[PATCH 1/2] drm/amdgpu: fix possible pstate switch race condition

Quan, Evan Evan.Quan at amd.com
Tue Nov 5 10:23:33 UTC 2019


Added lock protection so that the p-state switch will
be guarded to be sequential. Also update the hive
pstate only all device from the hive are in the same
state.

Change-Id: I165a6f44e8aec1e6da56eefa0fc49d36670e56fe
Signed-off-by: Evan Quan <evan.quan at amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu.h      |  3 +++
 drivers/gpu/drm/amd/amdgpu/amdgpu_xgmi.c | 34 ++++++++++++++++++++++--
 2 files changed, 35 insertions(+), 2 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
index 0469cc51a6fb..41cf2abd6209 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
@@ -1041,6 +1041,9 @@ struct amdgpu_device {
 
 	uint64_t			unique_id;
 	uint64_t	df_perfmon_config_assign_mask[AMDGPU_MAX_DF_PERFMONS];
+
+	/* device pstate */
+	int				pstate;
 };
 
 static inline struct amdgpu_device *amdgpu_ttm_adev(struct ttm_bo_device *bdev)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_xgmi.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_xgmi.c
index 167d9fbd2c4f..de20a9a1c444 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_xgmi.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_xgmi.c
@@ -274,12 +274,18 @@ int amdgpu_xgmi_set_pstate(struct amdgpu_device *adev, int pstate)
 {
 	int ret = 0;
 	struct amdgpu_hive_info *hive = amdgpu_get_xgmi_hive(adev, 0);
+	struct amdgpu_device *tmp_adev;
+	bool update_hive_pstate = true;
 
 	if (!hive)
 		return 0;
 
-	if (hive->pstate == pstate)
+	mutex_lock(&hive->hive_lock);
+
+	if (hive->pstate == pstate) {
+		mutex_unlock(&hive->hive_lock);
 		return 0;
+	}
 
 	dev_dbg(adev->dev, "Set xgmi pstate %d.\n", pstate);
 
@@ -290,11 +296,32 @@ int amdgpu_xgmi_set_pstate(struct amdgpu_device *adev, int pstate)
 		ret = adev->powerplay.pp_funcs->set_xgmi_pstate(adev->powerplay.pp_handle,
 								pstate);
 
-	if (ret)
+	if (ret) {
 		dev_err(adev->dev,
 			"XGMI: Set pstate failure on device %llx, hive %llx, ret %d",
 			adev->gmc.xgmi.node_id,
 			adev->gmc.xgmi.hive_id, ret);
+		goto out;
+	}
+
+	/* Update device pstate */
+	adev->pstate = pstate;
+
+	/*
+	 * Update the hive pstate only all devices of the hive
+	 * are in the same pstate
+	 */
+	list_for_each_entry(tmp_adev, &hive->device_list, gmc.xgmi.head) {
+		if (tmp_adev->pstate != adev->pstate) {
+			update_hive_pstate = false;
+			break;
+		}
+	}
+	if (update_hive_pstate)
+		hive->pstate = pstate;
+
+out:
+	mutex_unlock(&hive->hive_lock);
 
 	return ret;
 }
@@ -369,6 +396,9 @@ int amdgpu_xgmi_add_device(struct amdgpu_device *adev)
 		goto exit;
 	}
 
+	/* Set default device pstate */
+	adev->pstate = -1;
+
 	top_info = &adev->psp.xgmi_context.top_info;
 
 	list_add_tail(&adev->gmc.xgmi.head, &hive->device_list);
-- 
2.23.0



More information about the amd-gfx mailing list