From: Roger Tsang <roger.tsang@gm...> - 2006-07-15 02:34:54
Since developer CVS access is blocked I'll just post here until we get
cluster/ssi/cfs/read.c:cfs_async_handler_read() is missing a kunmap()
call in the if (status == -ECFS_TRYAGAIN) block which is likely
traversed during CFS failover.
>From preliminary tests this fix cures the problem with CFS failover
node using 100% CPU in IO Wait but idle backing devices and/or
kirqsoftd using 100% CPU after CFS failover.