barely
never        no     1      5332/5428       no     0MB no          50MB yes
never        no     1      5293/5429       no     0MB no          90MB yes

never        no     1      5001/5427       no     230MB yes       338MB yes
never        no     4*     4998/5424       no     230MB yes       338MB yes

* more memtesters were launched, able to allocate approximately another 100MB

Future Work

 - Test larger memory systems.

 - Test an embedded image.

 - Test other architectures.

 - Time malloc microbenchmarks.

 - Would it be useful to be able to set overcommit policy for
   each memory cgroup?

 - Some lines are slightly above 80 chars.
   Perhaps define a macro to convert between pages and kb?
   Other places in the kernel do this.

[akpm@linux-foundation.org: coding-style fixes]
[akpm@linux-foundation.org: make init_user_reserve() static]
Signed-off-by: NAndrew Shewmaker <agshew@gmail.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>
      c9b1d098
    • H
      mm: merging memory blocks resets mempolicy · 1444f92c
      由 Hampson, Steven T 提交于 
      Using mbind to change the mempolicy to MPOL_BIND on several adjacent
mmapped blocks may result in a reset of the mempolicy to MPOL_DEFAULT in
vma_adjust.

Test code.  Correct result is three lines containing "OK".

#include <stdio.h>
#include <unistd.h>
#include <sys/mman.h>
#include <numaif.h>
#include <errno.h>

/* gcc mbind_test.c -lnuma -o mbind_test -Wall */
#define MAXNODE 4096

void allocate()
{
	int ret;
	int len;
	int policy = -1;
	unsigned char *p;
	unsigned long mask[MAXNODE] = { 0 };
	unsigned long retmask[MAXNODE] = { 0 };

	len = getpagesize() * 0x2fc00;
	p = mmap(NULL, len, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS,
		 -1, 0);
	if (p == MAP_FAILED)
		printf("mbind err: %d\n", errno);

	mask[0] = 1;
	ret = mbind(p, len, MPOL_BIND, mask, MAXNODE, 0);
	if (ret < 0)
		printf("mbind err: %d %d\n", ret, errno);
	ret = get_mempolicy(&policy, retmask, MAXNODE, p, MPOL_F_ADDR);
	if (ret < 0)
		printf("get_mempolicy err: %d %d\n", ret, errno);

	if (policy == MPOL_BIND)
		printf("OK\n");
	else
		printf("ERROR: policy is %d\n", policy);
}

int main()
{
	allocate();
	allocate();
	allocate();
	return 0;
}
Signed-off-by: NSteven T Hampson <steven.t.hampson@intel.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Andi Kleen <andi@firstfloor.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>
      1444f92c
    • H
      mm: allow arch code to control the user page table ceiling · 6ee8630e
      由 Hugh Dickins 提交于 
      On architectures where a pgd entry may be shared between user and kernel
(e.g.  ARM+LPAE), freeing page tables needs a ceiling other than 0.
This patch introduces a generic USER_PGTABLES_CEILING that arch code can
override.  It is the responsibility of the arch code setting the ceiling
to ensure the complete freeing of the page tables (usually in
pgd_free()).

[catalin.marinas@arm.com: commit log; shift_arg_pages(), asm-generic/pgtables.h changes]
Signed-off-by: NHugh Dickins <hughd@google.com>
Signed-off-by: NCatalin Marinas <catalin.marinas@arm.com>
Cc: Russell King <linux@arm.linux.org.uk>
Cc: <stable@vger.kernel.org>	[3.3+]
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>
      6ee8630e
    • Z
      mmap: find_vma: remove the WARN_ON_ONCE(!mm) check · ee5df057
      由 Zhang Yanfei 提交于 
      Remove the WARN_ON_ONCE(!mm) check as the comment suggested.  Kernel
code calls find_vma only when it is absolutely sure that the mm_struct
arg to it is non-NULL.
Signed-off-by: NZhang Yanfei <zhangyanfei@cn.fujitsu.com>
Cc: k80c <k80ck80c@gmail.com>
Cc: Michel Lespinasse <walken@google.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>
      ee5df057
  39. 25 4月, 2013 1 次提交
  41. 05 4月, 2013 1 次提交
    • J
      mm: prevent mmap_cache race in find_vma() · b6a9b7f6
      由 Jan Stancek 提交于 
      find_vma() can be called by multiple threads with read lock
held on mm->mmap_sem and any of them can update mm->mmap_cache.
Prevent compiler from re-fetching mm->mmap_cache, because other
readers could update it in the meantime:

               thread 1                             thread 2
                                        |
  find_vma()                            |  find_vma()
    struct vm_area_struct *vma = NULL;  |
    vma = mm->mmap_cache;               |
    if (!(vma && vma->vm_end > addr     |
        && vma->vm_start <= addr)) {    |
                                        |    mm->mmap_cache = vma;
    return vma;                         |
     ^^ compiler may optimize this      |
        local variable out and re-read  |
        mm->mmap_cache                  |

This issue can be reproduced with gcc-4.8.0-1 on s390x by running
mallocstress testcase from LTP, which triggers:

  kernel BUG at mm/rmap.c:1088!
    Call Trace:
     ([<000003d100c57000>] 0x3d100c57000)
      [<000000000023a1c0>] do_wp_page+0x2fc/0xa88
      [<000000000023baae>] handle_pte_fault+0x41a/0xac8
      [<000000000023d832>] handle_mm_fault+0x17a/0x268
      [<000000000060507a>] do_protection_exception+0x1e2/0x394
      [<0000000000603a04>] pgm_check_handler+0x138/0x13c
      [<000003fffcf1f07a>] 0x3fffcf1f07a
    Last Breaking-Event-Address:
      [<000000000024755e>] page_add_new_anon_rmap+0xc2/0x168

Thanks to Jakub Jelinek for his insight on gcc and helping to
track this down.
Signed-off-by: NJan Stancek <jstancek@redhat.com>
Acked-by: NDavid Rientjes <rientjes@google.com>
Signed-off-by: NHugh Dickins <hughd@google.com>
Cc: stable@vger.kernel.org
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>
      b6a9b7f6
  43. 29 3月, 2013 1 次提交
  45. 28 2月, 2013 1 次提交
    • L
      mm: do not grow the stack vma just because of an overrun on preceding vma · 09884964
      由 Linus Torvalds 提交于 
      The stack vma is designed to grow automatically (marked with VM_GROWSUP
or VM_GROWSDOWN depending on architecture) when an access is made beyond
the existing boundary.  However, particularly if you have not limited
your stack at all ("ulimit -s unlimited"), this can cause the stack to
grow even if the access was really just one past *another* segment.

And that's wrong, especially since we first grow the segment, but then
immediately later enforce the stack guard page on the last page of the
segment.  So _despite_ first growing the stack segment as a result of
the access, the kernel will then make the access cause a SIGSEGV anyway!

So do the same logic as the guard page check does, and consider an
access to within one page of the next segment to be a bad access, rather
than growing the stack to abut the next segment.
Reported-and-tested-by: NHeiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>
      09884964
  47. 24 2月, 2013 1 次提交