[PATCH 1/8] drm/xe/oa: Separate batch submission from waiting for completion

[Matthew Brost]

On Thu, Aug 08, 2024 at 10:41:32AM -0700, Ashutosh Dixit wrote:
> When we introduce xe_syncs, we don't wait for internal OA programming
> batches to complete. That is, xe_syncs are signaled asynchronously. In
> anticipation for this, separate out batch submission from waiting for
> completion of those batches.
> 
> Signed-off-by: Ashutosh Dixit <ashutosh.dixit at intel.com>
> ---
>  drivers/gpu/drm/xe/xe_oa.c | 45 ++++++++++++++++++++++++--------------
>  1 file changed, 28 insertions(+), 17 deletions(-)
> 
> diff --git a/drivers/gpu/drm/xe/xe_oa.c b/drivers/gpu/drm/xe/xe_oa.c
> index 3ef92eb8fbb1e..d842c801fb9f1 100644
> --- a/drivers/gpu/drm/xe/xe_oa.c
> +++ b/drivers/gpu/drm/xe/xe_oa.c
> @@ -563,11 +563,10 @@ static __poll_t xe_oa_poll(struct file *file, poll_table *wait)
>  	return ret;
>  }
>  
> -static int xe_oa_submit_bb(struct xe_oa_stream *stream, struct xe_bb *bb)
> +static int xe_oa_submit_bb(struct xe_oa_stream *stream, struct xe_bb *bb,
> +			   struct dma_fence **fence)

static struct dma_fence *xe_oa_submit_bb(...)

Then use ERR_PTR, PTR_ERR semantics.

Matt

>  {
>  	struct xe_sched_job *job;
> -	struct dma_fence *fence;
> -	long timeout;
>  	int err = 0;
>  
>  	/* Kernel configuration is issued on stream->k_exec_q, not stream->exec_q */
> @@ -578,15 +577,8 @@ static int xe_oa_submit_bb(struct xe_oa_stream *stream, struct xe_bb *bb)
>  	}
>  
>  	xe_sched_job_arm(job);
> -	fence = dma_fence_get(&job->drm.s_fence->finished);
> +	*fence = dma_fence_get(&job->drm.s_fence->finished);
>  	xe_sched_job_push(job);
> -
> -	timeout = dma_fence_wait_timeout(fence, false, HZ);
> -	dma_fence_put(fence);
> -	if (timeout < 0)
> -		err = timeout;
> -	else if (!timeout)
> -		err = -ETIME;
>  exit:
>  	return err;
>  }
> @@ -652,6 +644,7 @@ static void xe_oa_store_flex(struct xe_oa_stream *stream, struct xe_lrc *lrc,
>  static int xe_oa_modify_ctx_image(struct xe_oa_stream *stream, struct xe_lrc *lrc,
>  				  const struct flex *flex, u32 count)
>  {
> +	struct dma_fence *fence;
>  	struct xe_bb *bb;
>  	int err;
>  
> @@ -663,14 +656,16 @@ static int xe_oa_modify_ctx_image(struct xe_oa_stream *stream, struct xe_lrc *lr
>  
>  	xe_oa_store_flex(stream, lrc, bb, flex, count);
>  
> -	err = xe_oa_submit_bb(stream, bb);
> -	xe_bb_free(bb, NULL);
> +	err = xe_oa_submit_bb(stream, bb, &fence);
> +	xe_bb_free(bb, fence);
> +	dma_fence_put(fence);
>  exit:
>  	return err;
>  }
>  
>  static int xe_oa_load_with_lri(struct xe_oa_stream *stream, struct xe_oa_reg *reg_lri)
>  {
> +	struct dma_fence *fence;
>  	struct xe_bb *bb;
>  	int err;
>  
> @@ -682,8 +677,9 @@ static int xe_oa_load_with_lri(struct xe_oa_stream *stream, struct xe_oa_reg *re
>  
>  	write_cs_mi_lri(bb, reg_lri, 1);
>  
> -	err = xe_oa_submit_bb(stream, bb);
> -	xe_bb_free(bb, NULL);
> +	err = xe_oa_submit_bb(stream, bb, &fence);
> +	xe_bb_free(bb, fence);
> +	dma_fence_put(fence);
>  exit:
>  	return err;
>  }
> @@ -913,15 +909,30 @@ static int xe_oa_emit_oa_config(struct xe_oa_stream *stream, struct xe_oa_config
>  {
>  #define NOA_PROGRAM_ADDITIONAL_DELAY_US 500
>  	struct xe_oa_config_bo *oa_bo;
> -	int err, us = NOA_PROGRAM_ADDITIONAL_DELAY_US;
> +	int err = 0, us = NOA_PROGRAM_ADDITIONAL_DELAY_US;
> +	struct dma_fence *fence;
> +	long timeout;
>  
> +	/* Emit OA configuration batch */
>  	oa_bo = xe_oa_alloc_config_buffer(stream, config);
>  	if (IS_ERR(oa_bo)) {
>  		err = PTR_ERR(oa_bo);
>  		goto exit;
>  	}
>  
> -	err = xe_oa_submit_bb(stream, oa_bo->bb);
> +	err = xe_oa_submit_bb(stream, oa_bo->bb, &fence);
> +	if (err)
> +		goto exit;
> +
> +	/* Wait till all previous batches have executed */
> +	timeout = dma_fence_wait_timeout(fence, false, 5 * HZ);
> +	dma_fence_put(fence);
> +	if (timeout < 0)
> +		err = timeout;
> +	else if (!timeout)
> +		err = -ETIME;
> +	if (err)
> +		drm_dbg(&stream->oa->xe->drm, "dma_fence_wait_timeout err %d\n", err);
>  
>  	/* Additional empirical delay needed for NOA programming after registers are written */
>  	usleep_range(us, 2 * us);
> -- 
> 2.41.0
>