From: Walter Alejandro Iglesias <wai@roquesor.com>
Subject: Re: Alder Lake ghosting after wscons screen burner (derived from "Re: Intel GPU, ghosting after zzz or ZZZ")
To: Vitaliy Makkoveev <mvs@openbsd.org>
Cc: Mark Kettenis <mark.kettenis@xs4all.nl>, Jonathan Gray <jsg@jsg.id.au>, tech@openbsd.org
Date: Thu, 28 May 2026 21:20:51 +0200

+kettenis@

On Thu, May 28, 2026 at 09:05:08PM +0300, Vitaliy Makkoveev wrote:
> On Thu, May 28, 2026 at 12:48:37PM +0200, Walter Alejandro Iglesias wrote:
> > On Wed, May 27, 2026 at 08:51:29PM +0300, Vitaliy Makkoveev wrote:
> > > On Wed, May 27, 2026 at 07:27:47PM +0200, Walter Alejandro Iglesias wrote:
> > > > 
> > > > I found a function defined in ../drm/drm_fb_helper.c,
> > > > drm_fb_helper_restore_fbdev_mode_unlocked() which restores the entire
> > > > fbdev display mode (including size).
> > > > 
> > > 
> > > Thanks, this diff completely fixes my issue. ok mvs@
> > > 
> > > > Index: i915_driver.c
> > > > ===================================================================
> > > > RCS file: /cvs/src/sys/dev/pci/drm/i915/i915_driver.c,v
> > > > diff -u -p -u -p -r1.30 i915_driver.c
> > > > --- i915_driver.c	30 Apr 2026 04:33:06 -0000	1.30
> > > > +++ i915_driver.c	27 May 2026 13:32:34 -0000
> > > > @@ -2154,8 +2154,14 @@ inteldrm_burner_cb(void *arg1)
> > > >  	struct drm_i915_private *dev_priv = arg1;
> > > >  	struct drm_device *dev = &dev_priv->drm;
> > > >  	struct drm_fb_helper *helper = dev->fb_helper;
> > > > +	struct rasops_info *ri = &dev_priv->ro;
> > > >  
> > > >  	drm_fb_helper_blank(dev_priv->burner_fblank, helper->info);
> > > > +
> > > > +	if (dev_priv->burner_fblank == FB_BLANK_UNBLANK) {
> > > > +		drm_fb_helper_restore_fbdev_mode_unlocked(helper);
> > > > +		rasops_show_screen(ri, ri->ri_active, 0, NULL, NULL);
> > > > +	}
> > > >  }
> > > >  
> > > >  int
> > 
> > 
> > To confirm your Alder Lake suffers from two different issues.  I'd like
> > to know what happens if we remove rasops_show_screen from the diff.
> > This should avoid the Alder Lake half-line on top but not the ghosting.
> > 
> 
> No ghosting, no half-line. On both machines.

Then that function alone does the trick.  It does a full reset.

If kettenis@ gives us at least some clue of where is the right place to
fix this, I'll try to investigate a new way.


> 
> > 
> > Index: i915_driver.c
> > ===================================================================
> > RCS file: /cvs/src/sys/dev/pci/drm/i915/i915_driver.c,v
> > diff -u -p -u -p -r1.30 i915_driver.c
> > --- i915_driver.c	30 Apr 2026 04:33:06 -0000	1.30
> > +++ i915_driver.c	28 May 2026 10:41:46 -0000
> > @@ -2156,6 +2156,8 @@ inteldrm_burner_cb(void *arg1)
> >  	struct drm_fb_helper *helper = dev->fb_helper;
> >  
> >  	drm_fb_helper_blank(dev_priv->burner_fblank, helper->info);
> > +	if (dev_priv->burner_fblank == FB_BLANK_UNBLANK && helper)
>                                                            ^^^^^^
>  This check is useless, we have `helper' dereference in previous line.

Yeah, right.


> > +		drm_fb_helper_restore_fbdev_mode_unlocked(helper);
> >  }
> >  
> >  int
> > 
> > 
> > 
> > 
> > -- 
> > Walter

-- 
Walter