[Intel-gfx] DMAR regression in 2.6.37

[Dave Airlie]

On Mon, Jan 10, 2011 at 12:46 PM, Dave Airlie <airlied at gmail.com> wrote:
> Hi guys,
>
> I bisected down to
>
> commit 450f2b3d51025a1749b694ee13f0e4e23ed58750
> Author: Daniel Vetter <daniel.vetter at ffwll.ch>
> Date:   Sat Sep 11 23:48:25 2010 +0200
>
>    intel-gtt: generic (insert|remove)_entries for g33/i965
>
>    Like for the i915.
>
>    Signed-off-by: Daniel Vetter <daniel.vetter at ffwll.ch>
>    Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
>
> Before this point I get no DMAR, after this I get DMAR, which isn't
> surprising since this commit is the one that removes the calls to the
> map/unmap via PCI layer calls.
>

This one is proving quite hard to nail down, I've been reading over
the codepaths and can't spot the major problem with it.

The code seq in these patches is a bit off in places, i.e. it removes
scratch page stuff but then still needs it in the i915_remove_entries
patch, so I've moved the bisection around a bit, but I can't get a
DMAR error with the code prior to this AGP rework and I can after the
patch above.

I'll keep digging and hope one of you guys has a good idea.

Dave.
>
> Dave.
>