Re: [PATCH 1/2] drm/i915: Track fence region ID in plane state

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

On Thu, Apr 02, 2026 at 05:54:05PM +0300, Jani Nikula wrote:
> On Tue, 31 Mar 2026, Ville Syrjala <ville.syrjala@linux.intel.com> wrote:
> > @@ -203,7 +204,7 @@ intel_fb_pin_to_ggtt(const struct drm_framebuffer *fb,
> >  		ret = 0;
> >  
> >  		if (vma->fence)
> > -			*out_flags |= PLANE_HAS_FENCE;
> > +			*out_fence_id |= vma->fence->id;
> 
> That should obviously be an assignment rather than OR.
> 
> I was reviewing the recently enabled Sashiko LLM reviews, and it had
> spotted this one [1].
> 
> > @@ -314,7 +314,7 @@ int intel_fbdev_driver_fbdev_probe(struct drm_fb_helper *helper,
> >  				   fb->min_alignment, 0,
> >  				   intel_fb_view_vtd_guard(&fb->base, &fb->normal_view,
> >  							   DRM_MODE_ROTATE_0),
> > -				   false, &flags);
> > +				   &fence_id);
> 
> This might also change behaviour as previously uses_fence == false.

Hmm. Seems harmless but I guess we don't need the fence here at all.
If a fence is needed i915_vma_pin_iomap() will anyway grab one.
And even if we didn't iomap the thing intel_plane_pin_fb() would still
grab a fence if the display hardware needs it.

> 
> BR,
> Jani.
> 
> 
> 
> [1] https://sashiko.dev/#/patchset/20260331162138.19258-1-ville.syrjala%40linux.intel.com
> 
> 
> 
> -- 
> Jani Nikula, Intel

-- 
Ville Syrjälä
Intel