[Intel-gfx] [PATCH] drm/i915: Fix __wait_seqno to use true infinite timeouts

[Chris Wilson]

On Fri, Aug 23, 2013 at 09:27:42AM +0200, Daniel Vetter wrote:
> On Fri, Aug 23, 2013 at 3:05 AM, Chris Wilson <chris at chris-wilson.co.uk> wrote:
> > Rather than continue to fix up the timeouts to work around the interface
> > impedence in wait_event_*(), open code the combination of
> > wait_event[_interruptible][_timeout]. And note the code size reduction,
> > and dare say readability?, in doing so..
> >
> > v2: In order to satisfy the debug requirement of logging missed
> > interrupts with the real world requirments of making machines work even
> > if interrupts are hosed, we revert to polling after detecting a missed
> > interrupt.
> >
> > Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
> 
> 
> Yeah, I think this approach should work. A few comments:
> - I think we need a debugfs file to shut the safety quirk off - when
> testing on a machine where we actually miss interrupts it might be
> usful to get the warning output every time.

I did consider a modparam. Exposing it would indeed necessitate some
protection against concurrent modificatin.

> - I'd go for a new bool dev_priv->unreliable_seqno_signalling or so to
> avoid any races due to the rmw cycle you now do on dev_priv->quirks.

There isn't a race during writing, as hangcheck should never be run
concurrently (or at least any concurrent calls filtered out at the start
of the function). The read side is inherently racey.

> - I'd have opted for a faster timeout of the fake irq, but one that rearms.

Whoops, that was a mistake. The intention was to run at 100Hz, do you
want even faster? We could switch to a hrtimer and kill two birds with
one stone (as timer is singleshot only).
 
> Also I'd love to be able to test all this (both the missed irq
> detection stuff and the fake irq) but I don't have a good idea right
> now ... So I guess we need to again hope that it won't break too
> quickly (since it eventually will break again).

Disable the call to ring->get_irq. Perhaps the high word of
dev_priv->stop_rings?
-Chris

-- 
Chris Wilson, Intel Open Source Technology Centre