On Wed, Nov 10, 2010 at 03:16:37PM -0800, Andrew Morton wrote: > On Thu, 04 Nov 2010 08:50:58 +0800 > Shaohua Li <shaohua.li@xxxxxxxxx> wrote: > > > nr_dirty and nr_congested are increased only when page is dirty. So if all pages > > are clean, both them will be zero. In this case, we should not mark the zone > > congested. > > > > Signed-off-by: Shaohua Li <shaohua.li@xxxxxxxxx> > > > > diff --git a/mm/vmscan.c b/mm/vmscan.c > > index b8a6fdc..d31d7ce 100644 > > --- a/mm/vmscan.c > > +++ b/mm/vmscan.c > > @@ -913,7 +913,7 @@ keep_lumpy: > > * back off and wait for congestion to clear because further reclaim > > * will encounter the same problem > > */ > > - if (nr_dirty == nr_congested) > > + if (nr_dirty == nr_congested && nr_dirty != 0) > > zone_set_flag(zone, ZONE_CONGESTED); > > > > free_page_list(&free_pages); > > In a way, this was a really big bug. Reclaim will set the zone as > congested a *lot* - when reclaiming simple, clean pagecache. This is true and you're right, it was a bad mistake on my part. My test machines are tied up at the moment but I intend to run this patch through the same tests as were used to introduce wait_iff_congested to ensure nothing bad has happened. > It does > appear that kswapd will unset it a lot too, so the net effect isn't > obvious. > The unset log is more straight-forward. The flag is cleared when the watermark is met. Granted, there might be still congestion in there but not congestion that a called of alloc_pages() should use congestion_wait() for. > However most of the time, the atomic_read(&nr_bdi_congested[sync]) in > wait_iff_congested() will prevent this bug from causing harm. > Bit of a happy coincidence though. You'd think that if the congestion settting/clearing logic was perfect that nr_bdi_congested[] might become unnecessary. > btw, it's irritating that we have this asymmetry: > > setter: zone_set_flag(zone, ZONE_CONGESTED) > getter: zone_is_reclaim_congested(zone) > I struggled with this. Early versions used a simple getter but I eventually decided to match functions like zone_is_reclaim_locked() and zone_is_oom_locked(). -- Mel Gorman Part-time Phd Student Linux Technology Center University of Limerick IBM Dublin Software Lab -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@xxxxxxxxxx For more info on Linux MM, see: http://www.linux-mm.org/ . Fight unfair telecom policy in Canada: sign http://dissolvethecrtc.ca/ Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>