Re: [PATCH 4/5] drm/i915/irq: Keep GSE interrupts masked if display is not present

["Ville Syrjälä" <ville.syrjala@linux.intel.com>]

On Fri, Nov 06, 2026 at 11:25:09AM +0200, Jani Nikula wrote:
> On Thu, 05 Nov 2026, Ville Syrjala <ville.syrjala@linux.intel.com> wrote:
> > From: Ville Syrjälä <ville.syrjala@linux.intel.com>
> >
> > Display is the only current user of GSE interrupts. Keep the
> > GSE interrupt masked on icl+ if there is no display.
> >
> > Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
> 
> xe could use the same treatment.

Didn't even realize xe was enabling this. Ugh. The GSE irq
handler is currently called xe_disaplay_irq_enable()? What?

xe's interrupt handlers also look quite expensive because
all the master bits are checked in the non-inlined sub handler
functions. So no matter how many actual interrupt bits are set
you always pay the cost of a dozen redundant function calls.
At least it filters out the entire tile if nothing happened
on it.

I suppose the "just function calls in the top level handler"
approach does keep the code looking fairly neat. But for
efficiency those should probably be static inlines so you
don't pay the extra function call cost for every bit you
want to check. Might have to think about using that approach
for the display irq code actually...

I also don't know why xe has all those weird probe_display
checks in the display irq stuff. Why would display irqs
even be unmasked if there is no display?

Anyways, for now I'll just stick to fixing up the GSE
stuff in xe.

> 
> Reviewed-by: Jani Nikula <jani.nikula@intel.com>
> 
> > ---
> >  drivers/gpu/drm/i915/i915_irq.c | 7 ++++++-
> >  1 file changed, 6 insertions(+), 1 deletion(-)
> >
> > diff --git a/drivers/gpu/drm/i915/i915_irq.c b/drivers/gpu/drm/i915/i915_irq.c
> > index 823d97284d78..5476472d33bd 100644
> > --- a/drivers/gpu/drm/i915/i915_irq.c
> > +++ b/drivers/gpu/drm/i915/i915_irq.c
> > @@ -36,6 +36,7 @@
> >  #include <drm/intel/display_parent_interface.h>
> >  #include <drm/intel/intel_gmd_interrupt_regs.h>
> >  
> > +#include "display/intel_display_device.h"
> >  #include "display/intel_display_irq.h"
> >  
> >  #include "gt/intel_breadcrumbs.h"
> > @@ -727,9 +728,13 @@ static void gen8_irq_postinstall(struct drm_i915_private *dev_priv)
> >  
> >  static void gen11_gu_misc_irq_postinstall(struct drm_i915_private *dev_priv)
> >  {
> > +	struct intel_display *display = dev_priv->display;
> >  	struct intel_gt *gt = to_gt(dev_priv);
> >  	struct intel_uncore *uncore = gt->uncore;
> > -	u32 gu_misc_masked = GEN11_GU_MISC_GSE;
> > +	u32 gu_misc_masked = 0;
> > +
> > +	if (intel_display_device_present(display))
> > +		gu_misc_masked |= GEN11_GU_MISC_GSE;
> >  
> >  	gen2_irq_init(uncore, GEN11_GU_MISC_IRQ_REGS, ~gu_misc_masked, gu_misc_masked);
> >  }
> 
> -- 
> Jani Nikula, Intel

-- 
Ville Syrjälä
Intel