Re: [PATCH 12/14] mm, oom: protect !costly allocations some more

[Joonsoo Kim]

2016-05-04 17:53 GMT+09:00 Michal Hocko <mhocko@xxxxxxxxxx>:
> On Wed 04-05-16 15:01:24, Joonsoo Kim wrote:
>> On Wed, Apr 20, 2016 at 03:47:25PM -0400, Michal Hocko wrote:
> [...]
>
> Please try to trim your responses it makes it much easier to follow the
> discussion

Okay.

>> > +static inline bool
>> > +should_compact_retry(unsigned int order, enum compact_result compact_result,
>> > +                enum migrate_mode *migrate_mode,
>> > +                int compaction_retries)
>> > +{
>> > +   if (!order)
>> > +           return false;
>> > +
>> > +   /*
>> > +    * compaction considers all the zone as desperately out of memory
>> > +    * so it doesn't really make much sense to retry except when the
>> > +    * failure could be caused by weak migration mode.
>> > +    */
>> > +   if (compaction_failed(compact_result)) {
>>
>> IIUC, this compaction_failed() means that at least one zone is
>> compacted and failed. This is not same with your assumption in the
>> comment. If compaction is done and failed on ZONE_DMA, it would be
>> premature decision.
>
> Not really, because if other zones are making some progress then their
> result will override COMPACT_COMPLETE

Think about the situation that DMA zone fails to compact and
the other zones are deferred or skipped. In this case, COMPACT_COMPLETE
will be returned as a final result and should_compact_retry() return false.
I don't think that it means all the zones are desperately out of memory.

Thanks.