[PATCH] drm/amd/powerplay: fix deadlock on setting power_dpm_force_performance_level

Quan, Evan Evan.Quan at amd.com
Wed Nov 6 06:57:21 UTC 2019


smu_enable_umd_pstate() will try to get the smu->mutex which was already
hold by its parent API smu_force_performance_level() on the call path.
Thus deadlock happens.

Change-Id: Ic4d3c7d06eb47eab2ea42b98f399cd95ab320f0c
Signed-off-by: Evan Quan <evan.quan at amd.com>
---
 drivers/gpu/drm/amd/powerplay/amdgpu_smu.c | 19 ++++++++++++++-----
 1 file changed, 14 insertions(+), 5 deletions(-)

diff --git a/drivers/gpu/drm/amd/powerplay/amdgpu_smu.c b/drivers/gpu/drm/amd/powerplay/amdgpu_smu.c
index facc19cae7e5..c21fe7ac5df8 100644
--- a/drivers/gpu/drm/amd/powerplay/amdgpu_smu.c
+++ b/drivers/gpu/drm/amd/powerplay/amdgpu_smu.c
@@ -383,14 +383,25 @@ bool smu_clk_dpm_is_enabled(struct smu_context *smu, enum smu_clk_type clk_type)
 	return true;
 }
 
-
+/**
+ * smu_dpm_set_power_gate - power gate/ungate the specific IP block
+ *
+ * @smu:        smu_context pointer
+ * @block_type: the IP block to power gate/ungate
+ * @gate:       to power gate if true, ungate otherwise
+ *
+ * This API uses no smu->mutex lock protection due to:
+ * 1. It is either called by other IP block(gfx/sdma/vcn/uvd/vce).
+ *    This is guarded to be race condition free by the caller.
+ * 2. Or get called on user setting request of power_dpm_force_performance_level.
+ *    Under this case, the smu->mutex lock protection is already enforced on
+ *    the parent API smu_force_performance_level of the call path.
+ */
 int smu_dpm_set_power_gate(struct smu_context *smu, uint32_t block_type,
 			   bool gate)
 {
 	int ret = 0;
 
-	mutex_lock(&smu->mutex);
-
 	switch (block_type) {
 	case AMD_IP_BLOCK_TYPE_UVD:
 		ret = smu_dpm_set_uvd_enable(smu, gate);
@@ -408,8 +419,6 @@ int smu_dpm_set_power_gate(struct smu_context *smu, uint32_t block_type,
 		break;
 	}
 
-	mutex_unlock(&smu->mutex);
-
 	return ret;
 }
 
-- 
2.23.0



More information about the amd-gfx mailing list