linux-stable.git/include/linux/mm_types.h, branch v3.14.78

mm: per-thread vma caching

2014-10-09T19:21:29+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: Greg Kroah-Hartman

mm: do not allocate page->ptl dynamically, if spinlock_t fits to long

2013-12-20T20:25:45+00:00

In struct page we have enough space to fit long-size page->ptl there,
but we use dynamically-allocated page->ptl if size(spinlock_t) is larger
than sizeof(int).

It hurts 64-bit architectures with CONFIG_GENERIC_LOCKBREAK, where
sizeof(spinlock_t) == 8, but it easily fits into struct page.

Signed-off-by: Kirill A. Shutemov 
Acked-by: Hugh Dickins 
Signed-off-by: Linus Torvalds

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

2013-12-19T03:04:51+00:00

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 
Cc: 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds

mm: fix TLB flush race between migration, and change_protection_range

2013-12-19T03:04:51+00:00

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 
Cc: 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds

Merge branch 'slab/next' of git://git.kernel.org/pub/scm/linux/kernel/git/penberg/linux

2013-11-22T16:10:34+00:00

Pull SLAB changes from Pekka Enberg:
 "The patches from Joonsoo Kim switch mm/slab.c to use 'struct page' for
  slab internals similar to mm/slub.c.  This reduces memory usage and
  improves performance:

    https://lkml.org/lkml/2013/10/16/155

  Rest of the changes are bug fixes from various people"

* 'slab/next' of git://git.kernel.org/pub/scm/linux/kernel/git/penberg/linux: (21 commits)
  mm, slub: fix the typo in mm/slub.c
  mm, slub: fix the typo in include/linux/slub_def.h
  slub: Handle NULL parameter in kmem_cache_flags
  slab: replace non-existing 'struct freelist *' with 'void *'
  slab: fix to calm down kmemleak warning
  slub: proper kmemleak tracking if CONFIG_SLUB_DEBUG disabled
  slab: rename slab_bufctl to slab_freelist
  slab: remove useless statement for checking pfmemalloc
  slab: use struct page for slab management
  slab: replace free and inuse in struct slab with newly introduced active
  slab: remove SLAB_LIMIT
  slab: remove kmem_bufctl_t
  slab: change the management method of free objects of the slab
  slab: use __GFP_COMP flag for allocating slab pages
  slab: use well-defined macro, virt_to_slab()
  slab: overloading the RCU head over the LRU for RCU free
  slab: remove cachep in struct slab_rcu
  slab: remove nodeid in struct slab
  slab: remove colouroff in struct slab
  slab: change return type of kmem_getpages() to struct page
  ...

mm: place page->pmd_huge_pte to right union

2013-11-22T00:42:28+00:00

I don't know what went wrong, mis-merge or something, but ->pmd_huge_pte
placed in wrong union within struct page.

In original patch[1] it's placed to union with ->lru and ->slab, but in
commit e009bb30c8df ("mm: implement split page table lock for PMD
level") it's in union with ->index and ->freelist.

That union seems also unused for pages with table tables and safe to
re-use, but it's not what I've tested.

Let's move it to original place.  It fixes indentation at least.  :)

[1] https://lkml.org/lkml/2013/10/7/288

Signed-off-by: Kirill A. Shutemov 
Reviewed-by: Naoya Horiguchi 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds

mm: properly separate the bloated ptl from the regular case

2013-11-15T00:32:20+00:00

Use kernel/bounds.c to convert build-time spinlock_t size check into a
preprocessor symbol and apply that to properly separate the page::ptl
situation.

Signed-off-by: Peter Zijlstra 
Signed-off-by: Kirill A. Shutemov 
Cc: Ingo Molnar 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds

mm: dynamically allocate page->ptl if it cannot be embedded to struct page

2013-11-15T00:32:20+00:00

If split page table lock is in use, we embed the lock into struct page
of table's page.  We have to disable split lock, if spinlock_t is too
big be to be embedded, like when DEBUG_SPINLOCK or DEBUG_LOCK_ALLOC
enabled.

This patch add support for dynamic allocation of split page table lock
if we can't embed it to struct page.

page->ptl is unsigned long now and we use it as spinlock_t if
sizeof(spinlock_t) <= sizeof(long), otherwise it's pointer to spinlock_t.

The spinlock_t allocated in pgtable_page_ctor() for PTE table and in
pgtable_pmd_page_ctor() for PMD table.  All other helpers converted to
support dynamically allocated page->ptl.

Signed-off-by: Kirill A. Shutemov 
Reviewed-by: Peter Zijlstra 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds

mm: implement split page table lock for PMD level

2013-11-15T00:32:15+00:00

The basic idea is the same as with PTE level: the lock is embedded into
struct page of table's page.

We can't use mm->pmd_huge_pte to store pgtables for THP, since we don't
take mm->page_table_lock anymore.  Let's reuse page->lru of table's page
for that.

pgtable_pmd_page_ctor() returns true, if initialization is successful
and false otherwise.  Current implementation never fails, but assumption
that constructor can fail will help to port it to -rt where spinlock_t
is rather huge and cannot be embedded into struct page -- dynamic
allocation is required.

Signed-off-by: Naoya Horiguchi 
Signed-off-by: Kirill A. Shutemov 
Tested-by: Alex Thorlton 
Cc: Ingo Molnar 
Cc: "Eric W . Biederman" 
Cc: "Paul E . McKenney" 
Cc: Al Viro 
Cc: Andi Kleen 
Cc: Andrea Arcangeli 
Cc: Dave Hansen 
Cc: Dave Jones 
Cc: David Howells 
Cc: Frederic Weisbecker 
Cc: Johannes Weiner 
Cc: Kees Cook 
Cc: Mel Gorman 
Cc: Michael Kerrisk 
Cc: Oleg Nesterov 
Cc: Peter Zijlstra 
Cc: Rik van Riel 
Cc: Robin Holt 
Cc: Sedat Dilek 
Cc: Srikar Dronamraju 
Cc: Thomas Gleixner 
Cc: Hugh Dickins 
Reviewed-by: Steven Rostedt 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds

mm: convert mm->nr_ptes to atomic_long_t

2013-11-15T00:32:14+00:00

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: Kirill A. Shutemov 
Tested-by: Alex Thorlton 
Cc: Ingo Molnar 
Cc: Naoya Horiguchi 
Cc: "Eric W . Biederman" 
Cc: "Paul E . McKenney" 
Cc: Al Viro 
Cc: Andi Kleen 
Cc: Andrea Arcangeli 
Cc: Dave Hansen 
Cc: Dave Jones 
Cc: David Howells 
Cc: Frederic Weisbecker 
Cc: Johannes Weiner 
Cc: Kees Cook 
Cc: Mel Gorman 
Cc: Michael Kerrisk 
Cc: Oleg Nesterov 
Cc: Peter Zijlstra 
Cc: Rik van Riel 
Cc: Robin Holt 
Cc: Sedat Dilek 
Cc: Srikar Dronamraju 
Cc: Thomas Gleixner 
Cc: Hugh Dickins 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds