Messages in this thread | | | Date | Tue, 26 Nov 2013 15:31:06 +1100 | From | NeilBrown <> | Subject | Re: kernel BUG at drivers/md/raid5.c:693! |
| |
On Mon, 25 Nov 2013 10:53:23 +0100 Peter Zijlstra <peterz@infradead.org> wrote:
> On Mon, Nov 25, 2013 at 10:35:22AM +1100, NeilBrown wrote: > > Ingo/Peter: is it considered OK to call wake_up while holding a spinlock? > > Yes, very much so. Doing a wakeup isn't _that_ expensive.
Oh good. Thanks.
> > > Could "sleeping spinlocks" affect this at all? (some sample stack traces are > > below). > > Not entirely sure, are you referencing to -rt where me make spinlock_t > pi-mutexes?
I'm not sure either. Just stabbing in the dark really.
The stack trace in the previous email seemed to suggest that a process was blocking inside a wake_up call, but it wasn't at all conclusive. And I've seen a few wake_ups in other stack traces which seem to be connected with other deadlock. Probably some sort of co-incidence.
Thanks, NeilBrown [unhandled content-type:application/pgp-signature] | |