Merge branch 'drm-fixes-4.4' of git://people.freedesktop.org/~agd5f/linux into drm-fixes

Radeon and amdgpu fixes for 4.4:
- DPM fixes for r7xx devices
- VCE fixes for Stoney
- GPUVM fixes
- Scheduler fixes

* 'drm-fixes-4.4' of git://people.freedesktop.org/~agd5f/linux:
  drm/radeon: make some dpm errors debug only
  drm/radeon: make rv770_set_sw_state failures non-fatal
  drm/amdgpu: move dependency handling out of atomic section v2
  drm/amdgpu: optimize scheduler fence handling
  drm/amdgpu: remove vm->mutex
  drm/amdgpu: add mutex for ba_va->valids/invalids
  drm/amdgpu: adapt vce session create interface changes
  drm/amdgpu: vce use multiple cache surface starting from stoney
  drm/amdgpu: reset vce trap interrupt flag

11 files changed, 145 insertions, 76 deletions

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
index 306f75700bf8..251b14736de9 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
@@ -496,6 +496,7 @@ struct amdgpu_bo_va_mapping {
 
 /* bo virtual addresses in a specific vm */
 struct amdgpu_bo_va {
+        struct mutex                    mutex;
         /* protected by bo being reserved */
         struct list_head                bo_list;
         struct fence                    *last_pt_update;
@@ -928,8 +929,6 @@ struct amdgpu_vm_id {
 };
 
 struct amdgpu_vm {
-        struct mutex            mutex;
-
         struct rb_root          va;
 
         /* protecting invalidated */
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index 3afcf0237c25..1d44d508d4d4 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -784,8 +784,6 @@ int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
 {
         struct amdgpu_device *adev = dev->dev_private;
         union drm_amdgpu_cs *cs = data;
-        struct amdgpu_fpriv *fpriv = filp->driver_priv;
-        struct amdgpu_vm *vm = &fpriv->vm;
         struct amdgpu_cs_parser parser = {};
         bool reserved_buffers = false;
         int i, r;
@@ -803,7 +801,6 @@ int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
                 r = amdgpu_cs_handle_lockup(adev, r);
                 return r;
         }
-        mutex_lock(&vm->mutex);
         r = amdgpu_cs_parser_relocs(&parser);
         if (r == -ENOMEM)
                 DRM_ERROR("Not enough memory for command submission!\n");
@@ -888,7 +885,6 @@ int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
 
 out:
         amdgpu_cs_parser_fini(&parser, r, reserved_buffers);
-        mutex_unlock(&vm->mutex);
         r = amdgpu_cs_handle_lockup(adev, r);
         return r;
 }
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
index 00c5b580f56c..fc32fc01a64b 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
@@ -115,12 +115,9 @@ int amdgpu_gem_object_open(struct drm_gem_object *obj, struct drm_file *file_pri
         struct amdgpu_vm *vm = &fpriv->vm;
         struct amdgpu_bo_va *bo_va;
         int r;
-        mutex_lock(&vm->mutex);
         r = amdgpu_bo_reserve(rbo, false);
-        if (r) {
-                mutex_unlock(&vm->mutex);
+        if (r)
                 return r;
-        }
 
         bo_va = amdgpu_vm_bo_find(vm, rbo);
         if (!bo_va) {
@@ -129,7 +126,6 @@ int amdgpu_gem_object_open(struct drm_gem_object *obj, struct drm_file *file_pri
                 ++bo_va->ref_count;
         }
         amdgpu_bo_unreserve(rbo);
-        mutex_unlock(&vm->mutex);
         return 0;
 }
 
@@ -142,10 +138,8 @@ void amdgpu_gem_object_close(struct drm_gem_object *obj,
         struct amdgpu_vm *vm = &fpriv->vm;
         struct amdgpu_bo_va *bo_va;
         int r;
-        mutex_lock(&vm->mutex);
         r = amdgpu_bo_reserve(rbo, true);
         if (r) {
-                mutex_unlock(&vm->mutex);
                 dev_err(adev->dev, "leaking bo va because "
                         "we fail to reserve bo (%d)\n", r);
                 return;
@@ -157,7 +151,6 @@ void amdgpu_gem_object_close(struct drm_gem_object *obj,
                 }
         }
         amdgpu_bo_unreserve(rbo);
-        mutex_unlock(&vm->mutex);
 }
 
 static int amdgpu_gem_handle_lockup(struct amdgpu_device *adev, int r)
@@ -553,7 +546,6 @@ int amdgpu_gem_va_ioctl(struct drm_device *dev, void *data,
         gobj = drm_gem_object_lookup(dev, filp, args->handle);
         if (gobj == NULL)
                 return -ENOENT;
-        mutex_lock(&fpriv->vm.mutex);
         rbo = gem_to_amdgpu_bo(gobj);
         INIT_LIST_HEAD(&list);
         INIT_LIST_HEAD(&duplicates);
@@ -568,7 +560,6 @@ int amdgpu_gem_va_ioctl(struct drm_device *dev, void *data,
         }
         r = ttm_eu_reserve_buffers(&ticket, &list, true, &duplicates);
         if (r) {
-                mutex_unlock(&fpriv->vm.mutex);
                 drm_gem_object_unreference_unlocked(gobj);
                 return r;
         }
@@ -577,7 +568,6 @@ int amdgpu_gem_va_ioctl(struct drm_device *dev, void *data,
         if (!bo_va) {
                 ttm_eu_backoff_reservation(&ticket, &list);
                 drm_gem_object_unreference_unlocked(gobj);
-                mutex_unlock(&fpriv->vm.mutex);
                 return -ENOENT;
         }
 
@@ -602,7 +592,7 @@ int amdgpu_gem_va_ioctl(struct drm_device *dev, void *data,
         ttm_eu_backoff_reservation(&ticket, &list);
         if (!r && !(args->flags & AMDGPU_VM_DELAY_UPDATE))
                 amdgpu_gem_va_update_vm(adev, bo_va, args->operation);
-        mutex_unlock(&fpriv->vm.mutex);
+
         drm_gem_object_unreference_unlocked(gobj);
         return r;
 }
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c
index 03f0c3bae516..a745eeeb5d82 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c
@@ -392,7 +392,10 @@ int amdgpu_vce_get_create_msg(struct amdgpu_ring *ring, uint32_t handle,
         ib->ptr[ib->length_dw++] = 0x00000001; /* session cmd */
         ib->ptr[ib->length_dw++] = handle;
 
-        ib->ptr[ib->length_dw++] = 0x00000030; /* len */
+        if ((ring->adev->vce.fw_version >> 24) >= 52)
+                ib->ptr[ib->length_dw++] = 0x00000040; /* len */
+        else
+                ib->ptr[ib->length_dw++] = 0x00000030; /* len */
         ib->ptr[ib->length_dw++] = 0x01000001; /* create cmd */
         ib->ptr[ib->length_dw++] = 0x00000000;
         ib->ptr[ib->length_dw++] = 0x00000042;
@@ -404,6 +407,12 @@ int amdgpu_vce_get_create_msg(struct amdgpu_ring *ring, uint32_t handle,
         ib->ptr[ib->length_dw++] = 0x00000100;
         ib->ptr[ib->length_dw++] = 0x0000000c;
         ib->ptr[ib->length_dw++] = 0x00000000;
+        if ((ring->adev->vce.fw_version >> 24) >= 52) {
+                ib->ptr[ib->length_dw++] = 0x00000000;
+                ib->ptr[ib->length_dw++] = 0x00000000;
+                ib->ptr[ib->length_dw++] = 0x00000000;
+                ib->ptr[ib->length_dw++] = 0x00000000;
+        }
 
         ib->ptr[ib->length_dw++] = 0x00000014; /* len */
         ib->ptr[ib->length_dw++] = 0x05000005; /* feedback buffer */
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index 159ce54bbd8d..ae037e5b6ad0 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -922,8 +922,9 @@ int amdgpu_vm_clear_invalids(struct amdgpu_device *adev,
                 bo_va = list_first_entry(&vm->invalidated,
                         struct amdgpu_bo_va, vm_status);
                 spin_unlock(&vm->status_lock);
-
+                mutex_lock(&bo_va->mutex);
                 r = amdgpu_vm_bo_update(adev, bo_va, NULL);
+                mutex_unlock(&bo_va->mutex);
                 if (r)
                         return r;
 
@@ -967,7 +968,7 @@ struct amdgpu_bo_va *amdgpu_vm_bo_add(struct amdgpu_device *adev,
         INIT_LIST_HEAD(&bo_va->valids);
         INIT_LIST_HEAD(&bo_va->invalids);
         INIT_LIST_HEAD(&bo_va->vm_status);
-
+        mutex_init(&bo_va->mutex);
         list_add_tail(&bo_va->bo_list, &bo->va);
 
         return bo_va;
@@ -1045,7 +1046,9 @@ int amdgpu_vm_bo_map(struct amdgpu_device *adev,
         mapping->offset = offset;
         mapping->flags = flags;
 
+        mutex_lock(&bo_va->mutex);
         list_add(&mapping->list, &bo_va->invalids);
+        mutex_unlock(&bo_va->mutex);
         spin_lock(&vm->it_lock);
         interval_tree_insert(&mapping->it, &vm->va);
         spin_unlock(&vm->it_lock);
@@ -1121,7 +1124,7 @@ int amdgpu_vm_bo_unmap(struct amdgpu_device *adev,
         bool valid = true;
 
         saddr /= AMDGPU_GPU_PAGE_SIZE;
-
+        mutex_lock(&bo_va->mutex);
         list_for_each_entry(mapping, &bo_va->valids, list) {
                 if (mapping->it.start == saddr)
                         break;
@@ -1135,10 +1138,12 @@ int amdgpu_vm_bo_unmap(struct amdgpu_device *adev,
                                 break;
                 }
 
-                if (&mapping->list == &bo_va->invalids)
+                if (&mapping->list == &bo_va->invalids) {
+                        mutex_unlock(&bo_va->mutex);
                         return -ENOENT;
+                }
         }
-
+        mutex_unlock(&bo_va->mutex);
         list_del(&mapping->list);
         spin_lock(&vm->it_lock);
         interval_tree_remove(&mapping->it, &vm->va);
@@ -1190,8 +1195,8 @@ void amdgpu_vm_bo_rmv(struct amdgpu_device *adev,
                 spin_unlock(&vm->it_lock);
                 kfree(mapping);
         }
-
         fence_put(bo_va->last_pt_update);
+        mutex_destroy(&bo_va->mutex);
         kfree(bo_va);
 }
 
@@ -1236,7 +1241,6 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm)
                 vm->ids[i].id = 0;
                 vm->ids[i].flushed_updates = NULL;
         }
-        mutex_init(&vm->mutex);
         vm->va = RB_ROOT;
         spin_lock_init(&vm->status_lock);
         INIT_LIST_HEAD(&vm->invalidated);
@@ -1320,7 +1324,6 @@ void amdgpu_vm_fini(struct amdgpu_device *adev, struct amdgpu_vm *vm)
                 fence_put(vm->ids[i].flushed_updates);
         }
 
-        mutex_destroy(&vm->mutex);
 }
 
 /**
diff --git a/drivers/gpu/drm/amd/amdgpu/vce_v3_0.c b/drivers/gpu/drm/amd/amdgpu/vce_v3_0.c
index 6a52db6ad8d7..370c6c9d81c2 100644
--- a/drivers/gpu/drm/amd/amdgpu/vce_v3_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/vce_v3_0.c
@@ -40,6 +40,9 @@
 
 #define GRBM_GFX_INDEX__VCE_INSTANCE__SHIFT     0x04
 #define GRBM_GFX_INDEX__VCE_INSTANCE_MASK       0x10
+#define mmVCE_LMI_VCPU_CACHE_40BIT_BAR0         0x8616
+#define mmVCE_LMI_VCPU_CACHE_40BIT_BAR1         0x8617
+#define mmVCE_LMI_VCPU_CACHE_40BIT_BAR2         0x8618
 
 #define VCE_V3_0_FW_SIZE        (384 * 1024)
 #define VCE_V3_0_STACK_SIZE     (64 * 1024)
@@ -130,9 +133,11 @@ static int vce_v3_0_start(struct amdgpu_device *adev)
 
                 /* set BUSY flag */
                 WREG32_P(mmVCE_STATUS, 1, ~1);
-
-                WREG32_P(mmVCE_VCPU_CNTL, VCE_VCPU_CNTL__CLK_EN_MASK,
-                        ~VCE_VCPU_CNTL__CLK_EN_MASK);
+                if (adev->asic_type >= CHIP_STONEY)
+                        WREG32_P(mmVCE_VCPU_CNTL, 1, ~0x200001);
+                else
+                        WREG32_P(mmVCE_VCPU_CNTL, VCE_VCPU_CNTL__CLK_EN_MASK,
+                                ~VCE_VCPU_CNTL__CLK_EN_MASK);
 
                 WREG32_P(mmVCE_SOFT_RESET,
                          VCE_SOFT_RESET__ECPU_SOFT_RESET_MASK,
@@ -391,8 +396,12 @@ static void vce_v3_0_mc_resume(struct amdgpu_device *adev, int idx)
         WREG32(mmVCE_LMI_SWAP_CNTL, 0);
         WREG32(mmVCE_LMI_SWAP_CNTL1, 0);
         WREG32(mmVCE_LMI_VM_CTRL, 0);
-
-        WREG32(mmVCE_LMI_VCPU_CACHE_40BIT_BAR, (adev->vce.gpu_addr >> 8));
+        if (adev->asic_type >= CHIP_STONEY) {
+                WREG32(mmVCE_LMI_VCPU_CACHE_40BIT_BAR0, (adev->vce.gpu_addr >> 8));
+                WREG32(mmVCE_LMI_VCPU_CACHE_40BIT_BAR1, (adev->vce.gpu_addr >> 8));
+                WREG32(mmVCE_LMI_VCPU_CACHE_40BIT_BAR2, (adev->vce.gpu_addr >> 8));
+        } else
+                WREG32(mmVCE_LMI_VCPU_CACHE_40BIT_BAR, (adev->vce.gpu_addr >> 8));
         offset = AMDGPU_VCE_FIRMWARE_OFFSET;
         size = VCE_V3_0_FW_SIZE;
         WREG32(mmVCE_VCPU_CACHE_OFFSET0, offset & 0x7fffffff);
@@ -576,6 +585,11 @@ static int vce_v3_0_process_interrupt(struct amdgpu_device *adev,
                                       struct amdgpu_iv_entry *entry)
 {
         DRM_DEBUG("IH: VCE\n");
+
+        WREG32_P(mmVCE_SYS_INT_STATUS,
+                VCE_SYS_INT_STATUS__VCE_SYS_INT_TRAP_INTERRUPT_INT_MASK,
+                ~VCE_SYS_INT_STATUS__VCE_SYS_INT_TRAP_INTERRUPT_INT_MASK);
+
         switch (entry->src_data) {
         case 0:
                 amdgpu_fence_process(&adev->vce.ring[0]);
diff --git a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c
index ea30d6ad4c13..651129f2ec1d 100644
--- a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c
+++ b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c
@@ -30,8 +30,7 @@
 #define CREATE_TRACE_POINTS
 #include "gpu_sched_trace.h"
 
-static struct amd_sched_job *
-amd_sched_entity_pop_job(struct amd_sched_entity *entity);
+static bool amd_sched_entity_is_ready(struct amd_sched_entity *entity);
 static void amd_sched_wakeup(struct amd_gpu_scheduler *sched);
 
 struct kmem_cache *sched_fence_slab;
@@ -64,36 +63,36 @@ static void amd_sched_rq_remove_entity(struct amd_sched_rq *rq,
 }
 
 /**
- * Select next job from a specified run queue with round robin policy.
- * Return NULL if nothing available.
+ * Select an entity which could provide a job to run
+ *
+ * @rq          The run queue to check.
+ *
+ * Try to find a ready entity, returns NULL if none found.
  */
-static struct amd_sched_job *
-amd_sched_rq_select_job(struct amd_sched_rq *rq)
+static struct amd_sched_entity *
+amd_sched_rq_select_entity(struct amd_sched_rq *rq)
 {
         struct amd_sched_entity *entity;
-        struct amd_sched_job *sched_job;
 
         spin_lock(&rq->lock);
 
         entity = rq->current_entity;
         if (entity) {
                 list_for_each_entry_continue(entity, &rq->entities, list) {
-                        sched_job = amd_sched_entity_pop_job(entity);
-                        if (sched_job) {
+                        if (amd_sched_entity_is_ready(entity)) {
                                 rq->current_entity = entity;
                                 spin_unlock(&rq->lock);
-                                return sched_job;
+                                return entity;
                         }
                 }
         }
 
         list_for_each_entry(entity, &rq->entities, list) {
 
-                sched_job = amd_sched_entity_pop_job(entity);
-                if (sched_job) {
+                if (amd_sched_entity_is_ready(entity)) {
                         rq->current_entity = entity;
                         spin_unlock(&rq->lock);
-                        return sched_job;
+                        return entity;
                 }
 
                 if (entity == rq->current_entity)
@@ -177,6 +176,24 @@ static bool amd_sched_entity_is_idle(struct amd_sched_entity *entity)
 }
 
 /**
+ * Check if entity is ready
+ *
+ * @entity      The pointer to a valid scheduler entity
+ *
+ * Return true if entity could provide a job.
+ */
+static bool amd_sched_entity_is_ready(struct amd_sched_entity *entity)
+{
+        if (kfifo_is_empty(&entity->job_queue))
+                return false;
+
+        if (ACCESS_ONCE(entity->dependency))
+                return false;
+
+        return true;
+}
+
+/**
  * Destroy a context entity
  *
  * @sched       Pointer to scheduler instance
@@ -211,32 +228,53 @@ static void amd_sched_entity_wakeup(struct fence *f, struct fence_cb *cb)
         amd_sched_wakeup(entity->sched);
 }
 
+static bool amd_sched_entity_add_dependency_cb(struct amd_sched_entity *entity)
+{
+        struct amd_gpu_scheduler *sched = entity->sched;
+        struct fence * fence = entity->dependency;
+        struct amd_sched_fence *s_fence;
+
+        if (fence->context == entity->fence_context) {
+                /* We can ignore fences from ourself */
+                fence_put(entity->dependency);
+                return false;
+        }
+
+        s_fence = to_amd_sched_fence(fence);
+        if (s_fence && s_fence->sched == sched) {
+                /* Fence is from the same scheduler */
+                if (test_bit(AMD_SCHED_FENCE_SCHEDULED_BIT, &fence->flags)) {
+                        /* Ignore it when it is already scheduled */
+                        fence_put(entity->dependency);
+                        return false;
+                }
+
+                /* Wait for fence to be scheduled */
+                entity->cb.func = amd_sched_entity_wakeup;
+                list_add_tail(&entity->cb.node, &s_fence->scheduled_cb);
+                return true;
+        }
+
+        if (!fence_add_callback(entity->dependency, &entity->cb,
+                                amd_sched_entity_wakeup))
+                return true;
+
+        fence_put(entity->dependency);
+        return false;
+}
+
 static struct amd_sched_job *
 amd_sched_entity_pop_job(struct amd_sched_entity *entity)
 {
         struct amd_gpu_scheduler *sched = entity->sched;
         struct amd_sched_job *sched_job;
 
-        if (ACCESS_ONCE(entity->dependency))
-                return NULL;
-
         if (!kfifo_out_peek(&entity->job_queue, &sched_job, sizeof(sched_job)))
                 return NULL;
 
-        while ((entity->dependency = sched->ops->dependency(sched_job))) {
-
-                if (entity->dependency->context == entity->fence_context) {
-                        /* We can ignore fences from ourself */
-                        fence_put(entity->dependency);
-                        continue;
-                }
-
-                if (fence_add_callback(entity->dependency, &entity->cb,
-                                       amd_sched_entity_wakeup))
-                        fence_put(entity->dependency);
-                else
+        while ((entity->dependency = sched->ops->dependency(sched_job)))
+                if (amd_sched_entity_add_dependency_cb(entity))
                         return NULL;
-        }
 
         return sched_job;
 }
@@ -304,22 +342,22 @@ static void amd_sched_wakeup(struct amd_gpu_scheduler *sched)
 }
 
 /**
- * Select next to run
+ * Select next entity to process
 */
-static struct amd_sched_job *
-amd_sched_select_job(struct amd_gpu_scheduler *sched)
+static struct amd_sched_entity *
+amd_sched_select_entity(struct amd_gpu_scheduler *sched)
 {
-        struct amd_sched_job *sched_job;
+        struct amd_sched_entity *entity;
 
         if (!amd_sched_ready(sched))
                 return NULL;
 
         /* Kernel run queue has higher priority than normal run queue*/
-        sched_job = amd_sched_rq_select_job(&sched->kernel_rq);
-        if (sched_job == NULL)
-                sched_job = amd_sched_rq_select_job(&sched->sched_rq);
+        entity = amd_sched_rq_select_entity(&sched->kernel_rq);
+        if (entity == NULL)
+                entity = amd_sched_rq_select_entity(&sched->sched_rq);
 
-        return sched_job;
+        return entity;
 }
 
 static void amd_sched_process_job(struct fence *f, struct fence_cb *cb)
@@ -381,13 +419,16 @@ static int amd_sched_main(void *param)
                 unsigned long flags;
 
                 wait_event_interruptible(sched->wake_up_worker,
-                        kthread_should_stop() ||
-                        (sched_job = amd_sched_select_job(sched)));
+                        (entity = amd_sched_select_entity(sched)) ||
+                        kthread_should_stop());
 
+                if (!entity)
+                        continue;
+
+                sched_job = amd_sched_entity_pop_job(entity);
                 if (!sched_job)
                         continue;
 
-                entity = sched_job->s_entity;
                 s_fence = sched_job->s_fence;
 
                 if (sched->timeout != MAX_SCHEDULE_TIMEOUT) {
@@ -400,6 +441,7 @@ static int amd_sched_main(void *param)
 
                 atomic_inc(&sched->hw_rq_count);
                 fence = sched->ops->run_job(sched_job);
+                amd_sched_fence_scheduled(s_fence);
                 if (fence) {
                         r = fence_add_callback(fence, &s_fence->cb,
                                                amd_sched_process_job);
diff --git a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h
index 939692b14f4b..a0f0ae53aacd 100644
--- a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h
+++ b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h
@@ -27,6 +27,8 @@
 #include <linux/kfifo.h>
 #include <linux/fence.h>
 
+#define AMD_SCHED_FENCE_SCHEDULED_BIT   FENCE_FLAG_USER_BITS
+
 struct amd_gpu_scheduler;
 struct amd_sched_rq;
 
@@ -68,6 +70,7 @@ struct amd_sched_rq {
 struct amd_sched_fence {
         struct fence                    base;
         struct fence_cb                 cb;
+        struct list_head                scheduled_cb;
         struct amd_gpu_scheduler        *sched;
         spinlock_t                      lock;
         void                            *owner;
@@ -134,7 +137,7 @@ void amd_sched_entity_push_job(struct amd_sched_job *sched_job);
 
 struct amd_sched_fence *amd_sched_fence_create(
         struct amd_sched_entity *s_entity, void *owner);
+void amd_sched_fence_scheduled(struct amd_sched_fence *fence);
 void amd_sched_fence_signal(struct amd_sched_fence *fence);
 
-
 #endif
diff --git a/drivers/gpu/drm/amd/scheduler/sched_fence.c b/drivers/gpu/drm/amd/scheduler/sched_fence.c
index 8d2130b9ff05..87c78eecea64 100644
--- a/drivers/gpu/drm/amd/scheduler/sched_fence.c
+++ b/drivers/gpu/drm/amd/scheduler/sched_fence.c
@@ -35,6 +35,8 @@ struct amd_sched_fence *amd_sched_fence_create(struct amd_sched_entity *s_entity
         fence = kmem_cache_zalloc(sched_fence_slab, GFP_KERNEL);
         if (fence == NULL)
                 return NULL;
+
+        INIT_LIST_HEAD(&fence->scheduled_cb);
         fence->owner = owner;
         fence->sched = s_entity->sched;
         spin_lock_init(&fence->lock);
@@ -55,6 +57,17 @@ void amd_sched_fence_signal(struct amd_sched_fence *fence)
                 FENCE_TRACE(&fence->base, "was already signaled\n");
 }
 
+void amd_sched_fence_scheduled(struct amd_sched_fence *s_fence)
+{
+        struct fence_cb *cur, *tmp;
+
+        set_bit(AMD_SCHED_FENCE_SCHEDULED_BIT, &s_fence->base.flags);
+        list_for_each_entry_safe(cur, tmp, &s_fence->scheduled_cb, node) {
+                list_del_init(&cur->node);
+                cur->func(&s_fence->base, cur);
+        }
+}
+
 static const char *amd_sched_fence_get_driver_name(struct fence *fence)
 {
         return "amd_sched";
diff --git a/drivers/gpu/drm/radeon/rv730_dpm.c b/drivers/gpu/drm/radeon/rv730_dpm.c
index 3f5e1cf138ba..d37ba2cb886e 100644
--- a/drivers/gpu/drm/radeon/rv730_dpm.c
+++ b/drivers/gpu/drm/radeon/rv730_dpm.c
@@ -464,7 +464,7 @@ void rv730_stop_dpm(struct radeon_device *rdev)
         result = rv770_send_msg_to_smc(rdev, PPSMC_MSG_TwoLevelsDisabled);
 
         if (result != PPSMC_Result_OK)
-                DRM_ERROR("Could not force DPM to low\n");
+                DRM_DEBUG("Could not force DPM to low\n");
 
         WREG32_P(GENERAL_PWRMGT, 0, ~GLOBAL_PWRMGT_EN);
 
diff --git a/drivers/gpu/drm/radeon/rv770_dpm.c b/drivers/gpu/drm/radeon/rv770_dpm.c
index b9c770745a7a..e830c8935db0 100644
--- a/drivers/gpu/drm/radeon/rv770_dpm.c
+++ b/drivers/gpu/drm/radeon/rv770_dpm.c
@@ -193,7 +193,7 @@ void rv770_stop_dpm(struct radeon_device *rdev)
         result = rv770_send_msg_to_smc(rdev, PPSMC_MSG_TwoLevelsDisabled);
 
         if (result != PPSMC_Result_OK)
-                DRM_ERROR("Could not force DPM to low.\n");
+                DRM_DEBUG("Could not force DPM to low.\n");
 
         WREG32_P(GENERAL_PWRMGT, 0, ~GLOBAL_PWRMGT_EN);
 
@@ -1418,7 +1418,7 @@ int rv770_resume_smc(struct radeon_device *rdev)
 int rv770_set_sw_state(struct radeon_device *rdev)
 {
         if (rv770_send_msg_to_smc(rdev, PPSMC_MSG_SwitchToSwState) != PPSMC_Result_OK)
-                return -EINVAL;
+                DRM_DEBUG("rv770_set_sw_state failed\n");
         return 0;
 }