Re: [PATCH v3 1/2] drm/i915: Use the correct IRQ during resume

[Daniel Vetter <daniel@ffwll.ch>]

On Wed, Jun 30, 2021 at 4:18 PM Thomas Zimmermann <tzimmermann@suse.de> wrote:
>
> Hi
>
> Am 30.06.21 um 12:49 schrieb Daniel Vetter:
> > On Wed, Jun 30, 2021 at 11:52:27AM +0200, Thomas Zimmermann wrote:
> >> The code in xcs_resume() probably didn't work as intended. It uses
> >> struct drm_device.irq, which is allocated to 0, but never initialized
> >> by i915 to the device's interrupt number.
> >>
> >> v3:
> >>      * also use intel_synchronize_hardirq() at another callsite
> >> v2:
> >>      * wrap irq code in intel_synchronize_hardirq() (Ville)
> >>
> >> Signed-off-by: Thomas Zimmermann <tzimmermann@suse.de>
> >> Fixes: 536f77b1caa0 ("drm/i915/gt: Call stop_ring() from ring resume, again")
> >> Cc: Chris Wilson <chris@chris-wilson.co.uk>
> >> Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com>
> >> Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
> >> Cc: Rodrigo Vivi <rodrigo.vivi@intel.com>
> >> Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
> >> Cc: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
> >> Cc: Lucas De Marchi <lucas.demarchi@intel.com>
> >> ---
> >>   drivers/gpu/drm/i915/gt/intel_engine_cs.c       | 2 +-
> >>   drivers/gpu/drm/i915/gt/intel_ring_submission.c | 2 +-
> >>   drivers/gpu/drm/i915/i915_irq.c                 | 5 +++++
> >>   drivers/gpu/drm/i915/i915_irq.h                 | 1 +
> >>   4 files changed, 8 insertions(+), 2 deletions(-)
> >>
> >> diff --git a/drivers/gpu/drm/i915/gt/intel_engine_cs.c b/drivers/gpu/drm/i915/gt/intel_engine_cs.c
> >> index 88694822716a..5ca3d1664335 100644
> >> --- a/drivers/gpu/drm/i915/gt/intel_engine_cs.c
> >> +++ b/drivers/gpu/drm/i915/gt/intel_engine_cs.c
> >> @@ -1229,7 +1229,7 @@ bool intel_engine_is_idle(struct intel_engine_cs *engine)
> >>              return true;
> >>
> >>      /* Waiting to drain ELSP? */
> >> -    synchronize_hardirq(to_pci_dev(engine->i915->drm.dev)->irq);
> >> +    intel_synchronize_hardirq(engine->i915);
> >>      intel_engine_flush_submission(engine);
> >>
> >>      /* ELSP is empty, but there are ready requests? E.g. after reset */
> >> diff --git a/drivers/gpu/drm/i915/gt/intel_ring_submission.c b/drivers/gpu/drm/i915/gt/intel_ring_submission.c
> >> index 5d42a12ef3d6..1b5a22a83db6 100644
> >> --- a/drivers/gpu/drm/i915/gt/intel_ring_submission.c
> >> +++ b/drivers/gpu/drm/i915/gt/intel_ring_submission.c
> >> @@ -185,7 +185,7 @@ static int xcs_resume(struct intel_engine_cs *engine)
> >>                   ring->head, ring->tail);
> >>
> >>      /* Double check the ring is empty & disabled before we resume */
> >> -    synchronize_hardirq(engine->i915->drm.irq);
> >> +    intel_synchronize_hardirq(engine->i915);
> >>      if (!stop_ring(engine))
> >>              goto err;
> >>
> >> diff --git a/drivers/gpu/drm/i915/i915_irq.c b/drivers/gpu/drm/i915/i915_irq.c
> >> index 7d0ce8b9f8ed..2203dca19895 100644
> >> --- a/drivers/gpu/drm/i915/i915_irq.c
> >> +++ b/drivers/gpu/drm/i915/i915_irq.c
> >> @@ -4575,3 +4575,8 @@ void intel_synchronize_irq(struct drm_i915_private *i915)
> >>   {
> >>      synchronize_irq(to_pci_dev(i915->drm.dev)->irq);
> >>   }
> >> +
> >> +void intel_synchronize_hardirq(struct drm_i915_private *i915)
> >> +{
> >> +    synchronize_hardirq(to_pci_dev(i915->drm.dev)->irq);
> >
> > I honestly think the hardirq here is about as much cargo-culted as using
> > the wrong irq number.
> >
> > I'd just use intel_synchronize_irq in both places and see whether CI
> > complains, then go with that.
>
> Well, ok. I don't think I have Sandybridge HW available. Would the Intel
> CI infrastructure catch any problems with such a change?

If there's anything obvious busted with it it should catch it (like if
we end up calling synchronize_irq from softirq context, where only
synchronize_hardirq is allowed, but I don't think that's the case).

And if I'm wrong we know what kind of comment to put there to explain
why things are different.
-Daniel

> Best regards
> Thomas
>
> > -Daniel
> >
> >> +}
> >> diff --git a/drivers/gpu/drm/i915/i915_irq.h b/drivers/gpu/drm/i915/i915_irq.h
> >> index db34d5dbe402..e43b6734f21b 100644
> >> --- a/drivers/gpu/drm/i915/i915_irq.h
> >> +++ b/drivers/gpu/drm/i915/i915_irq.h
> >> @@ -94,6 +94,7 @@ void intel_runtime_pm_disable_interrupts(struct drm_i915_private *dev_priv);
> >>   void intel_runtime_pm_enable_interrupts(struct drm_i915_private *dev_priv);
> >>   bool intel_irqs_enabled(struct drm_i915_private *dev_priv);
> >>   void intel_synchronize_irq(struct drm_i915_private *i915);
> >> +void intel_synchronize_hardirq(struct drm_i915_private *i915);
> >>
> >>   int intel_get_crtc_scanline(struct intel_crtc *crtc);
> >>   void gen8_irq_power_well_post_enable(struct drm_i915_private *dev_priv,
> >> --
> >> 2.32.0
> >>
> >
>
> --
> Thomas Zimmermann
> Graphics Driver Developer
> SUSE Software Solutions Germany GmbH
> Maxfeldstr. 5, 90409 Nürnberg, Germany
> (HRB 36809, AG Nürnberg)
> Geschäftsführer: Felix Imendörffer
>


-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch