Merge branch 'drm-next-4.3' of git://people.freedesktop.org/~agd5f/linux into drm-next

More fixes for radeon and amdgpu for 4.3:
- Send full DP aux address fixes for radeon and amdgpu
- Fix an HDMI display regression for pre-DCE5 parts
- UVD suspend fixes for amdgpu
- Add an rs480 suspend quirk
- Fix bo reserve handling in amdgpu GEM_OP ioctl
- GPU scheduler fixes
- SDMA optimizations
- MEC fix for Fiji

* 'drm-next-4.3' of git://people.freedesktop.org/~agd5f/linux: (21 commits)
  drm/amdgpu: set MEC doorbell range for Fiji
  drm/amdgpu: implement burst NOP for SDMA
  drm/amdgpu: add insert_nop ring func and default implementation
  drm/amdgpu: add amdgpu_get_sdma_instance helper function
  drm/amdgpu: add AMDGPU_MAX_SDMA_INSTANCES
  drm/amdgpu: add burst_nop flag for sdma
  drm/amdgpu: add count field for the SDMA NOP packet v2
  drm/amdgpu: use PT for VM sync on unmap
  drm/amdgpu: make wait_event uninterruptible in push_job
  drm/amdgpu: fix amdgpu_bo_unreserve order in GEM_OP IOCTL v2
  drm/amdgpu: partially revert "modify amdgpu_fence_wait_any() to amdgpu_fence_wait_multiple()" v2
  Add radeon suspend/resume quirk for HP Compaq dc5750.
  drm/amdgpu: re-work sync_resv
  drm/amdgpu/atom: Send out the full AUX address
  drm/radeon/native: Send out the full AUX address
  drm/radeon/atom: Send out the full AUX address
  drm/amdgpu: use IB for fill_buffer instead of direct command
  drm/amdgpu: stop trying to suspend UVD sessions v2
  drm/amdgpu: add scheduler dependency callback v2
  drm/amdgpu: let the scheduler work more with jobs v2
  ...

drivers/gpu/drm/amd/amdgpu/amdgpu.h

@@ -98,6 +98,9 @@ extern int amdgpu_sched_hw_submission;
 #define AMDGPU_MAX_COMPUTE_RINGS		8
 #define AMDGPU_MAX_VCE_RINGS			2
 
+/* max number of IP instances */
+#define AMDGPU_MAX_SDMA_INSTANCES		2
+
 /* number of hw syncs before falling back on blocking */
 #define AMDGPU_NUM_SYNCS			4
 
@@ -262,7 +265,7 @@ struct amdgpu_buffer_funcs {
 	unsigned	fill_num_dw;
 
 	/* used for buffer clearing */
-	void (*emit_fill_buffer)(struct amdgpu_ring *ring,
+	void (*emit_fill_buffer)(struct amdgpu_ib *ib,
 				 /* value to write to memory */
 				 uint32_t src_data,
 				 /* dst addr in bytes */
@@ -340,6 +343,8 @@ struct amdgpu_ring_funcs {
 	int (*test_ring)(struct amdgpu_ring *ring);
 	int (*test_ib)(struct amdgpu_ring *ring);
 	bool (*is_lockup)(struct amdgpu_ring *ring);
+	/* insert NOP packets */
+	void (*insert_nop)(struct amdgpu_ring *ring, uint32_t count);
 };
 
 /*
@@ -440,10 +445,9 @@ int amdgpu_fence_wait_next(struct amdgpu_ring *ring);
 int amdgpu_fence_wait_empty(struct amdgpu_ring *ring);
 unsigned amdgpu_fence_count_emitted(struct amdgpu_ring *ring);
 
-signed long amdgpu_fence_wait_multiple(struct amdgpu_device *adev,
+signed long amdgpu_fence_wait_any(struct amdgpu_device *adev,
 				  struct fence **array,
 				  uint32_t count,
-				       bool wait_all,
 				  bool intr,
 				  signed long t);
 struct amdgpu_fence *amdgpu_fence_ref(struct amdgpu_fence *fence);
@@ -717,6 +721,7 @@ int amdgpu_sync_resv(struct amdgpu_device *adev,
 		     void *owner);
 int amdgpu_sync_rings(struct amdgpu_sync *sync,
 		      struct amdgpu_ring *ring);
+struct fence *amdgpu_sync_get_fence(struct amdgpu_sync *sync);
 int amdgpu_sync_wait(struct amdgpu_sync *sync);
 void amdgpu_sync_free(struct amdgpu_device *adev, struct amdgpu_sync *sync,
 		      struct fence *fence);
@@ -1214,6 +1219,7 @@ int amdgpu_ib_ring_tests(struct amdgpu_device *adev);
 void amdgpu_ring_free_size(struct amdgpu_ring *ring);
 int amdgpu_ring_alloc(struct amdgpu_ring *ring, unsigned ndw);
 int amdgpu_ring_lock(struct amdgpu_ring *ring, unsigned ndw);
+void amdgpu_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count);
 void amdgpu_ring_commit(struct amdgpu_ring *ring);
 void amdgpu_ring_unlock_commit(struct amdgpu_ring *ring);
 void amdgpu_ring_undo(struct amdgpu_ring *ring);
@@ -1665,7 +1671,6 @@ struct amdgpu_uvd {
 	struct amdgpu_bo	*vcpu_bo;
 	void			*cpu_addr;
 	uint64_t		gpu_addr;
-	void			*saved_bo;
 	atomic_t		handles[AMDGPU_MAX_UVD_HANDLES];
 	struct drm_file		*filp[AMDGPU_MAX_UVD_HANDLES];
 	struct delayed_work	idle_work;
@@ -1709,6 +1714,7 @@ struct amdgpu_sdma {
 	uint32_t		feature_version;
 
 	struct amdgpu_ring	ring;
+	bool			burst_nop;
 };
 
 /*
@@ -2057,7 +2063,7 @@ struct amdgpu_device {
 	struct amdgpu_gfx		gfx;
 
 	/* sdma */
-	struct amdgpu_sdma		sdma[2];
+	struct amdgpu_sdma		sdma[AMDGPU_MAX_SDMA_INSTANCES];
 	struct amdgpu_irq_src		sdma_trap_irq;
 	struct amdgpu_irq_src		sdma_illegal_inst_irq;
 
@@ -2196,6 +2202,21 @@ static inline void amdgpu_ring_write(struct amdgpu_ring *ring, uint32_t v)
 	ring->ring_free_dw--;
 }
 
+static inline struct amdgpu_sdma * amdgpu_get_sdma_instance(struct amdgpu_ring *ring)
+{
+	struct amdgpu_device *adev = ring->adev;
+	int i;
+
+	for (i = 0; i < AMDGPU_MAX_SDMA_INSTANCES; i++)
+		if (&adev->sdma[i].ring == ring)
+			break;
+
+	if (i < AMDGPU_MAX_SDMA_INSTANCES)
+		return &adev->sdma[i];
+	else
+		return NULL;
+}
+
 /*
  * ASICs macro.
  */
@@ -2248,7 +2269,7 @@ static inline void amdgpu_ring_write(struct amdgpu_ring *ring, uint32_t v)
 #define amdgpu_display_stop_mc_access(adev, s) (adev)->mode_info.funcs->stop_mc_access((adev), (s))
 #define amdgpu_display_resume_mc_access(adev, s) (adev)->mode_info.funcs->resume_mc_access((adev), (s))
 #define amdgpu_emit_copy_buffer(adev, ib, s, d, b) (adev)->mman.buffer_funcs->emit_copy_buffer((ib),  (s), (d), (b))
-#define amdgpu_emit_fill_buffer(adev, r, s, d, b) (adev)->mman.buffer_funcs->emit_fill_buffer((r), (s), (d), (b))
+#define amdgpu_emit_fill_buffer(adev, ib, s, d, b) (adev)->mman.buffer_funcs->emit_fill_buffer((ib), (s), (d), (b))
 #define amdgpu_dpm_get_temperature(adev) (adev)->pm.funcs->get_temperature((adev))
 #define amdgpu_dpm_pre_set_power_state(adev) (adev)->pm.funcs->pre_set_power_state((adev))
 #define amdgpu_dpm_set_power_state(adev) (adev)->pm.funcs->set_power_state((adev))

drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c

@@ -851,22 +851,6 @@ static bool amdgpu_test_signaled_any(struct fence **fences, uint32_t count)
 	return false;
 }
 
-static bool amdgpu_test_signaled_all(struct fence **fences, uint32_t count)
-{
-	int idx;
-	struct fence *fence;
-
-	for (idx = 0; idx < count; ++idx) {
-		fence = fences[idx];
-		if (fence) {
-			if (!test_bit(FENCE_FLAG_SIGNALED_BIT, &fence->flags))
-				return false;
-		}
-	}
-
-	return true;
-}
-
 struct amdgpu_wait_cb {
 	struct fence_cb base;
 	struct task_struct *task;
@@ -885,7 +869,7 @@ static signed long amdgpu_fence_default_wait(struct fence *f, bool intr,
 	struct amdgpu_fence *fence = to_amdgpu_fence(f);
 	struct amdgpu_device *adev = fence->ring->adev;
 
-	return amdgpu_fence_wait_multiple(adev, &f, 1, false, intr, t);
+	return amdgpu_fence_wait_any(adev, &f, 1, intr, t);
 }
 
 /**
@@ -894,23 +878,18 @@ static signed long amdgpu_fence_default_wait(struct fence *f, bool intr,
  * @adev:     amdgpu device
  * @array:    the fence array with amdgpu fence pointer
  * @count:    the number of the fence array
- * @wait_all: the flag of wait all(true) or wait any(false)
  * @intr:     when sleep, set the current task interruptable or not
  * @t:        timeout to wait
  *
- * If wait_all is true, it will return when all fences are signaled or timeout.
- * If wait_all is false, it will return when any fence is signaled or timeout.
+ * It will return when any fence is signaled or timeout.
  */
-signed long amdgpu_fence_wait_multiple(struct amdgpu_device *adev,
-				       struct fence **array,
-				       uint32_t count,
-				       bool wait_all,
-				       bool intr,
-				       signed long t)
+signed long amdgpu_fence_wait_any(struct amdgpu_device *adev,
+				  struct fence **array, uint32_t count,
+				  bool intr, signed long t)
 {
-	long idx = 0;
 	struct amdgpu_wait_cb *cb;
 	struct fence *fence;
+	unsigned idx;
 
 	BUG_ON(!array);
 
@@ -927,9 +906,6 @@ signed long amdgpu_fence_wait_multiple(struct amdgpu_device *adev,
 			if (fence_add_callback(fence,
 					&cb[idx].base, amdgpu_fence_wait_cb)) {
 				/* The fence is already signaled */
-				if (wait_all)
-					continue;
-				else
 				goto fence_rm_cb;
 			}
 		}
@@ -945,9 +921,7 @@ signed long amdgpu_fence_wait_multiple(struct amdgpu_device *adev,
 		 * amdgpu_test_signaled_any must be called after
 		 * set_current_state to prevent a race with wake_up_process
 		 */
-		if (!wait_all && amdgpu_test_signaled_any(array, count))
-			break;
-		if (wait_all && amdgpu_test_signaled_all(array, count))
+		if (amdgpu_test_signaled_any(array, count))
 			break;
 
 		if (adev->needs_reset) {

drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c

@@ -615,6 +615,7 @@ int amdgpu_gem_op_ioctl(struct drm_device *dev, void *data,
 		info.alignment = robj->tbo.mem.page_alignment << PAGE_SHIFT;
 		info.domains = robj->initial_domain;
 		info.domain_flags = robj->flags;
+		amdgpu_bo_unreserve(robj);
 		if (copy_to_user(out, &info, sizeof(info)))
 			r = -EFAULT;
 		break;
@@ -622,17 +623,19 @@ int amdgpu_gem_op_ioctl(struct drm_device *dev, void *data,
 	case AMDGPU_GEM_OP_SET_PLACEMENT:
 		if (amdgpu_ttm_tt_has_userptr(robj->tbo.ttm)) {
 			r = -EPERM;
+			amdgpu_bo_unreserve(robj);
 			break;
 		}
 		robj->initial_domain = args->value & (AMDGPU_GEM_DOMAIN_VRAM |
 						      AMDGPU_GEM_DOMAIN_GTT |
 						      AMDGPU_GEM_DOMAIN_CPU);
+		amdgpu_bo_unreserve(robj);
 		break;
 	default:
+		amdgpu_bo_unreserve(robj);
 		r = -EINVAL;
 	}
 
-	amdgpu_bo_unreserve(robj);
 out:
 	drm_gem_object_unreference_unlocked(gobj);
 	return r;

drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c

@@ -131,6 +131,21 @@ int amdgpu_ring_lock(struct amdgpu_ring *ring, unsigned ndw)
 	return 0;
 }
 
+/** amdgpu_ring_insert_nop - insert NOP packets
+ *
+ * @ring: amdgpu_ring structure holding ring information
+ * @count: the number of NOP packets to insert
+ *
+ * This is the generic insert_nop function for rings except SDMA
+ */
+void amdgpu_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count)
+{
+	int i;
+
+	for (i = 0; i < count; i++)
+		amdgpu_ring_write(ring, ring->nop);
+}
+
 /**
  * amdgpu_ring_commit - tell the GPU to execute the new
  * commands on the ring buffer
@@ -143,10 +158,13 @@ int amdgpu_ring_lock(struct amdgpu_ring *ring, unsigned ndw)
  */
 void amdgpu_ring_commit(struct amdgpu_ring *ring)
 {
+	uint32_t count;
+
 	/* We pad to match fetch size */
-	while (ring->wptr & ring->align_mask) {
-		amdgpu_ring_write(ring, ring->nop);
-	}
+	count = ring->align_mask + 1 - (ring->wptr & ring->align_mask);
+	count %= ring->align_mask + 1;
+	ring->funcs->insert_nop(ring, count);
+
 	mb();
 	amdgpu_ring_set_wptr(ring);
 }

drivers/gpu/drm/amd/amdgpu/amdgpu_sa.c

@@ -367,8 +367,8 @@ int amdgpu_sa_bo_new(struct amdgpu_device *adev,
 		} while (amdgpu_sa_bo_next_hole(sa_manager, fences, tries));
 
 		spin_unlock(&sa_manager->wq.lock);
-		t = amdgpu_fence_wait_multiple(adev, fences, AMDGPU_MAX_RINGS, false, false,
-						MAX_SCHEDULE_TIMEOUT);
+		t = amdgpu_fence_wait_any(adev, fences, AMDGPU_MAX_RINGS,
+					  false, MAX_SCHEDULE_TIMEOUT);
 		r = (t > 0) ? 0 : t;
 		spin_lock(&sa_manager->wq.lock);
 		/* if we have nothing to wait for block */

drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c

@@ -27,6 +27,12 @@
 #include <drm/drmP.h>
 #include "amdgpu.h"
 
+static struct fence *amdgpu_sched_dependency(struct amd_sched_job *job)
+{
+	struct amdgpu_job *sched_job = (struct amdgpu_job *)job;
+	return amdgpu_sync_get_fence(&sched_job->ibs->sync);
+}
+
 static struct fence *amdgpu_sched_run_job(struct amd_sched_job *job)
 {
 	struct amdgpu_job *sched_job;
@@ -75,6 +81,7 @@ static void amdgpu_sched_process_job(struct amd_sched_job *job)
 }
 
 struct amd_sched_backend_ops amdgpu_sched_ops = {
+	.dependency = amdgpu_sched_dependency,
 	.run_job = amdgpu_sched_run_job,
 	.process_job = amdgpu_sched_process_job
 };

drivers/gpu/drm/amd/amdgpu/amdgpu_sync.c

@@ -142,6 +142,18 @@ int amdgpu_sync_fence(struct amdgpu_device *adev, struct amdgpu_sync *sync,
 	return 0;
 }
 
+static void *amdgpu_sync_get_owner(struct fence *f)
+{
+	struct amdgpu_fence *a_fence = to_amdgpu_fence(f);
+	struct amd_sched_fence *s_fence = to_amd_sched_fence(f);
+
+	if (s_fence)
+		return s_fence->owner;
+	else if (a_fence)
+		return a_fence->owner;
+	return AMDGPU_FENCE_OWNER_UNDEFINED;
+}
+
 /**
  * amdgpu_sync_resv - use the semaphores to sync to a reservation object
  *
@@ -158,7 +170,7 @@ int amdgpu_sync_resv(struct amdgpu_device *adev,
 {
 	struct reservation_object_list *flist;
 	struct fence *f;
-	struct amdgpu_fence *fence;
+	void *fence_owner;
 	unsigned i;
 	int r = 0;
 
@@ -176,22 +188,22 @@ int amdgpu_sync_resv(struct amdgpu_device *adev,
 	for (i = 0; i < flist->shared_count; ++i) {
 		f = rcu_dereference_protected(flist->shared[i],
 					      reservation_object_held(resv));
-		fence = f ? to_amdgpu_fence(f) : NULL;
-		if (fence && fence->ring->adev == adev) {
+		if (amdgpu_sync_same_dev(adev, f)) {
 			/* VM updates are only interesting
 			 * for other VM updates and moves.
 			 */
+			fence_owner = amdgpu_sync_get_owner(f);
 			if ((owner != AMDGPU_FENCE_OWNER_MOVE) &&
-			    (fence->owner != AMDGPU_FENCE_OWNER_MOVE) &&
+			    (fence_owner != AMDGPU_FENCE_OWNER_MOVE) &&
 			    ((owner == AMDGPU_FENCE_OWNER_VM) !=
-			     (fence->owner == AMDGPU_FENCE_OWNER_VM)))
+			     (fence_owner == AMDGPU_FENCE_OWNER_VM)))
 				continue;
 
 			/* Ignore fence from the same owner as
 			 * long as it isn't undefined.
 			 */
 			if (owner != AMDGPU_FENCE_OWNER_UNDEFINED &&
-			    fence->owner == owner)
+			    fence_owner == owner)
 				continue;
 		}
 
@@ -202,6 +214,28 @@ int amdgpu_sync_resv(struct amdgpu_device *adev,
 	return r;
 }
 
+struct fence *amdgpu_sync_get_fence(struct amdgpu_sync *sync)
+{
+	struct amdgpu_sync_entry *e;
+	struct hlist_node *tmp;
+	struct fence *f;
+	int i;
+
+	hash_for_each_safe(sync->fences, i, tmp, e, node) {
+
+		f = e->fence;
+
+		hash_del(&e->node);
+		kfree(e);
+
+		if (!fence_is_signaled(f))
+			return f;
+
+		fence_put(f);
+	}
+	return NULL;
+}
+
 int amdgpu_sync_wait(struct amdgpu_sync *sync)
 {
 	struct amdgpu_sync_entry *e;

drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c

@@ -221,31 +221,32 @@ int amdgpu_uvd_sw_fini(struct amdgpu_device *adev)
 
 int amdgpu_uvd_suspend(struct amdgpu_device *adev)
 {
-	unsigned size;
-	void *ptr;
-	const struct common_firmware_header *hdr;
-	int i;
+	struct amdgpu_ring *ring = &adev->uvd.ring;
+	int i, r;
 
 	if (adev->uvd.vcpu_bo == NULL)
 		return 0;
 
-	for (i = 0; i < AMDGPU_MAX_UVD_HANDLES; ++i)
-		if (atomic_read(&adev->uvd.handles[i]))
-			break;
-
-	if (i == AMDGPU_MAX_UVD_HANDLES)
-		return 0;
+	for (i = 0; i < AMDGPU_MAX_UVD_HANDLES; ++i) {
+		uint32_t handle = atomic_read(&adev->uvd.handles[i]);
+		if (handle != 0) {
+			struct fence *fence;
 
-	hdr = (const struct common_firmware_header *)adev->uvd.fw->data;
+			amdgpu_uvd_note_usage(adev);
 
-	size = amdgpu_bo_size(adev->uvd.vcpu_bo);
-	size -= le32_to_cpu(hdr->ucode_size_bytes);
+			r = amdgpu_uvd_get_destroy_msg(ring, handle, &fence);
+			if (r) {
+				DRM_ERROR("Error destroying UVD (%d)!\n", r);
+				continue;
+			}
 
-	ptr = adev->uvd.cpu_addr;
-	ptr += le32_to_cpu(hdr->ucode_size_bytes);
+			fence_wait(fence, false);
+			fence_put(fence);
 
-	adev->uvd.saved_bo = kmalloc(size, GFP_KERNEL);
-	memcpy(adev->uvd.saved_bo, ptr, size);
+			adev->uvd.filp[i] = NULL;
+			atomic_set(&adev->uvd.handles[i], 0);
+		}
+	}
 
 	return 0;
 }
@@ -270,11 +271,6 @@ int amdgpu_uvd_resume(struct amdgpu_device *adev)
 	ptr = adev->uvd.cpu_addr;
 	ptr += le32_to_cpu(hdr->ucode_size_bytes);
 
-	if (adev->uvd.saved_bo != NULL) {
-		memcpy(ptr, adev->uvd.saved_bo, size);
-		kfree(adev->uvd.saved_bo);
-		adev->uvd.saved_bo = NULL;
-	} else
 	memset(ptr, 0, size);
 
 	return 0;

drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c

@@ -627,9 +627,14 @@ static int amdgpu_vm_update_ptes(struct amdgpu_device *adev,
 {
 	uint64_t mask = AMDGPU_VM_PTE_COUNT - 1;
 	uint64_t last_pte = ~0, last_dst = ~0;
+	void *owner = AMDGPU_FENCE_OWNER_VM;
 	unsigned count = 0;
 	uint64_t addr;
 
+	/* sync to everything on unmapping */
+	if (!(flags & AMDGPU_PTE_VALID))
+		owner = AMDGPU_FENCE_OWNER_UNDEFINED;
+
 	/* walk over the address space and update the page tables */
 	for (addr = start; addr < end; ) {
 		uint64_t pt_idx = addr >> amdgpu_vm_block_size;
@@ -638,8 +643,7 @@ static int amdgpu_vm_update_ptes(struct amdgpu_device *adev,
 		uint64_t pte;
 		int r;
 
-		amdgpu_sync_resv(adev, &ib->sync, pt->tbo.resv,
-				 AMDGPU_FENCE_OWNER_VM);
+		amdgpu_sync_resv(adev, &ib->sync, pt->tbo.resv, owner);
 		r = reservation_object_reserve_shared(pt->tbo.resv);
 		if (r)
 			return r;
@@ -790,17 +794,6 @@ static int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev,
 
 	ib->length_dw = 0;
 
-	if (!(flags & AMDGPU_PTE_VALID)) {
-		unsigned i;
-
-		for (i = 0; i < AMDGPU_MAX_RINGS; ++i) {
-			struct amdgpu_fence *f = vm->ids[i].last_id_use;
-			r = amdgpu_sync_fence(adev, &ib->sync, &f->base);
-			if (r)
-				return r;
-		}
-	}
-
 	r = amdgpu_vm_update_ptes(adev, vm, ib, mapping->it.start,
 				  mapping->it.last + 1, addr + mapping->offset,
 				  flags, gtt_flags);

drivers/gpu/drm/amd/amdgpu/atombios_dp.c

@@ -139,7 +139,8 @@ amdgpu_atombios_dp_aux_transfer(struct drm_dp_aux *aux, struct drm_dp_aux_msg *m
 
 	tx_buf[0] = msg->address & 0xff;
 	tx_buf[1] = msg->address >> 8;
-	tx_buf[2] = msg->request << 4;
+	tx_buf[2] = (msg->request << 4) |
+		((msg->address >> 16) & 0xf);
 	tx_buf[3] = msg->size ? (msg->size - 1) : 0;
 
 	switch (msg->request & ~DP_AUX_I2C_MOT) {

drivers/gpu/drm/amd/amdgpu/cik_sdma.c

@@ -188,6 +188,19 @@ static void cik_sdma_ring_set_wptr(struct amdgpu_ring *ring)
 	WREG32(mmSDMA0_GFX_RB_WPTR + sdma_offsets[me], (ring->wptr << 2) & 0x3fffc);
 }
 
+static void cik_sdma_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count)
+{
+	struct amdgpu_sdma *sdma = amdgpu_get_sdma_instance(ring);
+	int i;
+
+	for (i = 0; i < count; i++)
+		if (sdma && sdma->burst_nop && (i == 0))
+			amdgpu_ring_write(ring, ring->nop |
+					  SDMA_NOP_COUNT(count - 1));
+		else
+			amdgpu_ring_write(ring, ring->nop);
+}
+
 /**
  * cik_sdma_ring_emit_ib - Schedule an IB on the DMA engine
  *
@@ -213,8 +226,8 @@ static void cik_sdma_ring_emit_ib(struct amdgpu_ring *ring,
 	amdgpu_ring_write(ring, next_rptr);
 
 	/* IB packet must end on a 8 DW boundary */
-	while ((ring->wptr & 7) != 4)
-		amdgpu_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_NOP, 0, 0));
+	cik_sdma_ring_insert_nop(ring, (12 - (ring->wptr & 7)) % 8);
+
 	amdgpu_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_INDIRECT_BUFFER, 0, extra_bits));
 	amdgpu_ring_write(ring, ib->gpu_addr & 0xffffffe0); /* base must be 32 byte aligned */
 	amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr) & 0xffffffff);
@@ -501,6 +514,8 @@ static int cik_sdma_load_microcode(struct amdgpu_device *adev)
 		fw_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4;
 		adev->sdma[i].fw_version = le32_to_cpu(hdr->header.ucode_version);
 		adev->sdma[i].feature_version = le32_to_cpu(hdr->ucode_feature_version);
+		if (adev->sdma[i].feature_version >= 20)
+			adev->sdma[i].burst_nop = true;
 		fw_data = (const __le32 *)
 			(adev->sdma[i].fw->data + le32_to_cpu(hdr->header.ucode_array_offset_bytes));
 		WREG32(mmSDMA0_UCODE_ADDR + sdma_offsets[i], 0);
@@ -815,8 +830,19 @@ static void cik_sdma_vm_set_pte_pde(struct amdgpu_ib *ib,
  */
 static void cik_sdma_vm_pad_ib(struct amdgpu_ib *ib)
 {
-	while (ib->length_dw & 0x7)
-		ib->ptr[ib->length_dw++] = SDMA_PACKET(SDMA_OPCODE_NOP, 0, 0);
+	struct amdgpu_sdma *sdma = amdgpu_get_sdma_instance(ib->ring);
+	u32 pad_count;
+	int i;
+
+	pad_count = (8 - (ib->length_dw & 0x7)) % 8;
+	for (i = 0; i < pad_count; i++)
+		if (sdma && sdma->burst_nop && (i == 0))
+			ib->ptr[ib->length_dw++] =
+					SDMA_PACKET(SDMA_OPCODE_NOP, 0, 0) |
+					SDMA_NOP_COUNT(pad_count - 1);
+		else
+			ib->ptr[ib->length_dw++] =
+					SDMA_PACKET(SDMA_OPCODE_NOP, 0, 0);
 }
 
 /**
@@ -1303,6 +1329,7 @@ static const struct amdgpu_ring_funcs cik_sdma_ring_funcs = {
 	.test_ring = cik_sdma_ring_test_ring,
 	.test_ib = cik_sdma_ring_test_ib,
 	.is_lockup = cik_sdma_ring_is_lockup,
+	.insert_nop = cik_sdma_ring_insert_nop,
 };
 
 static void cik_sdma_set_ring_funcs(struct amdgpu_device *adev)
@@ -1363,16 +1390,16 @@ static void cik_sdma_emit_copy_buffer(struct amdgpu_ib *ib,
  *
  * Fill GPU buffers using the DMA engine (CIK).
  */
-static void cik_sdma_emit_fill_buffer(struct amdgpu_ring *ring,
+static void cik_sdma_emit_fill_buffer(struct amdgpu_ib *ib,
 				      uint32_t src_data,
 				      uint64_t dst_offset,
 				      uint32_t byte_count)
 {
-	amdgpu_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_CONSTANT_FILL, 0, 0));
-	amdgpu_ring_write(ring, lower_32_bits(dst_offset));
-	amdgpu_ring_write(ring, upper_32_bits(dst_offset));
-	amdgpu_ring_write(ring, src_data);
-	amdgpu_ring_write(ring, byte_count);
+	ib->ptr[ib->length_dw++] = SDMA_PACKET(SDMA_OPCODE_CONSTANT_FILL, 0, 0);
+	ib->ptr[ib->length_dw++] = lower_32_bits(dst_offset);
+	ib->ptr[ib->length_dw++] = upper_32_bits(dst_offset);
+	ib->ptr[ib->length_dw++] = src_data;
+	ib->ptr[ib->length_dw++] = byte_count;
 }
 
 static const struct amdgpu_buffer_funcs cik_sdma_buffer_funcs = {

drivers/gpu/drm/amd/amdgpu/cikd.h

@@ -487,6 +487,7 @@
 					 (((op) & 0xFF) << 0))
 /* sDMA opcodes */
 #define	SDMA_OPCODE_NOP					  0
+#	define SDMA_NOP_COUNT(x)			  (((x) & 0x3FFF) << 16)
 #define	SDMA_OPCODE_COPY				  1
 #       define SDMA_COPY_SUB_OPCODE_LINEAR                0
 #       define SDMA_COPY_SUB_OPCODE_TILED                 1

drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c

@@ -5598,6 +5598,7 @@ static const struct amdgpu_ring_funcs gfx_v7_0_ring_funcs_gfx = {
 	.test_ring = gfx_v7_0_ring_test_ring,
 	.test_ib = gfx_v7_0_ring_test_ib,
 	.is_lockup = gfx_v7_0_ring_is_lockup,
+	.insert_nop = amdgpu_ring_insert_nop,
 };
 
 static const struct amdgpu_ring_funcs gfx_v7_0_ring_funcs_compute = {
@@ -5614,6 +5615,7 @@ static const struct amdgpu_ring_funcs gfx_v7_0_ring_funcs_compute = {
 	.test_ring = gfx_v7_0_ring_test_ring,
 	.test_ib = gfx_v7_0_ring_test_ib,
 	.is_lockup = gfx_v7_0_ring_is_lockup,
+	.insert_nop = amdgpu_ring_insert_nop,
 };
 
 static void gfx_v7_0_set_ring_funcs(struct amdgpu_device *adev)

drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c

@@ -3240,7 +3240,8 @@ static int gfx_v8_0_cp_compute_resume(struct amdgpu_device *adev)
 
 		/* enable the doorbell if requested */
 		if (use_doorbell) {
-			if (adev->asic_type == CHIP_CARRIZO) {
+			if ((adev->asic_type == CHIP_CARRIZO) ||
+			    (adev->asic_type == CHIP_FIJI)) {
 				WREG32(mmCP_MEC_DOORBELL_RANGE_LOWER,
 				       AMDGPU_DOORBELL_KIQ << 2);
 				WREG32(mmCP_MEC_DOORBELL_RANGE_UPPER,
@@ -4378,6 +4379,7 @@ static const struct amdgpu_ring_funcs gfx_v8_0_ring_funcs_gfx = {
 	.test_ring = gfx_v8_0_ring_test_ring,
 	.test_ib = gfx_v8_0_ring_test_ib,
 	.is_lockup = gfx_v8_0_ring_is_lockup,
+	.insert_nop = amdgpu_ring_insert_nop,
 };
 
 static const struct amdgpu_ring_funcs gfx_v8_0_ring_funcs_compute = {
@@ -4394,6 +4396,7 @@ static const struct amdgpu_ring_funcs gfx_v8_0_ring_funcs_compute = {
 	.test_ring = gfx_v8_0_ring_test_ring,
 	.test_ib = gfx_v8_0_ring_test_ib,
 	.is_lockup = gfx_v8_0_ring_is_lockup,
+	.insert_nop = amdgpu_ring_insert_nop,
 };
 
 static void gfx_v8_0_set_ring_funcs(struct amdgpu_device *adev)

drivers/gpu/drm/amd/amdgpu/iceland_sdma_pkt_open.h

@@ -2163,5 +2163,10 @@
 #define SDMA_PKT_NOP_HEADER_sub_op_shift  8
 #define SDMA_PKT_NOP_HEADER_SUB_OP(x) (((x) & SDMA_PKT_NOP_HEADER_sub_op_mask) << SDMA_PKT_NOP_HEADER_sub_op_shift)
 
+/*define for count field*/
+#define SDMA_PKT_NOP_HEADER_count_offset 0
+#define SDMA_PKT_NOP_HEADER_count_mask   0x00003FFF
+#define SDMA_PKT_NOP_HEADER_count_shift  16
+#define SDMA_PKT_NOP_HEADER_COUNT(x) (((x) & SDMA_PKT_NOP_HEADER_count_mask) << SDMA_PKT_NOP_HEADER_count_shift)
 
 #endif /* __ICELAND_SDMA_PKT_OPEN_H_ */

drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c

@@ -146,6 +146,8 @@ static int sdma_v2_4_init_microcode(struct amdgpu_device *adev)
 		hdr = (const struct sdma_firmware_header_v1_0 *)adev->sdma[i].fw->data;
 		adev->sdma[i].fw_version = le32_to_cpu(hdr->header.ucode_version);
 		adev->sdma[i].feature_version = le32_to_cpu(hdr->ucode_feature_version);
+		if (adev->sdma[i].feature_version >= 20)
+			adev->sdma[i].burst_nop = true;
 
 		if (adev->firmware.smu_load) {
 			info = &adev->firmware.ucode[AMDGPU_UCODE_ID_SDMA0 + i];
@@ -218,6 +220,19 @@ static void sdma_v2_4_ring_set_wptr(struct amdgpu_ring *ring)
 	WREG32(mmSDMA0_GFX_RB_WPTR + sdma_offsets[me], ring->wptr << 2);
 }
 
+static void sdma_v2_4_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count)
+{
+	struct amdgpu_sdma *sdma = amdgpu_get_sdma_instance(ring);
+	int i;
+
+	for (i = 0; i < count; i++)
+		if (sdma && sdma->burst_nop && (i == 0))
+			amdgpu_ring_write(ring, ring->nop |
+				SDMA_PKT_NOP_HEADER_COUNT(count - 1));
+		else
+			amdgpu_ring_write(ring, ring->nop);
+}
+
 /**
  * sdma_v2_4_ring_emit_ib - Schedule an IB on the DMA engine
  *
@@ -245,8 +260,8 @@ static void sdma_v2_4_ring_emit_ib(struct amdgpu_ring *ring,
 	amdgpu_ring_write(ring, next_rptr);
 
 	/* IB packet must end on a 8 DW boundary */
-	while ((ring->wptr & 7) != 2)
-		amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_NOP));
+	sdma_v2_4_ring_insert_nop(ring, (10 - (ring->wptr & 7)) % 8);
+
 	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_INDIRECT) |
 			  SDMA_PKT_INDIRECT_HEADER_VMID(vmid));
 	/* base must be 32 byte aligned */
@@ -879,8 +894,19 @@ static void sdma_v2_4_vm_set_pte_pde(struct amdgpu_ib *ib,
  */
 static void sdma_v2_4_vm_pad_ib(struct amdgpu_ib *ib)
 {
-	while (ib->length_dw & 0x7)
-		ib->ptr[ib->length_dw++] = SDMA_PKT_HEADER_OP(SDMA_OP_NOP);
+	struct amdgpu_sdma *sdma = amdgpu_get_sdma_instance(ib->ring);
+	u32 pad_count;
+	int i;
+
+	pad_count = (8 - (ib->length_dw & 0x7)) % 8;
+	for (i = 0; i < pad_count; i++)
+		if (sdma && sdma->burst_nop && (i == 0))
+			ib->ptr[ib->length_dw++] =
+				SDMA_PKT_HEADER_OP(SDMA_OP_NOP) |
+				SDMA_PKT_NOP_HEADER_COUNT(pad_count - 1);
+		else
+			ib->ptr[ib->length_dw++] =
+				SDMA_PKT_HEADER_OP(SDMA_OP_NOP);
 }
 
 /**
@@ -1314,6 +1340,7 @@ static const struct amdgpu_ring_funcs sdma_v2_4_ring_funcs = {
 	.test_ring = sdma_v2_4_ring_test_ring,
 	.test_ib = sdma_v2_4_ring_test_ib,
 	.is_lockup = sdma_v2_4_ring_is_lockup,
+	.insert_nop = sdma_v2_4_ring_insert_nop,
 };
 
 static void sdma_v2_4_set_ring_funcs(struct amdgpu_device *adev)
@@ -1375,16 +1402,16 @@ static void sdma_v2_4_emit_copy_buffer(struct amdgpu_ib *ib,
  *
  * Fill GPU buffers using the DMA engine (VI).
  */
-static void sdma_v2_4_emit_fill_buffer(struct amdgpu_ring *ring,
+static void sdma_v2_4_emit_fill_buffer(struct amdgpu_ib *ib,
 				       uint32_t src_data,
 				       uint64_t dst_offset,
 				       uint32_t byte_count)
 {
-	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_CONST_FILL));
-	amdgpu_ring_write(ring, lower_32_bits(dst_offset));
-	amdgpu_ring_write(ring, upper_32_bits(dst_offset));
-	amdgpu_ring_write(ring, src_data);
-	amdgpu_ring_write(ring, byte_count);
+	ib->ptr[ib->length_dw++] = SDMA_PKT_HEADER_OP(SDMA_OP_CONST_FILL);
+	ib->ptr[ib->length_dw++] = lower_32_bits(dst_offset);
+	ib->ptr[ib->length_dw++] = upper_32_bits(dst_offset);
+	ib->ptr[ib->length_dw++] = src_data;
+	ib->ptr[ib->length_dw++] = byte_count;
 }
 
 static const struct amdgpu_buffer_funcs sdma_v2_4_buffer_funcs = {

drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c

@@ -218,6 +218,8 @@ static int sdma_v3_0_init_microcode(struct amdgpu_device *adev)
 		hdr = (const struct sdma_firmware_header_v1_0 *)adev->sdma[i].fw->data;
 		adev->sdma[i].fw_version = le32_to_cpu(hdr->header.ucode_version);
 		adev->sdma[i].feature_version = le32_to_cpu(hdr->ucode_feature_version);
+		if (adev->sdma[i].feature_version >= 20)
+			adev->sdma[i].burst_nop = true;
 
 		if (adev->firmware.smu_load) {
 			info = &adev->firmware.ucode[AMDGPU_UCODE_ID_SDMA0 + i];
@@ -304,6 +306,19 @@ static void sdma_v3_0_ring_set_wptr(struct amdgpu_ring *ring)
 	}
 }
 
+static void sdma_v3_0_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count)
+{
+	struct amdgpu_sdma *sdma = amdgpu_get_sdma_instance(ring);
+	int i;
+
+	for (i = 0; i < count; i++)
+		if (sdma && sdma->burst_nop && (i == 0))
+			amdgpu_ring_write(ring, ring->nop |
+				SDMA_PKT_NOP_HEADER_COUNT(count - 1));
+		else
+			amdgpu_ring_write(ring, ring->nop);
+}
+
 /**
  * sdma_v3_0_ring_emit_ib - Schedule an IB on the DMA engine
  *
@@ -330,8 +345,7 @@ static void sdma_v3_0_ring_emit_ib(struct amdgpu_ring *ring,
 	amdgpu_ring_write(ring, next_rptr);
 
 	/* IB packet must end on a 8 DW boundary */
-	while ((ring->wptr & 7) != 2)
-		amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_NOP));
+	sdma_v3_0_ring_insert_nop(ring, (10 - (ring->wptr & 7)) % 8);
 
 	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_INDIRECT) |
 			  SDMA_PKT_INDIRECT_HEADER_VMID(vmid));
@@ -999,8 +1013,19 @@ static void sdma_v3_0_vm_set_pte_pde(struct amdgpu_ib *ib,
  */
 static void sdma_v3_0_vm_pad_ib(struct amdgpu_ib *ib)
 {
-	while (ib->length_dw & 0x7)
-		ib->ptr[ib->length_dw++] = SDMA_PKT_HEADER_OP(SDMA_OP_NOP);
+	struct amdgpu_sdma *sdma = amdgpu_get_sdma_instance(ib->ring);
+	u32 pad_count;
+	int i;
+
+	pad_count = (8 - (ib->length_dw & 0x7)) % 8;
+	for (i = 0; i < pad_count; i++)
+		if (sdma && sdma->burst_nop && (i == 0))
+			ib->ptr[ib->length_dw++] =
+				SDMA_PKT_HEADER_OP(SDMA_OP_NOP) |
+				SDMA_PKT_NOP_HEADER_COUNT(pad_count - 1);
+		else
+			ib->ptr[ib->length_dw++] =
+				SDMA_PKT_HEADER_OP(SDMA_OP_NOP);
 }
 
 /**
@@ -1438,6 +1463,7 @@ static const struct amdgpu_ring_funcs sdma_v3_0_ring_funcs = {
 	.test_ring = sdma_v3_0_ring_test_ring,
 	.test_ib = sdma_v3_0_ring_test_ib,
 	.is_lockup = sdma_v3_0_ring_is_lockup,
+	.insert_nop = sdma_v3_0_ring_insert_nop,
 };
 
 static void sdma_v3_0_set_ring_funcs(struct amdgpu_device *adev)
@@ -1499,16 +1525,16 @@ static void sdma_v3_0_emit_copy_buffer(struct amdgpu_ib *ib,
  *
  * Fill GPU buffers using the DMA engine (VI).
  */
-static void sdma_v3_0_emit_fill_buffer(struct amdgpu_ring *ring,
+static void sdma_v3_0_emit_fill_buffer(struct amdgpu_ib *ib,
 				       uint32_t src_data,
 				       uint64_t dst_offset,
 				       uint32_t byte_count)
 {
-	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_CONST_FILL));
-	amdgpu_ring_write(ring, lower_32_bits(dst_offset));
-	amdgpu_ring_write(ring, upper_32_bits(dst_offset));
-	amdgpu_ring_write(ring, src_data);
-	amdgpu_ring_write(ring, byte_count);
+	ib->ptr[ib->length_dw++] = SDMA_PKT_HEADER_OP(SDMA_OP_CONST_FILL);
+	ib->ptr[ib->length_dw++] = lower_32_bits(dst_offset);
+	ib->ptr[ib->length_dw++] = upper_32_bits(dst_offset);
+	ib->ptr[ib->length_dw++] = src_data;
+	ib->ptr[ib->length_dw++] = byte_count;
 }
 
 static const struct amdgpu_buffer_funcs sdma_v3_0_buffer_funcs = {

drivers/gpu/drm/amd/amdgpu/tonga_sdma_pkt_open.h

@@ -2236,5 +2236,10 @@
 #define SDMA_PKT_NOP_HEADER_sub_op_shift  8
 #define SDMA_PKT_NOP_HEADER_SUB_OP(x) (((x) & SDMA_PKT_NOP_HEADER_sub_op_mask) << SDMA_PKT_NOP_HEADER_sub_op_shift)
 
+/*define for count field*/
+#define SDMA_PKT_NOP_HEADER_count_offset 0
+#define SDMA_PKT_NOP_HEADER_count_mask   0x00003FFF
+#define SDMA_PKT_NOP_HEADER_count_shift  16
+#define SDMA_PKT_NOP_HEADER_COUNT(x) (((x) & SDMA_PKT_NOP_HEADER_count_mask) << SDMA_PKT_NOP_HEADER_count_shift)
 
 #endif /* __TONGA_SDMA_PKT_OPEN_H_ */

drivers/gpu/drm/amd/amdgpu/uvd_v4_2.c

@@ -886,6 +886,7 @@ static const struct amdgpu_ring_funcs uvd_v4_2_ring_funcs = {
 	.test_ring = uvd_v4_2_ring_test_ring,
 	.test_ib = uvd_v4_2_ring_test_ib,
 	.is_lockup = amdgpu_ring_test_lockup,
+	.insert_nop = amdgpu_ring_insert_nop,
 };
 
 static void uvd_v4_2_set_ring_funcs(struct amdgpu_device *adev)

drivers/gpu/drm/amd/amdgpu/uvd_v5_0.c

@@ -825,6 +825,7 @@ static const struct amdgpu_ring_funcs uvd_v5_0_ring_funcs = {
 	.test_ring = uvd_v5_0_ring_test_ring,
 	.test_ib = uvd_v5_0_ring_test_ib,
 	.is_lockup = amdgpu_ring_test_lockup,
+	.insert_nop = amdgpu_ring_insert_nop,
 };
 
 static void uvd_v5_0_set_ring_funcs(struct amdgpu_device *adev)

drivers/gpu/drm/amd/amdgpu/uvd_v6_0.c

@@ -805,6 +805,7 @@ static const struct amdgpu_ring_funcs uvd_v6_0_ring_funcs = {
 	.test_ring = uvd_v6_0_ring_test_ring,
 	.test_ib = uvd_v6_0_ring_test_ib,
 	.is_lockup = amdgpu_ring_test_lockup,
+	.insert_nop = amdgpu_ring_insert_nop,
 };
 
 static void uvd_v6_0_set_ring_funcs(struct amdgpu_device *adev)

drivers/gpu/drm/amd/amdgpu/vce_v2_0.c

@@ -643,6 +643,7 @@ static const struct amdgpu_ring_funcs vce_v2_0_ring_funcs = {
 	.test_ring = amdgpu_vce_ring_test_ring,
 	.test_ib = amdgpu_vce_ring_test_ib,
 	.is_lockup = amdgpu_ring_test_lockup,
+	.insert_nop = amdgpu_ring_insert_nop,
 };
 
 static void vce_v2_0_set_ring_funcs(struct amdgpu_device *adev)

drivers/gpu/drm/amd/amdgpu/vce_v3_0.c

@@ -608,6 +608,7 @@ static const struct amdgpu_ring_funcs vce_v3_0_ring_funcs = {
 	.test_ring = amdgpu_vce_ring_test_ring,
 	.test_ib = amdgpu_vce_ring_test_ib,
 	.is_lockup = amdgpu_ring_test_lockup,
+	.insert_nop = amdgpu_ring_insert_nop,
 };
 
 static void vce_v3_0_set_ring_funcs(struct amdgpu_device *adev)

drivers/gpu/drm/amd/scheduler/gpu_scheduler.c

@@ -27,6 +27,8 @@
 #include <drm/drmP.h>
 #include "gpu_scheduler.h"
 
+static struct amd_sched_job *
+amd_sched_entity_pop_job(struct amd_sched_entity *entity);
 static void amd_sched_wakeup(struct amd_gpu_scheduler *sched);
 
 /* Initialize a given run queue struct */
@@ -56,34 +58,36 @@ static void amd_sched_rq_remove_entity(struct amd_sched_rq *rq,
 }
 
 /**
- * Select next entity from a specified run queue with round robin policy.
- * It could return the same entity as current one if current is the only
- * available one in the queue. Return NULL if nothing available.
+ * Select next job from a specified run queue with round robin policy.
+ * Return NULL if nothing available.
  */
-static struct amd_sched_entity *
-amd_sched_rq_select_entity(struct amd_sched_rq *rq)
+static struct amd_sched_job *
+amd_sched_rq_select_job(struct amd_sched_rq *rq)
 {
 	struct amd_sched_entity *entity;
+	struct amd_sched_job *job;
 
 	spin_lock(&rq->lock);
 
 	entity = rq->current_entity;
 	if (entity) {
 		list_for_each_entry_continue(entity, &rq->entities, list) {
-			if (!kfifo_is_empty(&entity->job_queue)) {
+			job = amd_sched_entity_pop_job(entity);
+			if (job) {
 				rq->current_entity = entity;
 				spin_unlock(&rq->lock);
-				return rq->current_entity;
+				return job;
 			}
 		}
 	}
 
 	list_for_each_entry(entity, &rq->entities, list) {
 
-		if (!kfifo_is_empty(&entity->job_queue)) {
+		job = amd_sched_entity_pop_job(entity);
+		if (job) {
 			rq->current_entity = entity;
 			spin_unlock(&rq->lock);
-			return rq->current_entity;
+			return job;
 		}
 
 		if (entity == rq->current_entity)
@@ -188,6 +192,39 @@ void amd_sched_entity_fini(struct amd_gpu_scheduler *sched,
 	kfifo_free(&entity->job_queue);
 }
 
+static void amd_sched_entity_wakeup(struct fence *f, struct fence_cb *cb)
+{
+	struct amd_sched_entity *entity =
+		container_of(cb, struct amd_sched_entity, cb);
+	entity->dependency = NULL;
+	fence_put(f);
+	amd_sched_wakeup(entity->scheduler);
+}
+
+static struct amd_sched_job *
+amd_sched_entity_pop_job(struct amd_sched_entity *entity)
+{
+	struct amd_gpu_scheduler *sched = entity->scheduler;
+	struct amd_sched_job *job;
+
+	if (ACCESS_ONCE(entity->dependency))
+		return NULL;
+
+	if (!kfifo_out_peek(&entity->job_queue, &job, sizeof(job)))
+		return NULL;
+
+	while ((entity->dependency = sched->ops->dependency(job))) {
+
+		if (fence_add_callback(entity->dependency, &entity->cb,
+				       amd_sched_entity_wakeup))
+			fence_put(entity->dependency);
+		else
+			return NULL;
+	}
+
+	return job;
+}
+
 /**
  * Helper to submit a job to the job queue
  *
@@ -227,7 +264,6 @@ int amd_sched_entity_push_job(struct amd_sched_job *sched_job)
 	struct amd_sched_entity *entity = sched_job->s_entity;
 	struct amd_sched_fence *fence = amd_sched_fence_create(
 		entity, sched_job->owner);
-	int r;
 
 	if (!fence)
 		return -ENOMEM;
@@ -235,10 +271,10 @@ int amd_sched_entity_push_job(struct amd_sched_job *sched_job)
 	fence_get(&fence->base);
 	sched_job->s_fence = fence;
 
-	r = wait_event_interruptible(entity->scheduler->job_scheduled,
+	wait_event(entity->scheduler->job_scheduled,
 		   amd_sched_entity_in(sched_job));
 
-	return r;
+	return 0;
 }
 
 /**
@@ -260,22 +296,22 @@ static void amd_sched_wakeup(struct amd_gpu_scheduler *sched)
 }
 
 /**
- * Select next entity containing real IB submissions
+ * Select next to run
 */
-static struct amd_sched_entity *
-amd_sched_select_context(struct amd_gpu_scheduler *sched)
+static struct amd_sched_job *
+amd_sched_select_job(struct amd_gpu_scheduler *sched)
 {
-	struct amd_sched_entity *tmp;
+	struct amd_sched_job *job;
 
 	if (!amd_sched_ready(sched))
 		return NULL;
 
 	/* Kernel run queue has higher priority than normal run queue*/
-	tmp = amd_sched_rq_select_entity(&sched->kernel_rq);
-	if (tmp == NULL)
-		tmp = amd_sched_rq_select_entity(&sched->sched_rq);
+	job = amd_sched_rq_select_job(&sched->kernel_rq);
+	if (job == NULL)
+		job = amd_sched_rq_select_job(&sched->sched_rq);
 
-	return tmp;
+	return job;
 }
 
 static void amd_sched_process_job(struct fence *f, struct fence_cb *cb)
@@ -301,22 +337,19 @@ static int amd_sched_main(void *param)
 	sched_setscheduler(current, SCHED_FIFO, &sparam);
 
 	while (!kthread_should_stop()) {
-		struct amd_sched_entity *c_entity = NULL;
+		struct amd_sched_entity *entity;
 		struct amd_sched_job *job;
 		struct fence *fence;
 
 		wait_event_interruptible(sched->wake_up_worker,
 			kthread_should_stop() ||
-			(c_entity = amd_sched_select_context(sched)));
+			(job = amd_sched_select_job(sched)));
 
-		if (!c_entity)
+		if (!job)
 			continue;
 
-		r = kfifo_out(&c_entity->job_queue, &job, sizeof(void *));
-		if (r != sizeof(void *))
-			continue;
+		entity = job->s_entity;
 		atomic_inc(&sched->hw_rq_count);
-
 		fence = sched->ops->run_job(job);
 		if (fence) {
 			r = fence_add_callback(fence, &job->cb,
@@ -328,6 +361,7 @@ static int amd_sched_main(void *param)
 			fence_put(fence);
 		}
 
+		kfifo_out(&entity->job_queue, &job, sizeof(job));
 		wake_up(&sched->job_scheduled);
 	}
 	return 0;

drivers/gpu/drm/amd/scheduler/gpu_scheduler.h

@@ -45,6 +45,8 @@ struct amd_sched_entity {
 	spinlock_t			queue_lock;
 	struct amd_gpu_scheduler	*scheduler;
 	uint64_t                        fence_context;
+	struct fence			*dependency;
+	struct fence_cb			cb;
 };
 
 /**
@@ -89,6 +91,7 @@ static inline struct amd_sched_fence *to_amd_sched_fence(struct fence *f)
  * these functions should be implemented in driver side
 */
 struct amd_sched_backend_ops {
+	struct fence *(*dependency)(struct amd_sched_job *job);
 	struct fence *(*run_job)(struct amd_sched_job *job);
 	void (*process_job)(struct amd_sched_job *job);
 };

drivers/gpu/drm/radeon/atombios_dp.c

@@ -171,8 +171,9 @@ radeon_dp_aux_transfer_atom(struct drm_dp_aux *aux, struct drm_dp_aux_msg *msg)
 		return -E2BIG;
 
 	tx_buf[0] = msg->address & 0xff;
-	tx_buf[1] = msg->address >> 8;
-	tx_buf[2] = msg->request << 4;
+	tx_buf[1] = (msg->address >> 8) & 0xff;
+	tx_buf[2] = (msg->request << 4) |
+		((msg->address >> 16) & 0xf);
 	tx_buf[3] = msg->size ? (msg->size - 1) : 0;
 
 	switch (msg->request & ~DP_AUX_I2C_MOT) {

drivers/gpu/drm/radeon/radeon_audio.c

@@ -522,6 +522,7 @@ static int radeon_audio_set_avi_packet(struct drm_encoder *encoder,
 		return err;
 	}
 
+	if (radeon_encoder->output_csc != RADEON_OUTPUT_CSC_BYPASS) {
 		if (drm_rgb_quant_range_selectable(radeon_connector_edid(connector))) {
 			if (radeon_encoder->output_csc == RADEON_OUTPUT_CSC_TVRGB)
 				frame.quantization_range = HDMI_QUANTIZATION_RANGE_LIMITED;
@@ -530,6 +531,7 @@ static int radeon_audio_set_avi_packet(struct drm_encoder *encoder,
 		} else {
 			frame.quantization_range = HDMI_QUANTIZATION_RANGE_DEFAULT;
 		}
+	}
 
 	err = hdmi_avi_infoframe_pack(&frame, buffer, sizeof(buffer));
 	if (err < 0) {

drivers/gpu/drm/radeon/radeon_combios.c

@@ -3387,6 +3387,14 @@ void radeon_combios_asic_init(struct drm_device *dev)
 	    rdev->pdev->subsystem_device == 0x30ae)
 		return;
 
+	/* quirk for rs4xx HP Compaq dc5750 Small Form Factor to make it resume
+	 * - it hangs on resume inside the dynclk 1 table.
+	 */
+	if (rdev->family == CHIP_RS480 &&
+	    rdev->pdev->subsystem_vendor == 0x103c &&
+	    rdev->pdev->subsystem_device == 0x280a)
+		return;
+
 	/* DYN CLK 1 */
 	table = combios_get_table_offset(dev, COMBIOS_DYN_CLK_1_TABLE);
 	if (table)

drivers/gpu/drm/radeon/radeon_dp_auxch.c

@@ -116,8 +116,8 @@ radeon_dp_aux_transfer_native(struct drm_dp_aux *aux, struct drm_dp_aux_msg *msg
 	       AUX_SW_WR_BYTES(bytes));
 
 	/* write the data header into the registers */
-	/* request, addres, msg size */
-	byte = (msg->request << 4);
+	/* request, address, msg size */
+	byte = (msg->request << 4) | ((msg->address >> 16) & 0xf);
 	WREG32(AUX_SW_DATA + aux_offset[instance],
 	       AUX_SW_DATA_MASK(byte) | AUX_SW_AUTOINCREMENT_DISABLE);