kernel_optimize_test/mm
Minchan Kim 854e9ed09d mm: support madvise(MADV_FREE)
Linux doesn't have an ability to free pages lazy while other OS already
have been supported that named by madvise(MADV_FREE).

The gain is clear that kernel can discard freed pages rather than
swapping out or OOM if memory pressure happens.

Without memory pressure, freed pages would be reused by userspace
without another additional overhead(ex, page fault + allocation +
zeroing).

Jason Evans said:

: Facebook has been using MAP_UNINITIALIZED
: (https://lkml.org/lkml/2012/1/18/308) in some of its applications for
: several years, but there are operational costs to maintaining this
: out-of-tree in our kernel and in jemalloc, and we are anxious to retire it
: in favor of MADV_FREE.  When we first enabled MAP_UNINITIALIZED it
: increased throughput for much of our workload by ~5%, and although the
: benefit has decreased using newer hardware and kernels, there is still
: enough benefit that we cannot reasonably retire it without a replacement.
:
: Aside from Facebook operations, there are numerous broadly used
: applications that would benefit from MADV_FREE.  The ones that immediately
: come to mind are redis, varnish, and MariaDB.  I don't have much insight
: into Android internals and development process, but I would hope to see
: MADV_FREE support eventually end up there as well to benefit applications
: linked with the integrated jemalloc.
:
: jemalloc will use MADV_FREE once it becomes available in the Linux kernel.
: In fact, jemalloc already uses MADV_FREE or equivalent everywhere it's
: available: *BSD, OS X, Windows, and Solaris -- every platform except Linux
: (and AIX, but I'm not sure it even compiles on AIX).  The lack of
: MADV_FREE on Linux forced me down a long series of increasingly
: sophisticated heuristics for madvise() volume reduction, and even so this
: remains a common performance issue for people using jemalloc on Linux.
: Please integrate MADV_FREE; many people will benefit substantially.

How it works:

When madvise syscall is called, VM clears dirty bit of ptes of the
range.  If memory pressure happens, VM checks dirty bit of page table
and if it found still "clean", it means it's a "lazyfree pages" so VM
could discard the page instead of swapping out.  Once there was store
operation for the page before VM peek a page to reclaim, dirty bit is
set so VM can swap out the page instead of discarding.

One thing we should notice is that basically, MADV_FREE relies on dirty
bit in page table entry to decide whether VM allows to discard the page
or not.  IOW, if page table entry includes marked dirty bit, VM
shouldn't discard the page.

However, as a example, if swap-in by read fault happens, page table
entry doesn't have dirty bit so MADV_FREE could discard the page
wrongly.

For avoiding the problem, MADV_FREE did more checks with PageDirty and
PageSwapCache.  It worked out because swapped-in page lives on swap
cache and since it is evicted from the swap cache, the page has PG_dirty
flag.  So both page flags check effectively prevent wrong discarding by
MADV_FREE.

However, a problem in above logic is that swapped-in page has PG_dirty
still after they are removed from swap cache so VM cannot consider the
page as freeable any more even if madvise_free is called in future.

Look at below example for detail.

    ptr = malloc();
    memset(ptr);
    ..
    ..
    .. heavy memory pressure so all of pages are swapped out
    ..
    ..
    var = *ptr; -> a page swapped-in and could be removed from
                   swapcache. Then, page table doesn't mark
                   dirty bit and page descriptor includes PG_dirty
    ..
    ..
    madvise_free(ptr); -> It doesn't clear PG_dirty of the page.
    ..
    ..
    ..
    .. heavy memory pressure again.
    .. In this time, VM cannot discard the page because the page
    .. has *PG_dirty*

To solve the problem, this patch clears PG_dirty if only the page is
owned exclusively by current process when madvise is called because
PG_dirty represents ptes's dirtiness in several processes so we could
clear it only if we own it exclusively.

Firstly, heavy users would be general allocators(ex, jemalloc, tcmalloc
and hope glibc supports it) and jemalloc/tcmalloc already have supported
the feature for other OS(ex, FreeBSD)

  barrios@blaptop:~/benchmark/ebizzy$ lscpu
  Architecture:          x86_64
  CPU op-mode(s):        32-bit, 64-bit
  Byte Order:            Little Endian
  CPU(s):                12
  On-line CPU(s) list:   0-11
  Thread(s) per core:    1
  Core(s) per socket:    1
  Socket(s):             12
  NUMA node(s):          1
  Vendor ID:             GenuineIntel
  CPU family:            6
  Model:                 2
  Stepping:              3
  CPU MHz:               3200.185
  BogoMIPS:              6400.53
  Virtualization:        VT-x
  Hypervisor vendor:     KVM
  Virtualization type:   full
  L1d cache:             32K
  L1i cache:             32K
  L2 cache:              4096K
  NUMA node0 CPU(s):     0-11
  ebizzy benchmark(./ebizzy -S 10 -n 512)

  Higher avg is better.

   vanilla-jemalloc             MADV_free-jemalloc

  1 thread
  records: 10                   records: 10
  avg:   2961.90                avg:  12069.70
  std:     71.96(2.43%)         std:    186.68(1.55%)
  max:   3070.00                max:  12385.00
  min:   2796.00                min:  11746.00

  2 thread
  records: 10                   records: 10
  avg:   5020.00                avg:  17827.00
  std:    264.87(5.28%)         std:    358.52(2.01%)
  max:   5244.00                max:  18760.00
  min:   4251.00                min:  17382.00

  4 thread
  records: 10                   records: 10
  avg:   8988.80                avg:  27930.80
  std:   1175.33(13.08%)        std:   3317.33(11.88%)
  max:   9508.00                max:  30879.00
  min:   5477.00                min:  21024.00

  8 thread
  records: 10                   records: 10
  avg:  13036.50                avg:  33739.40
  std:    170.67(1.31%)         std:   5146.22(15.25%)
  max:  13371.00                max:  40572.00
  min:  12785.00                min:  24088.00

  16 thread
  records: 10                   records: 10
  avg:  11092.40                avg:  31424.20
  std:    710.60(6.41%)         std:   3763.89(11.98%)
  max:  12446.00                max:  36635.00
  min:   9949.00                min:  25669.00

  32 thread
  records: 10                   records: 10
  avg:  11067.00                avg:  34495.80
  std:    971.06(8.77%)         std:   2721.36(7.89%)
  max:  12010.00                max:  38598.00
  min:   9002.00                min:  30636.00

In summary, MADV_FREE is about much faster than MADV_DONTNEED.

This patch (of 12):

Add core MADV_FREE implementation.

[akpm@linux-foundation.org: small cleanups]
Signed-off-by: Minchan Kim <minchan@kernel.org>
Acked-by: Michal Hocko <mhocko@suse.com>
Acked-by: Hugh Dickins <hughd@google.com>
Cc: Mika Penttil <mika.penttila@nextfour.com>
Cc: Michael Kerrisk <mtk.manpages@gmail.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Rik van Riel <riel@redhat.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Jason Evans <je@fb.com>
Cc: Daniel Micay <danielmicay@gmail.com>
Cc: "Kirill A. Shutemov" <kirill@shutemov.name>
Cc: Shaohua Li <shli@kernel.org>
Cc: <yalin.wang2010@gmail.com>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: "James E.J. Bottomley" <jejb@parisc-linux.org>
Cc: "Kirill A. Shutemov" <kirill@shutemov.name>
Cc: "Shaohua Li" <shli@kernel.org>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Chen Gang <gang.chen.5i5j@gmail.com>
Cc: Chris Zankel <chris@zankel.net>
Cc: Darrick J. Wong <darrick.wong@oracle.com>
Cc: David S. Miller <davem@davemloft.net>
Cc: Helge Deller <deller@gmx.de>
Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru>
Cc: Matt Turner <mattst88@gmail.com>
Cc: Max Filippov <jcmvbkbc@gmail.com>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: Richard Henderson <rth@twiddle.net>
Cc: Roland Dreier <roland@kernel.org>
Cc: Russell King <rmk@arm.linux.org.uk>
Cc: Shaohua Li <shli@kernel.org>
Cc: Will Deacon <will.deacon@arm.com>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-01-15 17:56:32 -08:00
..
kasan kasan: fix kmemleak false-positive in kasan_module_alloc() 2015-11-20 16:17:32 -08:00
backing-dev.c mm: memcontrol: export root_mem_cgroup 2016-01-14 16:00:49 -08:00
balloon_compaction.c mm: page migration trylock newpage at same level as oldpage 2015-11-05 19:34:48 -08:00
bootmem.c x86/mm: Introduce max_possible_pfn 2015-12-06 12:46:31 +01:00
cleancache.c cleancache: remove limit on the number of cleancache enabled filesystems 2015-04-14 16:49:03 -07:00
cma_debug.c mm/cma_debug: correct size input to bitmap function 2015-07-17 16:39:54 -07:00
cma.c mm/cma.c: suppress warning 2015-11-05 19:34:48 -08:00
cma.h mm: cma: mark cma_bitmap_maxno() inline in header 2015-08-14 15:56:32 -07:00
compaction.c mm/compaction.c: __compact_pgdat() code cleanuup 2016-01-14 16:00:49 -08:00
debug-pagealloc.c mm/debug-pagealloc: make debug-pagealloc boottime configurable 2014-12-13 12:42:48 -08:00
debug.c mm: rework mapcount accounting to enable 4k mapping of THPs 2016-01-15 17:56:32 -08:00
dmapool.c mm, page_alloc: distinguish between being unable to sleep, unwilling to sleep and avoiding waking kswapd 2015-11-06 17:50:42 -08:00
early_ioremap.c mm/early_ioremap: use offset_in_page macro 2015-11-05 19:34:48 -08:00
fadvise.c writeback: implement and use inode_congested() 2015-06-02 08:33:35 -06:00
failslab.c mm, page_alloc: rename __GFP_WAIT to __GFP_RECLAIM 2015-11-06 17:50:42 -08:00
filemap.c mm: differentiate page_mapped() from page_mapcount() for compound pages 2016-01-15 17:56:32 -08:00
frame_vector.c mm: fix docbook comment for get_vaddr_frames() 2015-11-05 19:34:48 -08:00
frontswap.c frontswap: allow multiple backends 2015-06-24 17:49:45 -07:00
gup.c thp: allow mlocked THP again 2016-01-15 17:56:32 -08:00
highmem.c
huge_memory.c thp: increase split_huge_page() success rate 2016-01-15 17:56:32 -08:00
hugetlb_cgroup.c mm: make compound_head() robust 2015-11-06 17:50:42 -08:00
hugetlb.c mm: rework mapcount accounting to enable 4k mapping of THPs 2016-01-15 17:56:32 -08:00
hwpoison-inject.c hwpoison: use page_cgroup_ino for filtering by memcg 2015-09-10 13:29:01 -07:00
init-mm.c
internal.h thp: reintroduce split_huge_page() 2016-01-15 17:56:32 -08:00
interval_tree.c mm: replace vma->sharead.linear with vma->shared 2015-02-10 14:30:31 -08:00
Kconfig mm: re-enable THP 2016-01-15 17:56:32 -08:00
Kconfig.debug mm/debug_pagealloc: remove obsolete Kconfig options 2015-01-08 15:10:52 -08:00
kmemcheck.c mm/slab_common: move kmem_cache definition to internal header 2014-10-09 22:25:50 -04:00
kmemleak-test.c
kmemleak.c Revert "gfp: add __GFP_NOACCOUNT" 2016-01-14 16:00:49 -08:00
ksm.c ksm: prepare to new THP semantics 2016-01-15 17:56:32 -08:00
list_lru.c memcg: simplify and inline __mem_cgroup_from_kmem 2015-11-05 19:34:48 -08:00
maccess.c mm/maccess.c: actually return -EFAULT from strncpy_from_unsafe 2015-11-05 19:34:48 -08:00
madvise.c mm: support madvise(MADV_FREE) 2016-01-15 17:56:32 -08:00
Makefile media updates for v4.3-rc1 2015-09-11 16:42:39 -07:00
memblock.c mm/memblock: introduce for_each_memblock_type() 2016-01-14 16:00:49 -08:00
memcontrol.c mm: rework mapcount accounting to enable 4k mapping of THPs 2016-01-15 17:56:32 -08:00
memory_hotplug.c memory-hotplug: don't BUG() in register_memory_resource() 2016-01-14 16:00:49 -08:00
memory-failure.c mm: hwpoison: adjust for new thp refcounting 2016-01-15 17:56:32 -08:00
memory.c thp: allow mlocked THP again 2016-01-15 17:56:32 -08:00
mempolicy.c migrate_pages: try to split pages on queuing 2016-01-15 17:56:32 -08:00
mempool.c mm, page_alloc: distinguish between being unable to sleep, unwilling to sleep and avoiding waking kswapd 2015-11-06 17:50:42 -08:00
memtest.c memtest: remove unused header files 2015-09-08 15:35:28 -07:00
migrate.c thp: introduce deferred_split_huge_page() 2016-01-15 17:56:32 -08:00
mincore.c mm, thp: remove infrastructure for handling splitting PMDs 2016-01-15 17:56:32 -08:00
mlock.c thp: allow mlocked THP again 2016-01-15 17:56:32 -08:00
mm_init.c mm: meminit: remove mminit_verify_page_links 2015-06-30 19:44:56 -07:00
mmap.c mm: rework virtual memory accounting 2016-01-14 16:00:49 -08:00
mmu_context.c
mmu_notifier.c mmu-notifier: add clear_young callback 2015-09-10 13:29:01 -07:00
mmzone.c mm/mmzone.c: memmap_valid_within() can be boolean 2016-01-14 16:00:49 -08:00
mprotect.c thp: rename split_huge_page_pmd() to split_huge_pmd() 2016-01-15 17:56:32 -08:00
mremap.c mm, thp: remove infrastructure for handling splitting PMDs 2016-01-15 17:56:32 -08:00
msync.c mm/msync: use offset_in_page macro 2015-11-05 19:34:48 -08:00
nobootmem.c x86/mm: Introduce max_possible_pfn 2015-12-06 12:46:31 +01:00
nommu.c kmemcg: account certain kmem allocations to memcg 2016-01-14 16:00:49 -08:00
oom_kill.c mm, shmem: add internal shmem resident memory accounting 2016-01-14 16:00:49 -08:00
page_alloc.c thp: introduce deferred_split_huge_page() 2016-01-15 17:56:32 -08:00
page_counter.c mm: page_counter: let page_counter_try_charge() return bool 2015-11-05 19:34:48 -08:00
page_ext.c mm: introduce idle page tracking 2015-09-10 13:29:01 -07:00
page_idle.c mm: add page_check_address_transhuge() helper 2016-01-15 17:56:32 -08:00
page_io.c fs: use helper bio_add_page() instead of open coding on bi_io_vec 2015-08-13 12:32:00 -06:00
page_isolation.c mm/page_isolation: use macro to judge the alignment 2016-01-14 16:00:49 -08:00
page_owner.c mm/page_owner: set correct gfp_mask on page_owner 2015-07-17 16:39:54 -07:00
page-writeback.c mm: page_alloc: generalize the dirty balance reserve 2016-01-14 16:00:49 -08:00
pagewalk.c thp: rename split_huge_page_pmd() to split_huge_pmd() 2016-01-15 17:56:32 -08:00
percpu-km.c percpu: implmeent pcpu_nr_empty_pop_pages and chunk->nr_populated 2014-09-02 14:46:05 -04:00
percpu-vm.c percpu: move region iterations out of pcpu_[de]populate_chunk() 2014-09-02 14:46:02 -04:00
percpu.c mm/percpu: use offset_in_page macro 2015-11-05 19:34:48 -08:00
pgtable-generic.c mm, thp: remove infrastructure for handling splitting PMDs 2016-01-15 17:56:32 -08:00
process_vm_access.c process_vm_access: switch to {compat_,}import_iovec() 2015-04-11 22:27:12 -04:00
quicklist.c
readahead.c mm: move lru_to_page to mm_inline.h 2016-01-14 16:00:49 -08:00
rmap.c mm: support madvise(MADV_FREE) 2016-01-15 17:56:32 -08:00
shmem.c memcg: adjust to support new THP refcounting 2016-01-15 17:56:32 -08:00
slab_common.c slab: add SLAB_ACCOUNT flag 2016-01-14 16:00:49 -08:00
slab.c mm/slab.c: add a helper function get_first_slab 2016-01-14 16:00:49 -08:00
slab.h slab: add SLAB_ACCOUNT flag 2016-01-14 16:00:49 -08:00
slob.c slab/slub: adjust kmem_cache_alloc_bulk API 2015-11-22 11:58:44 -08:00
slub.c page-flags: define PG_locked behavior on compound pages 2016-01-15 17:56:32 -08:00
sparse-vmemmap.c
sparse.c
swap_cgroup.c mm: page_cgroup: rename file to mm/swap_cgroup.c 2014-12-10 17:41:09 -08:00
swap_state.c mm: support madvise(MADV_FREE) 2016-01-15 17:56:32 -08:00
swap.c thp: allow mlocked THP again 2016-01-15 17:56:32 -08:00
swapfile.c mm, thp: adjust conditions when we can reuse the page on WP fault 2016-01-15 17:56:32 -08:00
truncate.c memcg: add per cgroup dirty page accounting 2015-06-02 08:33:33 -06:00
userfaultfd.c memcg: adjust to support new THP refcounting 2016-01-15 17:56:32 -08:00
util.c mm: prepare page_referenced() and page_idle to new THP refcounting 2016-01-15 17:56:32 -08:00
vmacache.c mm/vmacache: inline vmacache_valid_mm() 2015-11-05 19:34:48 -08:00
vmalloc.c mm, vmalloc: remove VM_VPAGES 2016-01-14 16:00:49 -08:00
vmpressure.c memcg: avoid vmpressure oops when memcg disabled 2016-01-14 16:00:49 -08:00
vmscan.c mm: support madvise(MADV_FREE) 2016-01-15 17:56:32 -08:00
vmstat.c mm: support madvise(MADV_FREE) 2016-01-15 17:56:32 -08:00
workingset.c list_lru: add helpers to isolate items 2015-02-12 18:54:10 -08:00
zbud.c mm/zbud.c: use list_last_entry() instead of list_tail_entry() 2016-01-15 11:40:52 -08:00
zpool.c mm: zsmalloc: constify struct zs_pool name 2015-11-06 17:50:42 -08:00
zsmalloc.c zsmalloc: reorganize struct size_class to pack 4 bytes hole 2016-01-15 11:40:52 -08:00
zswap.c mm/zswap: change incorrect strncmp use to strcmp 2015-12-18 14:25:40 -08:00