Re: [Intel-gfx] [PATCH] drm/i915: Hold struct_mutex during hotplug processing

[Daniel Vetter <daniel@ffwll.ch>]

On Thu, Jul 28, 2011 at 03:50:00PM -0700, Keith Packard wrote:
> On Wed, 27 Jul 2011 09:03:31 -0700, Jesse Barnes <jbarnes@virtuousgeek.org> wrote:
> > On Wed, 27 Jul 2011 02:21:24 -0700
> > Keith Packard <keithp@keithp.com> wrote:
> > > So the work may get executed immediately rather than being run later at
> > > some point?
> > 
> > It sure looks that way... but I don't remember any rule about work
> > queue items having inter dependencies like this.

I've checked the workqueue code and haven't found it to run a work
immediately, it's always queued. Further this problem is very easy to
diagnose: Even without lockdep the scheduler will notice the stuck task
after about 120s and the backtrace should make matters extremely clear. On
the other hand if somebody adds some nice state clobbering in the drm
helper, we have a very hard bug to track down.

Generally modesetting isn't perf critical, so I vote for more locking,
just in case.
-Daniel
-- 
Daniel Vetter
Mail: daniel@ffwll.ch
Mobile: +41 (0)79 365 57 48