linux-stable.git/include/linux/mm_types.h, branch v3.12.64

mm: per-thread vma caching

2014-09-26T09:51:57+00:00

commit 615d6e8756c87149f2d4c1b93d471bca002bd849 upstream.

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: Davidlohr Bueso 
Reviewed-by: Rik van Riel 
Acked-by: Linus Torvalds 
Reviewed-by: Michel Lespinasse 
Cc: Oleg Nesterov 
Tested-by: Hugh Dickins 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds 
Signed-off-by: Mel Gorman 
Signed-off-by: Jiri Slaby

mm: numa: guarantee that tlb_flush_pending updates are visible before page table updates

2014-01-09T20:25:14+00:00

commit af2c1401e6f9177483be4fad876d0073669df9df upstream.

According to documentation on barriers, stores issued before a LOCK can
complete after the lock implying that it's possible tlb_flush_pending
can be visible after a page table update.  As per revised documentation,
this patch adds a smp_mb__before_spinlock to guarantee the correct
ordering.

Signed-off-by: Mel Gorman 
Acked-by: Paul E. McKenney 
Reviewed-by: Rik van Riel 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds 
Signed-off-by: Greg Kroah-Hartman

mm: fix TLB flush race between migration, and change_protection_range

2014-01-09T20:25:14+00:00

commit 20841405940e7be0617612d521e206e4b6b325db upstream.

There are a few subtle races, between change_protection_range (used by
mprotect and change_prot_numa) on one side, and NUMA page migration and
compaction on the other side.

The basic race is that there is a time window between when the PTE gets
made non-present (PROT_NONE or NUMA), and the TLB is flushed.

During that time, a CPU may continue writing to the page.

This is fine most of the time, however compaction or the NUMA migration
code may come in, and migrate the page away.

When that happens, the CPU may continue writing, through the cached
translation, to what is no longer the current memory location of the
process.

This only affects x86, which has a somewhat optimistic pte_accessible.
All other architectures appear to be safe, and will either always flush,
or flush whenever there is a valid mapping, even with no permissions
(SPARC).

The basic race looks like this:

CPU A			CPU B			CPU C

						load TLB entry
make entry PTE/PMD_NUMA
			fault on entry
						read/write old page
			start migrating page
			change PTE/PMD to new page
						read/write old page [*]
flush TLB
						reload TLB from new entry
						read/write new page
						lose data

[*] the old page may belong to a new user at this point!

The obvious fix is to flush remote TLB entries, by making sure that
pte_accessible aware of the fact that PROT_NONE and PROT_NUMA memory may
still be accessible if there is a TLB flush pending for the mm.

This should fix both NUMA migration and compaction.

[mgorman@suse.de: fix build]
Signed-off-by: Rik van Riel 
Signed-off-by: Mel Gorman 
Cc: Alex Thorlton 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds 
Signed-off-by: Greg Kroah-Hartman

Merge git://git.kvack.org/~bcrl/aio-next

2013-09-13T17:55:58+00:00

Pull aio changes from Ben LaHaise:
 "First off, sorry for this pull request being late in the merge window.
  Al had raised a couple of concerns about 2 items in the series below.
  I addressed the first issue (the race introduced by Gu's use of
  mm_populate()), but he has not provided any further details on how he
  wants to rework the anon_inode.c changes (which were sent out months
  ago but have yet to be commented on).

  The bulk of the changes have been sitting in the -next tree for a few
  months, with all the issues raised being addressed"

* git://git.kvack.org/~bcrl/aio-next: (22 commits)
  aio: rcu_read_lock protection for new rcu_dereference calls
  aio: fix race in ring buffer page lookup introduced by page migration support
  aio: fix rcu sparse warnings introduced by ioctx table lookup patch
  aio: remove unnecessary debugging from aio_free_ring()
  aio: table lookup: verify ctx pointer
  staging/lustre: kiocb->ki_left is removed
  aio: fix error handling and rcu usage in "convert the ioctx list to table lookup v3"
  aio: be defensive to ensure request batching is non-zero instead of BUG_ON()
  aio: convert the ioctx list to table lookup v3
  aio: double aio_max_nr in calculations
  aio: Kill ki_dtor
  aio: Kill ki_users
  aio: Kill unneeded kiocb members
  aio: Kill aio_rw_vect_retry()
  aio: Don't use ctx->tail unnecessarily
  aio: io_cancel() no longer returns the io_event
  aio: percpu ioctx refcount
  aio: percpu reqs_available
  aio: reqs_active -> reqs_available
  aio: fix build when migration is disabled
  ...

x86 get_unmapped_area: Access mmap_legacy_base through mm_struct member

2013-08-22T17:19:35+00:00

This is the updated version of df54d6fa5427 ("x86 get_unmapped_area():
use proper mmap base for bottom-up direction") that only randomizes the
mmap base address once.

Signed-off-by: Radu Caragea 
Reported-and-tested-by: Jeff Shorey 
Cc: Andrew Morton 
Cc: Michel Lespinasse 
Cc: Oleg Nesterov 
Cc: Rik van Riel 
Cc: Ingo Molnar 
Cc: Adrian Sendroiu 
Cc: Greg KH 
Cc: Kamal Mostafa 
Signed-off-by: Linus Torvalds

aio: convert the ioctx list to table lookup v3

2013-07-30T16:56:36+00:00

On Wed, Jun 12, 2013 at 11:14:40AM -0700, Kent Overstreet wrote:
> On Mon, Apr 15, 2013 at 02:40:55PM +0300, Octavian Purdila wrote:
> > When using a large number of threads performing AIO operations the
> > IOCTX list may get a significant number of entries which will cause
> > significant overhead. For example, when running this fio script:
> >
> > rw=randrw; size=256k ;directory=/mnt/fio; ioengine=libaio; iodepth=1
> > blocksize=1024; numjobs=512; thread; loops=100
> >
> > on an EXT2 filesystem mounted on top of a ramdisk we can observe up to
> > 30% CPU time spent by lookup_ioctx:
> >
> >  32.51%  [guest.kernel]  [g] lookup_ioctx
> >   9.19%  [guest.kernel]  [g] __lock_acquire.isra.28
> >   4.40%  [guest.kernel]  [g] lock_release
> >   4.19%  [guest.kernel]  [g] sched_clock_local
> >   3.86%  [guest.kernel]  [g] local_clock
> >   3.68%  [guest.kernel]  [g] native_sched_clock
> >   3.08%  [guest.kernel]  [g] sched_clock_cpu
> >   2.64%  [guest.kernel]  [g] lock_release_holdtime.part.11
> >   2.60%  [guest.kernel]  [g] memcpy
> >   2.33%  [guest.kernel]  [g] lock_acquired
> >   2.25%  [guest.kernel]  [g] lock_acquire
> >   1.84%  [guest.kernel]  [g] do_io_submit
> >
> > This patchs converts the ioctx list to a radix tree. For a performance
> > comparison the above FIO script was run on a 2 sockets 8 core
> > machine. This are the results (average and %rsd of 10 runs) for the
> > original list based implementation and for the radix tree based
> > implementation:
> >
> > cores         1         2         4         8         16        32
> > list       109376 ms  69119 ms  35682 ms  22671 ms  19724 ms  16408 ms
> > %rsd         0.69%      1.15%     1.17%     1.21%     1.71%     1.43%
> > radix       73651 ms  41748 ms  23028 ms  16766 ms  15232 ms   13787 ms
> > %rsd         1.19%      0.98%     0.69%     1.13%    0.72%      0.75%
> > % of radix
> > relative    66.12%     65.59%    66.63%    72.31%   77.26%     83.66%
> > to list
> >
> > To consider the impact of the patch on the typical case of having
> > only one ctx per process the following FIO script was run:
> >
> > rw=randrw; size=100m ;directory=/mnt/fio; ioengine=libaio; iodepth=1
> > blocksize=1024; numjobs=1; thread; loops=100
> >
> > on the same system and the results are the following:
> >
> > list        58892 ms
> > %rsd         0.91%
> > radix       59404 ms
> > %rsd         0.81%
> > % of radix
> > relative    100.87%
> > to list
>
> So, I was just doing some benchmarking/profiling to get ready to send
> out the aio patches I've got for 3.11 - and it looks like your patch is
> causing a ~1.5% throughput regression in my testing :/
... 

I've got an alternate approach for fixing this wart in lookup_ioctx()...
Instead of using an rbtree, just use the reserved id in the ring buffer
header to index an array pointing the ioctx.  It's not finished yet, and
it needs to be tidied up, but is most of the way there.

		-ben
--
"Thought is the essence of where you are now."
--
kmo> And, a rework of Ben's code, but this was entirely his idea
kmo>		-Kent

bcrl> And fix the code to use the right mm_struct in kill_ioctx(), actually
free memory.

Signed-off-by: Benjamin LaHaise

mm: remove free_area_cache

2013-07-11T01:11:34+00:00

Since all architectures have been converted to use vm_unmapped_area(),
there is no remaining use for the free_area_cache.

Signed-off-by: Michel Lespinasse 
Acked-by: Rik van Riel 
Cc: "James E.J. Bottomley" 
Cc: "Luck, Tony" 
Cc: Benjamin Herrenschmidt 
Cc: David Howells 
Cc: Helge Deller 
Cc: Ivan Kokshaysky 
Cc: Matt Turner 
Cc: Paul Mackerras 
Cc: Richard Henderson 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds

mm: fold page->_last_nid into page->flags where possible

2013-02-24T01:50:17+00:00

page->_last_nid fits into page->flags on 64-bit.  The unlikely 32-bit
NUMA configuration with NUMA Balancing will still need an extra page
field.  As Peter notes "Completely dropping 32bit support for
CONFIG_NUMA_BALANCING would simplify things, but it would also remove
the warning if we grow enough 64bit only page-flags to push the last-cpu
out."

[mgorman@suse.de: minor modifications]
Signed-off-by: Mel Gorman 
Cc: Peter Zijlstra 
Cc: Andrea Arcangeli 
Cc: Ingo Molnar 
Cc: Simon Jeons 
Cc: Wanpeng Li 
Cc: Hugh Dickins 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds

mm: move page flags layout to separate header

2013-02-24T01:50:17+00:00

This is a preparation patch for moving page->_last_nid into page->flags
that moves page flag layout information to a separate header.  This
patch is necessary because otherwise there would be a circular
dependency between mm_types.h and mm.h.

Signed-off-by: Mel Gorman 
Cc: Peter Zijlstra 
Cc: Andrea Arcangeli 
Cc: Ingo Molnar 
Cc: Simon Jeons 
Cc: Wanpeng Li 
Cc: Hugh Dickins 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds

mm: numa: fix minor typo in numa_next_scan

2013-02-24T01:50:16+00:00

s/me/be/ and clarify the comment a bit when we're changing it anyway.

Signed-off-by: Mel Gorman 
Suggested-by: Simon Jeons 
Cc: Peter Zijlstra 
Cc: Andrea Arcangeli 
Cc: Ingo Molnar 
Cc: Wanpeng Li 
Cc: Hugh Dickins 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds