[Intel-gfx] [PATCH] drm/i915: Do not access stolen memory directly by the CPU, even for error capture

[Ben Widawsky]

On Sun, Jul 20, 2014 at 09:29:55AM +0100, Chris Wilson wrote:
> On Tue, Jul 15, 2014 at 08:30:33PM -0700, Ben Widawsky wrote:
> > On Tue, Jul 15, 2014 at 04:15:08PM +0200, Daniel Vetter wrote:
> > > On Thu, Apr 24, 2014 at 02:47:48PM -0700, Ben Widawsky wrote:
> > > > On Wed, Feb 12, 2014 at 07:18:40PM +0000, Chris Wilson wrote:
> > > > > For stolen pages, since it is verboten to access them directly on many
> > > > > architectures, we have to read them through the GTT aperture. If they
> > > > > are not accessible through the aperture, then we have to abort.
> > > > > 
> > > > > This was complicated by
> > > > > 
> > > > > commit 8b6124a633d8095b0c8364f585edff9c59568a96
> > > > > Author: Chris Wilson <chris at chris-wilson.co.uk>
> > > > > Date:   Thu Jan 30 14:38:16 2014 +0000
> > > > > 
> > > > >     drm/i915: Don't access snooped pages through the GTT (even for error capture)
> > > > > 
> > > > > and the desire to use stolen memory for ringbuffers, contexts and
> > > > > batches in the future.
> > > > 
> > > > I am somewhat unclear as to whether we want to prefer the aperture for
> > > > reading back objects which may be mapped in multiple address spaces.
> > 
> > Can we just ioremap the physical address (at least for error capture)?
> 
> Do you want to hard hang the machine?
> -Chris
> 

What's the latest GEN you can hang the machine with? This is
ioremap_nocache we're talking about, right? I will try it on BDW
tomorrow.

I can't imagine anything but snoop cycles hanging the machine...

-- 
Ben Widawsky, Intel Open Source Technology Center