On Thu, Mar 31, 2011 at 03:32:46PM -0500, Christoph Lameter wrote: > On Thu, 31 Mar 2011, Tim Gardner wrote: > > > Christoph, > > > > commit c2222d66adedc567318a7c288726363daeb3374b in 2.6.32.y (commit > > aa45484031ddee09b06350ab8528bfe5b2c76d1c upstream) appears to have > > introduced > > a live-lock regression on UP machines running an SMP kernel. > > > > Pawel Jasnos has bisected his issue to that commit, which is consistent with > > the kernel version we've released where things started not working so well. > > > > https://bugs.launchpad.net/ubuntu/maverick/+source/linux/+bug/719446/comments/18 > > > > Any thoughts? > > Well we have discussed this repeatedly and the upstream solution was to > reduce the vm statistics thresholds. My contribution to the patch > mentioned above was only the zone_page_state_snapshot function. The > rest is from Mel. > > AFAICT The fixup to this was 88f5acf88ae6a9778f6d25d0d5d7ec2d57764a97 > > mm: page allocator: adjust the per-cpu counter threshold when memory is low >
Agreed, this was the fixup for reports similar, if not identical, to what is described in ubuntu#719446 but was not backported to 2.6.32.y. If you find the patch helps, it may be possible to still get it merged. Thanks. -- Mel Gorman SUSE Labs _______________________________________________ stable mailing list [email protected] http://linux.kernel.org/mailman/listinfo/stable
