1 files changed, 332 insertions, 37 deletions

diff --git a/drivers/gpu/drm/amd/amdgpu/ci_dpm.c b/drivers/gpu/drm/amd/amdgpu/ci_dpm.c
index bda9e3de191e..11ccda83d767 100644
--- a/drivers/gpu/drm/amd/amdgpu/ci_dpm.c
+++ b/drivers/gpu/drm/amd/amdgpu/ci_dpm.c
@@ -889,7 +889,16 @@ static void ci_dpm_powergate_uvd(struct amdgpu_device *adev, bool gate)
 
 	pi->uvd_power_gated = gate;
 
-	ci_update_uvd_dpm(adev, gate);
+	if (gate) {
+		/* stop the UVD block */
+		amdgpu_set_powergating_state(adev, AMD_IP_BLOCK_TYPE_UVD,
+							AMD_PG_STATE_GATE);
+		ci_update_uvd_dpm(adev, gate);
+	} else {
+		amdgpu_set_powergating_state(adev, AMD_IP_BLOCK_TYPE_UVD,
+							AMD_PG_STATE_UNGATE);
+		ci_update_uvd_dpm(adev, gate);
+	}
 }
 
 static bool ci_dpm_vblank_too_short(struct amdgpu_device *adev)
@@ -2201,7 +2210,6 @@ static void ci_clear_vc(struct amdgpu_device *adev)
 
 static int ci_upload_firmware(struct amdgpu_device *adev)
 {
-	struct ci_power_info *pi = ci_get_pi(adev);
 	int i, ret;
 
 	if (amdgpu_ci_is_smc_running(adev)) {
@@ -2218,7 +2226,7 @@ static int ci_upload_firmware(struct amdgpu_device *adev)
 	amdgpu_ci_stop_smc_clock(adev);
 	amdgpu_ci_reset_smc(adev);
 
-	ret = amdgpu_ci_load_smc_ucode(adev, pi->sram_end);
+	ret = amdgpu_ci_load_smc_ucode(adev, SMC_RAM_END);
 
 	return ret;
 
@@ -3673,6 +3681,40 @@ static int ci_find_boot_level(struct ci_single_dpm_table *table,
 	return ret;
 }
 
+static void ci_save_default_power_profile(struct amdgpu_device *adev)
+{
+	struct ci_power_info *pi = ci_get_pi(adev);
+	struct SMU7_Discrete_GraphicsLevel *levels =
+				pi->smc_state_table.GraphicsLevel;
+	uint32_t min_level = 0;
+
+	pi->default_gfx_power_profile.activity_threshold =
+			be16_to_cpu(levels[0].ActivityLevel);
+	pi->default_gfx_power_profile.up_hyst = levels[0].UpH;
+	pi->default_gfx_power_profile.down_hyst = levels[0].DownH;
+	pi->default_gfx_power_profile.type = AMD_PP_GFX_PROFILE;
+
+	pi->default_compute_power_profile = pi->default_gfx_power_profile;
+	pi->default_compute_power_profile.type = AMD_PP_COMPUTE_PROFILE;
+
+	/* Optimize compute power profile: Use only highest
+	 * 2 power levels (if more than 2 are available), Hysteresis:
+	 * 0ms up, 5ms down
+	 */
+	if (pi->smc_state_table.GraphicsDpmLevelCount > 2)
+		min_level = pi->smc_state_table.GraphicsDpmLevelCount - 2;
+	else if (pi->smc_state_table.GraphicsDpmLevelCount == 2)
+		min_level = 1;
+	pi->default_compute_power_profile.min_sclk =
+			be32_to_cpu(levels[min_level].SclkFrequency);
+
+	pi->default_compute_power_profile.up_hyst = 0;
+	pi->default_compute_power_profile.down_hyst = 5;
+
+	pi->gfx_power_profile = pi->default_gfx_power_profile;
+	pi->compute_power_profile = pi->default_compute_power_profile;
+}
+
 static int ci_init_smc_table(struct amdgpu_device *adev)
 {
 	struct ci_power_info *pi = ci_get_pi(adev);
@@ -3818,6 +3860,8 @@ static int ci_init_smc_table(struct amdgpu_device *adev)
 	if (ret)
 		return ret;
 
+	ci_save_default_power_profile(adev);
+
 	return 0;
 }
 
@@ -4248,12 +4292,6 @@ static int ci_update_vce_dpm(struct amdgpu_device *adev,
 
 	if (amdgpu_current_state->evclk != amdgpu_new_state->evclk) {
 		if (amdgpu_new_state->evclk) {
-			/* turn the clocks on when encoding */
-			ret = amdgpu_set_clockgating_state(adev, AMD_IP_BLOCK_TYPE_VCE,
-							    AMD_CG_STATE_UNGATE);
-			if (ret)
-				return ret;
-
 			pi->smc_state_table.VceBootLevel = ci_get_vce_boot_level(adev);
 			tmp = RREG32_SMC(ixDPM_TABLE_475);
 			tmp &= ~DPM_TABLE_475__VceBootLevel_MASK;
@@ -4265,9 +4303,6 @@ static int ci_update_vce_dpm(struct amdgpu_device *adev,
 			ret = ci_enable_vce_dpm(adev, false);
 			if (ret)
 				return ret;
-			/* turn the clocks off when not encoding */
-			ret = amdgpu_set_clockgating_state(adev, AMD_IP_BLOCK_TYPE_VCE,
-							    AMD_CG_STATE_GATE);
 		}
 	}
 	return ret;
@@ -4336,13 +4371,13 @@ static u32 ci_get_lowest_enabled_level(struct amdgpu_device *adev,
 
 
 static int ci_dpm_force_performance_level(struct amdgpu_device *adev,
-					  enum amdgpu_dpm_forced_level level)
+					  enum amd_dpm_forced_level level)
 {
 	struct ci_power_info *pi = ci_get_pi(adev);
 	u32 tmp, levels, i;
 	int ret;
 
-	if (level == AMDGPU_DPM_FORCED_LEVEL_HIGH) {
+	if (level == AMD_DPM_FORCED_LEVEL_HIGH) {
 		if ((!pi->pcie_dpm_key_disabled) &&
 		    pi->dpm_level_enable_mask.pcie_dpm_enable_mask) {
 			levels = 0;
@@ -4403,7 +4438,7 @@ static int ci_dpm_force_performance_level(struct amdgpu_device *adev,
 				}
 			}
 		}
-	} else if (level == AMDGPU_DPM_FORCED_LEVEL_LOW) {
+	} else if (level == AMD_DPM_FORCED_LEVEL_LOW) {
 		if ((!pi->sclk_dpm_key_disabled) &&
 		    pi->dpm_level_enable_mask.sclk_dpm_enable_mask) {
 			levels = ci_get_lowest_enabled_level(adev,
@@ -4452,7 +4487,7 @@ static int ci_dpm_force_performance_level(struct amdgpu_device *adev,
 				udelay(1);
 			}
 		}
-	} else if (level == AMDGPU_DPM_FORCED_LEVEL_AUTO) {
+	} else if (level == AMD_DPM_FORCED_LEVEL_AUTO) {
 		if (!pi->pcie_dpm_key_disabled) {
 			PPSMC_Result smc_result;
 
@@ -5805,9 +5840,7 @@ static int ci_dpm_init_microcode(struct amdgpu_device *adev)
 
 out:
 	if (err) {
-		printk(KERN_ERR
-		       "cik_smc: Failed to load firmware \"%s\"\n",
-		       fw_name);
+		pr_err("cik_smc: Failed to load firmware \"%s\"\n", fw_name);
 		release_firmware(adev->pm.fw);
 		adev->pm.fw = NULL;
 	}
@@ -6251,31 +6284,33 @@ static int ci_dpm_sw_init(void *handle)
 	int ret;
 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 
-	ret = amdgpu_irq_add_id(adev, 230, &adev->pm.dpm.thermal.irq);
+	ret = amdgpu_irq_add_id(adev, AMDGPU_IH_CLIENTID_LEGACY, 230,
+				&adev->pm.dpm.thermal.irq);
 	if (ret)
 		return ret;
 
-	ret = amdgpu_irq_add_id(adev, 231, &adev->pm.dpm.thermal.irq);
+	ret = amdgpu_irq_add_id(adev, AMDGPU_IH_CLIENTID_LEGACY, 231,
+				&adev->pm.dpm.thermal.irq);
 	if (ret)
 		return ret;
 
 	/* default to balanced state */
 	adev->pm.dpm.state = POWER_STATE_TYPE_BALANCED;
 	adev->pm.dpm.user_state = POWER_STATE_TYPE_BALANCED;
-	adev->pm.dpm.forced_level = AMDGPU_DPM_FORCED_LEVEL_AUTO;
+	adev->pm.dpm.forced_level = AMD_DPM_FORCED_LEVEL_AUTO;
 	adev->pm.default_sclk = adev->clock.default_sclk;
 	adev->pm.default_mclk = adev->clock.default_mclk;
 	adev->pm.current_sclk = adev->clock.default_sclk;
 	adev->pm.current_mclk = adev->clock.default_mclk;
 	adev->pm.int_thermal_type = THERMAL_TYPE_NONE;
 
-	if (amdgpu_dpm == 0)
-		return 0;
-
 	ret = ci_dpm_init_microcode(adev);
 	if (ret)
 		return ret;
 
+	if (amdgpu_dpm == 0)
+		return 0;
+
 	INIT_WORK(&adev->pm.dpm.thermal.work, amdgpu_dpm_thermal_work_handler);
 	mutex_lock(&adev->pm.mutex);
 	ret = ci_dpm_init(adev);
@@ -6319,8 +6354,15 @@ static int ci_dpm_hw_init(void *handle)
 
 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 
-	if (!amdgpu_dpm)
+	if (!amdgpu_dpm) {
+		ret = ci_upload_firmware(adev);
+		if (ret) {
+			DRM_ERROR("ci_upload_firmware failed\n");
+			return ret;
+		}
+		ci_dpm_start_smc(adev);
 		return 0;
+	}
 
 	mutex_lock(&adev->pm.mutex);
 	ci_dpm_setup_asic(adev);
@@ -6342,6 +6384,8 @@ static int ci_dpm_hw_fini(void *handle)
 		mutex_lock(&adev->pm.mutex);
 		ci_dpm_disable(adev);
 		mutex_unlock(&adev->pm.mutex);
+	} else {
+		ci_dpm_stop_smc(adev);
 	}
 
 	return 0;
@@ -6571,8 +6615,9 @@ static int ci_dpm_force_clock_level(struct amdgpu_device *adev,
 {
 	struct ci_power_info *pi = ci_get_pi(adev);
 
-	if (adev->pm.dpm.forced_level
-			!= AMDGPU_DPM_FORCED_LEVEL_MANUAL)
+	if (adev->pm.dpm.forced_level & (AMD_DPM_FORCED_LEVEL_AUTO |
+				AMD_DPM_FORCED_LEVEL_LOW |
+				AMD_DPM_FORCED_LEVEL_HIGH))
 		return -EINVAL;
 
 	switch (type) {
@@ -6679,6 +6724,260 @@ static int ci_dpm_set_mclk_od(struct amdgpu_device *adev, uint32_t value)
 	return 0;
 }
 
+static int ci_dpm_get_power_profile_state(struct amdgpu_device *adev,
+		struct amd_pp_profile *query)
+{
+	struct ci_power_info *pi = ci_get_pi(adev);
+
+	if (!pi || !query)
+		return -EINVAL;
+
+	if (query->type == AMD_PP_GFX_PROFILE)
+		memcpy(query, &pi->gfx_power_profile,
+				sizeof(struct amd_pp_profile));
+	else if (query->type == AMD_PP_COMPUTE_PROFILE)
+		memcpy(query, &pi->compute_power_profile,
+				sizeof(struct amd_pp_profile));
+	else
+		return -EINVAL;
+
+	return 0;
+}
+
+static int ci_populate_requested_graphic_levels(struct amdgpu_device *adev,
+		struct amd_pp_profile *request)
+{
+	struct ci_power_info *pi = ci_get_pi(adev);
+	struct ci_dpm_table *dpm_table = &(pi->dpm_table);
+	struct SMU7_Discrete_GraphicsLevel *levels =
+			pi->smc_state_table.GraphicsLevel;
+	uint32_t array = pi->dpm_table_start +
+			offsetof(SMU7_Discrete_DpmTable, GraphicsLevel);
+	uint32_t array_size = sizeof(struct SMU7_Discrete_GraphicsLevel) *
+			SMU7_MAX_LEVELS_GRAPHICS;
+	uint32_t i;
+
+	for (i = 0; i < dpm_table->sclk_table.count; i++) {
+		levels[i].ActivityLevel =
+				cpu_to_be16(request->activity_threshold);
+		levels[i].EnabledForActivity = 1;
+		levels[i].UpH = request->up_hyst;
+		levels[i].DownH = request->down_hyst;
+	}
+
+	return amdgpu_ci_copy_bytes_to_smc(adev, array, (uint8_t *)levels,
+				array_size, pi->sram_end);
+}
+
+static void ci_find_min_clock_masks(struct amdgpu_device *adev,
+		uint32_t *sclk_mask, uint32_t *mclk_mask,
+		uint32_t min_sclk, uint32_t min_mclk)
+{
+	struct ci_power_info *pi = ci_get_pi(adev);
+	struct ci_dpm_table *dpm_table = &(pi->dpm_table);
+	uint32_t i;
+
+	for (i = 0; i < dpm_table->sclk_table.count; i++) {
+		if (dpm_table->sclk_table.dpm_levels[i].enabled &&
+			dpm_table->sclk_table.dpm_levels[i].value >= min_sclk)
+			*sclk_mask |= 1 << i;
+	}
+
+	for (i = 0; i < dpm_table->mclk_table.count; i++) {
+		if (dpm_table->mclk_table.dpm_levels[i].enabled &&
+			dpm_table->mclk_table.dpm_levels[i].value >= min_mclk)
+			*mclk_mask |= 1 << i;
+	}
+}
+
+static int ci_set_power_profile_state(struct amdgpu_device *adev,
+		struct amd_pp_profile *request)
+{
+	struct ci_power_info *pi = ci_get_pi(adev);
+	int tmp_result, result = 0;
+	uint32_t sclk_mask = 0, mclk_mask = 0;
+
+	tmp_result = ci_freeze_sclk_mclk_dpm(adev);
+	if (tmp_result) {
+		DRM_ERROR("Failed to freeze SCLK MCLK DPM!");
+		result = tmp_result;
+	}
+
+	tmp_result = ci_populate_requested_graphic_levels(adev,
+			request);
+	if (tmp_result) {
+		DRM_ERROR("Failed to populate requested graphic levels!");
+		result = tmp_result;
+	}
+
+	tmp_result = ci_unfreeze_sclk_mclk_dpm(adev);
+	if (tmp_result) {
+		DRM_ERROR("Failed to unfreeze SCLK MCLK DPM!");
+		result = tmp_result;
+	}
+
+	ci_find_min_clock_masks(adev, &sclk_mask, &mclk_mask,
+			request->min_sclk, request->min_mclk);
+
+	if (sclk_mask) {
+		if (!pi->sclk_dpm_key_disabled)
+			amdgpu_ci_send_msg_to_smc_with_parameter(
+				adev,
+				PPSMC_MSG_SCLKDPM_SetEnabledMask,
+				pi->dpm_level_enable_mask.
+				sclk_dpm_enable_mask &
+				sclk_mask);
+	}
+
+	if (mclk_mask) {
+		if (!pi->mclk_dpm_key_disabled)
+			amdgpu_ci_send_msg_to_smc_with_parameter(
+				adev,
+				PPSMC_MSG_MCLKDPM_SetEnabledMask,
+				pi->dpm_level_enable_mask.
+				mclk_dpm_enable_mask &
+				mclk_mask);
+	}
+
+
+	return result;
+}
+
+static int ci_dpm_set_power_profile_state(struct amdgpu_device *adev,
+		struct amd_pp_profile *request)
+{
+	struct ci_power_info *pi = ci_get_pi(adev);
+	int ret = -1;
+
+	if (!pi || !request)
+		return -EINVAL;
+
+	if (adev->pm.dpm.forced_level !=
+			AMD_DPM_FORCED_LEVEL_AUTO)
+		return -EINVAL;
+
+	if (request->min_sclk ||
+		request->min_mclk ||
+		request->activity_threshold ||
+		request->up_hyst ||
+		request->down_hyst) {
+		if (request->type == AMD_PP_GFX_PROFILE)
+			memcpy(&pi->gfx_power_profile, request,
+					sizeof(struct amd_pp_profile));
+		else if (request->type == AMD_PP_COMPUTE_PROFILE)
+			memcpy(&pi->compute_power_profile, request,
+					sizeof(struct amd_pp_profile));
+		else
+			return -EINVAL;
+
+		if (request->type == pi->current_power_profile)
+			ret = ci_set_power_profile_state(
+					adev,
+					request);
+	} else {
+		/* set power profile if it exists */
+		switch (request->type) {
+		case AMD_PP_GFX_PROFILE:
+			ret = ci_set_power_profile_state(
+				adev,
+				&pi->gfx_power_profile);
+			break;
+		case AMD_PP_COMPUTE_PROFILE:
+			ret = ci_set_power_profile_state(
+				adev,
+				&pi->compute_power_profile);
+			break;
+		default:
+			return -EINVAL;
+		}
+	}
+
+	if (!ret)
+		pi->current_power_profile = request->type;
+
+	return 0;
+}
+
+static int ci_dpm_reset_power_profile_state(struct amdgpu_device *adev,
+		struct amd_pp_profile *request)
+{
+	struct ci_power_info *pi = ci_get_pi(adev);
+
+	if (!pi || !request)
+		return -EINVAL;
+
+	if (request->type == AMD_PP_GFX_PROFILE) {
+		pi->gfx_power_profile = pi->default_gfx_power_profile;
+		return ci_dpm_set_power_profile_state(adev,
+					  &pi->gfx_power_profile);
+	} else if (request->type == AMD_PP_COMPUTE_PROFILE) {
+		pi->compute_power_profile =
+			pi->default_compute_power_profile;
+		return ci_dpm_set_power_profile_state(adev,
+					  &pi->compute_power_profile);
+	} else
+		return -EINVAL;
+}
+
+static int ci_dpm_switch_power_profile(struct amdgpu_device *adev,
+		enum amd_pp_profile_type type)
+{
+	struct ci_power_info *pi = ci_get_pi(adev);
+	struct amd_pp_profile request = {0};
+
+	if (!pi)
+		return -EINVAL;
+
+	if (pi->current_power_profile != type) {
+		request.type = type;
+		return ci_dpm_set_power_profile_state(adev, &request);
+	}
+
+	return 0;
+}
+
+static int ci_dpm_read_sensor(struct amdgpu_device *adev, int idx,
+			      void *value, int *size)
+{
+	u32 activity_percent = 50;
+	int ret;
+
+	/* size must be at least 4 bytes for all sensors */
+	if (*size < 4)
+		return -EINVAL;
+
+	switch (idx) {
+	case AMDGPU_PP_SENSOR_GFX_SCLK:
+		*((uint32_t *)value) = ci_get_average_sclk_freq(adev);
+		*size = 4;
+		return 0;
+	case AMDGPU_PP_SENSOR_GFX_MCLK:
+		*((uint32_t *)value) = ci_get_average_mclk_freq(adev);
+		*size = 4;
+		return 0;
+	case AMDGPU_PP_SENSOR_GPU_TEMP:
+		*((uint32_t *)value) = ci_dpm_get_temp(adev);
+		*size = 4;
+		return 0;
+	case AMDGPU_PP_SENSOR_GPU_LOAD:
+		ret = ci_read_smc_soft_register(adev,
+						offsetof(SMU7_SoftRegisters,
+							 AverageGraphicsA),
+						&activity_percent);
+		if (ret == 0) {
+			activity_percent += 0x80;
+			activity_percent >>= 8;
+			activity_percent =
+				activity_percent > 100 ? 100 : activity_percent;
+		}
+		*((uint32_t *)value) = activity_percent;
+		*size = 4;
+		return 0;
+	default:
+		return -EINVAL;
+	}
+}
+
 const struct amd_ip_funcs ci_dpm_ip_funcs = {
 	.name = "ci_dpm",
 	.early_init = ci_dpm_early_init,
@@ -6721,6 +7020,11 @@ static const struct amdgpu_dpm_funcs ci_dpm_funcs = {
 	.set_mclk_od = ci_dpm_set_mclk_od,
 	.check_state_equal = ci_check_state_equal,
 	.get_vce_clock_state = amdgpu_get_vce_clock_state,
+	.get_power_profile_state = ci_dpm_get_power_profile_state,
+	.set_power_profile_state = ci_dpm_set_power_profile_state,
+	.reset_power_profile_state = ci_dpm_reset_power_profile_state,
+	.switch_power_profile = ci_dpm_switch_power_profile,
+	.read_sensor = ci_dpm_read_sensor,
 };
 
 static void ci_dpm_set_dpm_funcs(struct amdgpu_device *adev)
@@ -6739,12 +7043,3 @@ static void ci_dpm_set_irq_funcs(struct amdgpu_device *adev)
 	adev->pm.dpm.thermal.irq.num_types = AMDGPU_THERMAL_IRQ_LAST;
 	adev->pm.dpm.thermal.irq.funcs = &ci_dpm_irq_funcs;
 }
-
-const struct amdgpu_ip_block_version ci_dpm_ip_block =
-{
-	.type = AMD_IP_BLOCK_TYPE_SMC,
-	.major = 7,
-	.minor = 0,
-	.rev = 0,
-	.funcs = &ci_dpm_ip_funcs,
-};