Re: can't oom-kill zap the victim's memory?

[Vlastimil Babka <vbabka@suse.cz>]

On 10/07/2015 12:43 PM, Tetsuo Handa wrote:
> Vlastimil Babka wrote:
>> On 5.10.2015 16:44, Michal Hocko wrote:
>>> So I can see basically only few ways out of this deadlock situation.
>>> Either we face the reality and allow small allocations (withtout
>>> __GFP_NOFAIL) to fail after all attempts to reclaim memory have failed
>>> (so after even OOM killer hasn't made any progress).
>>
>> Note that small allocations already *can* fail if they are done in the context
>> of a task selected as OOM victim (i.e. TIF_MEMDIE). And yeah I've seen a case
>> when they failed in a code that "handled" the allocation failure with a
>> BUG_ON(!page).
>>
> Did You hit a race described below?

I don't know, I don't even have direct evidence of TIF_MEMDIE being set, 
but OOMs were happening all over the place, and I haven't found another 
reason why the allocation would not be too-small-to-fail otherwise.

> http://lkml.kernel.org/r/201508272249.HDH81838.FtQOLMFFOVSJOH@I-love.SAKURA.ne.jp
>
> Where was the BUG_ON(!page) ? Maybe it is a candidate for adding __GFP_NOFAIL.

Yes, I suggested so:
http://marc.info/?l=linux-kernel&m=144181523115244&w=2

--
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>