From c5093cddf56baceb1545028e8a5971d94cf59d25 Mon Sep 17 00:00:00 2001
From: Christian König <christian.koenig@amd.com>
Date: Wed, 28 Sep 2022 13:21:05 +0200
Subject: drm/amdgpu: drop the fence argument from amdgpu_vmid_grab
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

This is always the job anyway.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Luben Tuikov <luben.tuikov@amd.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20221014084641.128280-5-christian.koenig@amd.com
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_job.c | 4 +---
 1 file changed, 1 insertion(+), 3 deletions(-)

(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_job.c')

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c
index 46c99331d7f1..5aa053acc0b4 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c
@@ -256,9 +256,7 @@ static struct dma_fence *amdgpu_job_dependency(struct drm_sched_job *sched_job,
 	}
 
 	while (fence == NULL && vm && !job->vmid) {
-		r = amdgpu_vmid_grab(vm, ring, &job->sync,
-				     &job->base.s_fence->finished,
-				     job);
+		r = amdgpu_vmid_grab(vm, ring, &job->sync, job);
 		if (r)
 			DRM_ERROR("Error getting VM ID (%d)\n", r);
 
-- 
cgit 


From 940ca22b7ea9db6857ba7c6adb961b84d8cc28ff Mon Sep 17 00:00:00 2001
From: Christian König <christian.koenig@amd.com>
Date: Wed, 28 Sep 2022 14:00:57 +0200
Subject: drm/amdgpu: drop amdgpu_sync from amdgpu_vmid_grab v2
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Instead return the fence directly. Avoids memory allocation to store the
fence.

v2: cleanup coding style as well

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Luben Tuikov <luben.tuikov@amd.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20221014084641.128280-6-christian.koenig@amd.com
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ids.c | 42 +++++++++++++++++----------------
 drivers/gpu/drm/amd/amdgpu/amdgpu_ids.h |  2 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_job.c | 12 ++++------
 3 files changed, 28 insertions(+), 28 deletions(-)

(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_job.c')

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ids.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ids.c
index b76294d4275b..2a9a2593dc18 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ids.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ids.c
@@ -170,26 +170,27 @@ bool amdgpu_vmid_had_gpu_reset(struct amdgpu_device *adev,
  *
  * @vm: vm to allocate id for
  * @ring: ring we want to submit job to
- * @sync: sync object where we add dependencies
  * @idle: resulting idle VMID
+ * @fence: fence to wait for if no id could be grabbed
  *
  * Try to find an idle VMID, if none is idle add a fence to wait to the sync
  * object. Returns -ENOMEM when we are out of memory.
  */
 static int amdgpu_vmid_grab_idle(struct amdgpu_vm *vm,
 				 struct amdgpu_ring *ring,
-				 struct amdgpu_sync *sync,
-				 struct amdgpu_vmid **idle)
+				 struct amdgpu_vmid **idle,
+				 struct dma_fence **fence)
 {
 	struct amdgpu_device *adev = ring->adev;
 	unsigned vmhub = ring->funcs->vmhub;
 	struct amdgpu_vmid_mgr *id_mgr = &adev->vm_manager.id_mgr[vmhub];
 	struct dma_fence **fences;
 	unsigned i;
-	int r;
 
-	if (!dma_fence_is_signaled(ring->vmid_wait))
-		return amdgpu_sync_fence(sync, ring->vmid_wait);
+	if (!dma_fence_is_signaled(ring->vmid_wait)) {
+		*fence = dma_fence_get(ring->vmid_wait);
+		return 0;
+	}
 
 	fences = kmalloc_array(id_mgr->num_ids, sizeof(void *), GFP_KERNEL);
 	if (!fences)
@@ -228,10 +229,10 @@ static int amdgpu_vmid_grab_idle(struct amdgpu_vm *vm,
 			return -ENOMEM;
 		}
 
-		r = amdgpu_sync_fence(sync, &array->base);
+		*fence = dma_fence_get(&array->base);
 		dma_fence_put(ring->vmid_wait);
 		ring->vmid_wait = &array->base;
-		return r;
+		return 0;
 	}
 	kfree(fences);
 
@@ -243,17 +244,17 @@ static int amdgpu_vmid_grab_idle(struct amdgpu_vm *vm,
  *
  * @vm: vm to allocate id for
  * @ring: ring we want to submit job to
- * @sync: sync object where we add dependencies
  * @job: job who wants to use the VMID
  * @id: resulting VMID
+ * @fence: fence to wait for if no id could be grabbed
  *
  * Try to assign a reserved VMID.
  */
 static int amdgpu_vmid_grab_reserved(struct amdgpu_vm *vm,
 				     struct amdgpu_ring *ring,
-				     struct amdgpu_sync *sync,
 				     struct amdgpu_job *job,
-				     struct amdgpu_vmid **id)
+				     struct amdgpu_vmid **id,
+				     struct dma_fence **fence)
 {
 	struct amdgpu_device *adev = ring->adev;
 	unsigned vmhub = ring->funcs->vmhub;
@@ -280,7 +281,8 @@ static int amdgpu_vmid_grab_reserved(struct amdgpu_vm *vm,
 		tmp = amdgpu_sync_peek_fence(&(*id)->active, ring);
 		if (tmp) {
 			*id = NULL;
-			return amdgpu_sync_fence(sync, tmp);
+			*fence = dma_fence_get(tmp);
+			return 0;
 		}
 		needs_flush = true;
 	}
@@ -302,17 +304,17 @@ static int amdgpu_vmid_grab_reserved(struct amdgpu_vm *vm,
  *
  * @vm: vm to allocate id for
  * @ring: ring we want to submit job to
- * @sync: sync object where we add dependencies
  * @job: job who wants to use the VMID
  * @id: resulting VMID
+ * @fence: fence to wait for if no id could be grabbed
  *
  * Try to reuse a VMID for this submission.
  */
 static int amdgpu_vmid_grab_used(struct amdgpu_vm *vm,
 				 struct amdgpu_ring *ring,
-				 struct amdgpu_sync *sync,
 				 struct amdgpu_job *job,
-				 struct amdgpu_vmid **id)
+				 struct amdgpu_vmid **id,
+				 struct dma_fence **fence)
 {
 	struct amdgpu_device *adev = ring->adev;
 	unsigned vmhub = ring->funcs->vmhub;
@@ -367,13 +369,13 @@ static int amdgpu_vmid_grab_used(struct amdgpu_vm *vm,
  *
  * @vm: vm to allocate id for
  * @ring: ring we want to submit job to
- * @sync: sync object where we add dependencies
  * @job: job who wants to use the VMID
+ * @fence: fence to wait for if no id could be grabbed
  *
  * Allocate an id for the vm, adding fences to the sync obj as necessary.
  */
 int amdgpu_vmid_grab(struct amdgpu_vm *vm, struct amdgpu_ring *ring,
-		     struct amdgpu_sync *sync, struct amdgpu_job *job)
+		     struct amdgpu_job *job, struct dma_fence **fence)
 {
 	struct amdgpu_device *adev = ring->adev;
 	unsigned vmhub = ring->funcs->vmhub;
@@ -383,16 +385,16 @@ int amdgpu_vmid_grab(struct amdgpu_vm *vm, struct amdgpu_ring *ring,
 	int r = 0;
 
 	mutex_lock(&id_mgr->lock);
-	r = amdgpu_vmid_grab_idle(vm, ring, sync, &idle);
+	r = amdgpu_vmid_grab_idle(vm, ring, &idle, fence);
 	if (r || !idle)
 		goto error;
 
 	if (vm->reserved_vmid[vmhub]) {
-		r = amdgpu_vmid_grab_reserved(vm, ring, sync, job, &id);
+		r = amdgpu_vmid_grab_reserved(vm, ring, job, &id, fence);
 		if (r || !id)
 			goto error;
 	} else {
-		r = amdgpu_vmid_grab_used(vm, ring, sync, job, &id);
+		r = amdgpu_vmid_grab_used(vm, ring, job, &id, fence);
 		if (r)
 			goto error;
 
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ids.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_ids.h
index 1b1e7d04655c..57efe61dceed 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ids.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ids.h
@@ -84,7 +84,7 @@ void amdgpu_vmid_free_reserved(struct amdgpu_device *adev,
 			       struct amdgpu_vm *vm,
 			       unsigned vmhub);
 int amdgpu_vmid_grab(struct amdgpu_vm *vm, struct amdgpu_ring *ring,
-		     struct amdgpu_sync *sync, struct amdgpu_job *job);
+		     struct amdgpu_job *job, struct dma_fence **fence);
 void amdgpu_vmid_reset(struct amdgpu_device *adev, unsigned vmhub,
 		       unsigned vmid);
 void amdgpu_vmid_reset_all(struct amdgpu_device *adev);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c
index 5aa053acc0b4..384149a81978 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c
@@ -239,12 +239,12 @@ int amdgpu_job_submit_direct(struct amdgpu_job *job, struct amdgpu_ring *ring,
 	return 0;
 }
 
-static struct dma_fence *amdgpu_job_dependency(struct drm_sched_job *sched_job,
-					       struct drm_sched_entity *s_entity)
+static struct dma_fence *
+amdgpu_job_dependency(struct drm_sched_job *sched_job,
+		      struct drm_sched_entity *s_entity)
 {
 	struct amdgpu_ring *ring = to_amdgpu_ring(s_entity->rq->sched);
 	struct amdgpu_job *job = to_amdgpu_job(sched_job);
-	struct amdgpu_vm *vm = job->vm;
 	struct dma_fence *fence;
 	int r;
 
@@ -255,12 +255,10 @@ static struct dma_fence *amdgpu_job_dependency(struct drm_sched_job *sched_job,
 			DRM_ERROR("Error adding fence (%d)\n", r);
 	}
 
-	while (fence == NULL && vm && !job->vmid) {
-		r = amdgpu_vmid_grab(vm, ring, &job->sync, job);
+	while (!fence && job->vm && !job->vmid) {
+		r = amdgpu_vmid_grab(job->vm, ring, job, &fence);
 		if (r)
 			DRM_ERROR("Error getting VM ID (%d)\n", r);
-
-		fence = amdgpu_sync_get_fence(&job->sync);
 	}
 
 	if (!fence && job->gang_submit)
-- 
cgit 


From f7d66fb2ea43a3016e78a700a2ca6c77a74579f9 Mon Sep 17 00:00:00 2001
From: Christian König <christian.koenig@amd.com>
Date: Wed, 28 Sep 2022 20:31:38 +0200
Subject: drm/amdgpu: cleanup scheduler job initialization v2
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Init the DRM scheduler base class while allocating the job.

This makes the whole handling much more cleaner.

v2: fix coding style

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Luben Tuikov <luben.tuikov@amd.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20221014084641.128280-7-christian.koenig@amd.com
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c  |  2 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c      |  8 +---
 drivers/gpu/drm/amd/amdgpu/amdgpu_job.c     | 44 +++++++++++----------
 drivers/gpu/drm/amd/amdgpu/amdgpu_job.h     | 14 ++++---
 drivers/gpu/drm/amd/amdgpu/amdgpu_jpeg.c    |  7 ++--
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c     | 56 ++++++++++----------------
 drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c     |  9 ++---
 drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c     | 13 +++---
 drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c     | 22 ++++++-----
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm_sdma.c | 61 +++++++++++++++--------------
 drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c      | 12 ++----
 drivers/gpu/drm/amd/amdgpu/uvd_v6_0.c       |  8 ++--
 drivers/gpu/drm/amd/amdgpu/uvd_v7_0.c       | 12 +++---
 drivers/gpu/drm/amd/amdkfd/kfd_migrate.c    | 17 +++-----
 14 files changed, 135 insertions(+), 150 deletions(-)

(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_job.c')

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c
index 03bbfaa51cbc..cf58c1125e60 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c
@@ -674,7 +674,7 @@ int amdgpu_amdkfd_submit_ib(struct amdgpu_device *adev,
 		goto err;
 	}
 
-	ret = amdgpu_job_alloc(adev, 1, &job, NULL);
+	ret = amdgpu_job_alloc(adev, NULL, NULL, NULL, 1, &job);
 	if (ret)
 		goto err;
 
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index 1bbd39b3b0fc..aa6f6c428dbc 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -291,12 +291,8 @@ static int amdgpu_cs_pass1(struct amdgpu_cs_parser *p,
 		return -EINVAL;
 
 	for (i = 0; i < p->gang_size; ++i) {
-		ret = amdgpu_job_alloc(p->adev, num_ibs[i], &p->jobs[i], vm);
-		if (ret)
-			goto free_all_kdata;
-
-		ret = drm_sched_job_init(&p->jobs[i]->base, p->entities[i],
-					 &fpriv->vm);
+		ret = amdgpu_job_alloc(p->adev, vm, p->entities[i], vm,
+				       num_ibs[i], &p->jobs[i]);
 		if (ret)
 			goto free_all_kdata;
 	}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c
index 384149a81978..fa60d25dfbd1 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c
@@ -89,8 +89,9 @@ exit:
 	return DRM_GPU_SCHED_STAT_NOMINAL;
 }
 
-int amdgpu_job_alloc(struct amdgpu_device *adev, unsigned num_ibs,
-		     struct amdgpu_job **job, struct amdgpu_vm *vm)
+int amdgpu_job_alloc(struct amdgpu_device *adev, struct amdgpu_vm *vm,
+		     struct drm_sched_entity *entity, void *owner,
+		     unsigned int num_ibs, struct amdgpu_job **job)
 {
 	if (num_ibs == 0)
 		return -EINVAL;
@@ -111,23 +112,30 @@ int amdgpu_job_alloc(struct amdgpu_device *adev, unsigned num_ibs,
 	(*job)->vram_lost_counter = atomic_read(&adev->vram_lost_counter);
 	(*job)->vm_pd_addr = AMDGPU_BO_INVALID_OFFSET;
 
-	return 0;
+	if (!entity)
+		return 0;
+
+	return drm_sched_job_init(&(*job)->base, entity, owner);
 }
 
-int amdgpu_job_alloc_with_ib(struct amdgpu_device *adev, unsigned size,
-		enum amdgpu_ib_pool_type pool_type,
-		struct amdgpu_job **job)
+int amdgpu_job_alloc_with_ib(struct amdgpu_device *adev,
+			     struct drm_sched_entity *entity, void *owner,
+			     size_t size, enum amdgpu_ib_pool_type pool_type,
+			     struct amdgpu_job **job)
 {
 	int r;
 
-	r = amdgpu_job_alloc(adev, 1, job, NULL);
+	r = amdgpu_job_alloc(adev, NULL, entity, owner, 1, job);
 	if (r)
 		return r;
 
 	(*job)->num_ibs = 1;
 	r = amdgpu_ib_get(adev, NULL, size, pool_type, &(*job)->ibs[0]);
-	if (r)
+	if (r) {
+		if (entity)
+			drm_sched_job_cleanup(&(*job)->base);
 		kfree(*job);
+	}
 
 	return r;
 }
@@ -191,6 +199,9 @@ void amdgpu_job_set_gang_leader(struct amdgpu_job *job,
 
 void amdgpu_job_free(struct amdgpu_job *job)
 {
+	if (job->base.entity)
+		drm_sched_job_cleanup(&job->base);
+
 	amdgpu_job_free_resources(job);
 	amdgpu_sync_free(&job->sync);
 	amdgpu_sync_free(&job->sched_sync);
@@ -203,25 +214,16 @@ void amdgpu_job_free(struct amdgpu_job *job)
 		dma_fence_put(&job->hw_fence);
 }
 
-int amdgpu_job_submit(struct amdgpu_job *job, struct drm_sched_entity *entity,
-		      void *owner, struct dma_fence **f)
+struct dma_fence *amdgpu_job_submit(struct amdgpu_job *job)
 {
-	int r;
-
-	if (!f)
-		return -EINVAL;
-
-	r = drm_sched_job_init(&job->base, entity, owner);
-	if (r)
-		return r;
+	struct dma_fence *f;
 
 	drm_sched_job_arm(&job->base);
-
-	*f = dma_fence_get(&job->base.s_fence->finished);
+	f = dma_fence_get(&job->base.s_fence->finished);
 	amdgpu_job_free_resources(job);
 	drm_sched_entity_push_job(&job->base);
 
-	return 0;
+	return f;
 }
 
 int amdgpu_job_submit_direct(struct amdgpu_job *job, struct amdgpu_ring *ring,
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.h
index ab7b150e5d50..e2ecab977d0d 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.h
@@ -78,18 +78,20 @@ static inline struct amdgpu_ring *amdgpu_job_ring(struct amdgpu_job *job)
 	return to_amdgpu_ring(job->base.entity->rq->sched);
 }
 
-int amdgpu_job_alloc(struct amdgpu_device *adev, unsigned num_ibs,
-		     struct amdgpu_job **job, struct amdgpu_vm *vm);
-int amdgpu_job_alloc_with_ib(struct amdgpu_device *adev, unsigned size,
-		enum amdgpu_ib_pool_type pool, struct amdgpu_job **job);
+int amdgpu_job_alloc(struct amdgpu_device *adev, struct amdgpu_vm *vm,
+		     struct drm_sched_entity *entity, void *owner,
+		     unsigned int num_ibs, struct amdgpu_job **job);
+int amdgpu_job_alloc_with_ib(struct amdgpu_device *adev,
+			     struct drm_sched_entity *entity, void *owner,
+			     size_t size, enum amdgpu_ib_pool_type pool_type,
+			     struct amdgpu_job **job);
 void amdgpu_job_set_resources(struct amdgpu_job *job, struct amdgpu_bo *gds,
 			      struct amdgpu_bo *gws, struct amdgpu_bo *oa);
 void amdgpu_job_free_resources(struct amdgpu_job *job);
 void amdgpu_job_set_gang_leader(struct amdgpu_job *job,
 				struct amdgpu_job *leader);
 void amdgpu_job_free(struct amdgpu_job *job);
-int amdgpu_job_submit(struct amdgpu_job *job, struct drm_sched_entity *entity,
-		      void *owner, struct dma_fence **f);
+struct dma_fence *amdgpu_job_submit(struct amdgpu_job *job);
 int amdgpu_job_submit_direct(struct amdgpu_job *job, struct amdgpu_ring *ring,
 			     struct dma_fence **fence);
 
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_jpeg.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_jpeg.c
index 518eb0e40d32..de182bfcf85f 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_jpeg.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_jpeg.c
@@ -150,14 +150,15 @@ static int amdgpu_jpeg_dec_set_reg(struct amdgpu_ring *ring, uint32_t handle,
 	const unsigned ib_size_dw = 16;
 	int i, r;
 
-	r = amdgpu_job_alloc_with_ib(ring->adev, ib_size_dw * 4,
-					AMDGPU_IB_POOL_DIRECT, &job);
+	r = amdgpu_job_alloc_with_ib(ring->adev, NULL, NULL, ib_size_dw * 4,
+				     AMDGPU_IB_POOL_DIRECT, &job);
 	if (r)
 		return r;
 
 	ib = &job->ibs[0];
 
-	ib->ptr[0] = PACKETJ(adev->jpeg.internal.jpeg_pitch, 0, 0, PACKETJ_TYPE0);
+	ib->ptr[0] = PACKETJ(adev->jpeg.internal.jpeg_pitch, 0, 0,
+			     PACKETJ_TYPE0);
 	ib->ptr[1] = 0xDEADBEEF;
 	for (i = 2; i < 16; i += 2) {
 		ib->ptr[i] = PACKETJ(0, 0, 0, PACKETJ_TYPE6);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
index 5aae63891858..9f5c1d86d2ab 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
@@ -189,7 +189,6 @@ static int amdgpu_ttm_map_buffer(struct ttm_buffer_object *bo,
 	struct amdgpu_device *adev = ring->adev;
 	unsigned offset, num_pages, num_dw, num_bytes;
 	uint64_t src_addr, dst_addr;
-	struct dma_fence *fence;
 	struct amdgpu_job *job;
 	void *cpu_addr;
 	uint64_t flags;
@@ -229,7 +228,9 @@ static int amdgpu_ttm_map_buffer(struct ttm_buffer_object *bo,
 	num_dw = ALIGN(adev->mman.buffer_funcs->copy_num_dw, 8);
 	num_bytes = num_pages * 8 * AMDGPU_GPU_PAGES_IN_CPU_PAGE;
 
-	r = amdgpu_job_alloc_with_ib(adev, num_dw * 4 + num_bytes,
+	r = amdgpu_job_alloc_with_ib(adev, &adev->mman.entity,
+				     AMDGPU_FENCE_OWNER_UNDEFINED,
+				     num_dw * 4 + num_bytes,
 				     AMDGPU_IB_POOL_DELAYED, &job);
 	if (r)
 		return r;
@@ -269,18 +270,8 @@ static int amdgpu_ttm_map_buffer(struct ttm_buffer_object *bo,
 		}
 	}
 
-	r = amdgpu_job_submit(job, &adev->mman.entity,
-			      AMDGPU_FENCE_OWNER_UNDEFINED, &fence);
-	if (r)
-		goto error_free;
-
-	dma_fence_put(fence);
-
-	return r;
-
-error_free:
-	amdgpu_job_free(job);
-	return r;
+	dma_fence_put(amdgpu_job_submit(job));
+	return 0;
 }
 
 /**
@@ -1414,7 +1405,8 @@ static void amdgpu_ttm_vram_mm_access(struct amdgpu_device *adev, loff_t pos,
 }
 
 static int amdgpu_ttm_access_memory_sdma(struct ttm_buffer_object *bo,
-					unsigned long offset, void *buf, int len, int write)
+					unsigned long offset, void *buf,
+					int len, int write)
 {
 	struct amdgpu_bo *abo = ttm_to_amdgpu_bo(bo);
 	struct amdgpu_device *adev = amdgpu_ttm_adev(abo->tbo.bdev);
@@ -1438,26 +1430,27 @@ static int amdgpu_ttm_access_memory_sdma(struct ttm_buffer_object *bo,
 		memcpy(adev->mman.sdma_access_ptr, buf, len);
 
 	num_dw = ALIGN(adev->mman.buffer_funcs->copy_num_dw, 8);
-	r = amdgpu_job_alloc_with_ib(adev, num_dw * 4, AMDGPU_IB_POOL_DELAYED, &job);
+	r = amdgpu_job_alloc_with_ib(adev, &adev->mman.entity,
+				     AMDGPU_FENCE_OWNER_UNDEFINED,
+				     num_dw * 4, AMDGPU_IB_POOL_DELAYED,
+				     &job);
 	if (r)
 		goto out;
 
 	amdgpu_res_first(abo->tbo.resource, offset, len, &src_mm);
-	src_addr = amdgpu_ttm_domain_start(adev, bo->resource->mem_type) + src_mm.start;
+	src_addr = amdgpu_ttm_domain_start(adev, bo->resource->mem_type) +
+		src_mm.start;
 	dst_addr = amdgpu_bo_gpu_offset(adev->mman.sdma_access_bo);
 	if (write)
 		swap(src_addr, dst_addr);
 
-	amdgpu_emit_copy_buffer(adev, &job->ibs[0], src_addr, dst_addr, PAGE_SIZE, false);
+	amdgpu_emit_copy_buffer(adev, &job->ibs[0], src_addr, dst_addr,
+				PAGE_SIZE, false);
 
 	amdgpu_ring_pad_ib(adev->mman.buffer_funcs_ring, &job->ibs[0]);
 	WARN_ON(job->ibs[0].length_dw > num_dw);
 
-	r = amdgpu_job_submit(job, &adev->mman.entity, AMDGPU_FENCE_OWNER_UNDEFINED, &fence);
-	if (r) {
-		amdgpu_job_free(job);
-		goto out;
-	}
+	fence = amdgpu_job_submit(job);
 
 	if (!dma_fence_wait_timeout(fence, false, adev->sdma_timeout))
 		r = -ETIMEDOUT;
@@ -1956,7 +1949,9 @@ static int amdgpu_ttm_prepare_job(struct amdgpu_device *adev,
 		AMDGPU_IB_POOL_DELAYED;
 	int r;
 
-	r = amdgpu_job_alloc_with_ib(adev, num_dw * 4, pool, job);
+	r = amdgpu_job_alloc_with_ib(adev, &adev->mman.entity,
+				     AMDGPU_FENCE_OWNER_UNDEFINED,
+				     num_dw * 4, pool, job);
 	if (r)
 		return r;
 
@@ -2015,8 +2010,7 @@ int amdgpu_copy_buffer(struct amdgpu_ring *ring, uint64_t src_offset,
 	if (direct_submit)
 		r = amdgpu_job_submit_direct(job, ring, fence);
 	else
-		r = amdgpu_job_submit(job, &adev->mman.entity,
-				      AMDGPU_FENCE_OWNER_UNDEFINED, fence);
+		*fence = amdgpu_job_submit(job);
 	if (r)
 		goto error_free;
 
@@ -2061,16 +2055,8 @@ static int amdgpu_ttm_fill_mem(struct amdgpu_ring *ring, uint32_t src_data,
 
 	amdgpu_ring_pad_ib(ring, &job->ibs[0]);
 	WARN_ON(job->ibs[0].length_dw > num_dw);
-	r = amdgpu_job_submit(job, &adev->mman.entity,
-			      AMDGPU_FENCE_OWNER_UNDEFINED, fence);
-	if (r)
-		goto error_free;
-
+	*fence = amdgpu_job_submit(job);
 	return 0;
-
-error_free:
-	amdgpu_job_free(job);
-	return r;
 }
 
 int amdgpu_fill_buffer(struct amdgpu_bo *bo,
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c
index 6eac649499d3..8baddf79635b 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c
@@ -1132,7 +1132,9 @@ static int amdgpu_uvd_send_msg(struct amdgpu_ring *ring, struct amdgpu_bo *bo,
 	unsigned offset_idx = 0;
 	unsigned offset[3] = { UVD_BASE_SI, 0, 0 };
 
-	r = amdgpu_job_alloc_with_ib(adev, 64, direct ? AMDGPU_IB_POOL_DIRECT :
+	r = amdgpu_job_alloc_with_ib(ring->adev, &adev->uvd.entity,
+				     AMDGPU_FENCE_OWNER_UNDEFINED,
+				     64, direct ? AMDGPU_IB_POOL_DIRECT :
 				     AMDGPU_IB_POOL_DELAYED, &job);
 	if (r)
 		return r;
@@ -1181,10 +1183,7 @@ static int amdgpu_uvd_send_msg(struct amdgpu_ring *ring, struct amdgpu_bo *bo,
 		if (r)
 			goto err_free;
 
-		r = amdgpu_job_submit(job, &adev->uvd.entity,
-				      AMDGPU_FENCE_OWNER_UNDEFINED, &f);
-		if (r)
-			goto err_free;
+		f = amdgpu_job_submit(job);
 	}
 
 	amdgpu_bo_reserve(bo, true);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c
index 02cb3a12dd76..b239e874f2d5 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c
@@ -450,8 +450,10 @@ static int amdgpu_vce_get_create_msg(struct amdgpu_ring *ring, uint32_t handle,
 	uint64_t addr;
 	int i, r;
 
-	r = amdgpu_job_alloc_with_ib(ring->adev, ib_size_dw * 4,
-				     AMDGPU_IB_POOL_DIRECT, &job);
+	r = amdgpu_job_alloc_with_ib(ring->adev, &ring->adev->vce.entity,
+				     AMDGPU_FENCE_OWNER_UNDEFINED,
+				     ib_size_dw * 4, AMDGPU_IB_POOL_DIRECT,
+				     &job);
 	if (r)
 		return r;
 
@@ -538,7 +540,9 @@ static int amdgpu_vce_get_destroy_msg(struct amdgpu_ring *ring, uint32_t handle,
 	struct dma_fence *f = NULL;
 	int i, r;
 
-	r = amdgpu_job_alloc_with_ib(ring->adev, ib_size_dw * 4,
+	r = amdgpu_job_alloc_with_ib(ring->adev, &ring->adev->vce.entity,
+				     AMDGPU_FENCE_OWNER_UNDEFINED,
+				     ib_size_dw * 4,
 				     direct ? AMDGPU_IB_POOL_DIRECT :
 				     AMDGPU_IB_POOL_DELAYED, &job);
 	if (r)
@@ -570,8 +574,7 @@ static int amdgpu_vce_get_destroy_msg(struct amdgpu_ring *ring, uint32_t handle,
 	if (direct)
 		r = amdgpu_job_submit_direct(job, ring, &f);
 	else
-		r = amdgpu_job_submit(job, &ring->adev->vce.entity,
-				      AMDGPU_FENCE_OWNER_UNDEFINED, &f);
+		f = amdgpu_job_submit(job);
 	if (r)
 		goto err;
 
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c
index 0b52af415b28..3449145ab2bc 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c
@@ -600,15 +600,16 @@ static int amdgpu_vcn_dec_send_msg(struct amdgpu_ring *ring,
 				   struct amdgpu_ib *ib_msg,
 				   struct dma_fence **fence)
 {
+	u64 addr = AMDGPU_GPU_PAGE_ALIGN(ib_msg->gpu_addr);
 	struct amdgpu_device *adev = ring->adev;
 	struct dma_fence *f = NULL;
 	struct amdgpu_job *job;
 	struct amdgpu_ib *ib;
-	uint64_t addr = AMDGPU_GPU_PAGE_ALIGN(ib_msg->gpu_addr);
 	int i, r;
 
-	r = amdgpu_job_alloc_with_ib(adev, 64,
-					AMDGPU_IB_POOL_DIRECT, &job);
+	r = amdgpu_job_alloc_with_ib(ring->adev, NULL, NULL,
+				     64, AMDGPU_IB_POOL_DIRECT,
+				     &job);
 	if (r)
 		goto err;
 
@@ -787,8 +788,9 @@ static int amdgpu_vcn_dec_sw_send_msg(struct amdgpu_ring *ring,
 	if (sq)
 		ib_size_dw += 8;
 
-	r = amdgpu_job_alloc_with_ib(adev, ib_size_dw * 4,
-				AMDGPU_IB_POOL_DIRECT, &job);
+	r = amdgpu_job_alloc_with_ib(ring->adev, NULL, NULL,
+				     ib_size_dw * 4, AMDGPU_IB_POOL_DIRECT,
+				     &job);
 	if (r)
 		goto err;
 
@@ -916,8 +918,9 @@ static int amdgpu_vcn_enc_get_create_msg(struct amdgpu_ring *ring, uint32_t hand
 	if (sq)
 		ib_size_dw += 8;
 
-	r = amdgpu_job_alloc_with_ib(ring->adev, ib_size_dw * 4,
-					AMDGPU_IB_POOL_DIRECT, &job);
+	r = amdgpu_job_alloc_with_ib(ring->adev, NULL, NULL,
+				     ib_size_dw * 4, AMDGPU_IB_POOL_DIRECT,
+				     &job);
 	if (r)
 		return r;
 
@@ -982,8 +985,9 @@ static int amdgpu_vcn_enc_get_destroy_msg(struct amdgpu_ring *ring, uint32_t han
 	if (sq)
 		ib_size_dw += 8;
 
-	r = amdgpu_job_alloc_with_ib(ring->adev, ib_size_dw * 4,
-					AMDGPU_IB_POOL_DIRECT, &job);
+	r = amdgpu_job_alloc_with_ib(ring->adev, NULL, NULL,
+				     ib_size_dw * 4, AMDGPU_IB_POOL_DIRECT,
+				     &job);
 	if (r)
 		return r;
 
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm_sdma.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm_sdma.c
index 2b0669c464f6..6768af365f3b 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm_sdma.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm_sdma.c
@@ -47,6 +47,32 @@ static int amdgpu_vm_sdma_map_table(struct amdgpu_bo_vm *table)
 	return r;
 }
 
+/* Allocate a new job for @count PTE updates */
+static int amdgpu_vm_sdma_alloc_job(struct amdgpu_vm_update_params *p,
+				    unsigned int count)
+{
+	enum amdgpu_ib_pool_type pool = p->immediate ? AMDGPU_IB_POOL_IMMEDIATE
+		: AMDGPU_IB_POOL_DELAYED;
+	struct drm_sched_entity *entity = p->immediate ? &p->vm->immediate
+		: &p->vm->delayed;
+	unsigned int ndw;
+	int r;
+
+	/* estimate how many dw we need */
+	ndw = AMDGPU_VM_SDMA_MIN_NUM_DW;
+	if (p->pages_addr)
+		ndw += count * 2;
+	ndw = min(ndw, AMDGPU_VM_SDMA_MAX_NUM_DW);
+
+	r = amdgpu_job_alloc_with_ib(p->adev, entity, AMDGPU_FENCE_OWNER_VM,
+				     ndw * 4, pool, &p->job);
+	if (r)
+		return r;
+
+	p->num_dw_left = ndw;
+	return 0;
+}
+
 /**
  * amdgpu_vm_sdma_prepare - prepare SDMA command submission
  *
@@ -61,17 +87,12 @@ static int amdgpu_vm_sdma_prepare(struct amdgpu_vm_update_params *p,
 				  struct dma_resv *resv,
 				  enum amdgpu_sync_mode sync_mode)
 {
-	enum amdgpu_ib_pool_type pool = p->immediate ? AMDGPU_IB_POOL_IMMEDIATE
-		: AMDGPU_IB_POOL_DELAYED;
-	unsigned int ndw = AMDGPU_VM_SDMA_MIN_NUM_DW;
 	int r;
 
-	r = amdgpu_job_alloc_with_ib(p->adev, ndw * 4, pool, &p->job);
+	r = amdgpu_vm_sdma_alloc_job(p, 0);
 	if (r)
 		return r;
 
-	p->num_dw_left = ndw;
-
 	if (!resv)
 		return 0;
 
@@ -91,20 +112,16 @@ static int amdgpu_vm_sdma_commit(struct amdgpu_vm_update_params *p,
 				 struct dma_fence **fence)
 {
 	struct amdgpu_ib *ib = p->job->ibs;
-	struct drm_sched_entity *entity;
 	struct amdgpu_ring *ring;
 	struct dma_fence *f;
-	int r;
 
-	entity = p->immediate ? &p->vm->immediate : &p->vm->delayed;
-	ring = container_of(entity->rq->sched, struct amdgpu_ring, sched);
+	ring = container_of(p->vm->delayed.rq->sched, struct amdgpu_ring,
+			    sched);
 
 	WARN_ON(ib->length_dw == 0);
 	amdgpu_ring_pad_ib(ring, ib);
 	WARN_ON(ib->length_dw > p->num_dw_left);
-	r = amdgpu_job_submit(p->job, entity, AMDGPU_FENCE_OWNER_VM, &f);
-	if (r)
-		goto error;
+	f = amdgpu_job_submit(p->job);
 
 	if (p->unlocked) {
 		struct dma_fence *tmp = dma_fence_get(f);
@@ -120,10 +137,6 @@ static int amdgpu_vm_sdma_commit(struct amdgpu_vm_update_params *p,
 		swap(*fence, f);
 	dma_fence_put(f);
 	return 0;
-
-error:
-	amdgpu_job_free(p->job);
-	return r;
 }
 
 /**
@@ -203,8 +216,6 @@ static int amdgpu_vm_sdma_update(struct amdgpu_vm_update_params *p,
 				 uint64_t flags)
 {
 	struct amdgpu_bo *bo = &vmbo->bo;
-	enum amdgpu_ib_pool_type pool = p->immediate ? AMDGPU_IB_POOL_IMMEDIATE
-		: AMDGPU_IB_POOL_DELAYED;
 	struct dma_resv_iter cursor;
 	unsigned int i, ndw, nptes;
 	struct dma_fence *fence;
@@ -231,19 +242,9 @@ static int amdgpu_vm_sdma_update(struct amdgpu_vm_update_params *p,
 			if (r)
 				return r;
 
-			/* estimate how many dw we need */
-			ndw = 32;
-			if (p->pages_addr)
-				ndw += count * 2;
-			ndw = max(ndw, AMDGPU_VM_SDMA_MIN_NUM_DW);
-			ndw = min(ndw, AMDGPU_VM_SDMA_MAX_NUM_DW);
-
-			r = amdgpu_job_alloc_with_ib(p->adev, ndw * 4, pool,
-						     &p->job);
+			r = amdgpu_vm_sdma_alloc_job(p, count);
 			if (r)
 				return r;
-
-			p->num_dw_left = ndw;
 		}
 
 		if (!p->pages_addr) {
diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c b/drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c
index f513e2c2e964..657e53708248 100644
--- a/drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c
@@ -371,7 +371,9 @@ static void gmc_v10_0_flush_gpu_tlb(struct amdgpu_device *adev, uint32_t vmid,
 	 * translation. Avoid this by doing the invalidation from the SDMA
 	 * itself.
 	 */
-	r = amdgpu_job_alloc_with_ib(adev, 16 * 4, AMDGPU_IB_POOL_IMMEDIATE,
+	r = amdgpu_job_alloc_with_ib(ring->adev, &adev->mman.entity,
+				     AMDGPU_FENCE_OWNER_UNDEFINED,
+				     16 * 4, AMDGPU_IB_POOL_IMMEDIATE,
 				     &job);
 	if (r)
 		goto error_alloc;
@@ -380,10 +382,7 @@ static void gmc_v10_0_flush_gpu_tlb(struct amdgpu_device *adev, uint32_t vmid,
 	job->vm_needs_flush = true;
 	job->ibs->ptr[job->ibs->length_dw++] = ring->funcs->nop;
 	amdgpu_ring_pad_ib(ring, &job->ibs[0]);
-	r = amdgpu_job_submit(job, &adev->mman.entity,
-			      AMDGPU_FENCE_OWNER_UNDEFINED, &fence);
-	if (r)
-		goto error_submit;
+	fence = amdgpu_job_submit(job);
 
 	mutex_unlock(&adev->mman.gtt_window_lock);
 
@@ -392,9 +391,6 @@ static void gmc_v10_0_flush_gpu_tlb(struct amdgpu_device *adev, uint32_t vmid,
 
 	return;
 
-error_submit:
-	amdgpu_job_free(job);
-
 error_alloc:
 	mutex_unlock(&adev->mman.gtt_window_lock);
 	DRM_ERROR("Error flushing GPU TLB using the SDMA (%d)!\n", r);
diff --git a/drivers/gpu/drm/amd/amdgpu/uvd_v6_0.c b/drivers/gpu/drm/amd/amdgpu/uvd_v6_0.c
index 375c440957dc..5fe872f4bea7 100644
--- a/drivers/gpu/drm/amd/amdgpu/uvd_v6_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/uvd_v6_0.c
@@ -216,8 +216,8 @@ static int uvd_v6_0_enc_get_create_msg(struct amdgpu_ring *ring, uint32_t handle
 	uint64_t addr;
 	int i, r;
 
-	r = amdgpu_job_alloc_with_ib(ring->adev, ib_size_dw * 4,
-					AMDGPU_IB_POOL_DIRECT, &job);
+	r = amdgpu_job_alloc_with_ib(ring->adev, NULL, NULL, ib_size_dw * 4,
+				     AMDGPU_IB_POOL_DIRECT, &job);
 	if (r)
 		return r;
 
@@ -280,8 +280,8 @@ static int uvd_v6_0_enc_get_destroy_msg(struct amdgpu_ring *ring,
 	uint64_t addr;
 	int i, r;
 
-	r = amdgpu_job_alloc_with_ib(ring->adev, ib_size_dw * 4,
-					AMDGPU_IB_POOL_DIRECT, &job);
+	r = amdgpu_job_alloc_with_ib(ring->adev, NULL, NULL, ib_size_dw * 4,
+				     AMDGPU_IB_POOL_DIRECT, &job);
 	if (r)
 		return r;
 
diff --git a/drivers/gpu/drm/amd/amdgpu/uvd_v7_0.c b/drivers/gpu/drm/amd/amdgpu/uvd_v7_0.c
index e668b3baa8c6..e407be6cb63c 100644
--- a/drivers/gpu/drm/amd/amdgpu/uvd_v7_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/uvd_v7_0.c
@@ -213,7 +213,7 @@ static int uvd_v7_0_enc_ring_test_ring(struct amdgpu_ring *ring)
  *
  * Open up a stream for HW test
  */
-static int uvd_v7_0_enc_get_create_msg(struct amdgpu_ring *ring, uint32_t handle,
+static int uvd_v7_0_enc_get_create_msg(struct amdgpu_ring *ring, u32 handle,
 				       struct amdgpu_bo *bo,
 				       struct dma_fence **fence)
 {
@@ -224,8 +224,8 @@ static int uvd_v7_0_enc_get_create_msg(struct amdgpu_ring *ring, uint32_t handle
 	uint64_t addr;
 	int i, r;
 
-	r = amdgpu_job_alloc_with_ib(ring->adev, ib_size_dw * 4,
-					AMDGPU_IB_POOL_DIRECT, &job);
+	r = amdgpu_job_alloc_with_ib(ring->adev, NULL, NULL, ib_size_dw * 4,
+				     AMDGPU_IB_POOL_DIRECT, &job);
 	if (r)
 		return r;
 
@@ -276,7 +276,7 @@ err:
  *
  * Close up a stream for HW test or if userspace failed to do so
  */
-static int uvd_v7_0_enc_get_destroy_msg(struct amdgpu_ring *ring, uint32_t handle,
+static int uvd_v7_0_enc_get_destroy_msg(struct amdgpu_ring *ring, u32 handle,
 					struct amdgpu_bo *bo,
 					struct dma_fence **fence)
 {
@@ -287,8 +287,8 @@ static int uvd_v7_0_enc_get_destroy_msg(struct amdgpu_ring *ring, uint32_t handl
 	uint64_t addr;
 	int i, r;
 
-	r = amdgpu_job_alloc_with_ib(ring->adev, ib_size_dw * 4,
-					AMDGPU_IB_POOL_DIRECT, &job);
+	r = amdgpu_job_alloc_with_ib(ring->adev, NULL, NULL, ib_size_dw * 4,
+				     AMDGPU_IB_POOL_DIRECT, &job);
 	if (r)
 		return r;
 
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c b/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c
index 2797029bd500..79069c485386 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c
@@ -65,8 +65,11 @@ svm_migrate_gart_map(struct amdgpu_ring *ring, uint64_t npages,
 	num_dw = ALIGN(adev->mman.buffer_funcs->copy_num_dw, 8);
 	num_bytes = npages * 8;
 
-	r = amdgpu_job_alloc_with_ib(adev, num_dw * 4 + num_bytes,
-				     AMDGPU_IB_POOL_DELAYED, &job);
+	r = amdgpu_job_alloc_with_ib(adev, &adev->mman.entity,
+				     AMDGPU_FENCE_OWNER_UNDEFINED,
+				     num_dw * 4 + num_bytes,
+				     AMDGPU_IB_POOL_DELAYED,
+				     &job);
 	if (r)
 		return r;
 
@@ -89,18 +92,10 @@ svm_migrate_gart_map(struct amdgpu_ring *ring, uint64_t npages,
 	cpu_addr = &job->ibs[0].ptr[num_dw];
 
 	amdgpu_gart_map(adev, 0, npages, addr, pte_flags, cpu_addr);
-	r = amdgpu_job_submit(job, &adev->mman.entity,
-			      AMDGPU_FENCE_OWNER_UNDEFINED, &fence);
-	if (r)
-		goto error_free;
-
+	fence = amdgpu_job_submit(job);
 	dma_fence_put(fence);
 
 	return r;
-
-error_free:
-	amdgpu_job_free(job);
-	return r;
 }
 
 /**
-- 
cgit 


From 1b2d5eda5ad785d0dd13484141b78d2ac366c169 Mon Sep 17 00:00:00 2001
From: Christian König <christian.koenig@amd.com>
Date: Thu, 29 Sep 2022 13:05:56 +0200
Subject: drm/amdgpu: move explicit sync check into the CS
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

This moves the memory allocation out of the critical code path.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Luben Tuikov <luben.tuikov@amd.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20221014084641.128280-8-christian.koenig@amd.com
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c  | 13 ++++++++++++-
 drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c  |  2 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_job.c | 12 +++---------
 drivers/gpu/drm/amd/amdgpu/amdgpu_job.h |  2 +-
 4 files changed, 17 insertions(+), 12 deletions(-)

(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_job.c')

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index aa6f6c428dbc..d45b86bcf7fa 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -449,8 +449,19 @@ static int amdgpu_syncobj_lookup_and_add(struct amdgpu_cs_parser *p,
 	}
 
 	r = amdgpu_sync_fence(&p->gang_leader->sync, fence);
-	dma_fence_put(fence);
+	if (r)
+		goto error;
+
+	/*
+	 * When we have an explicit dependency it might be necessary to insert a
+	 * pipeline sync to make sure that all caches etc are flushed and the
+	 * next job actually sees the results from the previous one.
+	 */
+	if (fence->context == p->gang_leader->base.entity->fence_context)
+		r = amdgpu_sync_fence(&p->gang_leader->explicit_sync, fence);
 
+error:
+	dma_fence_put(fence);
 	return r;
 }
 
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c
index 258cffe3c06a..774c77bb8f4e 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c
@@ -182,7 +182,7 @@ int amdgpu_ib_schedule(struct amdgpu_ring *ring, unsigned num_ibs,
 
 	need_ctx_switch = ring->current_ctx != fence_ctx;
 	if (ring->funcs->emit_pipeline_sync && job &&
-	    ((tmp = amdgpu_sync_get_fence(&job->sched_sync)) ||
+	    ((tmp = amdgpu_sync_get_fence(&job->explicit_sync)) ||
 	     (amdgpu_sriov_vf(adev) && need_ctx_switch) ||
 	     amdgpu_vm_need_pipeline_sync(ring, job))) {
 		need_pipe_sync = true;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c
index fa60d25dfbd1..6c126797d6a4 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c
@@ -108,7 +108,7 @@ int amdgpu_job_alloc(struct amdgpu_device *adev, struct amdgpu_vm *vm,
 	(*job)->vm = vm;
 
 	amdgpu_sync_create(&(*job)->sync);
-	amdgpu_sync_create(&(*job)->sched_sync);
+	amdgpu_sync_create(&(*job)->explicit_sync);
 	(*job)->vram_lost_counter = atomic_read(&adev->vram_lost_counter);
 	(*job)->vm_pd_addr = AMDGPU_BO_INVALID_OFFSET;
 
@@ -176,7 +176,7 @@ static void amdgpu_job_free_cb(struct drm_sched_job *s_job)
 	drm_sched_job_cleanup(s_job);
 
 	amdgpu_sync_free(&job->sync);
-	amdgpu_sync_free(&job->sched_sync);
+	amdgpu_sync_free(&job->explicit_sync);
 
 	dma_fence_put(&job->hw_fence);
 }
@@ -204,7 +204,7 @@ void amdgpu_job_free(struct amdgpu_job *job)
 
 	amdgpu_job_free_resources(job);
 	amdgpu_sync_free(&job->sync);
-	amdgpu_sync_free(&job->sched_sync);
+	amdgpu_sync_free(&job->explicit_sync);
 	if (job->gang_submit != &job->base.s_fence->scheduled)
 		dma_fence_put(job->gang_submit);
 
@@ -251,12 +251,6 @@ amdgpu_job_dependency(struct drm_sched_job *sched_job,
 	int r;
 
 	fence = amdgpu_sync_get_fence(&job->sync);
-	if (fence && drm_sched_dependency_optimized(fence, s_entity)) {
-		r = amdgpu_sync_fence(&job->sched_sync, fence);
-		if (r)
-			DRM_ERROR("Error adding fence (%d)\n", r);
-	}
-
 	while (!fence && job->vm && !job->vmid) {
 		r = amdgpu_vmid_grab(job->vm, ring, job, &fence);
 		if (r)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.h
index e2ecab977d0d..ef4bdde7c918 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.h
@@ -48,7 +48,7 @@ struct amdgpu_job {
 	struct drm_sched_job    base;
 	struct amdgpu_vm	*vm;
 	struct amdgpu_sync	sync;
-	struct amdgpu_sync	sched_sync;
+	struct amdgpu_sync	explicit_sync;
 	struct dma_fence	hw_fence;
 	struct dma_fence	*gang_submit;
 	uint32_t		preamble_status;
-- 
cgit 


From 1728baa7e4e60054bf13dd9b1212d133cbd53b3f Mon Sep 17 00:00:00 2001
From: Christian König <christian.koenig@amd.com>
Date: Thu, 29 Sep 2022 14:04:01 +0200
Subject: drm/amdgpu: use scheduler dependencies for CS
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Entirely remove the sync obj in the job.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Luben Tuikov <luben.tuikov@amd.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20221014084641.128280-11-christian.koenig@amd.com
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c  | 21 ++++++++++-----------
 drivers/gpu/drm/amd/amdgpu/amdgpu_cs.h  |  2 ++
 drivers/gpu/drm/amd/amdgpu/amdgpu_job.c |  9 +--------
 drivers/gpu/drm/amd/amdgpu/amdgpu_job.h |  1 -
 4 files changed, 13 insertions(+), 20 deletions(-)

(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_job.c')

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index d45b86bcf7fa..0528c2b1db6e 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -426,7 +426,7 @@ static int amdgpu_cs_p2_dependencies(struct amdgpu_cs_parser *p,
 			dma_fence_put(old);
 		}
 
-		r = amdgpu_sync_fence(&p->gang_leader->sync, fence);
+		r = amdgpu_sync_fence(&p->sync, fence);
 		dma_fence_put(fence);
 		if (r)
 			return r;
@@ -448,7 +448,7 @@ static int amdgpu_syncobj_lookup_and_add(struct amdgpu_cs_parser *p,
 		return r;
 	}
 
-	r = amdgpu_sync_fence(&p->gang_leader->sync, fence);
+	r = amdgpu_sync_fence(&p->sync, fence);
 	if (r)
 		goto error;
 
@@ -1108,7 +1108,7 @@ static int amdgpu_cs_vm_handling(struct amdgpu_cs_parser *p)
 	if (r)
 		return r;
 
-	r = amdgpu_sync_fence(&job->sync, fpriv->prt_va->last_pt_update);
+	r = amdgpu_sync_fence(&p->sync, fpriv->prt_va->last_pt_update);
 	if (r)
 		return r;
 
@@ -1119,7 +1119,7 @@ static int amdgpu_cs_vm_handling(struct amdgpu_cs_parser *p)
 		if (r)
 			return r;
 
-		r = amdgpu_sync_fence(&job->sync, bo_va->last_pt_update);
+		r = amdgpu_sync_fence(&p->sync, bo_va->last_pt_update);
 		if (r)
 			return r;
 	}
@@ -1138,7 +1138,7 @@ static int amdgpu_cs_vm_handling(struct amdgpu_cs_parser *p)
 		if (r)
 			return r;
 
-		r = amdgpu_sync_fence(&job->sync, bo_va->last_pt_update);
+		r = amdgpu_sync_fence(&p->sync, bo_va->last_pt_update);
 		if (r)
 			return r;
 	}
@@ -1151,7 +1151,7 @@ static int amdgpu_cs_vm_handling(struct amdgpu_cs_parser *p)
 	if (r)
 		return r;
 
-	r = amdgpu_sync_fence(&job->sync, vm->last_update);
+	r = amdgpu_sync_fence(&p->sync, vm->last_update);
 	if (r)
 		return r;
 
@@ -1183,7 +1183,6 @@ static int amdgpu_cs_vm_handling(struct amdgpu_cs_parser *p)
 static int amdgpu_cs_sync_rings(struct amdgpu_cs_parser *p)
 {
 	struct amdgpu_fpriv *fpriv = p->filp->driver_priv;
-	struct amdgpu_job *leader = p->gang_leader;
 	struct amdgpu_bo_list_entry *e;
 	unsigned int i;
 	int r;
@@ -1195,14 +1194,14 @@ static int amdgpu_cs_sync_rings(struct amdgpu_cs_parser *p)
 
 		sync_mode = amdgpu_bo_explicit_sync(bo) ?
 			AMDGPU_SYNC_EXPLICIT : AMDGPU_SYNC_NE_OWNER;
-		r = amdgpu_sync_resv(p->adev, &leader->sync, resv, sync_mode,
+		r = amdgpu_sync_resv(p->adev, &p->sync, resv, sync_mode,
 				     &fpriv->vm);
 		if (r)
 			return r;
 	}
 
-	for (i = 0; i < p->gang_size - 1; ++i) {
-		r = amdgpu_sync_clone(&leader->sync, &p->jobs[i]->sync);
+	for (i = 0; i < p->gang_size; ++i) {
+		r = amdgpu_sync_push_to_job(&p->sync, p->jobs[i]);
 		if (r)
 			return r;
 	}
@@ -1248,7 +1247,7 @@ static int amdgpu_cs_submit(struct amdgpu_cs_parser *p,
 		struct dma_fence *fence;
 
 		fence = &p->jobs[i]->base.s_fence->scheduled;
-		r = amdgpu_sync_fence(&leader->sync, fence);
+		r = drm_sched_job_add_dependency(&leader->base, fence);
 		if (r)
 			goto error_cleanup;
 	}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.h
index cbaa19b2b8a3..207e801c24ed 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.h
@@ -75,6 +75,8 @@ struct amdgpu_cs_parser {
 
 	unsigned			num_post_deps;
 	struct amdgpu_cs_post_dep	*post_deps;
+
+	struct amdgpu_sync		sync;
 };
 
 int amdgpu_cs_find_mapping(struct amdgpu_cs_parser *parser,
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c
index 6c126797d6a4..16f7bcc9c654 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c
@@ -107,7 +107,6 @@ int amdgpu_job_alloc(struct amdgpu_device *adev, struct amdgpu_vm *vm,
 	(*job)->base.sched = &adev->rings[0]->sched;
 	(*job)->vm = vm;
 
-	amdgpu_sync_create(&(*job)->sync);
 	amdgpu_sync_create(&(*job)->explicit_sync);
 	(*job)->vram_lost_counter = atomic_read(&adev->vram_lost_counter);
 	(*job)->vm_pd_addr = AMDGPU_BO_INVALID_OFFSET;
@@ -175,9 +174,7 @@ static void amdgpu_job_free_cb(struct drm_sched_job *s_job)
 
 	drm_sched_job_cleanup(s_job);
 
-	amdgpu_sync_free(&job->sync);
 	amdgpu_sync_free(&job->explicit_sync);
-
 	dma_fence_put(&job->hw_fence);
 }
 
@@ -203,7 +200,6 @@ void amdgpu_job_free(struct amdgpu_job *job)
 		drm_sched_job_cleanup(&job->base);
 
 	amdgpu_job_free_resources(job);
-	amdgpu_sync_free(&job->sync);
 	amdgpu_sync_free(&job->explicit_sync);
 	if (job->gang_submit != &job->base.s_fence->scheduled)
 		dma_fence_put(job->gang_submit);
@@ -247,10 +243,9 @@ amdgpu_job_dependency(struct drm_sched_job *sched_job,
 {
 	struct amdgpu_ring *ring = to_amdgpu_ring(s_entity->rq->sched);
 	struct amdgpu_job *job = to_amdgpu_job(sched_job);
-	struct dma_fence *fence;
+	struct dma_fence *fence = NULL;
 	int r;
 
-	fence = amdgpu_sync_get_fence(&job->sync);
 	while (!fence && job->vm && !job->vmid) {
 		r = amdgpu_vmid_grab(job->vm, ring, job, &fence);
 		if (r)
@@ -274,8 +269,6 @@ static struct dma_fence *amdgpu_job_run(struct drm_sched_job *sched_job)
 	job = to_amdgpu_job(sched_job);
 	finished = &job->base.s_fence->finished;
 
-	BUG_ON(amdgpu_sync_peek_fence(&job->sync, NULL));
-
 	trace_amdgpu_sched_run_job(job);
 
 	/* Skip job if VRAM is lost and never resubmit gangs */
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.h
index ef4bdde7c918..a372802ea4e0 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.h
@@ -47,7 +47,6 @@ enum amdgpu_ib_pool_type;
 struct amdgpu_job {
 	struct drm_sched_job    base;
 	struct amdgpu_vm	*vm;
-	struct amdgpu_sync	sync;
 	struct amdgpu_sync	explicit_sync;
 	struct dma_fence	hw_fence;
 	struct dma_fence	*gang_submit;
-- 
cgit 


From a82f30b04c6aaefe62cbbfd297e1bb23435b6b3a Mon Sep 17 00:00:00 2001
From: Christian König <christian.koenig@amd.com>
Date: Thu, 29 Sep 2022 15:01:57 +0200
Subject: drm/scheduler: rename dependency callback into prepare_job
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

This now matches much better what this is doing.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Luben Tuikov <luben.tuikov@amd.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20221014084641.128280-14-christian.koenig@amd.com
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_job.c  |  4 ++--
 drivers/gpu/drm/scheduler/sched_entity.c |  4 ++--
 include/drm/gpu_scheduler.h              | 13 ++++++-------
 3 files changed, 10 insertions(+), 11 deletions(-)

(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_job.c')

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c
index 16f7bcc9c654..172572cfed36 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c
@@ -238,7 +238,7 @@ int amdgpu_job_submit_direct(struct amdgpu_job *job, struct amdgpu_ring *ring,
 }
 
 static struct dma_fence *
-amdgpu_job_dependency(struct drm_sched_job *sched_job,
+amdgpu_job_prepare_job(struct drm_sched_job *sched_job,
 		      struct drm_sched_entity *s_entity)
 {
 	struct amdgpu_ring *ring = to_amdgpu_ring(s_entity->rq->sched);
@@ -327,7 +327,7 @@ void amdgpu_job_stop_all_jobs_on_sched(struct drm_gpu_scheduler *sched)
 }
 
 const struct drm_sched_backend_ops amdgpu_sched_ops = {
-	.dependency = amdgpu_job_dependency,
+	.prepare_job = amdgpu_job_prepare_job,
 	.run_job = amdgpu_job_run,
 	.timedout_job = amdgpu_job_timedout,
 	.free_job = amdgpu_job_free_cb
diff --git a/drivers/gpu/drm/scheduler/sched_entity.c b/drivers/gpu/drm/scheduler/sched_entity.c
index 26ed81acb41e..52bf9056807a 100644
--- a/drivers/gpu/drm/scheduler/sched_entity.c
+++ b/drivers/gpu/drm/scheduler/sched_entity.c
@@ -396,8 +396,8 @@ drm_sched_job_dependency(struct drm_sched_job *job,
 	if (!xa_empty(&job->dependencies))
 		return xa_erase(&job->dependencies, job->last_dependency++);
 
-	if (job->sched->ops->dependency)
-		return job->sched->ops->dependency(job, entity);
+	if (job->sched->ops->prepare_job)
+		return job->sched->ops->prepare_job(job, entity);
 
 	return NULL;
 }
diff --git a/include/drm/gpu_scheduler.h b/include/drm/gpu_scheduler.h
index 09a1e86bb569..e40baefadc3a 100644
--- a/include/drm/gpu_scheduler.h
+++ b/include/drm/gpu_scheduler.h
@@ -320,7 +320,7 @@ struct drm_sched_job {
 	 */
 	union {
 		struct dma_fence_cb		finish_cb;
-		struct work_struct 		work;
+		struct work_struct		work;
 	};
 
 	uint64_t			id;
@@ -368,18 +368,17 @@ enum drm_gpu_sched_stat {
  */
 struct drm_sched_backend_ops {
 	/**
-	 * @dependency:
+	 * @prepare_job:
 	 *
 	 * Called when the scheduler is considering scheduling this job next, to
 	 * get another struct dma_fence for this job to block on.  Once it
 	 * returns NULL, run_job() may be called.
 	 *
-	 * If a driver exclusively uses drm_sched_job_add_dependency() and
-	 * drm_sched_job_add_implicit_dependencies() this can be ommitted and
-	 * left as NULL.
+	 * Can be NULL if no additional preparation to the dependencies are
+	 * necessary. Skipped when jobs are killed instead of run.
 	 */
-	struct dma_fence *(*dependency)(struct drm_sched_job *sched_job,
-					struct drm_sched_entity *s_entity);
+	struct dma_fence *(*prepare_job)(struct drm_sched_job *sched_job,
+					 struct drm_sched_entity *s_entity);
 
 	/**
          * @run_job: Called to execute the job once all of the dependencies
-- 
cgit