drm/amdgpu/gfx: fix ref counting for ring based profile handling

We need to make sure the workload profile ref counts are
balanced.  This isn't currently the case because we can
increment the count on submissions, but the decrement may
be delayed as work comes in.  Track when we enable the
workload profile so the references are balanced.

v2: switch to a mutex and active flag
v3: fix mutex init

Fixes: 8fdb3958e396 ("drm/amdgpu/gfx: add ring helpers for setting workload profile")
Cc: Yang Wang <kevinyang.wang@amd.com>
Cc: Kenneth Feng <kenneth.feng@amd.com>
Tested-by: Kenneth Feng <kenneth.feng@amd.com>
Reviewed-by: Kenneth Feng <kenneth.feng@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

3 files changed, 23 insertions, 10 deletions

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
index 35ace1368b9c..173048c46fc3 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
@@ -4282,6 +4282,7 @@ int amdgpu_device_init(struct amdgpu_device *adev,
 	/* Initialize the mutex for cleaner shader isolation between GFX and compute processes */
 	mutex_init(&adev->enforce_isolation_mutex);
 	mutex_init(&adev->gfx.kfd_sch_mutex);
+	mutex_init(&adev->gfx.workload_profile_mutex);
 
 	amdgpu_device_init_apu_flags(adev);
 
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c
index 984e6ff6e463..099329d15b9f 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c
@@ -2160,11 +2160,16 @@ void amdgpu_gfx_profile_idle_work_handler(struct work_struct *work)
 	for (i = 0; i < (AMDGPU_MAX_COMPUTE_RINGS * AMDGPU_MAX_GC_INSTANCES); ++i)
 		fences += amdgpu_fence_count_emitted(&adev->gfx.compute_ring[i]);
 	if (!fences && !atomic_read(&adev->gfx.total_submission_cnt)) {
-		r = amdgpu_dpm_switch_power_profile(adev, profile, false);
-		if (r)
-			dev_warn(adev->dev, "(%d) failed to disable %s power profile mode\n", r,
-				 profile == PP_SMC_POWER_PROFILE_FULLSCREEN3D ?
-				 "fullscreen 3D" : "compute");
+		mutex_lock(&adev->gfx.workload_profile_mutex);
+		if (adev->gfx.workload_profile_active) {
+			r = amdgpu_dpm_switch_power_profile(adev, profile, false);
+			if (r)
+				dev_warn(adev->dev, "(%d) failed to disable %s power profile mode\n", r,
+					 profile == PP_SMC_POWER_PROFILE_FULLSCREEN3D ?
+					 "fullscreen 3D" : "compute");
+			adev->gfx.workload_profile_active = false;
+		}
+		mutex_unlock(&adev->gfx.workload_profile_mutex);
 	} else {
 		schedule_delayed_work(&adev->gfx.idle_work, GFX_PROFILE_IDLE_TIMEOUT);
 	}
@@ -2184,11 +2189,16 @@ void amdgpu_gfx_profile_ring_begin_use(struct amdgpu_ring *ring)
 	atomic_inc(&adev->gfx.total_submission_cnt);
 
 	if (!cancel_delayed_work_sync(&adev->gfx.idle_work)) {
-		r = amdgpu_dpm_switch_power_profile(adev, profile, true);
-		if (r)
-			dev_warn(adev->dev, "(%d) failed to disable %s power profile mode\n", r,
-				 profile == PP_SMC_POWER_PROFILE_FULLSCREEN3D ?
-				 "fullscreen 3D" : "compute");
+		mutex_lock(&adev->gfx.workload_profile_mutex);
+		if (!adev->gfx.workload_profile_active) {
+			r = amdgpu_dpm_switch_power_profile(adev, profile, true);
+			if (r)
+				dev_warn(adev->dev, "(%d) failed to disable %s power profile mode\n", r,
+					 profile == PP_SMC_POWER_PROFILE_FULLSCREEN3D ?
+					 "fullscreen 3D" : "compute");
+			adev->gfx.workload_profile_active = true;
+		}
+		mutex_unlock(&adev->gfx.workload_profile_mutex);
 	}
 }
 
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h
index 3c030d3f39d4..9e2355a3b120 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h
@@ -482,6 +482,8 @@ struct amdgpu_gfx {
 
 	atomic_t			total_submission_cnt;
 	struct delayed_work		idle_work;
+	bool				workload_profile_active;
+	struct mutex                    workload_profile_mutex;
 };
 
 struct amdgpu_gfx_ras_reg_entry {