mm: introduce vm_ops->map_pages()
authorKirill A. Shutemov <kirill.shutemov@linux.intel.com>
Mon, 7 Apr 2014 22:37:18 +0000 (15:37 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Mon, 7 Apr 2014 23:35:52 +0000 (16:35 -0700)
commit8c6e50b0290c4c708a3e6462729e1e9151a9a7df
treebc858cbfab0be322dc802a41dd9e2c50270871a0
parent179e09637c1bb17c0b6fb347353c7c52e6931fa5
mm: introduce vm_ops->map_pages()

Here's new version of faultaround patchset.  It took a while to tune it
and collect performance data.

First patch adds new callback ->map_pages to vm_operations_struct.

->map_pages() is called when VM asks to map easy accessible pages.
Filesystem should find and map pages associated with offsets from
"pgoff" till "max_pgoff".  ->map_pages() is called with page table
locked and must not block.  If it's not possible to reach a page without
blocking, filesystem should skip it.  Filesystem should use do_set_pte()
to setup page table entry.  Pointer to entry associated with offset
"pgoff" is passed in "pte" field in vm_fault structure.  Pointers to
entries for other offsets should be calculated relative to "pte".

Currently VM use ->map_pages only on read page fault path.  We try to
map FAULT_AROUND_PAGES a time.  FAULT_AROUND_PAGES is 16 for now.
Performance data for different FAULT_AROUND_ORDER is below.

TODO:
 - implement ->map_pages() for shmem/tmpfs;
 - modify get_user_pages() to be able to use ->map_pages() and implement
   mmap(MAP_POPULATE|MAP_NONBLOCK) on top.

=========================================================================
Tested on 4-socket machine (120 threads) with 128GiB of RAM.

Few real-world workloads. The sweet spot for FAULT_AROUND_ORDER here is
somewhere between 3 and 5. Let's say 4 :)

Linux build (make -j60)
FAULT_AROUND_ORDER Baseline 1 3 4 5 7 9
minor-faults 283,301,572 247,151,987 212,215,789 204,772,882 199,568,944 194,703,779 193,381,485
time, seconds 151.227629483 153.920996480 151.356125472 150.863792049 150.879207877 151.150764954 151.450962358
Linux rebuild (make -j60)
FAULT_AROUND_ORDER Baseline 1 3 4 5 7 9
minor-faults 5,396,854 4,148,444 2,855,286 2,577,282 2,361,957 2,169,573 2,112,643
time, seconds 27.404543757 27.559725591 27.030057426 26.855045126 26.678618635 26.974523490 26.761320095
Git test suite (make -j60 test)
FAULT_AROUND_ORDER Baseline 1 3 4 5 7 9
minor-faults 129,591,823 99,200,751 66,106,718 57,606,410 51,510,808 45,776,813 44,085,515
time, seconds 66.087215026 64.784546905 64.401156567 65.282708668 66.034016829 66.793780811 67.237810413

Two synthetic tests: access every word in file in sequential/random order.
It doesn't improve much after FAULT_AROUND_ORDER == 4.

Sequential access 16GiB file
FAULT_AROUND_ORDER Baseline 1 3 4 5 7 9
 1 thread
minor-faults 4,195,437 2,098,275 525,068 262,251 131,170 32,856 8,282
time, seconds 7.250461742 6.461711074 5.493859139 5.488488147 5.707213983 5.898510832 5.109232856
 8 threads
minor-faults 33,557,540 16,892,728 4,515,848 2,366,999 1,423,382 442,732 142,339
time, seconds 16.649304881 9.312555263 6.612490639 6.394316732 6.669827501 6.75078944 6.371900528
 32 threads
minor-faults 134,228,222 67,526,810 17,725,386 9,716,537 4,763,731 1,668,921 537,200
time, seconds 49.164430543 29.712060103 12.938649729 10.175151004 11.840094583 9.594081325 9.928461797
 60 threads
minor-faults 251,687,988 126,146,952 32,919,406 18,208,804 10,458,947 2,733,907 928,217
time, seconds 86.260656897 49.626551828 22.335007632 17.608243696 16.523119035 16.339489186 16.326390902
 120 threads
minor-faults 503,352,863 252,939,677 67,039,168 35,191,827 19,170,091 4,688,357 1,471,862
time, seconds 124.589206333 79.757867787 39.508707872 32.167281632 29.972989292 28.729834575 28.042251622
Random access 1GiB file
 1 thread
minor-faults 262,636 132,743 34,369 17,299 8,527 3,451 1,222
time, seconds 15.351890914 16.613802482 16.569227308 15.179220992 16.557356122 16.578247824 15.365266994
 8 threads
minor-faults 2,098,948 1,061,871 273,690 154,501 87,110 25,663 7,384
time, seconds 15.040026343 15.096933500 14.474757288 14.289129964 14.411537468 14.296316837 14.395635804
 32 threads
minor-faults 8,390,734 4,231,023 1,054,432 528,847 269,242 97,746 26,881
time, seconds 20.430433109 21.585235358 22.115062928 14.872878951 14.880856305 14.883370649 14.821261690
 60 threads
minor-faults 15,733,258 7,892,809 1,973,393 988,266 594,789 164,994 51,691
time, seconds 26.577302548 25.692397770 18.728863715 20.153026398 21.619101933 17.745086260 17.613215273
 120 threads
minor-faults 31,471,111 15,816,616 3,959,209 1,978,685 1,008,299 264,635 96,010
time, seconds 41.835322703 40.459786095 36.085306105 35.313894834 35.814445675 36.552633793 34.289210594

Touch only one page in page table in 16GiB file
FAULT_AROUND_ORDER Baseline 1 3 4 5 7 9
 1 thread
minor-faults 8,372 8,324 8,270 8,260 8,249 8,239 8,237
time, seconds 0.039892712 0.045369149 0.051846126 0.063681685 0.079095975 0.17652406 0.541213386
 8 threads
minor-faults 65,731 65,681 65,628 65,620 65,608 65,599 65,596
time, seconds 0.124159196 0.488600638 0.156854426 0.191901957 0.242631486 0.543569456 1.677303984
 32 threads
minor-faults 262,388 262,341 262,285 262,276 262,266 262,257 263,183
time, seconds 0.452421421 0.488600638 0.565020946 0.648229739 0.789850823 1.651584361 5.000361559
 60 threads
minor-faults 491,822 491,792 491,723 491,711 491,701 491,691 491,825
time, seconds 0.763288616 0.869620515 0.980727360 1.161732354 1.466915814 3.04041448 9.308612938
 120 threads
minor-faults 983,466 983,655 983,366 983,372 983,363 984,083 984,164
time, seconds 1.595846553 1.667902182 2.008959376 2.425380942 2.941368804 5.977807890 18.401846125

This patch (of 2):

Introduce new vm_ops callback ->map_pages() and uses it for mapping easy
accessible pages around fault address.

On read page fault, if filesystem provides ->map_pages(), we try to map up
to FAULT_AROUND_PAGES pages around page fault address in hope to reduce
number of minor page faults.

We call ->map_pages first and use ->fault() as fallback if page by the
offset is not ready to be mapped (cold page cache or something).

Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Acked-by: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Rik van Riel <riel@redhat.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Matthew Wilcox <matthew.r.wilcox@intel.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Ning Qu <quning@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Documentation/filesystems/Locking
include/linux/mm.h
mm/memory.c