On Tue, Jan 19, 2016 at 01:09:28PM +0100, Daniel Vetter wrote:
> On Thu, Jan 14, 2016 at 10:49:45AM +0000, Arun Siluvery wrote:
> > Pending reset requests are cleared before suspending, they should be picked up
> > after resume when new work is submitted.
> >
> > This is originally added as part of TDR patches for Gen8 from Tomas Elf which
> > are under review, as suggested by Chris this is extracted as a separate patch
> > as it can be useful now.
> >
> > Cc: Mika Kuoppala <mika.kuoppala@xxxxxxxxx>
> > Cc: Chris Wilson <chris@xxxxxxxxxxxxxxxxxx>
> > Signed-off-by: Arun Siluvery <arun.siluvery@xxxxxxxxxxxxxxx>
>
> Pulling in the discussion we had from irc: Imo the right approach is to
> simply wait for gpu reset to finish it's job. Since that could in turn
> lead to a dead gpu (if we're unlucky and init_hw failed) we'd need to do
> that in a loop around gem_idle. And drop dev->struct_mutex in-between.
> E.g.
>
> while (busy) {
> mutex_lock();
> gpu_idle();
> mutex_unlock();
>
> flush_work(reset_work);
> }
Where does the requirement for gpu_idle come from? If there is a global
reset in progress, it cannot queue a request to flush the work and
waiting on the old results will be skipped. So just wait for the global
reset to complete, i.e. flush_work().
-Chris
--
Chris Wilson, Intel Open Source Technology Centre
_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
