Re: [RFC] [PATCH] memcg: fix infinite loop

[KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>]

On Thu, 15 Jan 2009 15:14:38 +0800
Li Zefan <lizf@cn.fujitsu.com> wrote:

> KAMEZAWA Hiroyuki wrote:
> > On Thu, 15 Jan 2009 11:45:57 +0530
> > Balbir Singh <balbir@linux.vnet.ibm.com> wrote:
> > 
> >> * Li Zefan <lizf@cn.fujitsu.com> [2009-01-15 14:07:51]:
> >>
> >>> 1. task p1 is in /memcg/0
> >>> 2. p1 does mmap(4096*2, MAP_LOCKED)
> >>> 3. echo 4096 > /memcg/0/memory.limit_in_bytes
> >>>
> >>> The above 'echo' will never return, unless p1 exited or freed the memory.
> >>> The cause is we can't reclaim memory from p1, so the while loop in
> >>> mem_cgroup_resize_limit() won't break.
> >>>
> >>> This patch fixes it by decrementing retry_count regardless the return value
> >>> of mem_cgroup_hierarchical_reclaim().
> >>>
> >> The problem definitely seems to exist, shouldn't we fix reclaim to
> >> return 0, so that we know progress is not made and retry count
> >> decrements? 
> >>
> > 
> > The behavior is correct. And we already check signal_pending() in the loop.
> > Ctrl-C or SIGALARM will works better than checking retry count.
> 
> But this behavior seems like a regression. Please try it in 2.6.28, you'll see
> it returns EBUSY immediately.
> 
> Looks like the return value of mem_cgroup_hierarchical_reclaim() is buggy ?
> 

This is intentional behavior change by
==
 memcg-make-oom-less-frequently.patch
==

try_to_free_page() returns positive value if try_to_free_page() reclaims at
least 1 pages. It itself doesn't seem to be buggy.

What buggy is resize_limit's retry-out check code, I think.

How about following ?
==
	while (1) {
		if (signal_pending())
			break;
		try to set limit ....
		...
		ret = mem_cgroup_hierarchical_reclaim(memcg,  GFP_KERNEL, false);
		total_progress += ret;	

		if (total_progress > (memcg->res.usage - val) * 2) {
			/*
			 * It seems we reclaimed twice of necessary
			 * pages...this memcg is busy
			 */
			ret = -EBUSY;
			break;
		}
	}
==

Thanks,
-Kame








--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>