[Intel-gfx] [PATCH v5 09/35] drm/i915: Force MMIO flips when scheduler enabled

[Jesse Barnes]

On 02/19/2016 11:53 AM, Ville Syrjälä wrote:
> On Fri, Feb 19, 2016 at 11:28:05AM -0800, Jesse Barnes wrote:
>> On 02/18/2016 06:26 AM, John.C.Harrison at Intel.com wrote:
>>> From: John Harrison <John.C.Harrison at Intel.com>
>>>
>>> MMIO flips are the preferred mechanism now but more importantly, pipe
>>> based flips cause issues for the scheduler. Specifically, submitting
>>> work to the rings around the side of the scheduler could cause that
>>> work to be lost if the scheduler generates a pre-emption event on that
>>> ring.
>>>
>>> For: VIZ-1587
>>> Signed-off-by: John Harrison <John.C.Harrison at Intel.com>
>>> ---
>>>  drivers/gpu/drm/i915/intel_display.c | 3 +++
>>>  1 file changed, 3 insertions(+)
>>>
>>> diff --git a/drivers/gpu/drm/i915/intel_display.c b/drivers/gpu/drm/i915/intel_display.c
>>> index 6e12ed7..731d20a 100644
>>> --- a/drivers/gpu/drm/i915/intel_display.c
>>> +++ b/drivers/gpu/drm/i915/intel_display.c
>>> @@ -46,6 +46,7 @@
>>>  #include <linux/dma_remapping.h>
>>>  #include <linux/reservation.h>
>>>  #include <linux/dma-buf.h>
>>> +#include "i915_scheduler.h"
>>>  
>>>  /* Primary plane formats for gen <= 3 */
>>>  static const uint32_t i8xx_primary_formats[] = {
>>> @@ -11330,6 +11331,8 @@ static bool use_mmio_flip(struct intel_engine_cs *ring,
>>>  		return true;
>>>  	else if (i915.enable_execlists)
>>>  		return true;
>>> +	else if (i915_scheduler_is_enabled(ring->dev))
>>> +		return true;
>>>  	else if (obj->base.dma_buf &&
>>>  		 !reservation_object_test_signaled_rcu(obj->base.dma_buf->resv,
>>>  						       false))
>>>
>>
>> Why can't we use mmio flips unconditionally?  Maarten or Ville?
> 
> We do when execlists are used, which is always on gen9+. So I guess I'm
> missing the point of this patch. For gen5+ we could also do it trivially.

didn't check if the scheduler is also enabled for gen8 (I guess it would be nice, that would cover BDW and BSW).

> 
> For older platforms it'd require a bit of work since we'd need to
> complete the flips from the vblank interrupt. Well, we actually do
> that already even with CS flips on those platforms, but we do look
> at the flip pending interrupt to figure out if CS already issued
> the flip or not. So that part would need changing.
> 
> I also think we should switch to using the vblank interrupt for this
> stuff on all platforms, mainly since the flip done interrupt is somewhat
> broken on at least BDW (no idea if it got fixed in SKL or later), and
> doing things in more than one way certainly won't decrease our bug count.

Yeah that's probably the way to go; I haven't checked the behavior on SKL either.