On Wed, Nov 30, 2016 at 08:55:30PM -0800, Darren Hart wrote: > On Thu, Nov 24, 2016 at 04:38:08PM +0100, Peter Zijlstra wrote:
> > > In this path the fixup can return -EFAIL as well, so it should drop > > > rtmutex > > > too if it owns it. We should move the rtmutex drop into the fixup > > > functions... > > > > Urgh, so would really like to avoid doing that, I'll have to instantly > > drag it back out again :/ > > Why would you have to drag it back out again? Something else you're working > on? Yeah, the very reason I've been staring at this mess in the first place :-) So I could point you at the patches; and I will, see: https://lkml.kernel.org/r/[email protected] but the TL;DR version is that we must not rt_mutex_unlock() while holding hb->lock, because on RT hb->lock is itself a rt_mutex which gives rise to some very fun prio inversions.
