[Intel-gfx] [PATCH] drm/i915: Disable stolen memory when DMAR is active

[Jani Nikula]

On Wed, 05 Mar 2014, Daniel Vetter <daniel at ffwll.ch> wrote:
> On Wed, Mar 05, 2014 at 03:32:54PM +0000, Chris Wilson wrote:
>> On Wed, Mar 05, 2014 at 04:17:17PM +0100, Daniel Vetter wrote:
>> > On Fri, Feb 28, 2014 at 12:14:02PM +0000, Chris Wilson wrote:
>> > > We have reports of heavy screen corruption if we try to use the stolen
>> > > memory reserved by the BIOS whilst the DMA-Remapper is active. This
>> > > quirk may be only specific to a few machines or BIOSes, but first lets
>> > > apply the big hammer and always disable use of stolen memory when DMAR
>> > > is active.
>> > > 
>> > > Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=68535
>> > > Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
>> > > Cc: Daniel Vetter <daniel.vetter at ffwll.ch>
>> > > ---
>> > >  drivers/gpu/drm/i915/i915_gem_stolen.c | 14 ++++++++++++++
>> > >  1 file changed, 14 insertions(+)
>> > > 
>> > > diff --git a/drivers/gpu/drm/i915/i915_gem_stolen.c b/drivers/gpu/drm/i915/i915_gem_stolen.c
>> > > index 984ada1b0084..13a849afe479 100644
>> > > --- a/drivers/gpu/drm/i915/i915_gem_stolen.c
>> > > +++ b/drivers/gpu/drm/i915/i915_gem_stolen.c
>> > > @@ -325,11 +325,25 @@ void i915_gem_cleanup_stolen(struct drm_device *dev)
>> > >  	drm_mm_takedown(&dev_priv->mm.stolen);
>> > >  }
>> > >  
>> > > +static bool need_dmar_wa(struct drm_device *dev)
>> > > +{
>> > > +#ifdef CONFIG_INTEL_IOMMU
>> > > +	if (INTEL_INFO(dev)->gen >= 6 && intel_iommu_gfx_mapped)
>> > 
>> > Why the gen check in here? Without that we could just check
>> > intel_iommu_gfx_mapped and shovel the #ifdef into stolen init. No need to
>> > spread it out like that imo.
>> 
>> It's almost as if it was cut'n'paste from another VTd w/a. I'm worried
>> that this is too big a hammer. I would have expected more complaints if
>> VTd was this broken since 3.9. Otoh, it is a severe bug.
>
> We've had piles of bugs in our stolen detection, combined with very few
> people having vt-d capable machines (thanks to intel's differentation) and
> probably even fewer having it enabled I'm not so much surprised.
>
> On top of that lots of noise on the interwebz that dmar is trouble for
> intel gfx on linux, so people probably just disable it and carry on ...

So what's the verdict on this one?

BR,
Jani.


-- 
Jani Nikula, Intel Open Source Technology Center