Skip to contentSkip to conversation

drm/amd/pm: Enable thermal alert on SMUv15.0.8

PATCH New Alex Deucher 1 participant 4 messages Apr 14

Subsystems (3)
lore
Share & tools
lore

Alex Deucher

author maintainer
#0 lore raw

PATCH 1/4 drm/amd/pm: Enable thermal alert on SMUv15.0.8

From: Lijo Lazar <[email protected]>

Enable thermal alert interrupts on SMUv15.0.8

v2: squash in cleanups

---
 drivers/gpu/drm/amd/pm/swsmu/smu15/smu_v15_0_8_ppt.c | 10 ++++++++++
 1 file changed, 10 insertions(+)
drivers/gpu/drm/amd/pm/swsmu/smu15/smu_v15_0_8_ppt.c +10 -0
index 25cfb051d0c6e..912a1e1e0cd6a 100644
@@ -2191,6 +2191,14 @@ static int smu_v15_0_8_get_ppt_limit(struct smu_context *smu,
 	return -EOPNOTSUPP;
 }
 
+static int smu_v15_0_8_enable_thermal_alert(struct smu_context *smu)
+{
+	if (!smu->irq_source.num_types)
+		return 0;
+
+	return amdgpu_irq_get(smu->adev, &smu->irq_source, 0);
+}
+
 static const struct pptable_funcs smu_v15_0_8_ppt_funcs = {
 	.init_allowed_features = smu_v15_0_8_init_allowed_features,
 	.set_default_dpm_table = smu_v15_0_8_set_default_dpm_table,
@@ -2208,6 +2216,8 @@ static const struct pptable_funcs smu_v15_0_8_ppt_funcs = {
 	.get_enabled_mask = smu_v15_0_8_get_enabled_mask,
 	.feature_is_enabled = smu_cmn_feature_is_enabled,
 	.register_irq_handler = smu_v15_0_8_register_irq_handler,
+	.enable_thermal_alert = smu_v15_0_8_enable_thermal_alert,
+	.disable_thermal_alert = smu_v15_0_disable_thermal_alert,
 	.setup_pptable = smu_v15_0_8_setup_pptable,
 	.get_pp_feature_mask = smu_cmn_get_pp_feature_mask,
 	.wait_for_event = smu_v15_0_wait_for_event,
Signature · 1 line
2.53.0
author maintainer
#1 lore raw

PATCH 2/4 drm/amd/pm: Add partition metrics support

From: Asad Kamal <[email protected]>

Add partition metrics support for smu_v15_0_8

---
 .../drm/amd/pm/swsmu/smu15/smu_v15_0_8_ppt.c  | 78 +++++++++++++++++++
 .../drm/amd/pm/swsmu/smu15/smu_v15_0_8_ppt.h  | 34 ++++++++
 2 files changed, 112 insertions(+)
drivers/gpu/drm/amd/pm/swsmu/smu15/smu_v15_0_8_ppt.c +78 -0
index 912a1e1e0cd6a..288fab5ee5e53 100644
@@ -63,6 +63,8 @@
 	for ((stack_idx) = 0; (umc_mask); \
 	     (umc_mask) >>= 4, (stack_idx)++) \
 
+#define NUM_JPEG_RINGS_FW	10
+
 #define to_amdgpu_device(x) (container_of(x, struct amdgpu_device, pm.smu_i2c))
 
 #define SMU_15_0_8_FEA_MAP(smu_feature, smu_15_0_8_feature)                    \
@@ -1593,6 +1595,81 @@ static ssize_t smu_v15_0_8_get_temp_metrics(struct smu_context *smu,
 	}
 }
 
+static ssize_t smu_v15_0_8_get_xcp_metrics(struct smu_context *smu, int xcp_id,
+					   void *table)
+{
+	struct smu_table_context *smu_table = &smu->smu_table;
+	const u8 num_jpeg_rings = NUM_JPEG_RINGS_FW;
+	struct smu_v15_0_8_partition_metrics *xcp_metrics;
+	MetricsTable_t *metrics;
+	struct amdgpu_device *adev = smu->adev;
+	int ret, inst, i, j, k, idx;
+	struct amdgpu_xcp *xcp;
+	u32 inst_mask;
+
+	if (!table)
+		return sizeof(*xcp_metrics);
+
+	for_each_xcp(adev->xcp_mgr, xcp, i) {
+		if (xcp->id == xcp_id)
+			break;
+	}
+	if (i == adev->xcp_mgr->num_xcps)
+		return -EINVAL;
+
+	xcp_metrics = (struct smu_v15_0_8_partition_metrics *)table;
+	smu_v15_0_8_partition_metrics_init(xcp_metrics, 1, 1);
+
+	ret = smu_v15_0_8_get_metrics_table_internal(smu, 1, NULL);
+	if (ret)
+		return ret;
+
+	metrics = (MetricsTable_t *)smu_table->metrics_table;
+
+	amdgpu_xcp_get_inst_details(xcp, AMDGPU_XCP_VCN, &inst_mask);
+	idx = 0;
+	for_each_inst(k, inst_mask) {
+		/* Both JPEG and VCN has same instances */
+		inst = GET_INST(VCN, k);
+		for (j = 0; j < num_jpeg_rings; ++j) {
+			xcp_metrics->jpeg_busy[(idx * num_jpeg_rings) + j] =
+				SMUQ10_ROUND(metrics->JpegBusy[(inst * num_jpeg_rings) + j]);
+		}
+		xcp_metrics->vcn_busy[idx] =
+			SMUQ10_ROUND(metrics->VcnBusy[inst]);
+
+		xcp_metrics->current_vclk[idx] = SMUQ10_ROUND(metrics->VclkFrequency[inst]);
+		xcp_metrics->current_dclk[idx] = SMUQ10_ROUND(metrics->DclkFrequency[inst]);
+
+		idx++;
+	}
+
+	amdgpu_xcp_get_inst_details(xcp, AMDGPU_XCP_GFX, &inst_mask);
+	idx = 0;
+	for_each_inst(k, inst_mask) {
+		inst = GET_INST(GC, k);
+		xcp_metrics->current_gfxclk[idx] =
+			SMUQ10_ROUND(metrics->GfxclkFrequency[inst]);
+
+		xcp_metrics->gfx_busy_inst[idx] = SMUQ10_ROUND(metrics->GfxBusy[inst]);
+		xcp_metrics->gfx_busy_acc[idx] = SMUQ10_ROUND(metrics->GfxBusyAcc[inst]);
+		xcp_metrics->gfx_below_host_limit_ppt_acc[idx] =
+			SMUQ10_ROUND(metrics->GfxclkBelowHostLimitPptAcc[inst]);
+		xcp_metrics->gfx_below_host_limit_thm_acc[idx] =
+			SMUQ10_ROUND(metrics->GfxclkBelowHostLimitThmAcc[inst]);
+		xcp_metrics->gfx_low_utilization_acc[idx] =
+			SMUQ10_ROUND(metrics->GfxclkLowUtilizationAcc[inst]);
+		xcp_metrics->gfx_below_host_limit_total_acc[idx] =
+			SMUQ10_ROUND(metrics->GfxclkBelowHostLimitTotalAcc[inst]);
+		idx++;
+	}
+
+	xcp_metrics->accumulation_counter = metrics->AccumulationCounter;
+	xcp_metrics->firmware_timestamp = metrics->Timestamp;
+
+	return sizeof(*xcp_metrics);
+}
+
 static ssize_t smu_v15_0_8_get_gpu_metrics(struct smu_context *smu, void **table)
 {
 	struct smu_table_context *smu_table = &smu->smu_table;
@@ -2222,6 +2299,7 @@ static const struct pptable_funcs smu_v15_0_8_ppt_funcs = {
 	.get_pp_feature_mask = smu_cmn_get_pp_feature_mask,
 	.wait_for_event = smu_v15_0_wait_for_event,
 	.get_pm_metrics = smu_v15_0_8_get_pm_metrics,
+	.get_xcp_metrics = smu_v15_0_8_get_xcp_metrics,
 	.mode2_reset = smu_v15_0_8_mode2_reset,
 	.get_dpm_ultimate_freq = smu_v15_0_8_get_dpm_ultimate_freq,
 	.get_gpu_metrics = smu_v15_0_8_get_gpu_metrics,
drivers/gpu/drm/amd/pm/swsmu/smu15/smu_v15_0_8_ppt.h +34 -0
index 398ce44821741..4dfc40aaffcb4 100644
@@ -309,5 +309,39 @@ DECLARE_SMU_METRICS_CLASS(smu_v15_0_8_gpuboard_temp_metrics,
 
 DECLARE_SMU_METRICS_CLASS(smu_v15_0_8_baseboard_temp_metrics,
 			  SMU_15_0_8_BASEBOARD_TEMP_METRICS_FIELDS);
+#define SMU_15_0_8_PARTITION_METRICS_FIELDS(SMU_SCALAR, SMU_ARRAY)             \
+	SMU_ARRAY(SMU_MATTR(CURRENT_GFXCLK), SMU_MUNIT(CLOCK_1),               \
+		  SMU_MTYPE(U16), current_gfxclk, SMU_15_0_8_MAX_XCC);         \
+	SMU_ARRAY(SMU_MATTR(CURRENT_VCLK0), SMU_MUNIT(CLOCK_1),                \
+		  SMU_MTYPE(U16), current_vclk, SMU_15_0_8_MAX_VCN);          \
+	SMU_ARRAY(SMU_MATTR(CURRENT_DCLK0), SMU_MUNIT(CLOCK_1),                \
+		  SMU_MTYPE(U16), current_dclk, SMU_15_0_8_MAX_VCN);          \
+	SMU_ARRAY(SMU_MATTR(GFX_BUSY_INST), SMU_MUNIT(PERCENT),                \
+		  SMU_MTYPE(U8), gfx_busy_inst, SMU_15_0_8_MAX_XCC);          \
+	SMU_ARRAY(SMU_MATTR(JPEG_BUSY), SMU_MUNIT(PERCENT), SMU_MTYPE(U8),    \
+		  jpeg_busy, SMU_15_0_8_MAX_JPEG);                             \
+	SMU_ARRAY(SMU_MATTR(VCN_BUSY), SMU_MUNIT(PERCENT), SMU_MTYPE(U8),     \
+		  vcn_busy, SMU_15_0_8_MAX_VCN);                               \
+	SMU_ARRAY(SMU_MATTR(GFX_BUSY_ACC), SMU_MUNIT(NONE), SMU_MTYPE(U64),    \
+		  gfx_busy_acc, SMU_15_0_8_MAX_XCC);                           \
+	SMU_ARRAY(SMU_MATTR(GFX_BELOW_HOST_LIMIT_PPT_ACC), SMU_MUNIT(NONE),    \
+		  SMU_MTYPE(U64), gfx_below_host_limit_ppt_acc,                \
+		  SMU_15_0_8_MAX_XCC);                                         \
+	SMU_ARRAY(SMU_MATTR(GFX_BELOW_HOST_LIMIT_THM_ACC), SMU_MUNIT(NONE),    \
+		  SMU_MTYPE(U64), gfx_below_host_limit_thm_acc,                \
+		  SMU_15_0_8_MAX_XCC);                                         \
+	SMU_ARRAY(SMU_MATTR(GFX_LOW_UTILIZATION_ACC), SMU_MUNIT(NONE),         \
+		  SMU_MTYPE(U64), gfx_low_utilization_acc,                     \
+		  SMU_15_0_8_MAX_XCC);                                         \
+	SMU_ARRAY(SMU_MATTR(GFX_BELOW_HOST_LIMIT_TOTAL_ACC), SMU_MUNIT(NONE),  \
+		  SMU_MTYPE(U64), gfx_below_host_limit_total_acc,              \
+		  SMU_15_0_8_MAX_XCC);					       \
+	SMU_SCALAR(SMU_MATTR(ACCUMULATION_COUNTER), SMU_MUNIT(NONE),           \
+		   SMU_MTYPE(U64), accumulation_counter);                      \
+	SMU_SCALAR(SMU_MATTR(FIRMWARE_TIMESTAMP), SMU_MUNIT(TIME_2),           \
+		   SMU_MTYPE(U64), firmware_timestamp);
+
+DECLARE_SMU_METRICS_CLASS(smu_v15_0_8_partition_metrics,
+			  SMU_15_0_8_PARTITION_METRICS_FIELDS);
 #endif
 #endif
Signature · 1 line
2.53.0
author maintainer
#2 lore raw

PATCH 3/4 drm/amd/pm: Add mem temp to gpu metrics for smu_v15_0_8

In reply to Alex Deucher

From: Asad Kamal <[email protected]>

Add mem temp to gpu metrics as maximum hbm stack temperature for smu_v15_0_8

---
 .../gpu/drm/amd/pm/swsmu/smu15/smu_v15_0_8_ppt.c   | 14 ++++++++++----
 1 file changed, 10 insertions(+), 4 deletions(-)
drivers/gpu/drm/amd/pm/swsmu/smu15/smu_v15_0_8_ppt.c +10 -4
index 288fab5ee5e53..e62494810b803 100644
@@ -1696,17 +1696,23 @@ static ssize_t smu_v15_0_8_get_gpu_metrics(struct smu_context *smu, void **table
 	if (adev->umc.active_mask) {
 		u64 mask = adev->umc.active_mask;
 		int out_idx = 0;
+		u16 max_hbm_temp = 0;
 		int stack_idx;
 
-		if (unlikely(hweight64(mask)/4 > SMU_15_0_8_MAX_HBM_STACKS))
+		if (unlikely(hweight64(mask) / 4 > SMU_15_0_8_MAX_HBM_STACKS)) {
 			dev_warn(adev->dev, "Invalid umc mask %lld\n", mask);
-		else  {
+		} else  {
 			for_each_hbm_stack(stack_idx, mask) {
+				u16 temp;
+
 				if (!hbm_stack_mask_valid(mask))
 					continue;
-				gpu_metrics->temperature_hbm[out_idx++] =
-					SMUQ10_ROUND(metrics->HbmTemperature[stack_idx]);
+				temp = SMUQ10_ROUND(metrics->HbmTemperature[stack_idx]);
+				gpu_metrics->temperature_hbm[out_idx++] = temp;
+				if (temp > max_hbm_temp)
+					max_hbm_temp = temp;
 			}
+			gpu_metrics->temperature_mem = max_hbm_temp;
 		}
 	}
 
Signature · 1 line
2.53.0
author maintainer
#3 lore raw

PATCH 4/4 drm/amd/pm: Add thermal alert logging to smu15.0.8

In reply to Alex Deucher

From: Lijo Lazar <[email protected]>

Add logging for thermal events in SMUv15.0.8 SOCs

---
 .../drm/amd/pm/swsmu/smu15/smu_v15_0_8_ppt.c  | 65 +++++++++++++++++++
 1 file changed, 65 insertions(+)
drivers/gpu/drm/amd/pm/swsmu/smu15/smu_v15_0_8_ppt.c +65 -0
index e62494810b803..977590377021e 100644
@@ -173,6 +173,13 @@ static const struct cmn2asic_mapping smu_v15_0_8_table_map[SMU_TABLE_COUNT] = {
 	TAB_MAP(I2C_COMMANDS),
 };
 
+static const uint8_t smu_v15_0_8_throttler_map[] = {
+	[THROTTLER_PROCHOT_BIT] = (SMU_THROTTLER_PROCHOT_GFX_BIT),
+	[THROTTLER_THERMAL_SOCKET_BIT] = (SMU_THROTTLER_TEMP_GPU_BIT),
+	[THROTTLER_THERMAL_VR_BIT] = (SMU_THROTTLER_TEMP_VR_GFX_BIT),
+	[THROTTLER_THERMAL_HBM_BIT] = (SMU_THROTTLER_TEMP_MEM_BIT),
+};
+
 static size_t smu_v15_0_8_get_system_metrics_size(void)
 {
 	return sizeof(SystemMetricsTable_t);
@@ -2274,6 +2281,63 @@ static int smu_v15_0_8_get_ppt_limit(struct smu_context *smu,
 	return -EOPNOTSUPP;
 }
 
+static uint32_t smu_v15_0_8_get_throttler_status(struct smu_context *smu)
+{
+	struct smu_power_context *smu_power = &smu->smu_power;
+	struct smu_15_0_power_context *power_context = smu_power->power_context;
+	uint32_t throttler_status = 0;
+
+	throttler_status = atomic_read(&power_context->throttle_status);
+	dev_dbg(smu->adev->dev, "SMU Throttler status: %u", throttler_status);
+
+	return throttler_status;
+}
+
+static const char *const throttling_logging_label[] = {
+	[THROTTLER_PROCHOT_BIT] = "Prochot",
+	[THROTTLER_THERMAL_SOCKET_BIT] = "SOC",
+	[THROTTLER_THERMAL_VR_BIT] = "VR",
+	[THROTTLER_THERMAL_HBM_BIT] = "HBM"
+};
+
+static void smu_v15_0_8_log_thermal_throttling_event(struct smu_context *smu)
+{
+	int throttler_idx, throttling_events = 0, buf_idx = 0;
+	struct amdgpu_device *adev = smu->adev;
+	uint32_t throttler_status;
+	char log_buf[256];
+
+	throttler_status = smu_v15_0_8_get_throttler_status(smu);
+	if (!throttler_status)
+		return;
+
+	memset(log_buf, 0, sizeof(log_buf));
+	for (throttler_idx = 0;
+	     throttler_idx < ARRAY_SIZE(throttling_logging_label);
+	     throttler_idx++) {
+		if (throttler_status & (1U << throttler_idx)) {
+			throttling_events++;
+			buf_idx += snprintf(
+				log_buf + buf_idx, sizeof(log_buf) - buf_idx,
+				"%s%s", throttling_events > 1 ? " and " : "",
+				throttling_logging_label[throttler_idx]);
+			if (buf_idx >= sizeof(log_buf)) {
+				dev_err(adev->dev, "buffer overflow!\n");
+				log_buf[sizeof(log_buf) - 1] = '\0';
+				break;
+			}
+		}
+	}
+
+	dev_warn(adev->dev,
+		 "WARN: GPU is throttled, expect performance decrease. %s.\n",
+		 log_buf);
+	kgd2kfd_smi_event_throttle(
+		smu->adev->kfd.dev,
+		smu_cmn_get_indep_throttler_status(throttler_status,
+						   smu_v15_0_8_throttler_map));
+}
+
 static int smu_v15_0_8_enable_thermal_alert(struct smu_context *smu)
 {
 	if (!smu->irq_source.num_types)
@@ -2301,6 +2365,7 @@ static const struct pptable_funcs smu_v15_0_8_ppt_funcs = {
 	.register_irq_handler = smu_v15_0_8_register_irq_handler,
 	.enable_thermal_alert = smu_v15_0_8_enable_thermal_alert,
 	.disable_thermal_alert = smu_v15_0_disable_thermal_alert,
+	.log_thermal_throttling_event = smu_v15_0_8_log_thermal_throttling_event,
 	.setup_pptable = smu_v15_0_8_setup_pptable,
 	.get_pp_feature_mask = smu_cmn_get_pp_feature_mask,
 	.wait_for_event = smu_v15_0_wait_for_event,
Signature · 1 line
2.53.0