memcg: fix oom kill behavior
authorKAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Wed, 10 Mar 2010 23:22:39 +0000 (15:22 -0800)
committerLinus Torvalds <torvalds@linux-foundation.org>
Fri, 12 Mar 2010 23:52:38 +0000 (15:52 -0800)
commit867578cbccb0893cc14fc29c670f7185809c90d6
treee4d0cefac265fc64399223bc82ed714a88ebe20c
parent0263c12c12ccc90edc9d856fa839f8936183e6d1
memcg: fix oom kill behavior

In current page-fault code,

handle_mm_fault()
-> ...
-> mem_cgroup_charge()
-> map page or handle error.
-> check return code.

If page fault's return code is VM_FAULT_OOM, page_fault_out_of_memory() is
called.  But if it's caused by memcg, OOM should have been already
invoked.

Then, I added a patch: a636b327f731143ccc544b966cfd8de6cb6d72c6.  That
patch records last_oom_jiffies for memcg's sub-hierarchy and prevents
page_fault_out_of_memory from being invoked in near future.

But Nishimura-san reported that check by jiffies is not enough when the
system is terribly heavy.

This patch changes memcg's oom logic as.
 * If memcg causes OOM-kill, continue to retry.
 * remove jiffies check which is used now.
 * add memcg-oom-lock which works like perzone oom lock.
 * If current is killed(as a process), bypass charge.

Something more sophisticated can be added but this pactch does
fundamental things.
TODO:
 - add oom notifier
 - add permemcg disable-oom-kill flag and freezer at oom.
 - more chances for wake up oom waiter (when changing memory limit etc..)

Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Tested-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: David Rientjes <rientjes@google.com>
Signed-off-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
include/linux/memcontrol.h
mm/memcontrol.c
mm/oom_kill.c