[Intel-gfx] [PATCH] drm/i915/bdw: Fix GEN8 GTT size calculation

[Daniel Vetter]

On Fri, Mar 14, 2014 at 11:05 PM, Woodhouse, David
<david.woodhouse at intel.com> wrote:
> On Fri, 2014-03-14 at 12:14 -0700, Ben Widawsky wrote:
>>
>> This was based off of linus master, since I assumed that is what David
>> was using.
>
> Indeed. Now it's much better. Now that nasty scary warning is gone away
> and now I just have a few hundred of these (from your other patch before
> which it crashed instead):
> [    6.167463] [drm:intel_set_cpu_fifo_underrun_reporting] *ERROR* Interrupt arrived before CRTCs were setup up
> ...
> [   35.443214] [drm:intel_set_cpu_fifo_underrun_reporting] *ERROR* Interrupt arrived before CRTCs were setup up

drm-intel-fixes should help:
http://cgit.freedesktop.org/drm-intel/commit/?id=5c673b60a9b3b23486f4eda75c72e91d31d26a2b

> ... followed by these until I turn it off:
> [   38.190001] dmar: DMAR:[DMA Read] Request device [00:02.0] fault addr ade47000
> [   38.190001] DMAR:[fault reason 06] PTE Read access is not set
> [   38.204801] dmar: DMAR:[DMA Read] Request device [00:02.0] fault addr add5e000
> [   38.204801] DMAR:[fault reason 06] PTE Read access is not set
> [   38.219598] dmar: DMAR:[DMA Read] Request device [00:02.0] fault addr adc75000
> [   38.219598] DMAR:[fault reason 06] PTE Read access is not set
>
> As I noted in private though, there's an RMRR instructing us to set up a
> 1:1 mapping for the entire ad000000-af7fffff range. An instruction which
> we appear to be following. So I have no idea why we're taking faults for
> those addresses; that might be a problem in IOMMU code. Will look into
> that further...

I've asked you on private irc whether this range matches/overlaps with
stolen - we know of things blowing up at least on earlier generations
in combination with dmar. Please boot with drm.debug=0xe and scan for
the stolen mem reporting:

[    9.459273] [drm:i915_gem_init_stolen], found 67108864 bytes of
stolen memory at 9ba00000

Your range has roughly an offset/size stolen usually has.

Thanks, Daniel
-- 
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch