tree 4cf646dbc3024f7747f43d96af16f739fc89620a
parent 455c6fdbd219161bd09b1165f11699d6d73de11c
author Minchan Kim <minchan@kernel.org> 1396414749 +0900
committer Minchan Kim <minchan@kernel.org> 1397433644 +0900

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).

How to work is following as.

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.

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):                4
On-line CPU(s) list:   0-3
Thread(s) per core:    2
Core(s) per socket:    2
Socket(s):             1
NUMA node(s):          1
Vendor ID:             GenuineIntel
CPU family:            6
Model:                 42
Stepping:              7
CPU MHz:               2801.000
BogoMIPS:              5581.64
Virtualization:        VT-x
L1d cache:             32K
L1i cache:             32K
L2 cache:              256K
L3 cache:              4096K
NUMA node0 CPU(s):     0-3

ebizzy benchmark(./ebizzy -S 10 -n 512)

 vanilla-jemalloc		MADV_free-jemalloc

1 thread
records:  10              records:  10
avg:      7436.70         avg:      15292.70
std:      48.01(0.65%)    std:      496.40(3.25%)
max:      7542.00         max:      15944.00
min:      7366.00         min:      14478.00

2 thread
records:  10              records:  10
avg:      12190.50        avg:      24975.50
std:      1011.51(8.30%)  std:      1127.22(4.51%)
max:      13012.00        max:      26382.00
min:      10192.00        min:      23265.00

4 thread
records:  10              records:  10
avg:      16875.30        avg:      36320.90
std:      562.59(3.33%)   std:      1503.75(4.14%)
max:      17465.00        max:      38314.00
min:      15552.00        min:      33863.00

8 thread
records:  10              records:  10
avg:      16966.80        avg:      35915.20
std:      229.35(1.35%)   std:      2153.89(6.00%)
max:      17456.00        max:      37943.00
min:      16742.00        min:      29891.00

16 thread
records:  10              records:  10
avg:      20590.90        avg:      37388.40
std:      362.33(1.76%)   std:      1282.59(3.43%)
max:      20954.00        max:      38911.00
min:      19985.00        min:      34928.00

32 thread
records:  10              records:  10
avg:      22633.40        avg:      37118.00
std:      413.73(1.83%)   std:      766.36(2.06%)
max:      23120.00        max:      38328.00
min:      22071.00        min:      35557.00

In summary, MADV_FREE is about 2 time faster than MADV_DONTNEED.

Patchset is based on 3.14
* From v3
 * Add "how to work part" in description - Zhang
 * Add page_discardable utility function - Zhang
 * Clean up

* From v2
 * Remove forceful dirty marking of swap-readed page - Johannes
 * Remove deactivation logic of lazyfreed page
 * Rebased on 3.14
 * Remove RFC tag

* From v1
 * Use custom page table walker for madvise_free - Johannes
 * Remove PG_lazypage flag - Johannes
 * Do madvise_dontneed instead of madvise_freein swapless system

Cc: Hugh Dickins <hughd@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Rik van Riel <riel@redhat.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Jason Evans <je@fb.com>
Signed-off-by: Minchan Kim <minchan@kernel.org>