drm/amd/amdgpu embed hw_fence into amdgpu_job

Why: Previously hw fence is alloced separately with job.
It caused historical lifetime issues and corner cases.
The ideal situation is to take fence to manage both job
and fence's lifetime, and simplify the design of gpu-scheduler.

How:
We propose to embed hw_fence into amdgpu_job.
1. We cover the normal job submission by this method.
2. For ib_test, and submit without a parent job keep the
legacy way to create a hw fence separately.
v2:
use AMDGPU_FENCE_FLAG_EMBED_IN_JOB_BIT to show that the fence is
embedded in a job.
v3:
remove redundant variable ring in amdgpu_job
v4:
add tdr sequence support for this feature. Add a job_run_counter to
indicate whether this job is a resubmit job.
v5
add missing handling in amdgpu_fence_enable_signaling

Signed-off-by: Jingwen Chen <Jingwen.Chen2@amd.com>
Signed-off-by: Jack Zhang <Jack.Zhang7@hotmail.com>
Reviewed-by: Andrey Grodzovsky <andrey.grodzovsky@amd.com>
Reviewed by: Monk Liu <monk.liu@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

1 files changed, 67 insertions, 19 deletions

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c
index 6ed53669f3e0..14499f0de32d 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c
@@ -129,30 +129,50 @@ static u32 amdgpu_fence_read(struct amdgpu_ring *ring)
  *
  * @ring: ring the fence is associated with
  * @f: resulting fence object
+ * @job: job the fence is embedded in
  * @flags: flags to pass into the subordinate .emit_fence() call
  *
  * Emits a fence command on the requested ring (all asics).
  * Returns 0 on success, -ENOMEM on failure.
  */
-int amdgpu_fence_emit(struct amdgpu_ring *ring, struct dma_fence **f,
+int amdgpu_fence_emit(struct amdgpu_ring *ring, struct dma_fence **f, struct amdgpu_job *job,
 		      unsigned flags)
 {
 	struct amdgpu_device *adev = ring->adev;
-	struct amdgpu_fence *fence;
+	struct dma_fence *fence;
+	struct amdgpu_fence *am_fence;
 	struct dma_fence __rcu **ptr;
 	uint32_t seq;
 	int r;
 
-	fence = kmem_cache_alloc(amdgpu_fence_slab, GFP_KERNEL);
-	if (fence == NULL)
-		return -ENOMEM;
+	if (job == NULL) {
+		/* create a sperate hw fence */
+		am_fence = kmem_cache_alloc(amdgpu_fence_slab, GFP_ATOMIC);
+		if (am_fence == NULL)
+			return -ENOMEM;
+		fence = &am_fence->base;
+		am_fence->ring = ring;
+	} else {
+		/* take use of job-embedded fence */
+		fence = &job->hw_fence;
+	}
 
 	seq = ++ring->fence_drv.sync_seq;
-	fence->ring = ring;
-	dma_fence_init(&fence->base, &amdgpu_fence_ops,
-		       &ring->fence_drv.lock,
-		       adev->fence_context + ring->idx,
-		       seq);
+	if (job != NULL && job->job_run_counter) {
+		/* reinit seq for resubmitted jobs */
+		fence->seqno = seq;
+	} else {
+		dma_fence_init(fence, &amdgpu_fence_ops,
+				&ring->fence_drv.lock,
+				adev->fence_context + ring->idx,
+				seq);
+	}
+
+	if (job != NULL) {
+		/* mark this fence has a parent job */
+		set_bit(AMDGPU_FENCE_FLAG_EMBED_IN_JOB_BIT, &fence->flags);
+	}
+
 	amdgpu_ring_emit_fence(ring, ring->fence_drv.gpu_addr,
 			       seq, flags | AMDGPU_FENCE_FLAG_INT);
 	pm_runtime_get_noresume(adev_to_drm(adev)->dev);
@@ -175,9 +195,9 @@ int amdgpu_fence_emit(struct amdgpu_ring *ring, struct dma_fence **f,
 	/* This function can't be called concurrently anyway, otherwise
 	 * emitting the fence would mess up the hardware ring buffer.
 	 */
-	rcu_assign_pointer(*ptr, dma_fence_get(&fence->base));
+	rcu_assign_pointer(*ptr, dma_fence_get(fence));
 
-	*f = &fence->base;
+	*f = fence;
 
 	return 0;
 }
@@ -621,8 +641,16 @@ static const char *amdgpu_fence_get_driver_name(struct dma_fence *fence)
 
 static const char *amdgpu_fence_get_timeline_name(struct dma_fence *f)
 {
-	struct amdgpu_fence *fence = to_amdgpu_fence(f);
-	return (const char *)fence->ring->name;
+	struct amdgpu_ring *ring;
+
+	if (test_bit(AMDGPU_FENCE_FLAG_EMBED_IN_JOB_BIT, &f->flags)) {
+		struct amdgpu_job *job = container_of(f, struct amdgpu_job, hw_fence);
+
+		ring = to_amdgpu_ring(job->base.sched);
+	} else {
+		ring = to_amdgpu_fence(f)->ring;
+	}
+	return (const char *)ring->name;
 }
 
 /**
@@ -635,13 +663,20 @@ static const char *amdgpu_fence_get_timeline_name(struct dma_fence *f)
  */
 static bool amdgpu_fence_enable_signaling(struct dma_fence *f)
 {
-	struct amdgpu_fence *fence = to_amdgpu_fence(f);
-	struct amdgpu_ring *ring = fence->ring;
+	struct amdgpu_ring *ring;
+
+	if (test_bit(AMDGPU_FENCE_FLAG_EMBED_IN_JOB_BIT, &f->flags)) {
+		struct amdgpu_job *job = container_of(f, struct amdgpu_job, hw_fence);
+
+		ring = to_amdgpu_ring(job->base.sched);
+	} else {
+		ring = to_amdgpu_fence(f)->ring;
+	}
 
 	if (!timer_pending(&ring->fence_drv.fallback_timer))
 		amdgpu_fence_schedule_fallback(ring);
 
-	DMA_FENCE_TRACE(&fence->base, "armed on ring %i!\n", ring->idx);
+	DMA_FENCE_TRACE(f, "armed on ring %i!\n", ring->idx);
 
 	return true;
 }
@@ -656,8 +691,20 @@ static bool amdgpu_fence_enable_signaling(struct dma_fence *f)
 static void amdgpu_fence_free(struct rcu_head *rcu)
 {
 	struct dma_fence *f = container_of(rcu, struct dma_fence, rcu);
-	struct amdgpu_fence *fence = to_amdgpu_fence(f);
-	kmem_cache_free(amdgpu_fence_slab, fence);
+
+	if (test_bit(AMDGPU_FENCE_FLAG_EMBED_IN_JOB_BIT, &f->flags)) {
+	/* free job if fence has a parent job */
+		struct amdgpu_job *job;
+
+		job = container_of(f, struct amdgpu_job, hw_fence);
+		kfree(job);
+	} else {
+	/* free fence_slab if it's separated fence*/
+		struct amdgpu_fence *fence;
+
+		fence = to_amdgpu_fence(f);
+		kmem_cache_free(amdgpu_fence_slab, fence);
+	}
 }
 
 /**
@@ -680,6 +727,7 @@ static const struct dma_fence_ops amdgpu_fence_ops = {
 	.release = amdgpu_fence_release,
 };
 
+
 /*
  * Fence debugfs
  */