On Mon, Jan 23, 2006 at 11:21:57PM -0800, Andrew Morton wrote: > > We have a set of machines that have been usually hanging seconds > > after reporting an oom-killer: gfp_mask=0xd1, order=0. We've tried > > variations of kernels from 2.6.11 through 2.6.16-rc1, with and > > without PAE enabled, all seem to want to hang after the syslog > > oom-killer message. > > This failure is due to a leak in the skbuff_head_cache slab. > > /proc/slabinfo: > skbuff_fclone_cache 259 340 384 10 1 : tunables 54 27 8 > : slabdata 34 34 173 > skbuff_head_cache 2259630 2280390 256 15 1 : tunables 120 60 8 > : slabdata 152026 152026 0
Ok, so I was mistaken, we decided to backdown the kernel on one of the hosts to 2.6.12.6 and the problem seems to have gone away. The skbuff_head_cache slab has been hovering around the 7k-8k range and never going over. Perhaps this leak was due to the introduction of the skbuff_fclone_cache patch set? We did try Herbert Xu's patch from the 22nd of this month (for alloc_skb) on netdev but that did not seem to make a difference with 2.6.16-rc1... Thanks. -mohan - To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html
