On Mon 16-05-16 09:31:44, Vlastimil Babka wrote: > On 05/13/2016 04:15 PM, Michal Hocko wrote: > > On Tue 10-05-16 09:36:02, Vlastimil Babka wrote: > > > > > > - should_compact_retry() is only called when should_reclaim_retry() returns > > > false. This means that compaction priority cannot get increased as long > > > as reclaim makes sufficient progress. Theoretically, reclaim should stop > > > retrying for high-order allocations as long as the high-order page doesn't > > > exist but due to races, this may result in spurious retries when the > > > high-order page momentarily does exist. > > > > This is intentional behavior and I would like to preserve it if it is > > possible. For higher order pages should_reclaim_retry retries as long > > as there are some eligible high order pages present which are just hidden > > by the watermark check. So this is mostly to get us over watermarks to > > start carrying about fragmentation. If we race there then nothing really > > terrible should happen and we should eventually converge to a terminal > > state. > > > > Does this make sense to you? > > Yeah it should work, my only worry was that this may get subtly wrong (as > experience shows us) and due to e.g. slightly different watermark checks > and/or a corner-case zone such as ZONE_DMA, should_reclaim_retry() would > keep returning true, even if reclaim couldn't/wouldn't help anything. Then > compaction would be needlessly kept at ineffective priority. watermark check for ZONE_DMA should always fail because it fails even when is completely free to the lowmem reserves. I had a subtle bug in the original code to check highzone_idx rather than classzone_idx but that should the fix has been posted recently: http://lkml.kernel.org/r/1463051677-29418-2-git-send-email-mhocko@xxxxxxxxxx > Also my understanding of the initial compaction priorities is to lower the > latency if fragmentation is just light and there's enough memory. Once we > start struggling, I don't see much point in not switching to the full > compaction priority quickly. That is true but why to compact when there are high order pages and they are just hidden by the watermark check. -- Michal Hocko SUSE Labs -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@xxxxxxxxx. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>