[Intel-gfx] [PATCH 14/56] drm/i915: Make retire condition check for requests not objects

[John Harrison]

Note that this is a new patch to the series. The issue was found when 
debugging a problem with the conversion to struct fence that is still in 
progress. Basically, it was possible to get continuous TDR timeouts on 
the ECS ring because the start of day initialisation request never got 
retired and the TDR got confused by this.

This patch should therefore be included in the anti-OLR series as a new 
patch #14 (immediately before 'update i915_gpu_idle() to ...'). As that 
is the point at which non-batch buffer requests start to appear in the 
system.


On 04/06/2015 19:23, John.C.Harrison at Intel.com wrote:
> From: John Harrison <John.C.Harrison at Intel.com>
>
> A previous patch (read-read optimisation) changed the early exit
> condition in i915_gem_retire_requests_ring() from checking the request
> list to checking the active list. This assumes that all requests have
> objects associated with them which are placed on the active list. The
> removal of the OLR means that non-batch buffer work is no longer
> tagged onto the nearest batch buffer submission and thus there are
> requests going through the system which do not have objects associated
> with them. This can therefore lead to the situation where an
> outstanding request never gets retired.
>
> This change reverts the early exit condition to check for requests.
> Given that the purpose of the function is to retire requests, this
> does seem to make much more sense.
>
> For: VIZ-5190
> Signed-off-by: John Harrison <John.C.Harrison at Intel.com>
> ---
>   drivers/gpu/drm/i915/i915_gem.c |    2 +-
>   1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
> index 7117659..4c5a6cd 100644
> --- a/drivers/gpu/drm/i915/i915_gem.c
> +++ b/drivers/gpu/drm/i915/i915_gem.c
> @@ -2859,7 +2859,7 @@ i915_gem_retire_requests_ring(struct intel_engine_cs *ring)
>   {
>   	WARN_ON(i915_verify_lists(ring->dev));
>   
> -	if (list_empty(&ring->active_list))
> +	if (list_empty(&ring->request_list))
>   		return;
>   
>   	/* Retire requests first as we use it above for the early return.