Skip to contentSkip to conversation

drm/amd/pm: enable GPO dynamic control support for SMU13.0.0

PATCH In review Evan Quan 2 participants 11 messages Dec 9, 2022

Subsystems (4)
lore
Share & tools
lore

Evan Quan

author
#0 lore raw

PATCH 1/10 drm/amd/pm: enable GPO dynamic control support for SMU13.0.0

To better support UMD pstate profilings, the GPO feature needs to be switched on/off accordingly.

Change-Id I162139a9e1ecac0a75d0fc3f93f608bf2223ac8c
---
 drivers/gpu/drm/amd/pm/swsmu/inc/smu_types.h      |  3 ++-
 drivers/gpu/drm/amd/pm/swsmu/inc/smu_v13_0.h      |  3 +++
 drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0.c    | 15 +++++++++++++++
 .../gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_0_ppt.c  |  2 ++
 4 files changed, 22 insertions(+), 1 deletion(-)
drivers/gpu/drm/amd/pm/swsmu/inc/smu_types.h +2 -1
index a4e3425b1027..4180c71d930f 100644
@@ -241,7 +241,8 @@
 	__SMU_DUMMY_MAP(GetGfxOffEntryCount),		 \
 	__SMU_DUMMY_MAP(LogGfxOffResidency),			\
 	__SMU_DUMMY_MAP(SetNumBadMemoryPagesRetired),		\
-	__SMU_DUMMY_MAP(SetBadMemoryPagesRetiredFlagsPerChannel),
+	__SMU_DUMMY_MAP(SetBadMemoryPagesRetiredFlagsPerChannel), \
+	__SMU_DUMMY_MAP(AllowGpo),
 
 #undef __SMU_DUMMY_MAP
 #define __SMU_DUMMY_MAP(type)	SMU_MSG_##type
drivers/gpu/drm/amd/pm/swsmu/inc/smu_v13_0.h +3 -0
index 9c7941ae3844..46d472310b7e 100644
@@ -273,6 +273,9 @@ int smu_v13_0_init_pptable_microcode(struct smu_context *smu);
 
 int smu_v13_0_run_btc(struct smu_context *smu);
 
+int smu_v13_0_gpo_control(struct smu_context *smu,
+			  bool enablement);
+
 int smu_v13_0_deep_sleep_control(struct smu_context *smu,
 				 bool enablement);
 
drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0.c +15 -0
index f9c8af984a45..38517fe3317d 100644
@@ -2180,6 +2180,21 @@ int smu_v13_0_run_btc(struct smu_context *smu)
 	return res;
 }
 
+int smu_v13_0_gpo_control(struct smu_context *smu,
+			  bool enablement)
+{
+	int res;
+
+	res = smu_cmn_send_smc_msg_with_param(smu,
+					      SMU_MSG_AllowGpo,
+					      enablement ? 1 : 0,
+					      NULL);
+	if (res)
+		dev_err(smu->adev->dev, "SetGpoAllow %d failed!\n", enablement);
+
+	return res;
+}
+
 int smu_v13_0_deep_sleep_control(struct smu_context *smu,
 				 bool enablement)
 {
drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_0_ppt.c +2 -0
index eba3d7754532..c2fcff384bab 100644
@@ -144,6 +144,7 @@ static struct cmn2asic_msg_mapping smu_v13_0_0_message_map[SMU_MSG_MAX_COUNT] =
 	MSG_MAP(SetNumBadMemoryPagesRetired,	PPSMC_MSG_SetNumBadMemoryPagesRetired,   0),
 	MSG_MAP(SetBadMemoryPagesRetiredFlagsPerChannel,
 			    PPSMC_MSG_SetBadMemoryPagesRetiredFlagsPerChannel,   0),
+	MSG_MAP(AllowGpo,			PPSMC_MSG_SetGpoAllow,           0),
 };
 
 static struct cmn2asic_mapping smu_v13_0_0_clk_map[SMU_CLK_COUNT] = {
@@ -1948,6 +1949,7 @@ static const struct pptable_funcs smu_v13_0_0_ppt_funcs = {
 	.set_df_cstate = smu_v13_0_0_set_df_cstate,
 	.send_hbm_bad_pages_num = smu_v13_0_0_smu_send_bad_mem_page_num,
 	.send_hbm_bad_channel_flag = smu_v13_0_0_send_bad_mem_channel_flag,
+	.gpo_control = smu_v13_0_gpo_control,
 };
 
 void smu_v13_0_0_set_ppt_funcs(struct smu_context *smu)
Signature · 1 line
2.34.1
author
#1 lore raw

PATCH 2/10 drm/amd/pm: enable GPO dynamic control support for SMU13.0.7

To better support UMD pstate profilings, the GPO feature needs to be switched on/off accordingly.

Change-Id Ia0236c05d507b6c253fd558482c665528aa0ceb8
---
 drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_7_ppt.c | 2 ++
 1 file changed, 2 insertions(+)
drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_7_ppt.c +2 -0
index 94c1aab7861f..60688d3619a4 100644
@@ -123,6 +123,7 @@ static struct cmn2asic_msg_mapping smu_v13_0_7_message_map[SMU_MSG_MAX_COUNT] =
 	MSG_MAP(SetMGpuFanBoostLimitRpm,	PPSMC_MSG_SetMGpuFanBoostLimitRpm,     0),
 	MSG_MAP(DFCstateControl,		PPSMC_MSG_SetExternalClientDfCstateAllow, 0),
 	MSG_MAP(ArmD3,				PPSMC_MSG_ArmD3,                       0),
+	MSG_MAP(AllowGpo,			PPSMC_MSG_SetGpoAllow,           0),
 };
 
 static struct cmn2asic_mapping smu_v13_0_7_clk_map[SMU_CLK_COUNT] = {
@@ -1689,6 +1690,7 @@ static const struct pptable_funcs smu_v13_0_7_ppt_funcs = {
 	.mode1_reset = smu_v13_0_mode1_reset,
 	.set_mp1_state = smu_v13_0_7_set_mp1_state,
 	.set_df_cstate = smu_v13_0_7_set_df_cstate,
+	.gpo_control = smu_v13_0_gpo_control,
 };
 
 void smu_v13_0_7_set_ppt_funcs(struct smu_context *smu)
Signature · 1 line
2.34.1
author
#2 lore raw

PATCH 3/10 drm/amd/pm: correct SMU13.0.0 pstate profiling clock settings

In reply to Evan Quan

Correct the pstate standard/peak profiling mode clock settings for SMU13.0.0.

Change-Id Ic47b18904b1fa021588b6d4b599f1cf0fe4b7a58
---
 .../drm/amd/pm/swsmu/smu13/smu_v13_0_0_ppt.c  | 22 +++++++++++++------
 1 file changed, 15 insertions(+), 7 deletions(-)
drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_0_ppt.c +15 -7
index c2fcff384bab..4adbab62e9dd 100644
@@ -1305,9 +1305,17 @@ static int smu_v13_0_0_populate_umd_state_clk(struct smu_context *smu)
 				&dpm_context->dpm_tables.fclk_table;
 	struct smu_umd_pstate_table *pstate_table =
 				&smu->pstate_table;
+	struct smu_table_context *table_context = &smu->smu_table;
+	PPTable_t *pptable = table_context->driver_pptable;
+	DriverReportedClocks_t driver_clocks =
+			pptable->SkuTable.DriverReportedClocks;
 
 	pstate_table->gfxclk_pstate.min = gfx_table->min;
-	pstate_table->gfxclk_pstate.peak = gfx_table->max;
+	if (driver_clocks.GameClockAc &&
+	    (driver_clocks.GameClockAc < gfx_table->max))
+		pstate_table->gfxclk_pstate.peak = driver_clocks.GameClockAc;
+	else
+		pstate_table->gfxclk_pstate.peak = gfx_table->max;
 
 	pstate_table->uclk_pstate.min = mem_table->min;
 	pstate_table->uclk_pstate.peak = mem_table->max;
@@ -1324,12 +1332,12 @@ static int smu_v13_0_0_populate_umd_state_clk(struct smu_context *smu)
 	pstate_table->fclk_pstate.min = fclk_table->min;
 	pstate_table->fclk_pstate.peak = fclk_table->max;
 
-	/*
-	 * For now, just use the mininum clock frequency.
-	 * TODO: update them when the real pstate settings available
-	 */
-	pstate_table->gfxclk_pstate.standard = gfx_table->min;
-	pstate_table->uclk_pstate.standard = mem_table->min;
+	if (driver_clocks.BaseClockAc &&
+	    driver_clocks.BaseClockAc < gfx_table->max)
+		pstate_table->gfxclk_pstate.standard = driver_clocks.BaseClockAc;
+	else
+		pstate_table->gfxclk_pstate.standard = gfx_table->max;
+	pstate_table->uclk_pstate.standard = mem_table->max;
 	pstate_table->socclk_pstate.standard = soc_table->min;
 	pstate_table->vclk_pstate.standard = vclk_table->min;
 	pstate_table->dclk_pstate.standard = dclk_table->min;
Signature · 1 line
2.34.1
author
#3 lore raw

PATCH 4/10 drm/amd/pm: drop unused SMU v13 API

In reply to Evan Quan

The API is not in use. And it's unlikely to be used in the future either.

Change-Id I7ace04dda5752d940fdd630e6b86879f748a3ac8
---
 drivers/gpu/drm/amd/pm/swsmu/inc/smu_v13_0.h  |  5 ---
 .../gpu/drm/amd/pm/swsmu/smu13/smu_v13_0.c    | 39 -------------------
 2 files changed, 44 deletions(-)
drivers/gpu/drm/amd/pm/swsmu/inc/smu_v13_0.h +0 -5
index 46d472310b7e..1c9c5dfc8b51 100644
@@ -244,11 +244,6 @@ int smu_v13_0_set_single_dpm_table(struct smu_context *smu,
 				   enum smu_clk_type clk_type,
 				   struct smu_13_0_dpm_table *single_dpm_table);
 
-int smu_v13_0_get_dpm_level_range(struct smu_context *smu,
-				  enum smu_clk_type clk_type,
-				  uint32_t *min_value,
-				  uint32_t *max_value);
-
 int smu_v13_0_get_current_pcie_link_width_level(struct smu_context *smu);
 
 int smu_v13_0_get_current_pcie_link_width(struct smu_context *smu);
drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0.c +0 -39
index 38517fe3317d..fbc138203fd2 100644
@@ -2062,45 +2062,6 @@ int smu_v13_0_set_single_dpm_table(struct smu_context *smu,
 	return 0;
 }
 
-int smu_v13_0_get_dpm_level_range(struct smu_context *smu,
-				  enum smu_clk_type clk_type,
-				  uint32_t *min_value,
-				  uint32_t *max_value)
-{
-	uint32_t level_count = 0;
-	int ret = 0;
-
-	if (!min_value && !max_value)
-		return -EINVAL;
-
-	if (min_value) {
-		/* by default, level 0 clock value as min value */
-		ret = smu_v13_0_get_dpm_freq_by_index(smu,
-						      clk_type,
-						      0,
-						      min_value);
-		if (ret)
-			return ret;
-	}
-
-	if (max_value) {
-		ret = smu_v13_0_get_dpm_level_count(smu,
-						    clk_type,
-						    &level_count);
-		if (ret)
-			return ret;
-
-		ret = smu_v13_0_get_dpm_freq_by_index(smu,
-						      clk_type,
-						      level_count - 1,
-						      max_value);
-		if (ret)
-			return ret;
-	}
-
-	return ret;
-}
-
 int smu_v13_0_get_current_pcie_link_width_level(struct smu_context *smu)
 {
 	struct amdgpu_device *adev = smu->adev;
Signature · 1 line
2.34.1
author
#4 lore raw

PATCH 5/10 drm/amd/pm: fulfill swsmu peak profiling mode shader/memory clock settings

In reply to Evan Quan

Enable peak profiling mode shader/memory clocks reporting for swsmu framework.

Change-Id Iebefecc9addd7352a50baec571791edc8d968375
---
 drivers/gpu/drm/amd/include/kgd_pp_interface.h | 2 ++
 drivers/gpu/drm/amd/pm/swsmu/amdgpu_smu.c      | 8 ++++++++
 2 files changed, 10 insertions(+)
drivers/gpu/drm/amd/include/kgd_pp_interface.h +2 -0
index 20e0272e46bb..d68bd7c1aa04 100644
@@ -139,6 +139,8 @@ enum amd_pp_sensors {
 	AMDGPU_PP_SENSOR_MIN_FAN_RPM,
 	AMDGPU_PP_SENSOR_MAX_FAN_RPM,
 	AMDGPU_PP_SENSOR_VCN_POWER_STATE,
+	AMDGPU_PP_SENSOR_PEAK_PSTATE_SCLK,
+	AMDGPU_PP_SENSOR_PEAK_PSTATE_MCLK,
 };
 
 enum amd_pp_task {
drivers/gpu/drm/amd/pm/swsmu/amdgpu_smu.c +8 -0
index 8650a1404e22..3421022bf707 100644
@@ -2473,6 +2473,14 @@ static int smu_read_sensor(void *handle,
 		*((uint32_t *)data) = pstate_table->uclk_pstate.standard * 100;
 		*size = 4;
 		break;
+	case AMDGPU_PP_SENSOR_PEAK_PSTATE_SCLK:
+		*((uint32_t *)data) = pstate_table->gfxclk_pstate.peak * 100;
+		*size = 4;
+		break;
+	case AMDGPU_PP_SENSOR_PEAK_PSTATE_MCLK:
+		*((uint32_t *)data) = pstate_table->uclk_pstate.peak * 100;
+		*size = 4;
+		break;
 	case AMDGPU_PP_SENSOR_ENABLED_SMC_FEATURES_MASK:
 		ret = smu_feature_get_enabled_mask(smu, (uint64_t *)data);
 		*size = 8;
Signature · 1 line
2.34.1
author
#5 lore raw

PATCH 6/10 drm/amd/pm: fulfill powerplay peak profiling mode shader/memory clock settings

In reply to Evan Quan

Enable peak profiling mode shader/memory clock reporting for powerplay framework.

Change-Id Ied28f5921c3e72d86d8e769908910f1b11a0100b
---
 .../gpu/drm/amd/pm/powerplay/amd_powerplay.c  | 10 ++-
 .../drm/amd/pm/powerplay/hwmgr/smu10_hwmgr.c  | 16 +++-
 .../drm/amd/pm/powerplay/hwmgr/smu7_hwmgr.c   | 76 +++++++++++++++----
 .../drm/amd/pm/powerplay/hwmgr/smu8_hwmgr.c   | 16 +++-
 .../drm/amd/pm/powerplay/hwmgr/vega10_hwmgr.c | 31 ++++++--
 .../drm/amd/pm/powerplay/hwmgr/vega12_hwmgr.c | 22 ++++++
 .../drm/amd/pm/powerplay/hwmgr/vega20_hwmgr.c | 20 ++---
 drivers/gpu/drm/amd/pm/powerplay/inc/hwmgr.h  |  2 +
 8 files changed, 155 insertions(+), 38 deletions(-)
drivers/gpu/drm/amd/pm/powerplay/amd_powerplay.c +8 -2
index 304190d5c9d2..8f2cc6310340 100644
@@ -769,10 +769,16 @@ static int pp_dpm_read_sensor(void *handle, int idx,
 
 	switch (idx) {
 	case AMDGPU_PP_SENSOR_STABLE_PSTATE_SCLK:
-		*((uint32_t *)value) = hwmgr->pstate_sclk;
+		*((uint32_t *)value) = hwmgr->pstate_sclk * 100;
 		return 0;
 	case AMDGPU_PP_SENSOR_STABLE_PSTATE_MCLK:
-		*((uint32_t *)value) = hwmgr->pstate_mclk;
+		*((uint32_t *)value) = hwmgr->pstate_mclk * 100;
+		return 0;
+	case AMDGPU_PP_SENSOR_PEAK_PSTATE_SCLK:
+		*((uint32_t *)value) = hwmgr->pstate_sclk_peak * 100;
+		return 0;
+	case AMDGPU_PP_SENSOR_PEAK_PSTATE_MCLK:
+		*((uint32_t *)value) = hwmgr->pstate_mclk_peak * 100;
 		return 0;
 	case AMDGPU_PP_SENSOR_MIN_FAN_RPM:
 		*((uint32_t *)value) = hwmgr->thermal_controller.fanInfo.ulMinRPM;
drivers/gpu/drm/amd/pm/powerplay/hwmgr/smu10_hwmgr.c +13 -3
index ede71de2343d..86d6e88c7386 100644
@@ -375,6 +375,17 @@ static int smu10_enable_gfx_off(struct pp_hwmgr *hwmgr)
 	return 0;
 }
 
+static void smu10_populate_umdpstate_clocks(struct pp_hwmgr *hwmgr)
+{
+	hwmgr->pstate_sclk = SMU10_UMD_PSTATE_GFXCLK;
+	hwmgr->pstate_mclk = SMU10_UMD_PSTATE_FCLK;
+
+	smum_send_msg_to_smc(hwmgr,
+			     PPSMC_MSG_GetMaxGfxclkFrequency,
+			     &hwmgr->pstate_sclk_peak);
+	hwmgr->pstate_mclk_peak = SMU10_UMD_PSTATE_PEAK_FCLK;
+}
+
 static int smu10_enable_dpm_tasks(struct pp_hwmgr *hwmgr)
 {
 	struct amdgpu_device *adev = hwmgr->adev;
@@ -398,6 +409,8 @@ static int smu10_enable_dpm_tasks(struct pp_hwmgr *hwmgr)
 			return ret;
 	}
 
+	smu10_populate_umdpstate_clocks(hwmgr);
+
 	return 0;
 }
 
@@ -574,9 +587,6 @@ static int smu10_hwmgr_backend_init(struct pp_hwmgr *hwmgr)
 
 	hwmgr->platform_descriptor.minimumClocksReductionPercentage = 50;
 
-	hwmgr->pstate_sclk = SMU10_UMD_PSTATE_GFXCLK * 100;
-	hwmgr->pstate_mclk = SMU10_UMD_PSTATE_FCLK * 100;
-
 	/* enable the pp_od_clk_voltage sysfs file */
 	hwmgr->od_enabled = 1;
 	/* disabled fine grain tuning function by default */
drivers/gpu/drm/amd/pm/powerplay/hwmgr/smu7_hwmgr.c +63 -13

122 lines not on this page · raw message

drivers/gpu/drm/amd/pm/powerplay/hwmgr/smu8_hwmgr.c +14 -2
index b50fd4a4a3d1..b015a601b385 100644
@@ -1016,6 +1016,18 @@ static void smu8_reset_acp_boot_level(struct pp_hwmgr *hwmgr)
 	data->acp_boot_level = 0xff;
 }
 
+static void smu8_populate_umdpstate_clocks(struct pp_hwmgr *hwmgr)
+{
+	struct phm_clock_voltage_dependency_table *table =
+				hwmgr->dyn_state.vddc_dependency_on_sclk;
+
+	hwmgr->pstate_sclk = table->entries[0].clk / 100;
+	hwmgr->pstate_mclk = 0;
+
+	hwmgr->pstate_sclk_peak = table->entries[table->count - 1].clk / 100;
+	hwmgr->pstate_mclk_peak = 0;
+}
+
 static int smu8_enable_dpm_tasks(struct pp_hwmgr *hwmgr)
 {
 	smu8_program_voting_clients(hwmgr);
@@ -1024,6 +1036,8 @@ static int smu8_enable_dpm_tasks(struct pp_hwmgr *hwmgr)
 	smu8_program_bootup_state(hwmgr);
 	smu8_reset_acp_boot_level(hwmgr);
 
+	smu8_populate_umdpstate_clocks(hwmgr);
+
 	return 0;
 }
 
@@ -1167,8 +1181,6 @@ static int smu8_phm_unforce_dpm_levels(struct pp_hwmgr *hwmgr)
 
 	data->sclk_dpm.soft_min_clk = table->entries[0].clk;
 	data->sclk_dpm.hard_min_clk = table->entries[0].clk;
-	hwmgr->pstate_sclk = table->entries[0].clk;
-	hwmgr->pstate_mclk = 0;
 
 	level = smu8_get_max_sclk_level(hwmgr) - 1;
 
drivers/gpu/drm/amd/pm/powerplay/hwmgr/vega10_hwmgr.c +26 -5
index c8c9fb827bda..6f5161738bf8 100644
@@ -3008,6 +3008,30 @@ static int vega10_enable_disable_PCC_limit_feature(struct pp_hwmgr *hwmgr, bool
 	return 0;
 }
 
+static void vega10_populate_umdpstate_clocks(struct pp_hwmgr *hwmgr)
+{
+	struct phm_ppt_v2_information *table_info =
+			(struct phm_ppt_v2_information *)(hwmgr->pptable);
+
+	if (table_info->vdd_dep_on_sclk->count > VEGA10_UMD_PSTATE_GFXCLK_LEVEL &&
+	    table_info->vdd_dep_on_mclk->count > VEGA10_UMD_PSTATE_MCLK_LEVEL) {
+		hwmgr->pstate_sclk = table_info->vdd_dep_on_sclk->entries[VEGA10_UMD_PSTATE_GFXCLK_LEVEL].clk;
+		hwmgr->pstate_mclk = table_info->vdd_dep_on_mclk->entries[VEGA10_UMD_PSTATE_MCLK_LEVEL].clk;
+	} else {
+		hwmgr->pstate_sclk = table_info->vdd_dep_on_sclk->entries[0].clk;
+		hwmgr->pstate_mclk = table_info->vdd_dep_on_mclk->entries[0].clk;
+	}
+
+	hwmgr->pstate_sclk_peak = table_info->vdd_dep_on_sclk->entries[table_info->vdd_dep_on_sclk->count - 1].clk;
+	hwmgr->pstate_mclk_peak = table_info->vdd_dep_on_mclk->entries[table_info->vdd_dep_on_mclk->count - 1].clk;
+
+	/* make sure the output is in Mhz */
+	hwmgr->pstate_sclk /= 100;
+	hwmgr->pstate_mclk /= 100;
+	hwmgr->pstate_sclk_peak /= 100;
+	hwmgr->pstate_mclk_peak /= 100;
+}
+
 static int vega10_enable_dpm_tasks(struct pp_hwmgr *hwmgr)
 {
 	struct vega10_hwmgr *data = hwmgr->backend;
@@ -3082,6 +3106,8 @@ static int vega10_enable_dpm_tasks(struct pp_hwmgr *hwmgr)
 				    result = tmp_result);
 	}
 
+	vega10_populate_umdpstate_clocks(hwmgr);
+
 	return result;
 }
 
@@ -4169,8 +4195,6 @@ static int vega10_get_profiling_clk_mask(struct pp_hwmgr *hwmgr, enum amd_dpm_fo
 		*sclk_mask = VEGA10_UMD_PSTATE_GFXCLK_LEVEL;
 		*soc_mask = VEGA10_UMD_PSTATE_SOCCLK_LEVEL;
 		*mclk_mask = VEGA10_UMD_PSTATE_MCLK_LEVEL;
-		hwmgr->pstate_sclk = table_info->vdd_dep_on_sclk->entries[VEGA10_UMD_PSTATE_GFXCLK_LEVEL].clk;
-		hwmgr->pstate_mclk = table_info->vdd_dep_on_mclk->entries[VEGA10_UMD_PSTATE_MCLK_LEVEL].clk;
 	}
 
 	if (level == AMD_DPM_FORCED_LEVEL_PROFILE_MIN_SCLK) {
@@ -4281,9 +4305,6 @@ static int vega10_dpm_force_dpm_level(struct pp_hwmgr *hwmgr,
 	uint32_t mclk_mask = 0;
 	uint32_t soc_mask = 0;
 
-	if (hwmgr->pstate_sclk == 0)
-		vega10_get_profiling_clk_mask(hwmgr, level, &sclk_mask, &mclk_mask, &soc_mask);
-
 	switch (level) {
 	case AMD_DPM_FORCED_LEVEL_HIGH:
 		ret = vega10_force_dpm_highest(hwmgr);
drivers/gpu/drm/amd/pm/powerplay/hwmgr/vega12_hwmgr.c +22 -0
index a2f4d6773d45..33f31461ea6c 100644
@@ -1026,6 +1026,25 @@ static int vega12_get_all_clock_ranges(struct pp_hwmgr *hwmgr)
 	return 0;
 }
 
+static void vega12_populate_umdpstate_clocks(struct pp_hwmgr *hwmgr)
+{
+	struct vega12_hwmgr *data = (struct vega12_hwmgr *)(hwmgr->backend);
+	struct vega12_single_dpm_table *gfx_dpm_table = &(data->dpm_table.gfx_table);
+	struct vega12_single_dpm_table *mem_dpm_table = &(data->dpm_table.mem_table);
+
+	if (gfx_dpm_table->count > VEGA12_UMD_PSTATE_GFXCLK_LEVEL &&
+	    mem_dpm_table->count > VEGA12_UMD_PSTATE_MCLK_LEVEL) {
+		hwmgr->pstate_sclk = gfx_dpm_table->dpm_levels[VEGA12_UMD_PSTATE_GFXCLK_LEVEL].value;
+		hwmgr->pstate_mclk = mem_dpm_table->dpm_levels[VEGA12_UMD_PSTATE_MCLK_LEVEL].value;
+	} else {
+		hwmgr->pstate_sclk = gfx_dpm_table->dpm_levels[0].value;
+		hwmgr->pstate_mclk = mem_dpm_table->dpm_levels[0].value;
+	}
+
+	hwmgr->pstate_sclk_peak = gfx_dpm_table->dpm_levels[gfx_dpm_table->count].value;
+	hwmgr->pstate_mclk_peak = mem_dpm_table->dpm_levels[mem_dpm_table->count].value;
+}
+
 static int vega12_enable_dpm_tasks(struct pp_hwmgr *hwmgr)
 {
 	int tmp_result, result = 0;
@@ -1077,6 +1096,9 @@ static int vega12_enable_dpm_tasks(struct pp_hwmgr *hwmgr)
 	PP_ASSERT_WITH_CODE(!result,
 			"Failed to setup default DPM tables!",
 			return result);
+
+	vega12_populate_umdpstate_clocks(hwmgr);
+
 	return result;
 }
 
drivers/gpu/drm/amd/pm/powerplay/hwmgr/vega20_hwmgr.c +7 -13
index b30684c84e20..2a5abac81b4a 100644
@@ -1555,26 +1555,23 @@ static int vega20_set_mclk_od(
 	return 0;
 }
 
-static int vega20_populate_umdpstate_clocks(
-		struct pp_hwmgr *hwmgr)
+static void vega20_populate_umdpstate_clocks(struct pp_hwmgr *hwmgr)
 {
 	struct vega20_hwmgr *data = (struct vega20_hwmgr *)(hwmgr->backend);
 	struct vega20_single_dpm_table *gfx_table = &(data->dpm_table.gfx_table);
 	struct vega20_single_dpm_table *mem_table = &(data->dpm_table.mem_table);
 
-	hwmgr->pstate_sclk = gfx_table->dpm_levels[0].value;
-	hwmgr->pstate_mclk = mem_table->dpm_levels[0].value;
-
 	if (gfx_table->count > VEGA20_UMD_PSTATE_GFXCLK_LEVEL &&
 	    mem_table->count > VEGA20_UMD_PSTATE_MCLK_LEVEL) {
 		hwmgr->pstate_sclk = gfx_table->dpm_levels[VEGA20_UMD_PSTATE_GFXCLK_LEVEL].value;
 		hwmgr->pstate_mclk = mem_table->dpm_levels[VEGA20_UMD_PSTATE_MCLK_LEVEL].value;
+	} else {
+		hwmgr->pstate_sclk = gfx_table->dpm_levels[0].value;
+		hwmgr->pstate_mclk = mem_table->dpm_levels[0].value;
 	}
 
-	hwmgr->pstate_sclk = hwmgr->pstate_sclk * 100;
-	hwmgr->pstate_mclk = hwmgr->pstate_mclk * 100;
-
-	return 0;
+	hwmgr->pstate_sclk_peak = gfx_table->dpm_levels[gfx_table->count - 1].value;
+	hwmgr->pstate_mclk_peak = mem_table->dpm_levels[mem_table->count - 1].value;
 }
 
 static int vega20_get_max_sustainable_clock(struct pp_hwmgr *hwmgr,
@@ -1753,10 +1750,7 @@ static int vega20_enable_dpm_tasks(struct pp_hwmgr *hwmgr)
 			"[EnableDPMTasks] Failed to initialize odn settings!",
 			return result);
 
-	result = vega20_populate_umdpstate_clocks(hwmgr);
-	PP_ASSERT_WITH_CODE(!result,
-			"[EnableDPMTasks] Failed to populate umdpstate clocks!",
-			return result);
+	vega20_populate_umdpstate_clocks(hwmgr);
 
 	result = smum_send_msg_to_smc_with_parameter(hwmgr, PPSMC_MSG_GetPptLimit,
 			POWER_SOURCE_AC << 16, &hwmgr->default_power_limit);
drivers/gpu/drm/amd/pm/powerplay/inc/hwmgr.h +2 -0
index 27f8d0e0e6a8..5ce433e2c16a 100644
@@ -809,6 +809,8 @@ struct pp_hwmgr {
 	uint32_t workload_prority[Workload_Policy_Max];
 	uint32_t workload_setting[Workload_Policy_Max];
 	bool gfxoff_state_changed_by_workload;
+	uint32_t pstate_sclk_peak;
+	uint32_t pstate_mclk_peak;
 };
 
 int hwmgr_early_init(struct pp_hwmgr *hwmgr);
Signature · 1 line
2.34.1
author
#6 lore raw

PATCH 7/10 drm/amdgpu: expose peak profiling mode shader/memory clocks

In reply to Evan Quan

Expose those informations to UMD who need them as for standard profiling mode.

Change-Id I7f60c17cb53846a73f43f0f847a32317f84b2861
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c | 18 ++++++++++++++++++
 include/uapi/drm/amdgpu_drm.h           |  4 ++++
 2 files changed, 22 insertions(+)
drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c +18 -0
index 150b431fc426..ccb80f0a7fe2 100644
@@ -1052,6 +1052,24 @@ int amdgpu_info_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
 			}
 			ui32 /= 100;
 			break;
+		case AMDGPU_INFO_SENSOR_PEAK_PSTATE_GFX_SCLK:
+			/* get peak pstate sclk in Mhz */
+			if (amdgpu_dpm_read_sensor(adev,
+						   AMDGPU_PP_SENSOR_PEAK_PSTATE_SCLK,
+						   (void *)&ui32, &ui32_size)) {
+				return -EINVAL;
+			}
+			ui32 /= 100;
+			break;
+		case AMDGPU_INFO_SENSOR_PEAK_PSTATE_GFX_MCLK:
+			/* get peak pstate mclk in Mhz */
+			if (amdgpu_dpm_read_sensor(adev,
+						   AMDGPU_PP_SENSOR_PEAK_PSTATE_MCLK,
+						   (void *)&ui32, &ui32_size)) {
+				return -EINVAL;
+			}
+			ui32 /= 100;
+			break;
 		default:
 			DRM_DEBUG_KMS("Invalid request %d\n",
 				      info->sensor_info.type);
include/uapi/drm/amdgpu_drm.h +4 -0
index 9bae36b82806..038153f2f4cd 100644
@@ -893,6 +893,10 @@ struct drm_amdgpu_cs_chunk_data {
 	#define AMDGPU_INFO_SENSOR_STABLE_PSTATE_GFX_SCLK		0x8
 	/* Subquery id: Query GPU stable pstate memory clock */
 	#define AMDGPU_INFO_SENSOR_STABLE_PSTATE_GFX_MCLK		0x9
+	/* Subquery id: Query GPU peak pstate shader clock */
+	#define AMDGPU_INFO_SENSOR_PEAK_PSTATE_GFX_SCLK			0xa
+	/* Subquery id: Query GPU peak pstate memory clock */
+	#define AMDGPU_INFO_SENSOR_PEAK_PSTATE_GFX_MCLK			0xb
 /* Number of VRAM page faults on CPU access. */
 #define AMDGPU_INFO_NUM_VRAM_CPU_PAGE_FAULTS	0x1E
 #define AMDGPU_INFO_VRAM_LOST_COUNTER		0x1F
Signature · 1 line
2.34.1
author
#7 lore raw

PATCH 8/10 drm/amdgpu: expose the minimum shader/memory clock frequency

In reply to Evan Quan

Otherwise, some UMD tools will treate them as 0 at default while actually they are not.

Change-Id I524a512eb538dba626edd6d2c73a370dd6a1b4a5
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c | 10 ++++++++--
 include/uapi/drm/amdgpu_drm.h           |  2 ++
 2 files changed, 10 insertions(+), 2 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c +8 -2
index ccb80f0a7fe2..dd1ebaa40e2e 100644
@@ -808,9 +808,15 @@ int amdgpu_info_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
 		if (adev->pm.dpm_enabled) {
 			dev_info->max_engine_clock = amdgpu_dpm_get_sclk(adev, false) * 10;
 			dev_info->max_memory_clock = amdgpu_dpm_get_mclk(adev, false) * 10;
+			dev_info->min_engine_clock = amdgpu_dpm_get_sclk(adev, true) * 10;
+			dev_info->min_memory_clock = amdgpu_dpm_get_mclk(adev, true) * 10;
 		} else {
-			dev_info->max_engine_clock = adev->clock.default_sclk * 10;
-			dev_info->max_memory_clock = adev->clock.default_mclk * 10;
+			dev_info->max_engine_clock =
+				dev_info->min_engine_clock =
+					adev->clock.default_sclk * 10;
+			dev_info->max_memory_clock =
+				dev_info->min_memory_clock =
+					adev->clock.default_mclk * 10;
 		}
 		dev_info->enabled_rb_pipes_mask = adev->gfx.config.backend_enable_mask;
 		dev_info->num_rb_pipes = adev->gfx.config.max_backends_per_se *
include/uapi/drm/amdgpu_drm.h +2 -0
index 038153f2f4cd..ab7895347979 100644
@@ -1189,6 +1189,8 @@ struct drm_amdgpu_info_device {
 	__u32 pa_sc_tile_steering_override;
 	/* disabled TCCs */
 	__u64 tcc_disabled_mask;
+	__u64 min_engine_clock;
+	__u64 min_memory_clock;
 };
 
 struct drm_amdgpu_info_hw_ip {
Signature · 1 line
2.34.1
author
#8 lore raw

PATCH 9/10 drm/amdgpu: bump minor version number for DEV_INFO and SENSOR IOCTLs update

In reply to Evan Quan

Update AMDGPU_INFO_DEV_INFO IOCTL for minimum engine and memory clock. And update AMDGPU_INFO_SENSOR IOCTL for PEAK_PSTATE engine and memory clock. User applications can better utilize these IOCTLs to get needed informations.

Increase the minor version number to indicate that the new flags are available.

Change-Id I2fc85ed8046391555282c0629d88c875488c2b97
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c +4 -2
index 0826b0bca328..fb1720d171de 100644
@@ -102,10 +102,12 @@
  * - 3.46.0 - To enable hot plug amdgpu tests in libdrm
  * - 3.47.0 - Add AMDGPU_GEM_CREATE_DISCARDABLE and AMDGPU_VM_NOALLOC flags
  * - 3.48.0 - Add IP discovery version info to HW INFO
- *   3.49.0 - Add gang submit into CS IOCTL
+ * - 3.49.0 - Add gang submit into CS IOCTL
+ * - 3.50.0 - Update AMDGPU_INFO_DEV_INFO IOCTL for minimum engine and memory clock
+ *            Update AMDGPU_INFO_SENSOR IOCTL for PEAK_PSTATE engine and memory clock
  */
 #define KMS_DRIVER_MAJOR	3
-#define KMS_DRIVER_MINOR	49
+#define KMS_DRIVER_MINOR	50
 #define KMS_DRIVER_PATCHLEVEL	0
 
 int amdgpu_vram_limit;
Signature · 1 line
2.34.1
Reviewed-by on v1 9/10
#9 lore raw

Re: [PATCH 09/10] drm/amdgpu: bump minor version number for DEV_INFO and SENSOR IOCTLs update

Quoted 7 lines · Evan Quan

Update AMDGPU_INFO_DEV_INFO IOCTL for minimum engine and memory clock. And update AMDGPU_INFO_SENSOR IOCTL for PEAK_PSTATE engine and memory clock. User applications can better utilize these IOCTLs to get needed informations.

Increase the minor version number to indicate that the new flags are available.

We need updated user space patches which use the IOCTL updates before they can land. Once we have those, the patches are:

Quoted 29 lines
Change-Id I2fc85ed8046391555282c0629d88c875488c2b97
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c +4 -3
index 0826b0bca328..fb1720d171de 100644
@@ -102,10 +102,12 @@
  * - 3.46.0 - To enable hot plug amdgpu tests in libdrm
  * - 3.47.0 - Add AMDGPU_GEM_CREATE_DISCARDABLE and AMDGPU_VM_NOALLOC flags
  * - 3.48.0 - Add IP discovery version info to HW INFO
- *   3.49.0 - Add gang submit into CS IOCTL
+ * - 3.49.0 - Add gang submit into CS IOCTL
+ * - 3.50.0 - Update AMDGPU_INFO_DEV_INFO IOCTL for minimum engine and memory clock
+ *            Update AMDGPU_INFO_SENSOR IOCTL for PEAK_PSTATE engine and memory clock
  */
 #define KMS_DRIVER_MAJOR       3
-#define KMS_DRIVER_MINOR       49
+#define KMS_DRIVER_MINOR       50
 #define KMS_DRIVER_PATCHLEVEL  0
 
 int amdgpu_vram_limit;
--

2.34.1

author
#10 lore raw

PATCH 10/10 drm/amd/pm: update SMU13.0.0 reported maximum shader clock

In reply to Evan Quan

Update the reported maximum shader clock to the value which can be guarded to be achieved on all cards. This is to align with Window setting.

Change-Id I83da5ec931664af36ee6db3cb13ef655db54220e
---
 .../drm/amd/pm/swsmu/smu13/smu_v13_0_0_ppt.c  | 70 ++++++++++++++++++-
 1 file changed, 69 insertions(+), 1 deletion(-)
drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_0_ppt.c +69 -1
index 4adbab62e9dd..5a0593d336be 100644
@@ -541,6 +541,23 @@ static int smu_v13_0_0_set_default_dpm_table(struct smu_context *smu)
 						     dpm_table);
 		if (ret)
 			return ret;
+
+		/*
+		 * Update the reported maximum shader clock to the value
+		 * which can be guarded to be achieved on all cards. This
+		 * is aligned with Window setting. And considering that value
+		 * might be not the peak frequency the card can achieve, it
+		 * is normal some real-time clock frequency can overtake this
+		 * labelled maximum clock frequency(for example in pp_dpm_sclk
+		 * sysfs output).
+		 */
+		if (skutable->DriverReportedClocks.GameClockAc &&
+		    (dpm_table->dpm_levels[dpm_table->count - 1].value >
+		    skutable->DriverReportedClocks.GameClockAc)) {
+			dpm_table->dpm_levels[dpm_table->count - 1].value =
+				skutable->DriverReportedClocks.GameClockAc;
+			dpm_table->max = skutable->DriverReportedClocks.GameClockAc;
+		}
 	} else {
 		dpm_table->count = 1;
 		dpm_table->dpm_levels[0].value = smu->smu_table.boot_values.gfxclk / 100;
@@ -803,6 +820,57 @@ static int smu_v13_0_0_get_smu_metrics_data(struct smu_context *smu,
 	return ret;
 }
 
+static int smu_v13_0_0_get_dpm_ultimate_freq(struct smu_context *smu,
+					     enum smu_clk_type clk_type,
+					     uint32_t *min,
+					     uint32_t *max)
+{
+	struct smu_13_0_dpm_context *dpm_context =
+		smu->smu_dpm.dpm_context;
+	struct smu_13_0_dpm_table *dpm_table;
+
+	switch (clk_type) {
+	case SMU_MCLK:
+	case SMU_UCLK:
+		/* uclk dpm table */
+		dpm_table = &dpm_context->dpm_tables.uclk_table;
+		break;
+	case SMU_GFXCLK:
+	case SMU_SCLK:
+		/* gfxclk dpm table */
+		dpm_table = &dpm_context->dpm_tables.gfx_table;
+		break;
+	case SMU_SOCCLK:
+		/* socclk dpm table */
+		dpm_table = &dpm_context->dpm_tables.soc_table;
+		break;
+	case SMU_FCLK:
+		/* fclk dpm table */
+		dpm_table = &dpm_context->dpm_tables.fclk_table;
+		break;
+	case SMU_VCLK:
+	case SMU_VCLK1:
+		/* vclk dpm table */
+		dpm_table = &dpm_context->dpm_tables.vclk_table;
+		break;
+	case SMU_DCLK:
+	case SMU_DCLK1:
+		/* dclk dpm table */
+		dpm_table = &dpm_context->dpm_tables.dclk_table;
+		break;
+	default:
+		dev_err(smu->adev->dev, "Unsupported clock type!\n");
+		return -EINVAL;
+	}
+
+	if (min)
+		*min = dpm_table->min;
+	if (max)
+		*max = dpm_table->max;
+
+	return 0;
+}
+
 static int smu_v13_0_0_read_sensor(struct smu_context *smu,
 				   enum amd_pp_sensors sensor,
 				   void *data,
@@ -1908,7 +1976,7 @@ static const struct pptable_funcs smu_v13_0_0_ppt_funcs = {
 	.get_enabled_mask = smu_cmn_get_enabled_mask,
 	.dpm_set_vcn_enable = smu_v13_0_set_vcn_enable,
 	.dpm_set_jpeg_enable = smu_v13_0_set_jpeg_enable,
-	.get_dpm_ultimate_freq = smu_v13_0_get_dpm_ultimate_freq,
+	.get_dpm_ultimate_freq = smu_v13_0_0_get_dpm_ultimate_freq,
 	.get_vbios_bootup_values = smu_v13_0_get_vbios_bootup_values,
 	.read_sensor = smu_v13_0_0_read_sensor,
 	.feature_is_enabled = smu_cmn_feature_is_enabled,
Signature · 1 line
2.34.1