drm/radeon/dpm: use multiple UVD power states (v3)

Use the UVD handle information to determine which
which power states to select when using UVD.  For
example, decoding a single SD stream requires much
lower clocks than multiple HD streams.

v2: switch to a cleaner dpm/uvd interface
v3: change the uvd power state while streams
are active if need be

Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

4 files changed, 57 insertions, 12 deletions

diff --git a/drivers/gpu/drm/radeon/radeon.h b/drivers/gpu/drm/radeon/radeon.h
index fc8d03b..a276f02 100644
--- a/drivers/gpu/drm/radeon/radeon.h
+++ b/drivers/gpu/drm/radeon/radeon.h
@@ -1360,11 +1360,14 @@ struct radeon_dpm {
 	struct radeon_dpm_thermal thermal;
 	/* forced levels */
 	enum radeon_dpm_forced_level forced_level;
+	/* track UVD streams */
+	unsigned sd;
+	unsigned hd;
 };
 
 void radeon_dpm_enable_power_state(struct radeon_device *rdev,
 				    enum radeon_pm_state_type dpm_state);
-
+void radeon_dpm_enable_uvd(struct radeon_device *rdev, bool enable);
 
 struct radeon_pm {
 	struct mutex		mutex;
diff --git a/drivers/gpu/drm/radeon/radeon_cs.c b/drivers/gpu/drm/radeon/radeon_cs.c
index 13a130f..5384fa4 100644
--- a/drivers/gpu/drm/radeon/radeon_cs.c
+++ b/drivers/gpu/drm/radeon/radeon_cs.c
@@ -383,6 +383,10 @@ static int radeon_cs_ib_chunk(struct radeon_device *rdev,
 		DRM_ERROR("Invalid command stream !\n");
 		return r;
 	}
+
+	if (parser->ring == R600_RING_TYPE_UVD_INDEX)
+		radeon_uvd_note_usage(rdev);
+
 	radeon_cs_sync_rings(parser);
 	r = radeon_ib_schedule(rdev, &parser->ib, NULL);
 	if (r) {
@@ -474,6 +478,9 @@ static int radeon_cs_ib_vm_chunk(struct radeon_device *rdev,
 		return r;
 	}
 
+	if (parser->ring == R600_RING_TYPE_UVD_INDEX)
+		radeon_uvd_note_usage(rdev);
+
 	mutex_lock(&rdev->vm_manager.lock);
 	mutex_lock(&vm->mutex);
 	r = radeon_vm_alloc_pt(rdev, vm);
@@ -552,10 +559,6 @@ int radeon_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
 		return r;
 	}
 
-	/* XXX pick SD/HD/MVC */
-	if (parser.ring == R600_RING_TYPE_UVD_INDEX)
-		radeon_uvd_note_usage(rdev);
-
 	r = radeon_cs_ib_chunk(rdev, &parser);
 	if (r) {
 		goto out;
diff --git a/drivers/gpu/drm/radeon/radeon_pm.c b/drivers/gpu/drm/radeon/radeon_pm.c
index c557850..59d7a0c 100644
--- a/drivers/gpu/drm/radeon/radeon_pm.c
+++ b/drivers/gpu/drm/radeon/radeon_pm.c
@@ -729,6 +729,8 @@ restart_search:
 	/* use a fallback state if we didn't match */
 	switch (dpm_state) {
 	case POWER_STATE_TYPE_INTERNAL_UVD_SD:
+		dpm_state = POWER_STATE_TYPE_INTERNAL_UVD_HD;
+		goto restart_search;
 	case POWER_STATE_TYPE_INTERNAL_UVD_HD:
 	case POWER_STATE_TYPE_INTERNAL_UVD_HD2:
 	case POWER_STATE_TYPE_INTERNAL_UVD_MVC:
@@ -884,6 +886,34 @@ void radeon_dpm_enable_power_state(struct radeon_device *rdev,
 	radeon_pm_compute_clocks(rdev);
 }
 
+void radeon_dpm_enable_uvd(struct radeon_device *rdev, bool enable)
+{
+	enum radeon_pm_state_type dpm_state;
+
+	if (enable) {
+		mutex_lock(&rdev->pm.mutex);
+		rdev->pm.dpm.uvd_active = true;
+		if ((rdev->pm.dpm.sd == 1) && (rdev->pm.dpm.hd == 0))
+			dpm_state = POWER_STATE_TYPE_INTERNAL_UVD_SD;
+		else if ((rdev->pm.dpm.sd == 2) && (rdev->pm.dpm.hd == 0))
+			dpm_state = POWER_STATE_TYPE_INTERNAL_UVD_HD;
+		else if ((rdev->pm.dpm.sd == 0) && (rdev->pm.dpm.hd == 1))
+			dpm_state = POWER_STATE_TYPE_INTERNAL_UVD_HD;
+		else if ((rdev->pm.dpm.sd == 0) && (rdev->pm.dpm.hd == 2))
+			dpm_state = POWER_STATE_TYPE_INTERNAL_UVD_HD2;
+		else
+			dpm_state = POWER_STATE_TYPE_INTERNAL_UVD;
+		rdev->pm.dpm.state = dpm_state;
+		mutex_unlock(&rdev->pm.mutex);
+	} else {
+		mutex_lock(&rdev->pm.mutex);
+		rdev->pm.dpm.uvd_active = false;
+		mutex_unlock(&rdev->pm.mutex);
+	}
+
+	radeon_pm_compute_clocks(rdev);
+}
+
 static void radeon_pm_suspend_old(struct radeon_device *rdev)
 {
 	mutex_lock(&rdev->pm.mutex);
diff --git a/drivers/gpu/drm/radeon/radeon_uvd.c b/drivers/gpu/drm/radeon/radeon_uvd.c
index 2a4cff1..1a01bbf 100644
--- a/drivers/gpu/drm/radeon/radeon_uvd.c
+++ b/drivers/gpu/drm/radeon/radeon_uvd.c
@@ -775,10 +775,7 @@ static void radeon_uvd_idle_work_handler(struct work_struct *work)
 
 	if (radeon_fence_count_emitted(rdev, R600_RING_TYPE_UVD_INDEX) == 0) {
 		if ((rdev->pm.pm_method == PM_METHOD_DPM) && rdev->pm.dpm_enabled) {
-			mutex_lock(&rdev->pm.mutex);
-			rdev->pm.dpm.uvd_active = false;
-			mutex_unlock(&rdev->pm.mutex);
-			radeon_pm_compute_clocks(rdev);
+			radeon_dpm_enable_uvd(rdev, false);
 		} else {
 			radeon_set_uvd_clocks(rdev, 0, 0);
 		}
@@ -790,13 +787,25 @@ static void radeon_uvd_idle_work_handler(struct work_struct *work)
 
 void radeon_uvd_note_usage(struct radeon_device *rdev)
 {
+	bool streams_changed = false;
 	bool set_clocks = !cancel_delayed_work_sync(&rdev->uvd.idle_work);
 	set_clocks &= schedule_delayed_work(&rdev->uvd.idle_work,
 					    msecs_to_jiffies(UVD_IDLE_TIMEOUT_MS));
-	if (set_clocks) {
+
+	if ((rdev->pm.pm_method == PM_METHOD_DPM) && rdev->pm.dpm_enabled) {
+		unsigned hd = 0, sd = 0;
+		radeon_uvd_count_handles(rdev, &sd, &hd);
+		if ((rdev->pm.dpm.sd != sd) ||
+		    (rdev->pm.dpm.hd != hd)) {
+			rdev->pm.dpm.sd = sd;
+			rdev->pm.dpm.hd = hd;
+			streams_changed = true;
+		}
+	}
+
+	if (set_clocks || streams_changed) {
 		if ((rdev->pm.pm_method == PM_METHOD_DPM) && rdev->pm.dpm_enabled) {
-			/* XXX pick SD/HD/MVC */
-			radeon_dpm_enable_power_state(rdev, POWER_STATE_TYPE_INTERNAL_UVD);
+			radeon_dpm_enable_uvd(rdev, true);
 		} else {
 			radeon_set_uvd_clocks(rdev, 53300, 40000);
 		}