[Intel-gfx] [PATCH 10/13] drm/i915: Nonblocking request submission

[John Harrison]

On 25/08/2016 10:08, Chris Wilson wrote:
> Now that we have fences in place to drive request submission, we can
> employ those to queue requests after their dependencies as opposed to
> stalling in the middle of an execbuf ioctl. (However, we still choose to
> spin before enabling the IRQ as that is faster - though contentious.)
>
> Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
> ---
>   drivers/gpu/drm/i915/i915_gem_execbuffer.c | 10 +++++++---
>   drivers/gpu/drm/i915/i915_gem_request.c    | 14 +++++++++++++-
>   2 files changed, 20 insertions(+), 4 deletions(-)
>
> diff --git a/drivers/gpu/drm/i915/i915_gem_execbuffer.c b/drivers/gpu/drm/i915/i915_gem_execbuffer.c
> index b7cc158733ad..104c713ec681 100644
> --- a/drivers/gpu/drm/i915/i915_gem_execbuffer.c
> +++ b/drivers/gpu/drm/i915/i915_gem_execbuffer.c
> @@ -1136,9 +1136,13 @@ __eb_sync(struct drm_i915_gem_request *to,
>   
>   	trace_i915_gem_ring_sync_to(to, from);
>   	if (!i915.semaphores) {
> -		ret = i915_wait_request(from, true, NULL, NO_WAITBOOST);
> -		if (ret)
> -			return ret;
> +		if (!i915_spin_request(from, TASK_INTERRUPTIBLE, 2)) {
> +			ret = i915_sw_fence_await_dma_fence(&to->submit,
> +							    &from->fence,
Why not use 'i915_sw_fence_await_sw_fence(from->submit)' as below? Or 
conversely, why not use '_await_dma_fence(prev->fence)' below?


> +							    GFP_KERNEL);
> +			if (ret < 0)
> +				return ret;
> +		}
>   	} else {
>   		ret = to->engine->semaphore.sync_to(to, from);
>   		if (ret)
> diff --git a/drivers/gpu/drm/i915/i915_gem_request.c b/drivers/gpu/drm/i915/i915_gem_request.c
> index db45482ea194..d3477dabb534 100644
> --- a/drivers/gpu/drm/i915/i915_gem_request.c
> +++ b/drivers/gpu/drm/i915/i915_gem_request.c
> @@ -345,7 +345,7 @@ i915_gem_request_alloc(struct intel_engine_cs *engine,
>   {
>   	struct drm_i915_private *dev_priv = engine->i915;
>   	unsigned int reset_counter = i915_reset_counter(&dev_priv->gpu_error);
> -	struct drm_i915_gem_request *req;
> +	struct drm_i915_gem_request *req, *prev;
>   	u32 seqno;
>   	int ret;
>   
> @@ -441,6 +441,18 @@ i915_gem_request_alloc(struct intel_engine_cs *engine,
>   	 */
>   	req->head = req->ring->tail;
>   
> +	prev = i915_gem_active_peek(&engine->last_request,
> +				    &req->i915->drm.struct_mutex);
> +	if (prev) {
> +		ret = i915_sw_fence_await_sw_fence(&req->submit,
> +						   &prev->submit,
> +						   GFP_KERNEL);
Is this only required to guarantee seqno ordering?

If the previous request is not sharing any objects, context, etc. then 
there is no fundamental reason why this request should be dependent upon 
the last one.


> +		if (ret < 0) {
> +			i915_add_request(req);
> +			return ERR_PTR(ret);
> +		}
> +	}
> +
>   	return req;
>   
>   err_ctx: