Skip to contentSkip to conversation

drm/amd/pm: correct the MGpuFanBoost support for Beige Goby

PATCH In review Evan Quan 2 participants 5 messages Jan 25, 2022 – Jan 26, 2022

Subsystems (3)
lore
Share & tools
lore

Evan Quan

author
#0 lore raw

PATCH 1/2 drm/amd/pm: correct the MGpuFanBoost support for Beige Goby

The existing way cannot handle Beige Goby well as a different PPTable data structure(PPTable_beige_goby_t instead of PPTable_t) is used there.

Change-Id I02208c011e93c4d37769bd022e65e9084faa97e4
---
 drivers/gpu/drm/amd/pm/swsmu/smu11/sienna_cichlid_ppt.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
drivers/gpu/drm/amd/pm/swsmu/smu11/sienna_cichlid_ppt.c +3 -3
index 651fe748e423..dcd35c68e59b 100644
@@ -3729,14 +3729,14 @@ static ssize_t sienna_cichlid_get_ecc_info(struct smu_context *smu,
 }
 static int sienna_cichlid_enable_mgpu_fan_boost(struct smu_context *smu)
 {
-	struct smu_table_context *table_context = &smu->smu_table;
-	PPTable_t *smc_pptable = table_context->driver_pptable;
+	uint16_t *mgpu_fan_boost_limit_rpm;
 
+	GET_PPTABLE_MEMBER(MGpuFanBoostLimitRpm, &mgpu_fan_boost_limit_rpm);
 	/*
 	 * Skip the MGpuFanBoost setting for those ASICs
 	 * which do not support it
 	 */
-	if (!smc_pptable->MGpuFanBoostLimitRpm)
+	if (*mgpu_fan_boost_limit_rpm == 0)
 		return 0;
 
 	return smu_cmn_send_smc_msg_with_param(smu,
Signature · 1 line
2.29.0
author
#1 lore raw

PATCH 2/2 drm/amd/pm: fix the deadlock observed on performance_level setting

The sub-routine(amdgpu_gfx_off_ctrl) tried to obtain the lock adev->pm.mutex which was actually hold by amdgpu_dpm_force_performance_level. A deadlock happened then.

Change-Id Id692829381dedc6380f5464d74107d696f7abca1
---
 drivers/gpu/drm/amd/pm/amdgpu_dpm.c | 50 ++++++++++-------------------
 1 file changed, 17 insertions(+), 33 deletions(-)
drivers/gpu/drm/amd/pm/amdgpu_dpm.c +17 -33
index 5fc33893a68c..ef574c96b41c 100644
@@ -692,25 +692,16 @@ void amdgpu_dpm_set_power_state(struct amdgpu_device *adev,
 		amdgpu_dpm_compute_clocks(adev);
 }
 
-static enum amd_dpm_forced_level amdgpu_dpm_get_performance_level_locked(struct amdgpu_device *adev)
+enum amd_dpm_forced_level amdgpu_dpm_get_performance_level(struct amdgpu_device *adev)
 {
 	const struct amd_pm_funcs *pp_funcs = adev->powerplay.pp_funcs;
 	enum amd_dpm_forced_level level;
 
+	mutex_lock(&adev->pm.mutex);
 	if (pp_funcs->get_performance_level)
 		level = pp_funcs->get_performance_level(adev->powerplay.pp_handle);
 	else
 		level = adev->pm.dpm.forced_level;
-
-	return level;
-}
-
-enum amd_dpm_forced_level amdgpu_dpm_get_performance_level(struct amdgpu_device *adev)
-{
-	enum amd_dpm_forced_level level;
-
-	mutex_lock(&adev->pm.mutex);
-	level = amdgpu_dpm_get_performance_level_locked(adev);
 	mutex_unlock(&adev->pm.mutex);
 
 	return level;
@@ -725,23 +716,16 @@ int amdgpu_dpm_force_performance_level(struct amdgpu_device *adev,
 					AMD_DPM_FORCED_LEVEL_PROFILE_MIN_SCLK |
 					AMD_DPM_FORCED_LEVEL_PROFILE_MIN_MCLK |
 					AMD_DPM_FORCED_LEVEL_PROFILE_PEAK;
-	int ret = 0;
 
 	if (!pp_funcs->force_performance_level)
 		return 0;
 
-	mutex_lock(&adev->pm.mutex);
-
-	if (adev->pm.dpm.thermal_active) {
-		ret = -EINVAL;
-		goto out;
-	}
+	if (adev->pm.dpm.thermal_active)
+		return -EINVAL;
 
-	current_level = amdgpu_dpm_get_performance_level_locked(adev);
-	if (current_level == level) {
-		ret = 0;
-		goto out;
-	}
+	current_level = amdgpu_dpm_get_performance_level(adev);
+	if (current_level == level)
+		return 0;
 
 	if (adev->asic_type == CHIP_RAVEN) {
 		if (!(adev->apu_flags & AMD_APU_IS_RAVEN2)) {
@@ -755,10 +739,8 @@ int amdgpu_dpm_force_performance_level(struct amdgpu_device *adev,
 	}
 
 	if (!(current_level & profile_mode_mask) &&
-	    (level == AMD_DPM_FORCED_LEVEL_PROFILE_EXIT)) {
-		ret = -EINVAL;
-		goto out;
-	}
+	    (level == AMD_DPM_FORCED_LEVEL_PROFILE_EXIT))
+		return -EINVAL;
 
 	if (!(current_level & profile_mode_mask) &&
 	      (level & profile_mode_mask)) {
@@ -780,17 +762,19 @@ int amdgpu_dpm_force_performance_level(struct amdgpu_device *adev,
 						       AMD_PG_STATE_GATE);
 	}
 
+	mutex_lock(&adev->pm.mutex);
+
 	if (pp_funcs->force_performance_level(adev->powerplay.pp_handle,
-					      level))
-		ret = -EINVAL;
+					      level)) {
+		mutex_unlock(&adev->pm.mutex);
+		return -EINVAL;
+	}
 
-	if (!ret)
-		adev->pm.dpm.forced_level = level;
+	adev->pm.dpm.forced_level = level;
 
-out:
 	mutex_unlock(&adev->pm.mutex);
 
-	return ret;
+	return 0;
 }
 
 int amdgpu_dpm_get_pp_num_states(struct amdgpu_device *adev,
Signature · 1 line
2.29.0
Reviewed-by on v1 2/2
#2 lore raw

Re: [PATCH 2/2] drm/amd/pm: fix the deadlock observed on performance_level setting

Quoted 7 lines · Evan Quan

The sub-routine(amdgpu_gfx_off_ctrl) tried to obtain the lock adev->pm.mutex which was actually hold by amdgpu_dpm_force_performance_level. A deadlock happened then.

Change-Id Id692829381dedc6380f5464d74107d696f7abca1

I think in the long term, we should fix up the logic to allow us to keep the lock across the whole function, but for now,

Quoted 107 lines
---
 drivers/gpu/drm/amd/pm/amdgpu_dpm.c | 50 ++++++++++-------------------
 1 file changed, 17 insertions(+), 33 deletions(-)
drivers/gpu/drm/amd/pm/amdgpu_dpm.c +17 -34
index 5fc33893a68c..ef574c96b41c 100644
@@ -692,25 +692,16 @@ void amdgpu_dpm_set_power_state(struct amdgpu_device *adev,
                amdgpu_dpm_compute_clocks(adev);
 }
 
-static enum amd_dpm_forced_level amdgpu_dpm_get_performance_level_locked(struct amdgpu_device *adev)
+enum amd_dpm_forced_level amdgpu_dpm_get_performance_level(struct amdgpu_device *adev)
 {
        const struct amd_pm_funcs *pp_funcs = adev->powerplay.pp_funcs;
        enum amd_dpm_forced_level level;
 
+       mutex_lock(&adev->pm.mutex);
        if (pp_funcs->get_performance_level)
                level = pp_funcs->get_performance_level(adev->powerplay.pp_handle);
        else
                level = adev->pm.dpm.forced_level;
-
-       return level;
-}
-
-enum amd_dpm_forced_level amdgpu_dpm_get_performance_level(struct amdgpu_device *adev)
-{
-       enum amd_dpm_forced_level level;
-
-       mutex_lock(&adev->pm.mutex);
-       level = amdgpu_dpm_get_performance_level_locked(adev);
        mutex_unlock(&adev->pm.mutex);
 
        return level;
@@ -725,23 +716,16 @@ int amdgpu_dpm_force_performance_level(struct amdgpu_device *adev,
                                        AMD_DPM_FORCED_LEVEL_PROFILE_MIN_SCLK |
                                        AMD_DPM_FORCED_LEVEL_PROFILE_MIN_MCLK |
                                        AMD_DPM_FORCED_LEVEL_PROFILE_PEAK;
-       int ret = 0;
 
        if (!pp_funcs->force_performance_level)
                return 0;
 
-       mutex_lock(&adev->pm.mutex);
-
-       if (adev->pm.dpm.thermal_active) {
-               ret = -EINVAL;
-               goto out;
-       }
+       if (adev->pm.dpm.thermal_active)
+               return -EINVAL;
 
-       current_level = amdgpu_dpm_get_performance_level_locked(adev);
-       if (current_level == level) {
-               ret = 0;
-               goto out;
-       }
+       current_level = amdgpu_dpm_get_performance_level(adev);
+       if (current_level == level)
+               return 0;
 
        if (adev->asic_type == CHIP_RAVEN) {
                if (!(adev->apu_flags & AMD_APU_IS_RAVEN2)) {
@@ -755,10 +739,8 @@ int amdgpu_dpm_force_performance_level(struct amdgpu_device *adev,
        }
 
        if (!(current_level & profile_mode_mask) &&
-           (level == AMD_DPM_FORCED_LEVEL_PROFILE_EXIT)) {
-               ret = -EINVAL;
-               goto out;
-       }
+           (level == AMD_DPM_FORCED_LEVEL_PROFILE_EXIT))
+               return -EINVAL;
 
        if (!(current_level & profile_mode_mask) &&
              (level & profile_mode_mask)) {
@@ -780,17 +762,19 @@ int amdgpu_dpm_force_performance_level(struct amdgpu_device *adev,
                                                       AMD_PG_STATE_GATE);
        }
 
+       mutex_lock(&adev->pm.mutex);
+
        if (pp_funcs->force_performance_level(adev->powerplay.pp_handle,
-                                             level))
-               ret = -EINVAL;
+                                             level)) {
+               mutex_unlock(&adev->pm.mutex);
+               return -EINVAL;
+       }
 
-       if (!ret)
-               adev->pm.dpm.forced_level = level;
+       adev->pm.dpm.forced_level = level;
 
-out:
        mutex_unlock(&adev->pm.mutex);
 
-       return ret;
+       return 0;
 }
 
 int amdgpu_dpm_get_pp_num_states(struct amdgpu_device *adev,
--

2.29.0

author
#3 lore raw

RE: [PATCH 2/2] drm/amd/pm: fix the deadlock observed on performance_level setting

[AMD Official Use Only]

Quoted 13 lines

-----Original Message-----
From: Alex Deucher <[email protected]>
Sent: Tuesday, January 25, 2022 11:35 PM
To: Quan, Evan <[email protected]>
Cc: amd-gfx list <[email protected]>; Deucher, Alexander <[email protected]>
Subject: Re: [PATCH 2/2] drm/amd/pm: fix the deadlock observed on performance_level setting

Quoted 3 lines · Evan Quan

The sub-routine(amdgpu_gfx_off_ctrl) tried to obtain the lock adev->pm.mutex which was actually hold by

Quoted 8 lines

amdgpu_dpm_force_performance_level.

Quoted 4 lines

A deadlock happened then.

Change-Id Id692829381dedc6380f5464d74107d696f7abca1

I think in the long term, we should fix up the logic to allow us to keep the lock across the whole function, but for now,

[Quan, Evan] Agreed. But I have not figured out what's the proper way to do that. In my opinion we could either enforce a different lock inside the function. Or we move this interface somewhere upper layer(e.g. amdgpu_device.c) and add some lock protections there.
But as you said, let's land this a temporary fix for now.

BR
Evan

Quoted 120 lines

drivers/gpu/drm/amd/pm/amdgpu_dpm.c void amdgpu_dpm_set_power_state(struct

Quoted 11 lines
---
 drivers/gpu/drm/amd/pm/amdgpu_dpm.c | 50
++++++++++-------------------
 1 file changed, 17 insertions(+), 33 deletions(-)
drivers/gpu/drm/amd/pm/amdgpu_dpm.c +0 -0
index 5fc33893a68c..ef574c96b41c 100644
@@ -692,25 +692,16 @@ void amdgpu_dpm_set_power_state(struct

amdgpu_device *adev,

Quoted 4 lines
drivers/gpu/drm/amd/pm/amdgpu_dpm.c +0 -1
                amdgpu_dpm_compute_clocks(adev);  }
 
-static enum amd_dpm_forced_level

amdgpu_dpm_get_performance_level_locked(struct amdgpu_device

*adev)

Quoted 1 line

+enum amd_dpm_forced_level

amdgpu_dpm_get_performance_level(struct

Quoted 11 lines
drivers/gpu/drm/amd/pm/amdgpu_dpm.c +2 -0
+amdgpu_device *adev)
 {
        const struct amd_pm_funcs *pp_funcs = adev->powerplay.pp_funcs;
        enum amd_dpm_forced_level level;
 
+       mutex_lock(&adev->pm.mutex);
        if (pp_funcs->get_performance_level)
                level = pp_funcs->get_performance_level(adev-
powerplay.pp_handle);
        else
                level = adev->pm.dpm.forced_level;

drivers/gpu/drm/amd/pm/amdgpu_dpm.c void amdgpu_dpm_set_power_state(struct amdgpu_device *adev,

Quoted 5 lines
drivers/gpu/drm/amd/pm/amdgpu_dpm.c +0 -5
-
-       return level;
-}
-
-enum amd_dpm_forced_level

amdgpu_dpm_get_performance_level(struct

drivers/gpu/drm/amd/pm/amdgpu_dpm.c int

Quoted 9 lines

amdgpu_device *adev) -{

drivers/gpu/drm/amd/pm/amdgpu_dpm.c +0 -4
-       enum amd_dpm_forced_level level;
-
-       mutex_lock(&adev->pm.mutex);
-       level = amdgpu_dpm_get_performance_level_locked(adev);
        mutex_unlock(&adev->pm.mutex);
 
        return level;
@@ -725,23 +716,16 @@ int

amdgpu_dpm_force_performance_level(struct amdgpu_device *adev,

Quoted 19 lines
drivers/gpu/drm/amd/pm/amdgpu_dpm.c +2 -9
                                        AMD_DPM_FORCED_LEVEL_PROFILE_MIN_SCLK |
                                        AMD_DPM_FORCED_LEVEL_PROFILE_MIN_MCLK |
                                        AMD_DPM_FORCED_LEVEL_PROFILE_PEAK;
-       int ret = 0;
 
        if (!pp_funcs->force_performance_level)
                return 0;
 
-       mutex_lock(&adev->pm.mutex);
-
-       if (adev->pm.dpm.thermal_active) {
-               ret = -EINVAL;
-               goto out;
-       }
+       if (adev->pm.dpm.thermal_active)
+               return -EINVAL;
 
-       current_level = amdgpu_dpm_get_performance_level_locked(adev);
-       if (current_level == level) {

drivers/gpu/drm/amd/pm/amdgpu_dpm.c int amdgpu_dpm_force_performance_level(struct amdgpu_device *adev,

Quoted 10 lines
drivers/gpu/drm/amd/pm/amdgpu_dpm.c +3 -4
-               ret = 0;
-               goto out;
-       }
+       current_level = amdgpu_dpm_get_performance_level(adev);
+       if (current_level == level)
+               return 0;
 
        if (adev->asic_type == CHIP_RAVEN) {
                if (!(adev->apu_flags & AMD_APU_IS_RAVEN2)) { @@
-755,10 +739,8 @@ int amdgpu_dpm_force_performance_level(struct

amdgpu_device *adev,

Quoted 41 lines
        }

        if (!(current_level & profile_mode_mask) &&
drivers/gpu/drm/amd/pm/amdgpu_dpm.c +2 -4
-           (level == AMD_DPM_FORCED_LEVEL_PROFILE_EXIT)) {
-               ret = -EINVAL;
-               goto out;
-       }
+           (level == AMD_DPM_FORCED_LEVEL_PROFILE_EXIT))
+               return -EINVAL;
 
        if (!(current_level & profile_mode_mask) &&
              (level & profile_mode_mask)) { @@ -780,17 +762,19 @@ int

amdgpu_dpm_force_performance_level(struct amdgpu_device *adev,

                                                       AMD_PG_STATE_GATE);
        }
drivers/gpu/drm/amd/pm/amdgpu_dpm.c +8 -7
+       mutex_lock(&adev->pm.mutex);
+
        if (pp_funcs->force_performance_level(adev->powerplay.pp_handle,
-                                             level))
-               ret = -EINVAL;
+                                             level)) {
+               mutex_unlock(&adev->pm.mutex);
+               return -EINVAL;
+       }
 
-       if (!ret)
-               adev->pm.dpm.forced_level = level;
+       adev->pm.dpm.forced_level = level;
 
-out:
        mutex_unlock(&adev->pm.mutex);
 
-       return ret;
+       return 0;
 }
 
 int amdgpu_dpm_get_pp_num_states(struct amdgpu_device *adev,
--

2.29.0

Reviewed-by on v1 1/2
#4 lore raw
In reply to Evan Quan
Quoted 36 lines · Evan Quan

The existing way cannot handle Beige Goby well as a different PPTable data structure(PPTable_beige_goby_t instead of PPTable_t) is used there.

Change-Id I02208c011e93c4d37769bd022e65e9084faa97e4
---
 drivers/gpu/drm/amd/pm/swsmu/smu11/sienna_cichlid_ppt.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
drivers/gpu/drm/amd/pm/swsmu/smu11/sienna_cichlid_ppt.c +3 -4
index 651fe748e423..dcd35c68e59b 100644
@@ -3729,14 +3729,14 @@ static ssize_t sienna_cichlid_get_ecc_info(struct smu_context *smu,
 }
 static int sienna_cichlid_enable_mgpu_fan_boost(struct smu_context *smu)
 {
-       struct smu_table_context *table_context = &smu->smu_table;
-       PPTable_t *smc_pptable = table_context->driver_pptable;
+       uint16_t *mgpu_fan_boost_limit_rpm;
 
+       GET_PPTABLE_MEMBER(MGpuFanBoostLimitRpm, &mgpu_fan_boost_limit_rpm);
        /*
         * Skip the MGpuFanBoost setting for those ASICs
         * which do not support it
         */
-       if (!smc_pptable->MGpuFanBoostLimitRpm)
+       if (*mgpu_fan_boost_limit_rpm == 0)
                return 0;
 
        return smu_cmn_send_smc_msg_with_param(smu,
--

2.29.0