Messages in this thread | | | From | Chris Mason <> | Subject | Re: [BUG] fatal hang untarring 90GB file, possibly writeback related. | Date | Wed, 27 Apr 2011 12:54:26 -0400 |
| |
Excerpts from James Bottomley's message of 2011-04-27 12:50:00 -0400: > On Wed, 2011-04-27 at 12:33 -0400, Chris Mason wrote: > > Probably easier to debug with a sysrq-l and sysrq-w. > > Yes, just enabling them for the repro ... I didn't echo 1 > > /proc/sys/kernel/sysrq first time around > > > If you get stuck > > on the filesystem, it is probably waiting on ram, which it probably > > can't get because kswapd is spinning. Eventually everyone backs up > > waiting for the transaction that never ends. If we're really lucky it > > is just GFP_KERNEL where it should NOFS. > > > > Since you're often stuck in different spots inside shrink_slab, we're > > probably not stuck on a lock. But, trying with lock debugging, lockdep > > enabled and preempt on is a good idea to rule out locking mistakes. > > > > Does the fedora debug kernel enable preempt? > > No ... I've taken the Fedora 15 kernel config directly from their > install CD for this (with a bit of munging to get it to work on -rc4)
Ok, I'd try turning it on so we catch the sleeping with a spinlock held case better.
-chris
| |