On Wed, Apr 11, 2012 at 02:06:11PM -0400, Rik van Riel wrote: > On 04/11/2012 01:52 PM, Mel Gorman wrote: > >On Wed, Apr 11, 2012 at 01:17:02PM -0400, Rik van Riel wrote: > > >>Next step: get rid of __GFP_NO_KSWAPD for THP, first > >>in the -mm kernel > >> > > > >Initially the flag was introduced because kswapd reclaimed too > >aggressively. One would like to believe that it would be less of a problem > >now but we must avoid a situation where the CPU and reclaim cost of kswapd > >exceeds the benefit of allocating a THP. > > Since kswapd and the direct reclaim code now use > the same conditionals for calling compaction, > the cost ought to be identical. > kswapd has different retry logic for reclaim and can stay awake if there are continual calls to wakeup_kswapd() setting pgdat->kswapd_max_order and kswapd makes forward progress. It's not identical enough that I would express 100% confidence that it will be free of problems. -- Mel Gorman 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/ . Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/ Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>