(cherry picked from commit 6bc2b856bb7c49f238914d965c0b1057ec78226e)
Set TIF_MEMDIE tsk_thread flag before send kill signal to the
selected thread. This is to fit a usual code sequence and avoid
potential race issue.
Signed-off-by: Weijie Yang <weijie.yang@samsung.com>
Acked-by: David Rientjes <rientjes@google.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Git-commit: 76fc4536faaac9644151fb8a38731c8cf4ab85c4
Git-repo: https://android.googlesource.com/kernel/common.git
Signed-off-by: Srinivasarao P <spathi@codeaurora.org>
The grouping of tasks based on oom_score_adj values change from
one framework to another. This requires corresponding changes in
the threshold values set for almk and per process reclaim.
Introduce sysfs nodes to set threshold adj for process reclaim
and adaptive LMK dynamically.
Change-Id: Ib7565bfd5d2e93aa4ff8fdd20414cac0a0f38bf7
Signed-off-by: Suyog Sarda <ssarda@codeaurora.org>
Signed-off-by: Vinayak Menon <vinmenon@codeaurora.org>
In vmpressure notifier of LMK, shift_adj would have been set
by a previous invocation of notifier, which is not followed by
a lowmem_shrink yet. If this is follwed by a lower vmpressure,
and then by a lowmem_shrink, ALMK still triggers because of the
previous higher vmpressure notification. This is wrong.
Since vmpressure has improved, reset shift_adj to avoid false
adaptive LMK trigger.
CRs-fixed: 893699
Change-Id: I2d77103d7c8f4d8a66e4652cba78e619a7bcef9a
Signed-off-by: Vinayak Menon <vinmenon@codeaurora.org>
Couple of cases were reported few months ago, where the cpu was blocked
on the following call stack for /seconds/ after which the watchdog fires.
test_task_flag(p = 0xE14ABF00, ?)
lowmem_shrink(?, sc = 0xD7A03C04)
shrink_slab(shrink = 0xD7A03C04, nr_pages_scanned = 0, lru_pages = 120)
try_to_free_pages(zonelist = 0xC1116440, ?, ?, ?)
__alloc_pages_nodemask(?, order = 0, ?, nodemask = 0x0)
__do_page_cache_readahead(mapping = 0xEB819364, filp = 0xCC16DC00, offset =
ra_submit(?, ?, ?)
filemap_fault(vma = 0xC105D240, vmf = 0xD7A03DC8)
There weren't any dumps to analyse the case, but this can be a possible
reason. while_each_thread is known to be buggy and can result in the
function looping forever if the task exits, even when protected with
rcu_read_lock. Use for_each_thread instead.
More details on the problems with while_each_thread can be found
at https://lkml.org/lkml/2013/12/2/320
Change-Id: I5eb6e4b463f81142a2a7824db389201357432ec7
Signed-off-by: Vinayak Menon <vinmenon@codeaurora.org>
There were issues reported, where page cache thrashing was
observed because of LMK not killing tasks when required,
resulting in sluggishness and higher app launch latency.
LMK does not kill a task for the following reasons.
1. The free and file pages are above the LMK thresholds
2. LMK tries to pick task with an adj level corresponding
to current thresholds, but fails to do so because of the
absence of tasks in that level.
But sometimes it is better to kill a lower adj task, than thrashing.
And there are cases where the number of file pages are huge, though
we dont thrash, the reclaim process becomes time consuming, since
LMK triggers will be delayed because of higher number of file
pages. Even in such cases, when reclaim path finds it difficult
to reclaim pages, it is better to trigger lmk to free up some memory
faster.
The basic idea here is to make LMK more aggressive dynamically
when such a thrashing scenario is detected.
To detect thrashing, this patch uses vmpressure events.
The values of vmpressure upon which an action has to be taken,
was derived empirically.
This patch also adds tracepoints to validate this feature,
almk_shrink and almk_vmpressure.
Two knobs are available for the user to tune adaptive lmk
behaviour.
/sys/module/lowmemorykiller/parameters/adaptive_lmk - Write
1 to enable the feature, 0 to disable. By default disabled.
/sys/module/lowmemorykiller/parameters/vmpressure_file_min -
This parameter controls the behaviour of LMK when vmpressure
is in the range of 90-94. Adaptive lmk triggers based on number file
pages wrt vmpressure_file_min, when vmpressure is in the range of
90-94. Usually this is a pseudo minfree value, higher than the
highest configured value in minfree array.
Change-Id: I1a08160c35d3e33bdfd1d2c789c288fc07d0f0d3
Signed-off-by: Vinayak Menon <vinmenon@codeaurora.org>
The lowmem_shrink function discounts all the swap cache pages from
the file cache count. The zone aware code also discounts all file
cache pages from a certain zone. This results in some swap cache
pages being discounted twice, which can result in the low memory
killer being unnecessarily aggressive.
Fix the low memory killer to only discount the swap cache pages
once.
Change-Id: I650bbfbf0fbbabd01d82bdb3502b57ff59c3e14f
Signed-off-by: Liam Mark <lmark@codeaurora.org>
Pointer other_free is getting dereferenced without
performing proper NULL checks which may cause issue.
Do proper NULL checks at all points before dereferencing
it.
Change-Id: I88515703d64730e42598ab16136dcce4c18b099c
Signed-off-by: Susheel Khiani <skhiani@codeaurora.org>
When the lowmemory killer starts killing processes, it's useful to
be able to see the state of other memory in the system. Call the
lowmemory notifier to get other clients to dump out their state.
Change-Id: Ie3b1d47149f0108b42b7961fd47a8cc5efe2c590
Signed-off-by: Laura Abbott <lauraa@codeaurora.org>
Leaks from the slab allocator are a very common cause of
out of memory issues and excessive lowmemorykiller killing.
Dump out the slab statistics as well to help with debugging.
Change-Id: Ia5e4eeb66092c76f2b761321c88fa75c4138927c
Signed-off-by: Laura Abbott <lauraa@codeaurora.org>
Add extra debug information to make it easier to both determine
why the lowmemorykiller killed a process and to help find the source
of memory leaks.
Also increase the debug level for "select" statements to help prevent
flooding the log.
Change-Id: I3b6876c5ecdf192ecc271aed3f37579f66d47a08
Signed-off-by: Liam Mark <lmark@codeaurora.org>
This is essentially revert of
dc41a1c61a
Revert "lowmemorykiller: Don't count reserve page twice"
We need to resolve double accounting of the reserve
pages. Lastly, we have reverted this suspecting this
as culprit of some issue. But the nature of this patch
is cleaner and solving real issue of mis-accounting.
Change-Id: I29debaa1bc05703205b323b458f2dfe2938fdfa9
Signed-off-by: Chintan Pandya <cpandya@codeaurora.org>
This reverts commit d73f05ceab.
The patch is causing device crash while running
kernel tests on mainline. The change was to
make LMK less aggressive.
Change-Id: I272bf5ccd54a29de1c770d69af7fea2963f46333
Signed-off-by: Susheel Khiani <skhiani@codeaurora.org>
This is essentially revert of
eeb0f4f35f
Justification:
Zone aware LMK logic already discounts on the reserved
free pages. Hence, present code does double discount
and that leads to improper accounting of the free memory.
Fix this by not discounting reserved_free pages.
CRs-Fixed: 648978
Change-Id: If0536d4a71b36edf0f46f3db723b6d59e36a89ee
Signed-off-by: Chintan Pandya <cpandya@codeaurora.org>
Currenlty most memory reclaim is done through kswapd.
Since kswapd uses a gfp mask of GFP_KERNEL, and because
the lowmemorykiller is zone aware, the lowmemorykiller will
ignore highmem most of the time.
This results in the lowmemorykiller being overly aggressive.
The fix to this issue is to allow the lowmemorykiller to
count highmem when being called by the kswapd if the lowmem
watermarks are satisfied.
Change-Id: I938644584f374763d10d429d835e74daa4854a38
Signed-off-by: Liam Mark <lmark@codeaurora.org>
The lowmemorykiller relies on NR_FILE_PAGES when measureing
the amount of reclaimable memory in the system.
However when swap is enabled swap cache pages are counted
in NR_FILE_PAGES, and swap cache pages aren't as reclaimable
in low memory as file cache pages. Therefore a large swap
cache can result in the lowmemorykiller not running and
an OOM occuring.
In order to ensure the lowmemorykiller properly evaluates the
amount of reclaimable memory don't count the swap cache pages.
Change-Id: I38239283e572f814b277c718eaf6be7f92abacbb
Signed-off-by: Liam Mark <lmark@codeaurora.org>
In certain memory configurations there can be a large number of
CMA pages which are not suitable to satisfy certain memory
requests.
This large number of unsuitable pages can cause the
lowmemorykiller to not kill any tasks because the
lowmemorykiller counts all free pages.
In order to ensure the lowmemorykiller properly evaluates the
free memory only count the free CMA pages if they are suitable
for satisfying the memory request.
Change-Id: I7f06d53e2d8cfe7439e5561fe6e5209ce73b1c90
CRs-fixed: 437016
Signed-off-by: Liam Mark <lmark@codeaurora.org>
There are cases that LMK doesn't run, even when it must run.
It is due to LMK shrinker not considering memory status per zone.
So add LMK parameters(other_free, other_file) tunnig code to
consider target zone of LMK shrinker.
Change-Id: I6f1f8660d5da920a0e3af45a160499965032081d
Git-commit: 22d990a58fc17b3f0155e15eb2dc3efa037bea1c
Git-repo: https://android.googlesource.com/kernel/common/
[ohaugan@codeaurora.org: Fix compilation issues]
Signed-off-by: Olav Haugan <ohaugan@codeaurora.org>
This change is not needed anymore due to a new and improved change to the
android low memory killer that is incompatible with this change. To be able
to apply the new change this change needs to be removed.
This reverts commit cc4baf70665f54305182bfaf98c6236d129e6dad.
Change-Id: I33801faf9ac7371f46cf409009939d853adee95e
Signed-off-by: Olav Haugan <ohaugan@codeaurora.org>
A killed task can stay in the task list long after its
memory has been returned to the system, therefore
ignore any tasks whose mm struct has been freed.
Change-Id: I76394b203b4ab2312437c839976f0ecb7b6dde4e
CRs-fixed: 450383
Signed-off-by: Liam Mark <lmark@codeaurora.org>
The memory reclaim code needs to give time to the system to
return the memory from a killed process otherwise the memory
reclaim code could run continuously, in multiple threads,
which could starve both the watchdog thread and the thread
which is responsible for returning the memory from the
killed process.
Change-Id: Ieded4bfe038ca936247fa4b638070e979b02eaa1
CRs-fixed: 447740
Signed-off-by: Liam Mark <lmark@codeaurora.org>
In certain memory configurations there can be a large number of
CMA pages which are not suitable to satisfy certain memory
requests.
This large number of unsuitable pages can cause the
lowmemorykiller to not kill any tasks because the
lowmemorykiller counts all free pages.
In order to ensure the lowmemorykiller properly evaluates the
free memory only count the free pages which are suitable for
satisfying the memory request.
Change-Id: Iabc3bebc54be9dec7fcbffb94a2fbf18dd684669
Signed-off-by: Liam Mark <lmark@codeaurora.org>
Signed-off-by: Laura Abbott <lauraa@codeaurora.org>
The lowmemorykiller uses the TIF_MEMDIE flag to help ensure it doesn't
kill another task until the memory from the previously killed task has
been returned to the system.
However the lowmemorykiller does not currently look at tasks who do not
have a tasks->mm, but just because a process doesn't have a tasks->mm
does not mean that the task's memory has been fully returned to the
system yet.
In order to prevent the lowmemorykiller from unnecessarily killing
multiple applications in a row the lowmemorykiller has been changed to
ensure that previous killed tasks are no longer in the process list
before attempting to kill another task.
Change-Id: I7d8a8fd39ca5625e6448ed2efebfb621f6e93845
Signed-off-by: Liam Mark <lmark@codeaurora.org>
lowmemorykiller debug messages are inscrutable and mostly useful
for debugging the lowmemorykiller, not explaining why a process
was killed. Make the messages more useful by prefixing them
with "lowmemorykiller: " and explaining in more readable terms
what was killed, who it was killed for, and why it was killed.
The messages now look like:
[ 76.997631] lowmemorykiller: Killing 'droid.gallery3d' (2172), adj 1000,
[ 76.997635] to free 27436kB on behalf of 'kswapd0' (29) because
[ 76.997638] cache 122624kB is below limit 122880kB for oom_score_adj 1000
[ 76.997641] Free memory is -53356kB above reserved
A negative number for free memory above reserved means some of the
reserved memory has been used and is being regenerated by kswapd,
which is likely what called the shrinkers.
Change-Id: I1fe983381e73e124b90aa5d91cb66e55eaca390f
Signed-off-by: Colin Cross <ccross@android.com>
The conversion to use oom_score_adj instead of the deprecated oom_adj
values breaks existing user-space code. Add a config option to convert
oom_adj values written to oom_score_adj values if they appear to be
valid oom_adj values.
Change-Id: I68308125059b802ee2991feefb07e9703bc48549
Signed-off-by: Arve Hjønnevåg <arve@android.com>
As it turns out, the CONFIG_PROFILING interfaces leak a
task struct if the notifier chain returns NOTIFY_OK.. doh.
This patch reworks lowmemkiller to use the new generic task
free notifier chain.
Signed-off-by: San Mehat <san@google.com>
The select...to kill messages are not very useful when not debugging
the lowmemorykiller itself. After the change to check TIF_MEMDIE
instead of using a task notifer this message can also get very
noisy.
Cc: Android Kernel Team <kernel-team@android.com>
Cc: Arve Hjønnevåg <arve@android.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Arve Hjønnevåg <arve@android.com>
Signed-off-by: John Stultz <john.stultz@linaro.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The amount of reserved memory varies between devices. Subtract it
here to reduce the amount of devices specific tuning needed for the
minfree values.
Cc: Android Kernel Team <kernel-team@android.com>
Cc: Arve Hjønnevåg <arve@android.com>
Signed-off-by: Arve Hjønnevåg <arve@android.com>
Signed-off-by: John Stultz <john.stultz@linaro.org>
Acked-by: David Rientjes <rientjes@google.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The maximum oom_score_adj is 1000 and the minimum oom_score_adj is -1000,
so this range can be represented by the signed short type with no
functional change. The extra space this frees up in struct signal_struct
will be used for per-thread oom kill flags in the next patch.
Signed-off-by: David Rientjes <rientjes@google.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Reviewed-by: Michal Hocko <mhocko@suse.cz>
Cc: Anton Vorontsov <anton.vorontsov@linaro.org>
Cc: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The task handoff notifier leaks task_struct since it never gets freed
after the callback returns NOTIFY_OK, which means it is responsible for
doing so.
It turns out the lowmemorykiller actually doesn't need this notifier at
all. It's used to prevent unnecessary killing by waiting for a thread
to exit as a result of lowmem_shrink(), however, it's possible to do
this in the same way the kernel oom killer works by setting TIF_MEMDIE
and avoid killing if we're still waiting for it to exit.
The kernel oom killer will already automatically set TIF_MEMDIE for
threads that are attempting to allocate memory that have a fatal signal.
The thread selected by lowmem_shrink() will have such a signal after the
lowmemorykiller sends it a SIGKILL, so this won't result in an
unnecessary use of memory reserves for the thread to exit.
This has the added benefit that we don't have to rely on
CONFIG_PROFILING to prevent needlessly killing tasks.
Reported-by: Werner Landgraf <w.landgraf@ru.ru>
Cc: stable@vger.kernel.org
Signed-off-by: David Rientjes <rientjes@google.com>
Acked-by: Colin Cross <ccross@android.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Fix compiler warning about the type of the module parameter.
Cc: San Mehat <san@google.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The lowmemorykiller registers an atomic notifier for notfication of when
the task is freed. From this atomic notifier callback, it removes the
atomic notifier via task_free_unregister(). This is incorrect because
atomic_notifier_chain_unregister() calls syncronize_rcu(), which can
sleep, which shouldn't be done from an atomic notifier.
Fix this by registering the notifier during init, and only unregister it
if the lowmemorykiller is unloaded.
Rebased to -next by Paul E. McKenney.
Rebased to -next again by Anton Vorontsov.
Signed-off-by: Rabin Vincent <rabin.vincent@stericsson.com>
Signed-off-by: Christian Bejram <christian.bejram@stericsson.com>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Reported-by: John Stultz <john.stultz@linaro.org>
Signed-off-by: Anton Vorontsov <anton.vorontsov@linaro.org>
Cc: stable <stable@vger.kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
/proc/pid/oom_adj is deprecated and will be removed in August 2012
according to Documentation/feature-removal-schedule.txt. Convert its
usage in the lowmemorykiller to use the new interface, oom_score_adj,
instead.
Signed-off-by: David Rientjes <rientjes@google.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
This was done to resolve some merge issues with the following files that
had changed in both branches:
drivers/staging/rtl8712/rtl871x_sta_mgt.c
drivers/staging/tidspbridge/rmgr/drv_interface.c
drivers/staging/zcache/zcache-main.c
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
task->signal == NULL is not possible, so no need for these checks.
Suggested-by: Oleg Nesterov <oleg@redhat.com>
Reviewed-by: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Anton Vorontsov <anton.vorontsov@linaro.org>
Acked-by: KOSAKI Motohiro <kosaki.motohiro@gmail.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
LMK should not directly check for task->mm. The reason is that the
process' threads may exit or detach its mm via use_mm(), but other
threads may still have a valid mm. To catch this we use
find_lock_task_mm(), which walks up all threads and returns an
appropriate task (with lock held).
Suggested-by: Oleg Nesterov <oleg@redhat.com>
Reviewed-by: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Anton Vorontsov <anton.vorontsov@linaro.org>
Acked-by: KOSAKI Motohiro <kosaki.motohiro@gmail.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Grabbing tasklist_lock has its disadvantages, i.e. it blocks
process creation and destruction. If there are lots of processes,
blocking doesn't sound as a great idea.
For LMK, it is sufficient to surround tasks list traverse with
rcu_read_{,un}lock().
>From now on using force_sig() is not safe, as it can race with an
already exiting task, so we use send_sig() now. As a downside, it
won't kill PID namespace init processes, but that's not what we
want anyway.
Suggested-by: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Anton Vorontsov <anton.vorontsov@linaro.org>
Reviewed-by: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
If a process forked and the child process was killed by the
lowmemorykiller, the lowmemory killer would be disabled until
the parent process reaped the child or it died itself.
Signed-off-by: Arve Hjønnevåg <arve@android.com>
Signed-off-by: Anton Vorontsov <anton.vorontsov@linaro.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
This patch fixes some 80 chatacters limit warnings in the lowmemorykiller.c file
Signed-off-by: Marco Navarra <fromenglish@gmail.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
The arguments to shrink functions have changed, update
lowmem_shrink to match.
Signed-off-by: Colin Cross <ccross@android.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
syphyr
Thierry Strudel
Weijie Yang
LuK1337
Suyog Sarda
Martijn Coenen
Vinayak Menon
Linux Build Service Account
Liam Mark
Susheel Khiani
Laura Abbott
Chintan Pandya
seungho1.park
Olav Haugan
Colin Cross
Arve Hjønnevåg
San Mehat
Dmitry Voytik
David Rientjes
Greg Kroah-Hartman
Paul E. McKenney
Anton Vorontsov
Marco Navarra