Age | Commit message (Collapse) | Author |
|
It's possible a zone watermark is ok when entering the balance_pgdat()
loop, while the zone is within the requested classzone_idx. Count pages
from this zone into `balanced'. In this way, we can skip shrinking zones
too much for high order allocation.
Signed-off-by: Shaohua Li <shaohua.li@intel.com>
Acked-by: Mel Gorman <mgorman@suse.de>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|
|
When direct reclaim encounters a dirty page, it gets recycled around the
LRU for another cycle. This patch marks the page PageReclaim similar to
deactivate_page() so that the page gets reclaimed almost immediately after
the page gets cleaned. This is to avoid reclaiming clean pages that are
younger than a dirty page encountered at the end of the LRU that might
have been something like a use-once page.
Signed-off-by: Mel Gorman <mgorman@suse.de>
Acked-by: Johannes Weiner <jweiner@redhat.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Cc: Jan Kara <jack@suse.cz>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Alex Elder <aelder@sgi.com>
Cc: Theodore Ts'o <tytso@mit.edu>
Cc: Chris Mason <chris.mason@oracle.com>
Cc: Dave Hansen <dave@linux.vnet.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|
|
writeback
Workloads that are allocating frequently and writing files place a large
number of dirty pages on the LRU. With use-once logic, it is possible for
them to reach the end of the LRU quickly requiring the reclaimer to scan
more to find clean pages. Ordinarily, processes that are dirtying memory
will get throttled by dirty balancing but this is a global heuristic and
does not take into account that LRUs are maintained on a per-zone basis.
This can lead to a situation whereby reclaim is scanning heavily, skipping
over a large number of pages under writeback and recycling them around the
LRU consuming CPU.
This patch checks how many of the number of pages isolated from the LRU
were dirty and under writeback. If a percentage of them under writeback,
the process will be throttled if a backing device or the zone is
congested. Note that this applies whether it is anonymous or file-backed
pages that are under writeback meaning that swapping is potentially
throttled. This is intentional due to the fact if the swap device is
congested, scanning more pages and dispatching more IO is not going to
help matters.
The percentage that must be in writeback depends on the priority. At
default priority, all of them must be dirty. At DEF_PRIORITY-1, 50% of
them must be, DEF_PRIORITY-2, 25% etc. i.e. as pressure increases the
greater the likelihood the process will get throttled to allow the flusher
threads to make some progress.
Signed-off-by: Mel Gorman <mgorman@suse.de>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Acked-by: Johannes Weiner <jweiner@redhat.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Cc: Jan Kara <jack@suse.cz>
Cc: Rik van Riel <riel@redhat.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Alex Elder <aelder@sgi.com>
Cc: Theodore Ts'o <tytso@mit.edu>
Cc: Chris Mason <chris.mason@oracle.com>
Cc: Dave Hansen <dave@linux.vnet.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|
|
It is preferable that no dirty pages are dispatched for cleaning from the
page reclaim path. At normal priorities, this patch prevents kswapd
writing pages.
However, page reclaim does have a requirement that pages be freed in a
particular zone. If it is failing to make sufficient progress (reclaiming
< SWAP_CLUSTER_MAX at any priority priority), the priority is raised to
scan more pages. A priority of DEF_PRIORITY - 3 is considered to be the
point where kswapd is getting into trouble reclaiming pages. If this
priority is reached, kswapd will dispatch pages for writing.
Signed-off-by: Mel Gorman <mgorman@suse.de>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Johannes Weiner <jweiner@redhat.com>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Cc: Jan Kara <jack@suse.cz>
Cc: Rik van Riel <riel@redhat.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Alex Elder <aelder@sgi.com>
Cc: Theodore Ts'o <tytso@mit.edu>
Cc: Chris Mason <chris.mason@oracle.com>
Cc: Dave Hansen <dave@linux.vnet.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|
|
writeback
Lumpy reclaim worked with two passes - the first which queued pages for IO
and the second which waited on writeback. As direct reclaim can no longer
write pages there is some dead code. This patch removes it but direct
reclaim will continue to wait on pages under writeback while in
synchronous reclaim mode.
Signed-off-by: Mel Gorman <mgorman@suse.de>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Johannes Weiner <jweiner@redhat.com>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Cc: Jan Kara <jack@suse.cz>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Alex Elder <aelder@sgi.com>
Cc: Theodore Ts'o <tytso@mit.edu>
Cc: Chris Mason <chris.mason@oracle.com>
Cc: Dave Hansen <dave@linux.vnet.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|
|
Testing from the XFS folk revealed that there is still too much I/O from
the end of the LRU in kswapd. Previously it was considered acceptable by
VM people for a small number of pages to be written back from reclaim with
testing generally showing about 0.3% of pages reclaimed were written back
(higher if memory was low). That writing back a small number of pages is
ok has been heavily disputed for quite some time and Dave Chinner
explained it well;
It doesn't have to be a very high number to be a problem. IO
is orders of magnitude slower than the CPU time it takes to
flush a page, so the cost of making a bad flush decision is
very high. And single page writeback from the LRU is almost
always a bad flush decision.
To complicate matters, filesystems respond very differently to requests
from reclaim according to Christoph Hellwig;
xfs tries to write it back if the requester is kswapd
ext4 ignores the request if it's a delayed allocation
btrfs ignores the request
As a result, each filesystem has different performance characteristics
when under memory pressure and there are many pages being dirtied. In
some cases, the request is ignored entirely so the VM cannot depend on the
IO being dispatched.
The objective of this series is to reduce writing of filesystem-backed
pages from reclaim, play nicely with writeback that is already in progress
and throttle reclaim appropriately when writeback pages are encountered.
The assumption is that the flushers will always write pages faster than if
reclaim issues the IO.
A secondary goal is to avoid the problem whereby direct reclaim splices
two potentially deep call stacks together.
There is a potential new problem as reclaim has less control over how long
before a page in a particularly zone or container is cleaned and direct
reclaimers depend on kswapd or flusher threads to do the necessary work.
However, as filesystems sometimes ignore direct reclaim requests already,
it is not expected to be a serious issue.
Patch 1 disables writeback of filesystem pages from direct reclaim
entirely. Anonymous pages are still written.
Patch 2 removes dead code in lumpy reclaim as it is no longer able
to synchronously write pages. This hurts lumpy reclaim but
there is an expectation that compaction is used for hugepage
allocations these days and lumpy reclaim's days are numbered.
Patches 3-4 add warnings to XFS and ext4 if called from
direct reclaim. With patch 1, this "never happens" and is
intended to catch regressions in this logic in the future.
Patch 5 disables writeback of filesystem pages from kswapd unless
the priority is raised to the point where kswapd is considered
to be in trouble.
Patch 6 throttles reclaimers if too many dirty pages are being
encountered and the zones or backing devices are congested.
Patch 7 invalidates dirty pages found at the end of the LRU so they
are reclaimed quickly after being written back rather than
waiting for a reclaimer to find them
I consider this series to be orthogonal to the writeback work but it is
worth noting that the writeback work affects the viability of patch 8 in
particular.
I tested this on ext4 and xfs using fs_mark, a simple writeback test based
on dd and a micro benchmark that does a streaming write to a large mapping
(exercises use-once LRU logic) followed by streaming writes to a mix of
anonymous and file-backed mappings. The command line for fs_mark when
botted with 512M looked something like
./fs_mark -d /tmp/fsmark-2676 -D 100 -N 150 -n 150 -L 25 -t 1 -S0 -s 10485760
The number of files was adjusted depending on the amount of available
memory so that the files created was about 3xRAM. For multiple threads,
the -d switch is specified multiple times.
The test machine is x86-64 with an older generation of AMD processor with
4 cores. The underlying storage was 4 disks configured as RAID-0 as this
was the best configuration of storage I had available. Swap is on a
separate disk. Dirty ratio was tuned to 40% instead of the default of
20%.
Testing was run with and without monitors to both verify that the patches
were operating as expected and that any performance gain was real and not
due to interference from monitors.
Here is a summary of results based on testing XFS.
512M1P-xfs Files/s mean 32.69 ( 0.00%) 34.44 ( 5.08%)
512M1P-xfs Elapsed Time fsmark 51.41 48.29
512M1P-xfs Elapsed Time simple-wb 114.09 108.61
512M1P-xfs Elapsed Time mmap-strm 113.46 109.34
512M1P-xfs Kswapd efficiency fsmark 62% 63%
512M1P-xfs Kswapd efficiency simple-wb 56% 61%
512M1P-xfs Kswapd efficiency mmap-strm 44% 42%
512M-xfs Files/s mean 30.78 ( 0.00%) 35.94 (14.36%)
512M-xfs Elapsed Time fsmark 56.08 48.90
512M-xfs Elapsed Time simple-wb 112.22 98.13
512M-xfs Elapsed Time mmap-strm 219.15 196.67
512M-xfs Kswapd efficiency fsmark 54% 56%
512M-xfs Kswapd efficiency simple-wb 54% 55%
512M-xfs Kswapd efficiency mmap-strm 45% 44%
512M-4X-xfs Files/s mean 30.31 ( 0.00%) 33.33 ( 9.06%)
512M-4X-xfs Elapsed Time fsmark 63.26 55.88
512M-4X-xfs Elapsed Time simple-wb 100.90 90.25
512M-4X-xfs Elapsed Time mmap-strm 261.73 255.38
512M-4X-xfs Kswapd efficiency fsmark 49% 50%
512M-4X-xfs Kswapd efficiency simple-wb 54% 56%
512M-4X-xfs Kswapd efficiency mmap-strm 37% 36%
512M-16X-xfs Files/s mean 60.89 ( 0.00%) 65.22 ( 6.64%)
512M-16X-xfs Elapsed Time fsmark 67.47 58.25
512M-16X-xfs Elapsed Time simple-wb 103.22 90.89
512M-16X-xfs Elapsed Time mmap-strm 237.09 198.82
512M-16X-xfs Kswapd efficiency fsmark 45% 46%
512M-16X-xfs Kswapd efficiency simple-wb 53% 55%
512M-16X-xfs Kswapd efficiency mmap-strm 33% 33%
Up until 512-4X, the FSmark improvements were statistically significant.
For the 4X and 16X tests the results were within standard deviations but
just barely. The time to completion for all tests is improved which is an
important result. In general, kswapd efficiency is not affected by
skipping dirty pages.
1024M1P-xfs Files/s mean 39.09 ( 0.00%) 41.15 ( 5.01%)
1024M1P-xfs Elapsed Time fsmark 84.14 80.41
1024M1P-xfs Elapsed Time simple-wb 210.77 184.78
1024M1P-xfs Elapsed Time mmap-strm 162.00 160.34
1024M1P-xfs Kswapd efficiency fsmark 69% 75%
1024M1P-xfs Kswapd efficiency simple-wb 71% 77%
1024M1P-xfs Kswapd efficiency mmap-strm 43% 44%
1024M-xfs Files/s mean 35.45 ( 0.00%) 37.00 ( 4.19%)
1024M-xfs Elapsed Time fsmark 94.59 91.00
1024M-xfs Elapsed Time simple-wb 229.84 195.08
1024M-xfs Elapsed Time mmap-strm 405.38 440.29
1024M-xfs Kswapd efficiency fsmark 79% 71%
1024M-xfs Kswapd efficiency simple-wb 74% 74%
1024M-xfs Kswapd efficiency mmap-strm 39% 42%
1024M-4X-xfs Files/s mean 32.63 ( 0.00%) 35.05 ( 6.90%)
1024M-4X-xfs Elapsed Time fsmark 103.33 97.74
1024M-4X-xfs Elapsed Time simple-wb 204.48 178.57
1024M-4X-xfs Elapsed Time mmap-strm 528.38 511.88
1024M-4X-xfs Kswapd efficiency fsmark 81% 70%
1024M-4X-xfs Kswapd efficiency simple-wb 73% 72%
1024M-4X-xfs Kswapd efficiency mmap-strm 39% 38%
1024M-16X-xfs Files/s mean 42.65 ( 0.00%) 42.97 ( 0.74%)
1024M-16X-xfs Elapsed Time fsmark 103.11 99.11
1024M-16X-xfs Elapsed Time simple-wb 200.83 178.24
1024M-16X-xfs Elapsed Time mmap-strm 397.35 459.82
1024M-16X-xfs Kswapd efficiency fsmark 84% 69%
1024M-16X-xfs Kswapd efficiency simple-wb 74% 73%
1024M-16X-xfs Kswapd efficiency mmap-strm 39% 40%
All FSMark tests up to 16X had statistically significant improvements.
For the most part, tests are completing faster with the exception of the
streaming writes to a mixture of anonymous and file-backed mappings which
were slower in two cases
In the cases where the mmap-strm tests were slower, there was more
swapping due to dirty pages being skipped. The number of additional pages
swapped is almost identical to the fewer number of pages written from
reclaim. In other words, roughly the same number of pages were reclaimed
but swapping was slower. As the test is a bit unrealistic and stresses
memory heavily, the small shift is acceptable.
4608M1P-xfs Files/s mean 29.75 ( 0.00%) 30.96 ( 3.91%)
4608M1P-xfs Elapsed Time fsmark 512.01 492.15
4608M1P-xfs Elapsed Time simple-wb 618.18 566.24
4608M1P-xfs Elapsed Time mmap-strm 488.05 465.07
4608M1P-xfs Kswapd efficiency fsmark 93% 86%
4608M1P-xfs Kswapd efficiency simple-wb 88% 84%
4608M1P-xfs Kswapd efficiency mmap-strm 46% 45%
4608M-xfs Files/s mean 27.60 ( 0.00%) 28.85 ( 4.33%)
4608M-xfs Elapsed Time fsmark 555.96 532.34
4608M-xfs Elapsed Time simple-wb 659.72 571.85
4608M-xfs Elapsed Time mmap-strm 1082.57 1146.38
4608M-xfs Kswapd efficiency fsmark 89% 91%
4608M-xfs Kswapd efficiency simple-wb 88% 82%
4608M-xfs Kswapd efficiency mmap-strm 48% 46%
4608M-4X-xfs Files/s mean 26.00 ( 0.00%) 27.47 ( 5.35%)
4608M-4X-xfs Elapsed Time fsmark 592.91 564.00
4608M-4X-xfs Elapsed Time simple-wb 616.65 575.07
4608M-4X-xfs Elapsed Time mmap-strm 1773.02 1631.53
4608M-4X-xfs Kswapd efficiency fsmark 90% 94%
4608M-4X-xfs Kswapd efficiency simple-wb 87% 82%
4608M-4X-xfs Kswapd efficiency mmap-strm 43% 43%
4608M-16X-xfs Files/s mean 26.07 ( 0.00%) 26.42 ( 1.32%)
4608M-16X-xfs Elapsed Time fsmark 602.69 585.78
4608M-16X-xfs Elapsed Time simple-wb 606.60 573.81
4608M-16X-xfs Elapsed Time mmap-strm 1549.75 1441.86
4608M-16X-xfs Kswapd efficiency fsmark 98% 98%
4608M-16X-xfs Kswapd efficiency simple-wb 88% 82%
4608M-16X-xfs Kswapd efficiency mmap-strm 44% 42%
Unlike the other tests, the fsmark results are not statistically
significant but the min and max times are both improved and for the most
part, tests completed faster.
There are other indications that this is an improvement as well. For
example, in the vast majority of cases, there were fewer pages scanned by
direct reclaim implying in many cases that stalls due to direct reclaim
are reduced. KSwapd is scanning more due to skipping dirty pages which is
unfortunate but the CPU usage is still acceptable
In an earlier set of tests, I used blktrace and in almost all cases
throughput throughout the entire test was higher. However, I ended up
discarding those results as recording blktrace data was too heavy for my
liking.
On a laptop, I plugged in a USB stick and ran a similar tests of tests
using it as backing storage. A desktop environment was running and for
the entire duration of the tests, firefox and gnome terminal were
launching and exiting to vaguely simulate a user.
1024M-xfs Files/s mean 0.41 ( 0.00%) 0.44 ( 6.82%)
1024M-xfs Elapsed Time fsmark 2053.52 1641.03
1024M-xfs Elapsed Time simple-wb 1229.53 768.05
1024M-xfs Elapsed Time mmap-strm 4126.44 4597.03
1024M-xfs Kswapd efficiency fsmark 84% 85%
1024M-xfs Kswapd efficiency simple-wb 92% 81%
1024M-xfs Kswapd efficiency mmap-strm 60% 51%
1024M-xfs Avg wait ms fsmark 5404.53 4473.87
1024M-xfs Avg wait ms simple-wb 2541.35 1453.54
1024M-xfs Avg wait ms mmap-strm 3400.25 3852.53
The mmap-strm results were hurt because firefox launching had a tendency
to push the test out of memory. On the postive side, firefox launched
marginally faster with the patches applied. Time to completion for many
tests was faster but more importantly - the "Avg wait" time as measured by
iostat was far lower implying the system would be more responsive. It was
also the case that "Avg wait ms" on the root filesystem was lower. I
tested it manually and while the system felt slightly more responsive
while copying data to a USB stick, it was marginal enough that it could be
my imagination.
This patch: do not writeback filesystem pages in direct reclaim.
When kswapd is failing to keep zones above the min watermark, a process
will enter direct reclaim in the same manner kswapd does. If a dirty page
is encountered during the scan, this page is written to backing storage
using mapping->writepage.
This causes two problems. First, it can result in very deep call stacks,
particularly if the target storage or filesystem are complex. Some
filesystems ignore write requests from direct reclaim as a result. The
second is that a single-page flush is inefficient in terms of IO. While
there is an expectation that the elevator will merge requests, this does
not always happen. Quoting Christoph Hellwig;
The elevator has a relatively small window it can operate on,
and can never fix up a bad large scale writeback pattern.
This patch prevents direct reclaim writing back filesystem pages by
checking if current is kswapd. Anonymous pages are still written to swap
as there is not the equivalent of a flusher thread for anonymous pages.
If the dirty pages cannot be written back, they are placed back on the LRU
lists. There is now a direct dependency on dirty page balancing to
prevent too many pages in the system being dirtied which would prevent
reclaim making forward progress.
Signed-off-by: Mel Gorman <mgorman@suse.de>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Johannes Weiner <jweiner@redhat.com>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Cc: Jan Kara <jack@suse.cz>
Cc: Rik van Riel <riel@redhat.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Alex Elder <aelder@sgi.com>
Cc: Theodore Ts'o <tytso@mit.edu>
Cc: Chris Mason <chris.mason@oracle.com>
Cc: Dave Hansen <dave@linux.vnet.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|
|
The nr_force_scan[] tuple holds the effective scan numbers for anon and
file pages in case the situation called for a forced scan and the
regularly calculated scan numbers turned out zero.
However, the effective scan number can always be assumed to be
SWAP_CLUSTER_MAX right before the division into anon and file. The
numerators and denominator are properly set up for all cases, be it force
scan for just file, just anon, or both, to do the right thing.
Signed-off-by: Johannes Weiner <jweiner@redhat.com>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Reviewed-by: Michal Hocko <mhocko@suse.cz>
Cc: Ying Han <yinghan@google.com>
Cc: Balbir Singh <bsingharora@gmail.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Acked-by: Mel Gorman <mel@csn.ul.ie>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|
|
When we get a bad_page bug report, it's useful to see what modules the
user had loaded.
Signed-off-by: Dave Jones <davej@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|
|
test_set_oom_score_adj() was introduced in 72788c385604 ("oom: replace
PF_OOM_ORIGIN with toggling oom_score_adj") to temporarily elevate
current's oom_score_adj for ksm and swapoff without requiring an
additional per-process flag.
Using that function to both set oom_score_adj to OOM_SCORE_ADJ_MAX and
then reinstate the previous value is racy since it's possible that
userspace can set the value to something else itself before the old value
is reinstated. That results in userspace setting current's oom_score_adj
to a different value and then the kernel immediately setting it back to
its previous value without notification.
To fix this, a new compare_swap_oom_score_adj() function is introduced
with the same semantics as the compare and swap CAS instruction, or
CMPXCHG on x86. It is used to reinstate the previous value of
oom_score_adj if and only if the present value is the same as the old
value.
Signed-off-by: David Rientjes <rientjes@google.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Ying Han <yinghan@google.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|
|
This removes mm->oom_disable_count entirely since it's unnecessary and
currently buggy. The counter was intended to be per-process but it's
currently decremented in the exit path for each thread that exits, causing
it to underflow.
The count was originally intended to prevent oom killing threads that
share memory with threads that cannot be killed since it doesn't lead to
future memory freeing. The counter could be fixed to represent all
threads sharing the same mm, but it's better to remove the count since:
- it is possible that the OOM_DISABLE thread sharing memory with the
victim is waiting on that thread to exit and will actually cause
future memory freeing, and
- there is no guarantee that a thread is disabled from oom killing just
because another thread sharing its mm is oom disabled.
Signed-off-by: David Rientjes <rientjes@google.com>
Reported-by: Oleg Nesterov <oleg@redhat.com>
Reviewed-by: Oleg Nesterov <oleg@redhat.com>
Cc: Ying Han <yinghan@google.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|
|
After selecting a task to kill, the oom killer iterates all processes and
kills all other threads that share the same mm_struct in different thread
groups. It would not otherwise be helpful to kill a thread if its memory
would not be subsequently freed.
A kernel thread, however, may assume a user thread's mm by using
use_mm(). This is only temporary and should not result in sending a
SIGKILL to that kthread.
This patch ensures that only user threads and not kthreads are sent a
SIGKILL if they share the same mm_struct as the oom killed task.
Signed-off-by: David Rientjes <rientjes@google.com>
Reviewed-by: Michal Hocko <mhocko@suse.cz>
Reviewed-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|
|
If a thread has been oom killed and is frozen, thaw it before returning to
the page allocator. Otherwise, it can stay frozen indefinitely and no
memory will be freed.
Signed-off-by: David Rientjes <rientjes@google.com>
Reported-by: Konstantin Khlebnikov <khlebnikov@openvz.org>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: "Rafael J. Wysocki" <rjw@sisk.pl>
Acked-by: Michal Hocko <mhocko@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|
|
Looks like someone got distracted after adding the comment characters.
Signed-off-by: Johannes Weiner <jweiner@redhat.com>
Acked-by: Peter Zijlstra <peterz@infradead.org>
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>
|
|
per-task block plug can reduce block queue lock contention and increase
request merge. Currently page reclaim doesn't support it. I originally
thought page reclaim doesn't need it, because kswapd thread count is
limited and file cache write is done at flusher mostly.
When I test a workload with heavy swap in a 4-node machine, each CPU is
doing direct page reclaim and swap. This causes block queue lock
contention. In my test, without below patch, the CPU utilization is about
2% ~ 7%. With the patch, the CPU utilization is about 1% ~ 3%. Disk
throughput isn't changed. This should improve normal kswapd write and
file cache write too (increase request merge for example), but might not
be so obvious as I explain above.
Signed-off-by: Shaohua Li <shaohua.li@intel.com>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Minchan Kim <minchan.kim@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|
|
unmap_and_move() is one a big messy function. Clean it up.
Signed-off-by: Minchan Kim <minchan.kim@gmail.com>
Reviewed-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Rik van Riel <riel@redhat.com>
Cc: Michal Hocko <mhocko@suse.cz>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|
|
In __zone_reclaim case, we don't want to shrink mapped page. Nonetheless,
we have isolated mapped page and re-add it into LRU's head. It's
unnecessary CPU overhead and makes LRU churning.
Of course, when we isolate the page, the page might be mapped but when we
try to migrate the page, the page would be not mapped. So it could be
migrated. But race is rare and although it happens, it's no big deal.
Signed-off-by: Minchan Kim <minchan.kim@gmail.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Reviewed-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Reviewed-by: Michal Hocko <mhocko@suse.cz>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Rik van Riel <riel@redhat.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|
|
In async mode, compaction doesn't migrate dirty or writeback pages. So,
it's meaningless to pick the page and re-add it to lru list.
Of course, when we isolate the page in compaction, the page might be dirty
or writeback but when we try to migrate the page, the page would be not
dirty, writeback. So it could be migrated. But it's very unlikely as
isolate and migration cycle is much faster than writeout.
So, this patch helps cpu overhead and prevent unnecessary LRU churning.
Signed-off-by: Minchan Kim <minchan.kim@gmail.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Reviewed-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Acked-by: Mel Gorman <mgorman@suse.de>
Acked-by: Rik van Riel <riel@redhat.com>
Reviewed-by: Michal Hocko <mhocko@suse.cz>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|
|
Change ISOLATE_XXX macro with bitwise isolate_mode_t type. Normally,
macro isn't recommended as it's type-unsafe and making debugging harder as
symbol cannot be passed throught to the debugger.
Quote from Johannes
" Hmm, it would probably be cleaner to fully convert the isolation mode
into independent flags. INACTIVE, ACTIVE, BOTH is currently a
tri-state among flags, which is a bit ugly."
This patch moves isolate mode from swap.h to mmzone.h by memcontrol.h
Signed-off-by: Minchan Kim <minchan.kim@gmail.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Rik van Riel <riel@redhat.com>
Cc: Michal Hocko <mhocko@suse.cz>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|
|
acct_isolated of compaction uses page_lru_base_type which returns only
base type of LRU list so it never returns LRU_ACTIVE_ANON or
LRU_ACTIVE_FILE. In addtion, cc->nr_[anon|file] is used in only
acct_isolated so it doesn't have fields in conpact_control.
This patch removes fields from compact_control and makes clear function of
acct_issolated which counts the number of anon|file pages isolated.
Signed-off-by: Minchan Kim <minchan.kim@gmail.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Reviewed-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Acked-by: Mel Gorman <mgorman@suse.de>
Acked-by: Rik van Riel <riel@redhat.com>
Reviewed-by: Michal Hocko <mhocko@suse.cz>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|
|
The basic idea behind cross memory attach is to allow MPI programs doing
intra-node communication to do a single copy of the message rather than a
double copy of the message via shared memory.
The following patch attempts to achieve this by allowing a destination
process, given an address and size from a source process, to copy memory
directly from the source process into its own address space via a system
call. There is also a symmetrical ability to copy from the current
process's address space into a destination process's address space.
- Use of /proc/pid/mem has been considered, but there are issues with
using it:
- Does not allow for specifying iovecs for both src and dest, assuming
preadv or pwritev was implemented either the area read from or
written to would need to be contiguous.
- Currently mem_read allows only processes who are currently
ptrace'ing the target and are still able to ptrace the target to read
from the target. This check could possibly be moved to the open call,
but its not clear exactly what race this restriction is stopping
(reason appears to have been lost)
- Having to send the fd of /proc/self/mem via SCM_RIGHTS on unix
domain socket is a bit ugly from a userspace point of view,
especially when you may have hundreds if not (eventually) thousands
of processes that all need to do this with each other
- Doesn't allow for some future use of the interface we would like to
consider adding in the future (see below)
- Interestingly reading from /proc/pid/mem currently actually
involves two copies! (But this could be fixed pretty easily)
As mentioned previously use of vmsplice instead was considered, but has
problems. Since you need the reader and writer working co-operatively if
the pipe is not drained then you block. Which requires some wrapping to
do non blocking on the send side or polling on the receive. In all to all
communication it requires ordering otherwise you can deadlock. And in the
example of many MPI tasks writing to one MPI task vmsplice serialises the
copying.
There are some cases of MPI collectives where even a single copy interface
does not get us the performance gain we could. For example in an
MPI_Reduce rather than copy the data from the source we would like to
instead use it directly in a mathops (say the reduce is doing a sum) as
this would save us doing a copy. We don't need to keep a copy of the data
from the source. I haven't implemented this, but I think this interface
could in the future do all this through the use of the flags - eg could
specify the math operation and type and the kernel rather than just
copying the data would apply the specified operation between the source
and destination and store it in the destination.
Although we don't have a "second user" of the interface (though I've had
some nibbles from people who may be interested in using it for intra
process messaging which is not MPI). This interface is something which
hardware vendors are already doing for their custom drivers to implement
fast local communication. And so in addition to this being useful for
OpenMPI it would mean the driver maintainers don't have to fix things up
when the mm changes.
There was some discussion about how much faster a true zero copy would
go. Here's a link back to the email with some testing I did on that:
http://marc.info/?l=linux-mm&m=130105930902915&w=2
There is a basic man page for the proposed interface here:
http://ozlabs.org/~cyeoh/cma/process_vm_readv.txt
This has been implemented for x86 and powerpc, other architecture should
mainly (I think) just need to add syscall numbers for the process_vm_readv
and process_vm_writev. There are 32 bit compatibility versions for
64-bit kernels.
For arch maintainers there are some simple tests to be able to quickly
verify that the syscalls are working correctly here:
http://ozlabs.org/~cyeoh/cma/cma-test-20110718.tgz
Signed-off-by: Chris Yeoh <yeohc@au1.ibm.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: David Howells <dhowells@redhat.com>
Cc: James Morris <jmorris@namei.org>
Cc: <linux-man@vger.kernel.org>
Cc: <linux-arch@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|
|
The removal of the implicitly everywhere module.h and its child includes
will reveal this implicit stat.h usage:
mm/dmapool.c:108: error: ‘S_IRUGO’ undeclared here (not in a function)
Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>
|
|
The files changed within are only using the EXPORT_SYMBOL
macro variants. They are not using core modular infrastructure
and hence don't need module.h but only the export.h header.
Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>
|
|
These files were getting <linux/module.h> via an implicit include
path, but we want to crush those out of existence since they cost
time during compiles of processing thousands of lines of headers
for no reason. Give them the lightweight header that just contains
the EXPORT_SYMBOL infrastructure.
Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>
|
|
There is nothing modular in these files, and no reason to drag
in all the 357 headers that module.h brings with it, since
it just slows down compiles.
Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>
|
|
This creates a new 'reason' field in a wb_writeback_work
structure, which unambiguously identifies who initiates
writeback activity. A 'wb_reason' enumeration has been
added to writeback.h, to enumerate the possible reasons.
The 'writeback_work_class' and tracepoint event class and
'writeback_queue_io' tracepoints are updated to include the
symbolic 'reason' in all trace events.
And the 'writeback_inodes_sbXXX' family of routines has had
a wb_stats parameter added to them, so callers can specify
why writeback is being started.
Acked-by: Jan Kara <jack@suse.cz>
Signed-off-by: Curt Wohlgemuth <curtw@google.com>
Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
|
|
Useful for analyzing the dynamics of the throttling algorithms and
debugging user reported problems.
Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
|
|
It helps understand how various throttle bandwidths are updated.
Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
|
|
git://git.kernel.org/pub/scm/linux/kernel/git/hch/vfs-queue
* 'for-next' of git://git.kernel.org/pub/scm/linux/kernel/git/hch/vfs-queue: (21 commits)
leases: fix write-open/read-lease race
nfs: drop unnecessary locking in llseek
ext4: replace cut'n'pasted llseek code with generic_file_llseek_size
vfs: add generic_file_llseek_size
vfs: do (nearly) lockless generic_file_llseek
direct-io: merge direct_io_walker into __blockdev_direct_IO
direct-io: inline the complete submission path
direct-io: separate map_bh from dio
direct-io: use a slab cache for struct dio
direct-io: rearrange fields in dio/dio_submit to avoid holes
direct-io: fix a wrong comment
direct-io: separate fields only used in the submission path from struct dio
vfs: fix spinning prevention in prune_icache_sb
vfs: add a comment to inode_permission()
vfs: pass all mask flags check_acl and posix_acl_permission
vfs: add hex format for MAY_* flag values
vfs: indicate that the permission functions take all the MAY_* flags
compat: sync compat_stats with statfs.
vfs: add "device" tag to /proc/self/mountstats
cleanup: vfs: small comment fix for block_invalidatepage
...
Fix up trivial conflict in fs/gfs2/file.c (llseek changes)
|
|
Currently, when you call iov_iter_advance, then the pointer to the iovec
array can be incremented, but it does not decrement the nr_segs value in
the iov_iter struct. The result is a iov_iter struct with a nr_segs
value that goes beyond the end of the array.
While I'm not aware of anything that's specifically broken by this, it
seems odd and a bit dangerous not to decrement that value. If someone
were to trust the nr_segs value to be correct, then they could end up
walking off the end of the array.
Changing this might also provide some micro-optimization when dealing
with the last iovec in an array. Many of the other routines that deal
with iov_iter have optimized codepaths when nr_segs == 1.
Cc: Nick Piggin <npiggin@suse.de>
Signed-off-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
|
|
|
|
git://git.kernel.org/pub/scm/linux/kernel/git/jikos/trivial
* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jikos/trivial: (59 commits)
MAINTAINERS: linux-m32r is moderated for non-subscribers
linux@lists.openrisc.net is moderated for non-subscribers
Drop default from "DM365 codec select" choice
parisc: Kconfig: cleanup Kernel page size default
Kconfig: remove redundant CONFIG_ prefix on two symbols
cris: remove arch/cris/arch-v32/lib/nand_init.S
microblaze: add missing CONFIG_ prefixes
h8300: drop puzzling Kconfig dependencies
MAINTAINERS: microblaze-uclinux@itee.uq.edu.au is moderated for non-subscribers
tty: drop superfluous dependency in Kconfig
ARM: mxc: fix Kconfig typo 'i.MX51'
Fix file references in Kconfig files
aic7xxx: fix Kconfig references to READMEs
Fix file references in drivers/ide/
thinkpad_acpi: Fix printk typo 'bluestooth'
bcmring: drop commented out line in Kconfig
btmrvl_sdio: fix typo 'btmrvl_sdio_sd6888'
doc: raw1394: Trivial typo fix
CIFS: Don't free volume_info->UNC until we are entirely done with it.
treewide: Correct spelling of successfully in comments
...
|
|
* 'next' of git://selinuxproject.org/~jmorris/linux-security: (95 commits)
TOMOYO: Fix incomplete read after seek.
Smack: allow to access /smack/access as normal user
TOMOYO: Fix unused kernel config option.
Smack: fix: invalid length set for the result of /smack/access
Smack: compilation fix
Smack: fix for /smack/access output, use string instead of byte
Smack: domain transition protections (v3)
Smack: Provide information for UDS getsockopt(SO_PEERCRED)
Smack: Clean up comments
Smack: Repair processing of fcntl
Smack: Rule list lookup performance
Smack: check permissions from user space (v2)
TOMOYO: Fix quota and garbage collector.
TOMOYO: Remove redundant tasklist_lock.
TOMOYO: Fix domain transition failure warning.
TOMOYO: Remove tomoyo_policy_memory_lock spinlock.
TOMOYO: Simplify garbage collector.
TOMOYO: Fix make namespacecheck warnings.
target: check hex2bin result
encrypted-keys: check hex2bin result
...
|
|
init_emergency_pool() does not create the page pool for bouncing block
requests if the current count of high pages is zero. If high memory
may be added later (either via memory hotplug or a balloon driver in a
virtualized system) then a oops occurs if a request with a high page
need bouncing because the pool does not exist.
So, always create the pool if memory hotplug is enabled and change the
test so it's valid even if all high pages are currently in the balloon
(the balloon drivers adjust totalhigh_pages but not max_pfn).
Signed-off-by: David Vrabel <david.vrabel@citrix.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
|
|
I don't usually pay much attention to the stale "? " addresses in
stack backtraces, but this lucky report from Pawel Sikora hints that
mremap's move_ptes() has inadequate locking against page migration.
3.0 BUG_ON(!PageLocked(p)) in migration_entry_to_page():
kernel BUG at include/linux/swapops.h:105!
RIP: 0010:[<ffffffff81127b76>] [<ffffffff81127b76>]
migration_entry_wait+0x156/0x160
[<ffffffff811016a1>] handle_pte_fault+0xae1/0xaf0
[<ffffffff810feee2>] ? __pte_alloc+0x42/0x120
[<ffffffff8112c26b>] ? do_huge_pmd_anonymous_page+0xab/0x310
[<ffffffff81102a31>] handle_mm_fault+0x181/0x310
[<ffffffff81106097>] ? vma_adjust+0x537/0x570
[<ffffffff81424bed>] do_page_fault+0x11d/0x4e0
[<ffffffff81109a05>] ? do_mremap+0x2d5/0x570
[<ffffffff81421d5f>] page_fault+0x1f/0x30
mremap's down_write of mmap_sem, together with i_mmap_mutex or lock,
and pagetable locks, were good enough before page migration (with its
requirement that every migration entry be found) came in, and enough
while migration always held mmap_sem; but not enough nowadays, when
there's memory hotremove and compaction.
The danger is that move_ptes() lets a migration entry dodge around
behind remove_migration_pte()'s back, so it's in the old location when
looking at the new, then in the new location when looking at the old.
Either mremap's move_ptes() must additionally take anon_vma lock(), or
migration's remove_migration_pte() must stop peeking for is_swap_entry()
before it takes pagetable lock.
Consensus chooses the latter: we prefer to add overhead to migration
than to mremapping, which gets used by JVMs and by exec stack setup.
Reported-and-tested-by: Paweł Sikora <pluto@agmk.net>
Signed-off-by: Hugh Dickins <hughd@google.com>
Acked-by: Andrea Arcangeli <aarcange@redhat.com>
Acked-by: Mel Gorman <mgorman@suse.de>
Cc: stable@vger.kernel.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|
|
Fix powerpc compile warnings
mm/page-writeback.c: In function 'bdi_position_ratio':
mm/page-writeback.c:622:3: warning: comparison of distinct pointer types lacks a cast [enabled by default]
page-writeback.c:635:4: warning: comparison of distinct pointer types lacks a cast [enabled by default]
Also fix gcc "uninitialized var" warnings.
Reported-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
|
|
Keep a minimal pool of dirty pages for each bdi, so that the disk IO
queues won't underrun. Also gently increase a small bdi_thresh to avoid
it stuck in 0 for some light dirtied bdi.
It's particularly useful for JBOD and small memory system.
It may result in (pos_ratio > 1) at the setpoint and push the dirty
pages high. This is more or less intended because the bdi is in the
danger of IO queue underflow.
Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
|
|
The dirty pause time shall ultimately be controlled by adjusting
nr_dirtied_pause, since there is relationship
pause = pages_dirtied / task_ratelimit
Assuming
pages_dirtied ~= nr_dirtied_pause
task_ratelimit ~= dirty_ratelimit
We get
nr_dirtied_pause ~= dirty_ratelimit * desired_pause
Here dirty_ratelimit is preferred over task_ratelimit because it's
more stable.
It's also important to limit possible large transitional errors:
- bw is changing quickly
- pages_dirtied << nr_dirtied_pause on entering dirty exceeded area
- pages_dirtied >> nr_dirtied_pause on btrfs (to be improved by a
separate fix, but still expect non-trivial errors)
So we end up using the above formula inside clamp_val().
The best test case for this code is to run 100 "dd bs=4M" tasks on
btrfs and check its pause time distribution.
Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
|
|
Apply two policies to scale down the max pause time for
1) small number of concurrent dirtiers
2) small memory system (comparing to storage bandwidth)
MAX_PAUSE=200ms may only be suitable for high end servers with lots of
concurrent dirtiers, where the large pause time can reduce much overheads.
Otherwise, smaller pause time is desirable whenever possible, so as to
get good responsiveness and smooth user experiences. It's actually
required for good disk utilization in the case when all the dirty pages
can be synced to disk within MAX_PAUSE=200ms.
Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
|
|
As proposed by Chris, Dave and Jan, don't start foreground writeback IO
inside balance_dirty_pages(). Instead, simply let it idle sleep for some
time to throttle the dirtying task. In the mean while, kick off the
per-bdi flusher thread to do background writeback IO.
RATIONALS
=========
- disk seeks on concurrent writeback of multiple inodes (Dave Chinner)
If every thread doing writes and being throttled start foreground
writeback, it leads to N IO submitters from at least N different
inodes at the same time, end up with N different sets of IO being
issued with potentially zero locality to each other, resulting in
much lower elevator sort/merge efficiency and hence we seek the disk
all over the place to service the different sets of IO.
OTOH, if there is only one submission thread, it doesn't jump between
inodes in the same way when congestion clears - it keeps writing to
the same inode, resulting in large related chunks of sequential IOs
being issued to the disk. This is more efficient than the above
foreground writeback because the elevator works better and the disk
seeks less.
- lock contention and cache bouncing on concurrent IO submitters (Dave Chinner)
With this patchset, the fs_mark benchmark on a 12-drive software RAID0 goes
from CPU bound to IO bound, freeing "3-4 CPUs worth of spinlock contention".
* "CPU usage has dropped by ~55%", "it certainly appears that most of
the CPU time saving comes from the removal of contention on the
inode_wb_list_lock" (IMHO at least 10% comes from the reduction of
cacheline bouncing, because the new code is able to call much less
frequently into balance_dirty_pages() and hence access the global
page states)
* the user space "App overhead" is reduced by 20%, by avoiding the
cacheline pollution by the complex writeback code path
* "for a ~5% throughput reduction", "the number of write IOs have
dropped by ~25%", and the elapsed time reduced from 41:42.17 to
40:53.23.
* On a simple test of 100 dd, it reduces the CPU %system time from 30% to 3%,
and improves IO throughput from 38MB/s to 42MB/s.
- IO size too small for fast arrays and too large for slow USB sticks
The write_chunk used by current balance_dirty_pages() cannot be
directly set to some large value (eg. 128MB) for better IO efficiency.
Because it could lead to more than 1 second user perceivable stalls.
Even the current 4MB write size may be too large for slow USB sticks.
The fact that balance_dirty_pages() starts IO on itself couples the
IO size to wait time, which makes it hard to do suitable IO size while
keeping the wait time under control.
Now it's possible to increase writeback chunk size proportional to the
disk bandwidth. In a simple test of 50 dd's on XFS, 1-HDD, 3GB ram,
the larger writeback size dramatically reduces the seek count to 1/10
(far beyond my expectation) and improves the write throughput by 24%.
- long block time in balance_dirty_pages() hurts desktop responsiveness
Many of us may have the experience: it often takes a couple of seconds
or even long time to stop a heavy writing dd/cp/tar command with
Ctrl-C or "kill -9".
- IO pipeline broken by bumpy write() progress
There are a broad class of "loop {read(buf); write(buf);}" applications
whose read() pipeline will be under-utilized or even come to a stop if
the write()s have long latencies _or_ don't progress in a constant rate.
The current threshold based throttling inherently transfers the large
low level IO completion fluctuations to bumpy application write()s,
and further deteriorates with increasing number of dirtiers and/or bdi's.
For example, when doing 50 dd's + 1 remote rsync to an XFS partition,
the rsync progresses very bumpy in legacy kernel, and throughput is
improved by 67% by this patchset. (plus the larger write chunk size,
it will be 93% speedup).
The new rate based throttling can support 1000+ dd's with excellent
smoothness, low latency and low overheads.
For the above reasons, it's much better to do IO-less and low latency
pauses in balance_dirty_pages().
Jan Kara, Dave Chinner and me explored the scheme to let
balance_dirty_pages() wait for enough writeback IO completions to
safeguard the dirty limit. However it's found to have two problems:
- in large NUMA systems, the per-cpu counters may have big accounting
errors, leading to big throttle wait time and jitters.
- NFS may kill large amount of unstable pages with one single COMMIT.
Because NFS server serves COMMIT with expensive fsync() IOs, it is
desirable to delay and reduce the number of COMMITs. So it's not
likely to optimize away such kind of bursty IO completions, and the
resulted large (and tiny) stall times in IO completion based throttling.
So here is a pause time oriented approach, which tries to control the
pause time in each balance_dirty_pages() invocations, by controlling
the number of pages dirtied before calling balance_dirty_pages(), for
smooth and efficient dirty throttling:
- avoid useless (eg. zero pause time) balance_dirty_pages() calls
- avoid too small pause time (less than 4ms, which burns CPU power)
- avoid too large pause time (more than 200ms, which hurts responsiveness)
- avoid big fluctuations of pause times
It can control pause times at will. The default policy (in a followup
patch) will be to do ~10ms pauses in 1-dd case, and increase to ~100ms
in 1000-dd case.
BEHAVIOR CHANGE
===============
(1) dirty threshold
Users will notice that the applications will get throttled once crossing
the global (background + dirty)/2=15% threshold, and then balanced around
17.5%. Before patch, the behavior is to just throttle it at 20% dirtyable
memory in 1-dd case.
Since the task will be soft throttled earlier than before, it may be
perceived by end users as performance "slow down" if his application
happens to dirty more than 15% dirtyable memory.
(2) smoothness/responsiveness
Users will notice a more responsive system during heavy writeback.
"killall dd" will take effect instantly.
Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
|
|
Add two fields to task_struct.
1) account dirtied pages in the individual tasks, for accuracy
2) per-task balance_dirty_pages() call intervals, for flexibility
The balance_dirty_pages() call interval (ie. nr_dirtied_pause) will
scale near-sqrt to the safety gap between dirty pages and threshold.
The main problem of per-task nr_dirtied is, if 1k+ tasks start dirtying
pages at exactly the same time, each task will be assigned a large
initial nr_dirtied_pause, so that the dirty threshold will be exceeded
long before each task reached its nr_dirtied_pause and hence call
balance_dirty_pages().
The solution is to watch for the number of pages dirtied on each CPU in
between the calls into balance_dirty_pages(). If it exceeds ratelimit_pages
(3% dirty threshold), force call balance_dirty_pages() for a chance to
set bdi->dirty_exceeded. In normal situations, this safeguarding
condition is not expected to trigger at all.
On the sqrt in dirty_poll_interval():
It will serve as an initial guess when dirty pages are still in the
freerun area.
When dirty pages are floating inside the dirty control scope [freerun,
limit], a followup patch will use some refined dirty poll interval to
get the desired pause time.
thresh-dirty (MB) sqrt
1 16
2 22
4 32
8 45
16 64
32 90
64 128
128 181
256 256
512 362
1024 512
The above table means, given 1MB (or 1GB) gap and the dd tasks polling
balance_dirty_pages() on every 16 (or 512) pages, the dirty limit won't
be exceeded as long as there are less than 16 (or 512) concurrent dd's.
So sqrt naturally leads to less overheads and more safe concurrent tasks
for large memory servers, which have large (thresh-freerun) gaps.
peter: keep the per-CPU ratelimit for safeguarding the 1k+ tasks case
CC: Peter Zijlstra <a.p.zijlstra@chello.nl>
Reviewed-by: Andrea Righi <andrea@betterlinux.com>
Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
|
|
There are some imperfections in balanced_dirty_ratelimit.
1) large fluctuations
The dirty_rate used for computing balanced_dirty_ratelimit is merely
averaged in the past 200ms (very small comparing to the 3s estimation
period for write_bw), which makes rather dispersed distribution of
balanced_dirty_ratelimit.
It's pretty hard to average out the singular points by increasing the
estimation period. Considering that the averaging technique will
introduce very undesirable time lags, I give it up totally. (btw, the 3s
write_bw averaging time lag is much more acceptable because its impact
is one-way and therefore won't lead to oscillations.)
The more practical way is filtering -- most singular
balanced_dirty_ratelimit points can be filtered out by remembering some
prev_balanced_rate and prev_prev_balanced_rate. However the more
reliable way is to guard balanced_dirty_ratelimit with task_ratelimit.
2) due to truncates and fs redirties, the (write_bw <=> dirty_rate)
match could become unbalanced, which may lead to large systematical
errors in balanced_dirty_ratelimit. The truncates, due to its possibly
bumpy nature, can hardly be compensated smoothly. So let's face it. When
some over-estimated balanced_dirty_ratelimit brings dirty_ratelimit
high, dirty pages will go higher than the setpoint. task_ratelimit will
in turn become lower than dirty_ratelimit. So if we consider both
balanced_dirty_ratelimit and task_ratelimit and update dirty_ratelimit
only when they are on the same side of dirty_ratelimit, the systematical
errors in balanced_dirty_ratelimit won't be able to bring
dirty_ratelimit far away.
The balanced_dirty_ratelimit estimation may also be inaccurate near
@limit or @freerun, however is less an issue.
3) since we ultimately want to
- keep the fluctuations of task ratelimit as small as possible
- keep the dirty pages around the setpoint as long time as possible
the update policy used for (2) also serves the above goals nicely:
if for some reason the dirty pages are high (task_ratelimit < dirty_ratelimit),
and dirty_ratelimit is low (dirty_ratelimit < balanced_dirty_ratelimit),
there is no point to bring up dirty_ratelimit in a hurry only to hurt
both the above two goals.
So, we make use of task_ratelimit to limit the update of dirty_ratelimit
in two ways:
1) avoid changing dirty rate when it's against the position control target
(the adjusted rate will slow down the progress of dirty pages going
back to setpoint).
2) limit the step size. task_ratelimit is changing values step by step,
leaving a consistent trace comparing to the randomly jumping
balanced_dirty_ratelimit. task_ratelimit also has the nice smaller
errors in stable state and typically larger errors when there are big
errors in rate. So it's a pretty good limiting factor for the step
size of dirty_ratelimit.
Note that bdi->dirty_ratelimit is always tracking balanced_dirty_ratelimit.
task_ratelimit is merely used as a limiting factor.
Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
|
|
It's all about bdi->dirty_ratelimit, which aims to be (write_bw / N)
when there are N dd tasks.
On write() syscall, use bdi->dirty_ratelimit
============================================
balance_dirty_pages(pages_dirtied)
{
task_ratelimit = bdi->dirty_ratelimit * bdi_position_ratio();
pause = pages_dirtied / task_ratelimit;
sleep(pause);
}
On every 200ms, update bdi->dirty_ratelimit
===========================================
bdi_update_dirty_ratelimit()
{
task_ratelimit = bdi->dirty_ratelimit * bdi_position_ratio();
balanced_dirty_ratelimit = task_ratelimit * write_bw / dirty_rate;
bdi->dirty_ratelimit = balanced_dirty_ratelimit
}
Estimation of balanced bdi->dirty_ratelimit
===========================================
balanced task_ratelimit
-----------------------
balance_dirty_pages() needs to throttle tasks dirtying pages such that
the total amount of dirty pages stays below the specified dirty limit in
order to avoid memory deadlocks. Furthermore we desire fairness in that
tasks get throttled proportionally to the amount of pages they dirty.
IOW we want to throttle tasks such that we match the dirty rate to the
writeout bandwidth, this yields a stable amount of dirty pages:
dirty_rate == write_bw (1)
The fairness requirement gives us:
task_ratelimit = balanced_dirty_ratelimit
== write_bw / N (2)
where N is the number of dd tasks. We don't know N beforehand, but
still can estimate balanced_dirty_ratelimit within 200ms.
Start by throttling each dd task at rate
task_ratelimit = task_ratelimit_0 (3)
(any non-zero initial value is OK)
After 200ms, we measured
dirty_rate = # of pages dirtied by all dd's / 200ms
write_bw = # of pages written to the disk / 200ms
For the aggressive dd dirtiers, the equality holds
dirty_rate == N * task_rate
== N * task_ratelimit_0 (4)
Or
task_ratelimit_0 == dirty_rate / N (5)
Now we conclude that the balanced task ratelimit can be estimated by
write_bw
balanced_dirty_ratelimit = task_ratelimit_0 * ---------- (6)
dirty_rate
Because with (4) and (5) we can get the desired equality (1):
write_bw
balanced_dirty_ratelimit == (dirty_rate / N) * ----------
dirty_rate
== write_bw / N
Then using the balanced task ratelimit we can compute task pause times like:
task_pause = task->nr_dirtied / task_ratelimit
task_ratelimit with position control
------------------------------------
However, while the above gives us means of matching the dirty rate to
the writeout bandwidth, it at best provides us with a stable dirty page
count (assuming a static system). In order to control the dirty page
count such that it is high enough to provide performance, but does not
exceed the specified limit we need another control.
The dirty position control works by extending (2) to
task_ratelimit = balanced_dirty_ratelimit * pos_ratio (7)
where pos_ratio is a negative feedback function that subjects to
1) f(setpoint) = 1.0
2) df/dx < 0
That is, if the dirty pages are ABOVE the setpoint, we throttle each
task a bit more HEAVY than balanced_dirty_ratelimit, so that the dirty
pages are created less fast than they are cleaned, thus DROP to the
setpoints (and the reverse).
Based on (7) and the assumption that both dirty_ratelimit and pos_ratio
remains CONSTANT for the past 200ms, we get
task_ratelimit_0 = balanced_dirty_ratelimit * pos_ratio (8)
Putting (8) into (6), we get the formula used in
bdi_update_dirty_ratelimit():
write_bw
balanced_dirty_ratelimit *= pos_ratio * ---------- (9)
dirty_rate
Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
|
|
No behavior change.
Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
|
|
bdi_position_ratio() provides a scale factor to bdi->dirty_ratelimit, so
that the resulted task rate limit can drive the dirty pages back to the
global/bdi setpoints.
Old scheme is,
|
free run area | throttle area
----------------------------------------+---------------------------->
thresh^ dirty pages
New scheme is,
^ task rate limit
|
| *
| *
| *
|[free run] * [smooth throttled]
| *
| *
| *
..bdi->dirty_ratelimit..........*
| . *
| . *
| . *
| . *
| . *
+-------------------------------.-----------------------*------------>
setpoint^ limit^ dirty pages
The slope of the bdi control line should be
1) large enough to pull the dirty pages to setpoint reasonably fast
2) small enough to avoid big fluctuations in the resulted pos_ratio and
hence task ratelimit
Since the fluctuation range of the bdi dirty pages is typically observed
to be within 1-second worth of data, the bdi control line's slope is
selected to be a linear function of bdi write bandwidth, so that it can
adapt to slow/fast storage devices well.
Assume the bdi control line
pos_ratio = 1.0 + k * (dirty - bdi_setpoint)
where k is the negative slope.
If targeting for 12.5% fluctuation range in pos_ratio when dirty pages
are fluctuating in range
[bdi_setpoint - write_bw/2, bdi_setpoint + write_bw/2],
we get slope
k = - 1 / (8 * write_bw)
Let pos_ratio(x_intercept) = 0, we get the parameter used in code:
x_intercept = bdi_setpoint + 8 * write_bw
The global/bdi slopes are nicely complementing each other when the
system has only one major bdi (indicated by bdi_thresh ~= thresh):
1) slope of global control line => scaling to the control scope size
2) slope of main bdi control line => scaling to the writeout bandwidth
so that
- in memory tight systems, (1) becomes strong enough to squeeze dirty
pages inside the control scope
- in large memory systems where the "gravity" of (1) for pulling the
dirty pages to setpoint is too weak, (2) can back (1) up and drive
dirty pages to bdi_setpoint ~= setpoint reasonably fast.
Unfortunately in JBOD setups, the fluctuation range of bdi threshold
is related to memory size due to the interferences between disks. In
this case, the bdi slope will be weighted sum of write_bw and bdi_thresh.
Given equations
span = x_intercept - bdi_setpoint
k = df/dx = - 1 / span
and the extremum values
span = bdi_thresh
dx = bdi_thresh
we get
df = - dx / span = - 1.0
That means, when bdi_dirty deviates bdi_thresh up, pos_ratio and hence
task ratelimit will fluctuate by -100%.
peter: use 3rd order polynomial for the global control line
CC: Peter Zijlstra <a.p.zijlstra@chello.nl>
Acked-by: Jan Kara <jack@suse.cz>
Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
|
|
Introduce the BDI_DIRTIED counter. It will be used for estimating the
bdi's dirty bandwidth.
CC: Jan Kara <jack@suse.cz>
CC: Michael Rubin <mrubin@google.com>
CC: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
|
|
Discarding slab should be done when node partial > min_partial. Otherwise,
node partial slab may eat up all memory.
Signed-off-by: Alex Shi <alex.shi@intel.com>
Acked-by: Christoph Lameter <cl@linux.com>
Signed-off-by: Pekka Enberg <penberg@kernel.org>
|
|
Correct comment errors, that mistake cpu partial objects number as pages
number, may make reader misunderstand.
Signed-off-by: Alex Shi <alex.shi@intel.com>
Reviewed-by: Christoph Lameter <cl@linux.com>
Signed-off-by: Pekka Enberg <penberg@kernel.org>
|
|
Historically /proc/slabinfo and files under /sys/kernel/slab/* have
world read permissions and are accessible to the world. slabinfo
contains rather private information related both to the kernel and
userspace tasks. Depending on the situation, it might reveal either
private information per se or information useful to make another
targeted attack. Some examples of what can be learned by
reading/watching for /proc/slabinfo entries:
1) dentry (and different *inode*) number might reveal other processes fs
activity. The number of dentry "active objects" doesn't strictly show
file count opened/touched by a process, however, there is a good
correlation between them. The patch "proc: force dcache drop on
unauthorized access" relies on the privacy of dentry count.
2) different inode entries might reveal the same information as (1), but
these are more fine granted counters. If a filesystem is mounted in a
private mount point (or even a private namespace) and fs type differs from
other mounted fs types, fs activity in this mount point/namespace is
revealed. If there is a single ecryptfs mount point, the whole fs
activity of a single user is revealed. Number of files in ecryptfs
mount point is a private information per se.
3) fuse_* reveals number of files / fs activity of a user in a user
private mount point. It is approx. the same severity as ecryptfs
infoleak in (2).
4) sysfs_dir_cache similar to (2) reveals devices' addition/removal,
which can be otherwise hidden by "chmod 0700 /sys/". With 0444 slabinfo
the precise number of sysfs files is known to the world.
5) buffer_head might reveal some kernel activity. With other
information leaks an attacker might identify what specific kernel
routines generate buffer_head activity.
6) *kmalloc* infoleaks are very situational. Attacker should watch for
the specific kmalloc size entry and filter the noise related to the unrelated
kernel activity. If an attacker has relatively silent victim system, he
might get rather precise counters.
Additional information sources might significantly increase the slabinfo
infoleak benefits. E.g. if an attacker knows that the processes
activity on the system is very low (only core daemons like syslog and
cron), he may run setxid binaries / trigger local daemon activity /
trigger network services activity / await sporadic cron jobs activity
/ etc. and get rather precise counters for fs and network activity of
these privileged tasks, which is unknown otherwise.
Also hiding slabinfo and /sys/kernel/slab/* is a one step to complicate
exploitation of kernel heap overflows (and possibly, other bugs). The
related discussion:
http://thread.gmane.org/gmane.linux.kernel/1108378
To keep compatibility with old permission model where non-root
monitoring daemon could watch for kernel memleaks though slabinfo one
should do:
groupadd slabinfo
usermod -a -G slabinfo $MONITOR_USER
And add the following commands to init scripts (to mountall.conf in
Ubuntu's upstart case):
chmod g+r /proc/slabinfo /sys/kernel/slab/*/*
chgrp slabinfo /proc/slabinfo /sys/kernel/slab/*/*
Signed-off-by: Vasiliy Kulikov <segoon@openwall.com>
Reviewed-by: Kees Cook <kees@ubuntu.com>
Reviewed-by: Dave Hansen <dave@linux.vnet.ibm.com>
Acked-by: Christoph Lameter <cl@gentwo.org>
Acked-by: David Rientjes <rientjes@google.com>
CC: Valdis.Kletnieks@vt.edu
CC: Linus Torvalds <torvalds@linux-foundation.org>
CC: Alan Cox <alan@linux.intel.com>
Signed-off-by: Pekka Enberg <penberg@kernel.org>
|
|
* 'for-linus' of git://git.kernel.dk/linux-block:
floppy: use del_timer_sync() in init cleanup
blk-cgroup: be able to remove the record of unplugged device
block: Don't check QUEUE_FLAG_SAME_COMP in __blk_complete_request
mm: Add comment explaining task state setting in bdi_forker_thread()
mm: Cleanup clearing of BDI_pending bit in bdi_forker_thread()
block: simplify force plug flush code a little bit
block: change force plug flush call order
block: Fix queue_flag update when rq_affinity goes from 2 to 1
block: separate priority boosting from REQ_META
block: remove READ_META and WRITE_META
xen-blkback: fixed indentation and comments
xen-blkback: Don't disconnect backend until state switched to XenbusStateClosed.
|
|
* 'slab/urgent' of git://github.com/penberg/linux:
slub: add slab with one free object to partial list tail
|