[Intel-gfx] [PATCH v3 1/2] drm/i915/gtt: Do not initialize drm_mm twice.

[Jesse Barnes]

On 09/23/2015 11:55 AM, Paulo Zanoni wrote:
> 2015-09-23 13:48 GMT-03:00 Imre Deak <imre.deak at intel.com>:
>> On ke, 2015-09-16 at 11:49 +0200, Michał Winiarski wrote:
>>> It would be initialized just moments later by i915_init_vm.
>>>
>>> v2: Commit msg update,
>>>     s/i915_init_vm/i915_address_space_init, move to i915_gem_gtt.c,
>>>     init address_space during i915_gem_setup_global_gtt for ggtt.
>>> v3: Do not init global_link - we are adding it to vm_list moments later,
>>>     make i915_address_space_init static, use OOP style parameter order.
>>>
>>> Cc: Chris Wilson <chris at chris-wilson.co.uk>
>>> Cc: Michel Thierry <michel.thierry at intel.com>
>>> Cc: Mika Kuoppala <mika.kuoppala at intel.com>
>>> Signed-off-by: Michał Winiarski <michal.winiarski at intel.com>
>>
>> With this patch I get the oops below when booting. Looking at it it's
>> caused by dev_priv->gtt being initialized now too late for
>> i915_gem_object_create_stolen_for_preallocated().
> 
> My BDW machine stopped booting, and git bisect tells me this patch is
> the cause. I didn't setup anything to help me grab the logs before the
> crash, so I can't confirm the backtrace below.
> 
> Something interesting is that if I boot with i915.modeset=0, then
> later "rmmod i915; modprobe i915 modeset=1", the driver loads
> successfully.
> 

I see this too on BYT.  Can we get this reverted or get a fix?

Michal?

Thanks,
Jesse