On Wed, Jan 11, 2012 at 12:44 AM, Mel Gorman <mgorman@xxxxxxx> wrote: > On Tue, Jan 10, 2012 at 11:58:03PM +0800, Hillf Danton wrote: >> From: Hillf Danton <dhillf@xxxxxxxxx> >> [PATCH] mm: vmscan: fix setting reclaim mode >> >> The comment says, initially assume we are entering either lumpy reclaim or >> reclaim/compaction, and depending on the reclaim order, we will either set the >> sync mode or just reclaim order-0 pages later. >> >> On other hand, order-0 reclaim, instead of sync reclaim, is expected when >> under memory pressure, but the check for memory pressure is incorrect, >> leading to sync reclaim at low reclaim priorities. >> >> And the result is sync reclaim is set for high priorities. >> > > RECLAIM_MODE_SYNC is only set for RECLAIM_MODE_LUMPYRECLAIM. Even when > using RECLAIM_MODE_LUMPYRECLAIM, it should only be set when reclaim > is under memory pressure and failing to reclaim the necessry pages > (priority < DEF_PRIORITY - 2). Once in symc reclaim, reclaim will call > wait_on_page_writeback() on dirty pages which potentially leads to > significant stalls (one of the reasons why RECLAIM_MODE_LUMPYRECLAIM > sucks and why compaction is preferred). Your patch means sync reclaim > is used even when priority == DEF_PRIORITY. This is unexpected. > > Your changelog really needs to explain what the problem is that you > have encountered and why this patch fixes it. It's not like some of > your other patches which were minor performance optimisations that > were self-evident. > Hi Mel To avoid another step in wrong direction, I'd goto bed now, it is about 1:09 AM. When sane again, I will try to understand your comment. Best Hillf -- 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/ . Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/ Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>