diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
index 2fc58e6..95d4969 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
@@ -183,6 +183,7 @@
 struct amdgpu_ring;
 struct amdgpu_semaphore;
 struct amdgpu_cs_parser;
+struct amdgpu_job;
 struct amdgpu_irq_src;
 struct amdgpu_fpriv;
 
@@ -871,7 +872,7 @@
 					 struct amdgpu_ring *ring,
 					 struct amdgpu_ib *ibs,
 					 unsigned num_ibs,
-					 int (*free_job)(struct amdgpu_cs_parser *),
+					 int (*free_job)(struct amdgpu_job *),
 					 void *owner,
 					 struct fence **fence);
 
@@ -1040,6 +1041,7 @@
 
 struct amdgpu_ctx *amdgpu_ctx_get(struct amdgpu_fpriv *fpriv, uint32_t id);
 int amdgpu_ctx_put(struct amdgpu_ctx *ctx);
+struct amdgpu_ctx *amdgpu_ctx_get_ref(struct amdgpu_ctx *ctx);
 
 uint64_t amdgpu_ctx_add_fence(struct amdgpu_ctx *ctx, struct amdgpu_ring *ring,
 			      struct fence *fence, uint64_t queued_seq);
@@ -1265,6 +1267,18 @@
 	struct amd_sched_fence *s_fence;
 };
 
+struct amdgpu_job {
+	struct amd_sched_job    base;
+	struct amdgpu_device	*adev;
+	struct amdgpu_ctx	*ctx;
+	struct drm_file		*owner;
+	struct amdgpu_ib	*ibs;
+	uint32_t		num_ibs;
+	struct mutex            job_lock;
+	struct amdgpu_user_fence uf;
+	int (*free_job)(struct amdgpu_job *sched_job);
+};
+
 static inline u32 amdgpu_get_ib_value(struct amdgpu_cs_parser *p, uint32_t ib_idx, int idx)
 {
 	return p->ibs[ib_idx].ptr[idx];
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index e4424b4..c8de4b6 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -126,19 +126,6 @@
 	return 0;
 }
 
-static void amdgpu_job_work_func(struct work_struct *work)
-{
-	struct amdgpu_cs_parser *sched_job =
-		container_of(work, struct amdgpu_cs_parser,
-			     job_work);
-	mutex_lock(&sched_job->job_lock);
-	if (sched_job->free_job)
-		sched_job->free_job(sched_job);
-	mutex_unlock(&sched_job->job_lock);
-	/* after processing job, free memory */
-	fence_put(&sched_job->s_fence->base);
-	kfree(sched_job);
-}
 struct amdgpu_cs_parser *amdgpu_cs_parser_create(struct amdgpu_device *adev,
                                                struct drm_file *filp,
                                                struct amdgpu_ctx *ctx,
@@ -157,10 +144,6 @@
 	parser->ctx = ctx;
 	parser->ibs = ibs;
 	parser->num_ibs = num_ibs;
-	if (amdgpu_enable_scheduler) {
-		mutex_init(&parser->job_lock);
-		INIT_WORK(&parser->job_work, amdgpu_job_work_func);
-	}
 	for (i = 0; i < num_ibs; i++)
 		ibs[i].ctx = ctx;
 
@@ -508,15 +491,17 @@
 	for (i = 0; i < parser->nchunks; i++)
 		drm_free_large(parser->chunks[i].kdata);
 	kfree(parser->chunks);
-	if (parser->ibs)
-		for (i = 0; i < parser->num_ibs; i++)
-			amdgpu_ib_free(parser->adev, &parser->ibs[i]);
-	kfree(parser->ibs);
-	if (parser->uf.bo)
-		drm_gem_object_unreference_unlocked(&parser->uf.bo->gem_base);
-
 	if (!amdgpu_enable_scheduler)
-		kfree(parser);
+	{
+		if (parser->ibs)
+			for (i = 0; i < parser->num_ibs; i++)
+				amdgpu_ib_free(parser->adev, &parser->ibs[i]);
+		kfree(parser->ibs);
+		if (parser->uf.bo)
+			drm_gem_object_unreference_unlocked(&parser->uf.bo->gem_base);
+	}
+
+	kfree(parser);
 }
 
 /**
@@ -533,12 +518,6 @@
        amdgpu_cs_parser_fini_late(parser);
 }
 
-static int amdgpu_cs_parser_free_job(struct amdgpu_cs_parser *sched_job)
-{
-       amdgpu_cs_parser_fini_late(sched_job);
-       return 0;
-}
-
 static int amdgpu_bo_vm_update_pte(struct amdgpu_cs_parser *p,
 				   struct amdgpu_vm *vm)
 {
@@ -874,6 +853,19 @@
 	return ring;
 }
 
+static int amdgpu_cs_free_job(struct amdgpu_job *sched_job)
+{
+	int i;
+	amdgpu_ctx_put(sched_job->ctx);
+	if (sched_job->ibs)
+		for (i = 0; i < sched_job->num_ibs; i++)
+			amdgpu_ib_free(sched_job->adev, &sched_job->ibs[i]);
+	kfree(sched_job->ibs);
+	if (sched_job->uf.bo)
+		drm_gem_object_unreference_unlocked(&sched_job->uf.bo->gem_base);
+	return 0;
+}
+
 int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
 {
 	struct amdgpu_device *adev = dev->dev_private;
@@ -900,33 +892,50 @@
 	}
 
 	if (amdgpu_enable_scheduler && parser->num_ibs) {
+		struct amdgpu_job *job;
 		struct amdgpu_ring * ring =
 			amdgpu_cs_parser_get_ring(adev, parser);
 		r = amdgpu_cs_parser_prepare_job(parser);
 		if (r)
 			goto out;
-		parser->ring = ring;
-		parser->free_job = amdgpu_cs_parser_free_job;
-		mutex_lock(&parser->job_lock);
-		r = amd_sched_push_job(ring->scheduler,
-				       &parser->ctx->rings[ring->idx].entity,
-				       parser,
-				       &parser->s_fence);
+		job = kzalloc(sizeof(struct amdgpu_job), GFP_KERNEL);
+		if (!job)
+			return -ENOMEM;
+		job->base.sched = ring->scheduler;
+		job->base.s_entity = &parser->ctx->rings[ring->idx].entity;
+		job->adev = parser->adev;
+		job->ibs = parser->ibs;
+		job->num_ibs = parser->num_ibs;
+		job->owner = parser->filp;
+		job->ctx = amdgpu_ctx_get_ref(parser->ctx);
+		mutex_init(&job->job_lock);
+		if (job->ibs[job->num_ibs - 1].user) {
+			memcpy(&job->uf,  &parser->uf,
+			       sizeof(struct amdgpu_user_fence));
+			job->ibs[job->num_ibs - 1].user = &job->uf;
+		}
+
+		job->free_job = amdgpu_cs_free_job;
+		mutex_lock(&job->job_lock);
+		r = amd_sched_push_job((struct amd_sched_job *)job);
 		if (r) {
-			mutex_unlock(&parser->job_lock);
+			mutex_unlock(&job->job_lock);
+			amdgpu_cs_free_job(job);
+			kfree(job);
 			goto out;
 		}
-		parser->ibs[parser->num_ibs - 1].sequence =
-			amdgpu_ctx_add_fence(parser->ctx, ring,
-					     &parser->s_fence->base,
-					     parser->s_fence->v_seq);
-		cs->out.handle = parser->s_fence->v_seq;
+		job->ibs[parser->num_ibs - 1].sequence =
+			amdgpu_ctx_add_fence(job->ctx, ring,
+					     &job->base.s_fence->base,
+					     job->base.s_fence->v_seq);
+		cs->out.handle = job->base.s_fence->v_seq;
 		list_sort(NULL, &parser->validated, cmp_size_smaller_first);
 		ttm_eu_fence_buffer_objects(&parser->ticket,
 				&parser->validated,
-				&parser->s_fence->base);
+				&job->base.s_fence->base);
 
-		mutex_unlock(&parser->job_lock);
+		mutex_unlock(&job->job_lock);
+		amdgpu_cs_parser_fini_late(parser);
 		up_read(&adev->exclusive_lock);
 		return 0;
 	}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
index 08bc772..8660c08 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
@@ -219,6 +219,13 @@
 	return ctx;
 }
 
+struct amdgpu_ctx *amdgpu_ctx_get_ref(struct amdgpu_ctx *ctx)
+{
+	if (ctx)
+		kref_get(&ctx->refcount);
+	return ctx;
+}
+
 int amdgpu_ctx_put(struct amdgpu_ctx *ctx)
 {
 	if (ctx == NULL)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c
index a86e381..5b1ae18 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c
@@ -27,81 +27,58 @@
 #include <drm/drmP.h>
 #include "amdgpu.h"
 
-static int amdgpu_sched_prepare_job(struct amd_gpu_scheduler *sched,
-				    struct amd_sched_entity *entity,
-				    struct amd_sched_job *job)
-{
-	int r = 0;
-	struct amdgpu_cs_parser *sched_job;
-	if (!job || !job->data) {
-		DRM_ERROR("job is null\n");
-		return -EINVAL;
-	}
-
-	sched_job = (struct amdgpu_cs_parser *)job->data;
-	if (sched_job->prepare_job) {
-		r = sched_job->prepare_job(sched_job);
-		if (r) {
-			DRM_ERROR("Prepare job error\n");
-			schedule_work(&sched_job->job_work);
-		}
-	}
-	return r;
-}
-
 static struct fence *amdgpu_sched_run_job(struct amd_gpu_scheduler *sched,
 					  struct amd_sched_entity *entity,
 					  struct amd_sched_job *job)
 {
 	int r = 0;
-	struct amdgpu_cs_parser *sched_job;
+	struct amdgpu_job *sched_job;
 	struct amdgpu_fence *fence;
 
-	if (!job || !job->data) {
+	if (!job) {
 		DRM_ERROR("job is null\n");
 		return NULL;
 	}
-	sched_job = (struct amdgpu_cs_parser *)job->data;
+	sched_job = (struct amdgpu_job *)job;
 	mutex_lock(&sched_job->job_lock);
 	r = amdgpu_ib_schedule(sched_job->adev,
 			       sched_job->num_ibs,
 			       sched_job->ibs,
-			       sched_job->filp);
+			       sched_job->owner);
 	if (r)
 		goto err;
 	fence = amdgpu_fence_ref(sched_job->ibs[sched_job->num_ibs - 1].fence);
 
-	if (sched_job->run_job) {
-		r = sched_job->run_job(sched_job);
-		if (r)
-			goto err;
-	}
-
 	mutex_unlock(&sched_job->job_lock);
 	return &fence->base;
 
 err:
 	DRM_ERROR("Run job error\n");
 	mutex_unlock(&sched_job->job_lock);
-	schedule_work(&sched_job->job_work);
+	sched->ops->process_job(sched, (struct amd_sched_job *)sched_job);
 	return NULL;
 }
 
 static void amdgpu_sched_process_job(struct amd_gpu_scheduler *sched,
 				     struct amd_sched_job *job)
 {
-	struct amdgpu_cs_parser *sched_job;
+	struct amdgpu_job *sched_job;
 
-	if (!job || !job->data) {
+	if (!job) {
 		DRM_ERROR("job is null\n");
 		return;
 	}
-	sched_job = (struct amdgpu_cs_parser *)job->data;
-	schedule_work(&sched_job->job_work);
+	sched_job = (struct amdgpu_job *)job;
+	mutex_lock(&sched_job->job_lock);
+	if (sched_job->free_job)
+		sched_job->free_job(sched_job);
+	mutex_unlock(&sched_job->job_lock);
+	/* after processing job, free memory */
+	fence_put(&sched_job->base.s_fence->base);
+	kfree(sched_job);
 }
 
 struct amd_sched_backend_ops amdgpu_sched_ops = {
-	.prepare_job = amdgpu_sched_prepare_job,
 	.run_job = amdgpu_sched_run_job,
 	.process_job = amdgpu_sched_process_job
 };
@@ -110,31 +87,34 @@
 					 struct amdgpu_ring *ring,
 					 struct amdgpu_ib *ibs,
 					 unsigned num_ibs,
-					 int (*free_job)(struct amdgpu_cs_parser *),
+					 int (*free_job)(struct amdgpu_job *),
 					 void *owner,
 					 struct fence **f)
 {
 	int r = 0;
 	if (amdgpu_enable_scheduler) {
-		struct amdgpu_cs_parser *sched_job =
-			amdgpu_cs_parser_create(adev, owner, &adev->kernel_ctx,
-						ibs, num_ibs);
-		if(!sched_job) {
+		struct amdgpu_job *job =
+			kzalloc(sizeof(struct amdgpu_job), GFP_KERNEL);
+		if (!job)
 			return -ENOMEM;
-		}
-		sched_job->free_job = free_job;
-		mutex_lock(&sched_job->job_lock);
-		r = amd_sched_push_job(ring->scheduler,
-				       &adev->kernel_ctx.rings[ring->idx].entity,
-				       sched_job, &sched_job->s_fence);
+		job->base.sched = ring->scheduler;
+		job->base.s_entity = &adev->kernel_ctx.rings[ring->idx].entity;
+		job->adev = adev;
+		job->ibs = ibs;
+		job->num_ibs = num_ibs;
+		job->owner = owner;
+		mutex_init(&job->job_lock);
+		job->free_job = free_job;
+		mutex_lock(&job->job_lock);
+		r = amd_sched_push_job((struct amd_sched_job *)job);
 		if (r) {
-			mutex_unlock(&sched_job->job_lock);
-			kfree(sched_job);
+			mutex_unlock(&job->job_lock);
+			kfree(job);
 			return r;
 		}
-		ibs[num_ibs - 1].sequence = sched_job->s_fence->v_seq;
-		*f = fence_get(&sched_job->s_fence->base);
-		mutex_unlock(&sched_job->job_lock);
+		ibs[num_ibs - 1].sequence = job->base.s_fence->v_seq;
+		*f = fence_get(&job->base.s_fence->base);
+		mutex_unlock(&job->job_lock);
 	} else {
 		r = amdgpu_ib_schedule(adev, num_ibs, ibs, owner);
 		if (r)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c
index 68369cf..b87355c 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c
@@ -807,7 +807,7 @@
 }
 
 static int amdgpu_uvd_free_job(
-	struct amdgpu_cs_parser *sched_job)
+	struct amdgpu_job *sched_job)
 {
 	amdgpu_ib_free(sched_job->adev, sched_job->ibs);
 	kfree(sched_job->ibs);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c
index 33ee6ae..1a984c9 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c
@@ -340,7 +340,7 @@
 }
 
 static int amdgpu_vce_free_job(
-	struct amdgpu_cs_parser *sched_job)
+	struct amdgpu_job *sched_job)
 {
 	amdgpu_ib_free(sched_job->adev, sched_job->ibs);
 	kfree(sched_job->ibs);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index a78a206..5b99214 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -307,7 +307,7 @@
 }
 
 static int amdgpu_vm_free_job(
-	struct amdgpu_cs_parser *sched_job)
+	struct amdgpu_job *sched_job)
 {
 	int i;
 	for (i = 0; i < sched_job->num_ibs; i++)
diff --git a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c
index 265d3e2..462c161 100644
--- a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c
+++ b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c
@@ -282,30 +282,18 @@
  *	     scheduler consum some queued command.
  *	  -1 other fail.
 */
-int amd_sched_push_job(struct amd_gpu_scheduler *sched,
-		       struct amd_sched_entity *c_entity,
-		       void *data,
-		       struct amd_sched_fence **fence)
+int amd_sched_push_job(struct amd_sched_job *sched_job)
 {
-	struct amd_sched_job *job;
-
+	struct amd_sched_fence 	*fence =
+		amd_sched_fence_create(sched_job->s_entity);
 	if (!fence)
 		return -EINVAL;
-	job = kzalloc(sizeof(struct amd_sched_job), GFP_KERNEL);
-	if (!job)
-		return -ENOMEM;
-	job->sched = sched;
-	job->s_entity = c_entity;
-	job->data = data;
-	*fence = amd_sched_fence_create(c_entity);
-	if ((*fence) == NULL) {
-		kfree(job);
-		return -EINVAL;
-	}
-	fence_get(&(*fence)->base);
-	job->s_fence = *fence;
-	while (kfifo_in_spinlocked(&c_entity->job_queue, &job, sizeof(void *),
-				   &c_entity->queue_lock) != sizeof(void *)) {
+	fence_get(&fence->base);
+	sched_job->s_fence = fence;
+	while (kfifo_in_spinlocked(&sched_job->s_entity->job_queue,
+				   &sched_job, sizeof(void *),
+				   &sched_job->s_entity->queue_lock) !=
+	       sizeof(void *)) {
 		/**
 		 * Current context used up all its IB slots
 		 * wait here, or need to check whether GPU is hung
@@ -313,8 +301,8 @@
 		schedule();
 	}
 	/* first job wake up scheduler */
-	if ((kfifo_len(&c_entity->job_queue) / sizeof(void *)) == 1)
-		wake_up_interruptible(&sched->wait_queue);
+	if ((kfifo_len(&sched_job->s_entity->job_queue) / sizeof(void *)) == 1)
+		wake_up_interruptible(&sched_job->sched->wait_queue);
 	return 0;
 }
 
@@ -333,10 +321,8 @@
 	list_del(&sched_job->list);
 	atomic64_dec(&sched->hw_rq_count);
 	spin_unlock_irqrestore(&sched->queue_lock, flags);
-
-	sched->ops->process_job(sched, sched_job);
 	fence_put(&sched_job->s_fence->base);
-	kfree(sched_job);
+	sched->ops->process_job(sched, sched_job);
 	wake_up_interruptible(&sched->wait_queue);
 }
 
@@ -359,7 +345,9 @@
 		r = kfifo_out(&c_entity->job_queue, &job, sizeof(void *));
 		if (r != sizeof(void *))
 			continue;
-		r = sched->ops->prepare_job(sched, c_entity, job);
+		r = 0;
+		if (sched->ops->prepare_job)
+			r = sched->ops->prepare_job(sched, c_entity, job);
 		if (!r) {
 			unsigned long flags;
 			spin_lock_irqsave(&sched->queue_lock, flags);
diff --git a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h
index ceb5918..25e38d0 100644
--- a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h
+++ b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h
@@ -81,7 +81,6 @@
 	struct fence_cb                 cb;
 	struct amd_gpu_scheduler        *sched;
 	struct amd_sched_entity         *s_entity;
-	void                            *data;
 	struct amd_sched_fence          *s_fence;
 };
 
@@ -140,10 +139,7 @@
 				uint32_t hw_submission);
 int amd_sched_destroy(struct amd_gpu_scheduler *sched);
 
-int amd_sched_push_job(struct amd_gpu_scheduler *sched,
-		       struct amd_sched_entity *c_entity,
-		       void *data,
-		       struct amd_sched_fence **fence);
+int amd_sched_push_job(struct amd_sched_job *sched_job);
 
 int amd_sched_entity_init(struct amd_gpu_scheduler *sched,
 			  struct amd_sched_entity *entity,