Re: [PATCH] drm/i915: WARN if interrupts aren't on in en/disable_pipestat

[Jesse Barnes <jbarnes@virtuousgeek.org>]

On Thu, 4 Sep 2014 18:59:55 +0200
Daniel Vetter <daniel.vetter@ffwll.ch> wrote:

> On Thu, Sep 4, 2014 at 6:24 PM, Jesse Barnes <jbarnes@virtuousgeek.org> wrote:
> > On Thu, 4 Sep 2014 17:59:18 +0200
> > Daniel Vetter <daniel.vetter@ffwll.ch> wrote:
> >
> >> On Thu, Aug 28, 2014 at 1:00 AM, Jesse Barnes <jbarnes@virtuousgeek.org> wrote:
> >> >> diff --git a/drivers/gpu/drm/i915/i915_irq.c b/drivers/gpu/drm/i915/i915_irq.c
> >> >> index 9eb303c1b621..76bc4d0de5a4 100644
> >> >> --- a/drivers/gpu/drm/i915/i915_irq.c
> >> >> +++ b/drivers/gpu/drm/i915/i915_irq.c
> >> >> @@ -589,6 +589,7 @@ __i915_enable_pipestat(struct drm_i915_private *dev_priv, enum pipe pipe,
> >> >>       u32 pipestat = I915_READ(reg) & PIPESTAT_INT_ENABLE_MASK;
> >> >>
> >> >>       assert_spin_locked(&dev_priv->irq_lock);
> >> >> +     WARN_ON(!intel_irqs_enabled(dev_priv));
> >> >>
> >> >>       if (WARN_ONCE(enable_mask & ~PIPESTAT_INT_ENABLE_MASK ||
> >> >>                     status_mask & ~PIPESTAT_INT_STATUS_MASK,
> >> >> @@ -615,6 +616,7 @@ __i915_disable_pipestat(struct drm_i915_private *dev_priv, enum pipe pipe,
> >> >>       u32 pipestat = I915_READ(reg) & PIPESTAT_INT_ENABLE_MASK;
> >> >>
> >> >>       assert_spin_locked(&dev_priv->irq_lock);
> >> >> +     WARN_ON(!intel_irqs_enabled(dev_priv));
> >> >>
> >> >>       if (WARN_ONCE(enable_mask & ~PIPESTAT_INT_ENABLE_MASK ||
> >> >>                     status_mask & ~PIPESTAT_INT_STATUS_MASK,
> >> >
> >> > Yeah looks good, wonder if it'll trigger any new warnings.
> >>
> >> It will blow up in a bunch of postinstall hooks, just like the one for
> >> ilk. At least without my patch to shuffle the pm._irqs_disabled
> >> assignment around.
> >>
> >> > Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
> >>
> >> ... so does that count as an implicit r-b on my other patch?
> >
> > Sure, though didn't Jani find some issues with it?
> 
> QA claims that both my and your patch break the display on hsw, bdw
> and snb or something like that. Which either means I'm blind (since
> either patch should only affect ilk in a functional way) or they're
> doing something really strange. I have no idea what's actually going
> on there, but we have a regular stream of reports of this one here ...

Yeah I agree the results are weird; don't see how we'd affect BDW
display, esp with my patch, so I'd say go ahead with either.  Though if
you do mine I guess you'd drop the above.

-- 
Jesse Barnes, Intel Open Source Technology Center