drm/amd/powerplay: dynamically disable ds and ulv for compute

This is to improve the performance in the compute mode
for vega10. For example, the original performance for a rocm
bandwidth test: 2G internal GPU copy, is about 99GB/s.
With the idle power features disabled dynamically, the porformance
is promoted to about 215GB/s.

Signed-off-by: Kenneth Feng <kenneth.feng@amd.com>
Reviewed-by: Evan Quan <evan.quan@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

3 files changed, 65 insertions, 0 deletions

diff --git a/drivers/gpu/drm/amd/powerplay/amd_powerplay.c b/drivers/gpu/drm/amd/powerplay/amd_powerplay.c
index 031447675203..7932eb163a00 100644
--- a/drivers/gpu/drm/amd/powerplay/amd_powerplay.c
+++ b/drivers/gpu/drm/amd/powerplay/amd_powerplay.c
@@ -969,6 +969,14 @@ static int pp_dpm_switch_power_profile(void *handle,
 		workload = hwmgr->workload_setting[index];
 	}
 
+	if (type == PP_SMC_POWER_PROFILE_COMPUTE &&
+		hwmgr->hwmgr_func->disable_power_features_for_compute_performance) {
+			if (hwmgr->hwmgr_func->disable_power_features_for_compute_performance(hwmgr, en)) {
+				mutex_unlock(&hwmgr->smu_lock);
+				return -EINVAL;
+			}
+	}
+
 	if (hwmgr->dpm_level != AMD_DPM_FORCED_LEVEL_MANUAL)
 		hwmgr->hwmgr_func->set_power_profile_mode(hwmgr, &workload, 0);
 	mutex_unlock(&hwmgr->smu_lock);
diff --git a/drivers/gpu/drm/amd/powerplay/hwmgr/vega10_hwmgr.c b/drivers/gpu/drm/amd/powerplay/hwmgr/vega10_hwmgr.c
index 4ea63a2e17da..d3229c274a36 100644
--- a/drivers/gpu/drm/amd/powerplay/hwmgr/vega10_hwmgr.c
+++ b/drivers/gpu/drm/amd/powerplay/hwmgr/vega10_hwmgr.c
@@ -5263,6 +5263,59 @@ static int vega10_get_performance_level(struct pp_hwmgr *hwmgr, const struct pp_
 	return 0;
 }
 
+static int vega10_disable_power_features_for_compute_performance(struct pp_hwmgr *hwmgr, bool disable)
+{
+	struct vega10_hwmgr *data = hwmgr->backend;
+	uint32_t feature_mask = 0;
+
+	if (disable) {
+		feature_mask |= data->smu_features[GNLD_ULV].enabled ?
+			data->smu_features[GNLD_ULV].smu_feature_bitmap : 0;
+		feature_mask |= data->smu_features[GNLD_DS_GFXCLK].enabled ?
+			data->smu_features[GNLD_DS_GFXCLK].smu_feature_bitmap : 0;
+		feature_mask |= data->smu_features[GNLD_DS_SOCCLK].enabled ?
+			data->smu_features[GNLD_DS_SOCCLK].smu_feature_bitmap : 0;
+		feature_mask |= data->smu_features[GNLD_DS_LCLK].enabled ?
+			data->smu_features[GNLD_DS_LCLK].smu_feature_bitmap : 0;
+		feature_mask |= data->smu_features[GNLD_DS_DCEFCLK].enabled ?
+			data->smu_features[GNLD_DS_DCEFCLK].smu_feature_bitmap : 0;
+	} else {
+		feature_mask |= (!data->smu_features[GNLD_ULV].enabled) ?
+			data->smu_features[GNLD_ULV].smu_feature_bitmap : 0;
+		feature_mask |= (!data->smu_features[GNLD_DS_GFXCLK].enabled) ?
+			data->smu_features[GNLD_DS_GFXCLK].smu_feature_bitmap : 0;
+		feature_mask |= (!data->smu_features[GNLD_DS_SOCCLK].enabled) ?
+			data->smu_features[GNLD_DS_SOCCLK].smu_feature_bitmap : 0;
+		feature_mask |= (!data->smu_features[GNLD_DS_LCLK].enabled) ?
+			data->smu_features[GNLD_DS_LCLK].smu_feature_bitmap : 0;
+		feature_mask |= (!data->smu_features[GNLD_DS_DCEFCLK].enabled) ?
+			data->smu_features[GNLD_DS_DCEFCLK].smu_feature_bitmap : 0;
+	}
+
+	if (feature_mask)
+		PP_ASSERT_WITH_CODE(!vega10_enable_smc_features(hwmgr,
+				!disable, feature_mask),
+				"enable/disable power features for compute performance Failed!",
+				return -EINVAL);
+
+	if (disable) {
+		data->smu_features[GNLD_ULV].enabled = false;
+		data->smu_features[GNLD_DS_GFXCLK].enabled = false;
+		data->smu_features[GNLD_DS_SOCCLK].enabled = false;
+		data->smu_features[GNLD_DS_LCLK].enabled = false;
+		data->smu_features[GNLD_DS_DCEFCLK].enabled = false;
+	} else {
+		data->smu_features[GNLD_ULV].enabled = true;
+		data->smu_features[GNLD_DS_GFXCLK].enabled = true;
+		data->smu_features[GNLD_DS_SOCCLK].enabled = true;
+		data->smu_features[GNLD_DS_LCLK].enabled = true;
+		data->smu_features[GNLD_DS_DCEFCLK].enabled = true;
+	}
+
+	return 0;
+
+}
+
 static const struct pp_hwmgr_func vega10_hwmgr_funcs = {
 	.backend_init = vega10_hwmgr_backend_init,
 	.backend_fini = vega10_hwmgr_backend_fini,
@@ -5330,6 +5383,8 @@ static const struct pp_hwmgr_func vega10_hwmgr_funcs = {
 	.get_ppfeature_status = vega10_get_ppfeature_status,
 	.set_ppfeature_status = vega10_set_ppfeature_status,
 	.set_mp1_state = vega10_set_mp1_state,
+	.disable_power_features_for_compute_performance =
+			vega10_disable_power_features_for_compute_performance,
 };
 
 int vega10_hwmgr_init(struct pp_hwmgr *hwmgr)
diff --git a/drivers/gpu/drm/amd/powerplay/inc/hwmgr.h b/drivers/gpu/drm/amd/powerplay/inc/hwmgr.h
index 40403bc76f1b..af977675fd33 100644
--- a/drivers/gpu/drm/amd/powerplay/inc/hwmgr.h
+++ b/drivers/gpu/drm/amd/powerplay/inc/hwmgr.h
@@ -357,6 +357,8 @@ struct pp_hwmgr_func {
 	int (*smu_i2c_bus_access)(struct pp_hwmgr *hwmgr, bool aquire);
 	int (*set_df_cstate)(struct pp_hwmgr *hwmgr, enum pp_df_cstate state);
 	int (*set_xgmi_pstate)(struct pp_hwmgr *hwmgr, uint32_t pstate);
+	int (*disable_power_features_for_compute_performance)(struct pp_hwmgr *hwmgr,
+					bool disable);
 };
 
 struct pp_table_func {