Fill and publish GPU metrics in v1.9 format for SMUv13.0.6 SOCs
Signed-off-by: Lijo Lazar <lijo.la...@amd.com>
---
.../drm/amd/pm/swsmu/smu13/smu_v13_0_6_ppt.c | 113 +++++++++--------
.../drm/amd/pm/swsmu/smu13/smu_v13_0_6_ppt.h | 118 ++++++++++++++++++
2 files changed, 177 insertions(+), 54 deletions(-)
diff --git a/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_6_ppt.c
b/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_6_ppt.c
index ebee659f8a1c..654612623341 100644
--- a/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_6_ppt.c
+++ b/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_6_ppt.c
@@ -539,6 +539,7 @@ static int smu_v13_0_6_tables_init(struct smu_context *smu)
struct smu_table_context *smu_table = &smu->smu_table;
struct smu_table *tables = smu_table->tables;
void *gpu_metrics_table __free(kfree) = NULL;
+ struct smu_v13_0_6_gpu_metrics *gpu_metrics;
void *driver_pptable __free(kfree) = NULL;
void *metrics_table __free(kfree) = NULL;
struct amdgpu_device *adev = smu->adev;
@@ -578,11 +579,22 @@ static int smu_v13_0_6_tables_init(struct smu_context
*smu)
if (!driver_pptable)
return -ENOMEM;
+ ret = smu_table_cache_init(smu, SMU_TABLE_SMU_METRICS,
+ sizeof(struct smu_v13_0_6_gpu_metrics), 1);
+ if (ret)
+ return ret;
+
+ gpu_metrics = (struct smu_v13_0_6_gpu_metrics
+ *)(tables[SMU_TABLE_SMU_METRICS].cache.buffer);
+
+ smu_v13_0_6_gpu_metrics_init(gpu_metrics, 1, 9);
if (amdgpu_ip_version(smu->adev, MP1_HWIP, 0) ==
IP_VERSION(13, 0, 12)) {
ret = smu_v13_0_12_tables_init(smu);
- if (ret)
+ if (ret) {
+ smu_table_cache_fini(smu, SMU_TABLE_SMU_METRICS);
return ret;
+ }
}
smu_table->gpu_metrics_table = no_free_ptr(gpu_metrics_table);
@@ -721,6 +733,7 @@ static int smu_v13_0_6_fini_smc_tables(struct smu_context
*smu)
{
if (amdgpu_ip_version(smu->adev, MP1_HWIP, 0) == IP_VERSION(13, 0, 12))
smu_v13_0_12_tables_fini(smu);
+ smu_table_cache_fini(smu, SMU_TABLE_SMU_METRICS);
return smu_v13_0_fini_smc_tables(smu);
}
@@ -2736,18 +2749,16 @@ static ssize_t smu_v13_0_6_get_xcp_metrics(struct
smu_context *smu, int xcp_id,
static ssize_t smu_v13_0_6_get_gpu_metrics(struct smu_context *smu, void
**table)
{
struct smu_table_context *smu_table = &smu->smu_table;
- struct gpu_metrics_v1_8 *gpu_metrics =
- (struct gpu_metrics_v1_8 *)smu_table->gpu_metrics_table;
+ struct smu_table *tables = smu_table->tables;
+ struct smu_v13_0_6_gpu_metrics *gpu_metrics;
int version = smu_v13_0_6_get_metrics_version(smu);
MetricsTableV0_t *metrics_v0 __free(kfree) = NULL;
- int ret = 0, xcc_id, inst, i, j, k, idx;
struct amdgpu_device *adev = smu->adev;
+ int ret = 0, xcc_id, inst, i, j;
MetricsTableV1_t *metrics_v1;
MetricsTableV2_t *metrics_v2;
- struct amdgpu_xcp *xcp;
u16 link_width_level;
u8 num_jpeg_rings;
- u32 inst_mask;
bool per_inst;
metrics_v0 = kzalloc(METRICS_TABLE_SIZE, GFP_KERNEL);
@@ -2762,8 +2773,8 @@ static ssize_t smu_v13_0_6_get_gpu_metrics(struct
smu_context *smu, void **table
metrics_v1 = (MetricsTableV1_t *)metrics_v0;
metrics_v2 = (MetricsTableV2_t *)metrics_v0;
-
- smu_cmn_init_soft_gpu_metrics(gpu_metrics, 1, 8);
+ gpu_metrics = (struct smu_v13_0_6_gpu_metrics
+ *)(tables[SMU_TABLE_SMU_METRICS].cache.buffer);
gpu_metrics->temperature_hotspot =
SMUQ10_ROUND(GET_METRIC_FIELD(MaxSocketTemperature, version));
@@ -2885,55 +2896,49 @@ static ssize_t smu_v13_0_6_get_gpu_metrics(struct
smu_context *smu, void **table
gpu_metrics->xgmi_link_status[j] = ret;
}
- gpu_metrics->num_partition = adev->xcp_mgr->num_xcps;
-
per_inst = smu_v13_0_6_cap_supported(smu, SMU_CAP(PER_INST_METRICS));
num_jpeg_rings = AMDGPU_MAX_JPEG_RINGS_4_0_3;
- for_each_xcp(adev->xcp_mgr, xcp, i) {
- 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) {
- gpu_metrics->xcp_stats[i].jpeg_busy
- [(idx * num_jpeg_rings) + j] =
- SMUQ10_ROUND(GET_METRIC_FIELD(JpegBusy,
version)
- [(inst *
num_jpeg_rings) + j]);
- }
- gpu_metrics->xcp_stats[i].vcn_busy[idx] =
- SMUQ10_ROUND(GET_METRIC_FIELD(VcnBusy,
version)[inst]);
- idx++;
-
- }
+ for (i = 0; i < adev->jpeg.num_jpeg_inst; ++i) {
+ inst = GET_INST(JPEG, i);
+ for (j = 0; j < num_jpeg_rings; ++j)
+ gpu_metrics->jpeg_busy[(i * num_jpeg_rings) + j] =
+ SMUQ10_ROUND(GET_METRIC_FIELD(
+ JpegBusy,
+ version)[(inst * num_jpeg_rings) + j]);
+ }
+ for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
+ inst = GET_INST(VCN, i);
+ gpu_metrics->vcn_busy[i] =
+ SMUQ10_ROUND(GET_METRIC_FIELD(VcnBusy, version)[inst]);
+ }
- if (per_inst) {
- amdgpu_xcp_get_inst_details(xcp, AMDGPU_XCP_GFX,
&inst_mask);
- idx = 0;
- for_each_inst(k, inst_mask) {
- inst = GET_INST(GC, k);
- gpu_metrics->xcp_stats[i].gfx_busy_inst[idx] =
-
SMUQ10_ROUND(GET_GPU_METRIC_FIELD(GfxBusy, version)[inst]);
- gpu_metrics->xcp_stats[i].gfx_busy_acc[idx] =
-
SMUQ10_ROUND(GET_GPU_METRIC_FIELD(GfxBusyAcc,
-
version)[inst]);
- if (smu_v13_0_6_cap_supported(smu,
SMU_CAP(HST_LIMIT_METRICS))) {
-
gpu_metrics->xcp_stats[i].gfx_below_host_limit_ppt_acc[idx] =
- SMUQ10_ROUND
-
(metrics_v0->GfxclkBelowHostLimitPptAcc[inst]);
-
gpu_metrics->xcp_stats[i].gfx_below_host_limit_thm_acc[idx] =
- SMUQ10_ROUND
-
(metrics_v0->GfxclkBelowHostLimitThmAcc[inst]);
-
gpu_metrics->xcp_stats[i].gfx_low_utilization_acc[idx] =
- SMUQ10_ROUND
-
(metrics_v0->GfxclkLowUtilizationAcc[inst]);
-
gpu_metrics->xcp_stats[i].gfx_below_host_limit_total_acc[idx] =
- SMUQ10_ROUND
-
(metrics_v0->GfxclkBelowHostLimitTotalAcc[inst]);
- }
- idx++;
+ if (per_inst) {
+ for (i = 0; i < NUM_XCC(adev->gfx.xcc_mask); ++i) {
+ inst = GET_INST(GC, i);
+ gpu_metrics->gfx_busy_inst[i] = SMUQ10_ROUND(
+ GET_GPU_METRIC_FIELD(GfxBusy, version)[inst]);
+ gpu_metrics->gfx_busy_acc[i] = SMUQ10_ROUND(
+ GET_GPU_METRIC_FIELD(GfxBusyAcc,
+ version)[inst]);
+ if (smu_v13_0_6_cap_supported(
+ smu, SMU_CAP(HST_LIMIT_METRICS))) {
+ gpu_metrics->gfx_below_host_limit_ppt_acc
+ [i] = SMUQ10_ROUND(
+ metrics_v0->GfxclkBelowHostLimitPptAcc
+ [inst]);
+ gpu_metrics->gfx_below_host_limit_thm_acc
+ [i] = SMUQ10_ROUND(
+ metrics_v0->GfxclkBelowHostLimitThmAcc
+ [inst]);
+ gpu_metrics->gfx_low_utilization_acc
+ [i] = SMUQ10_ROUND(
+ metrics_v0
+
->GfxclkLowUtilizationAcc[inst]);
+ gpu_metrics->gfx_below_host_limit_total_acc
+ [i] = SMUQ10_ROUND(
+ metrics_v0->GfxclkBelowHostLimitTotalAcc
+ [inst]);
}
}
}
@@ -2943,7 +2948,7 @@ static ssize_t smu_v13_0_6_get_gpu_metrics(struct
smu_context *smu, void **table
gpu_metrics->firmware_timestamp = GET_METRIC_FIELD(Timestamp, version);
- *table = (void *)gpu_metrics;
+ *table = tables[SMU_TABLE_SMU_METRICS].cache.buffer;
return sizeof(*gpu_metrics);
}
diff --git a/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_6_ppt.h
b/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_6_ppt.h
index aae9a546a67e..0e336cd5e8cc 100644
--- a/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_6_ppt.h
+++ b/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_6_ppt.h
@@ -73,6 +73,13 @@ enum smu_v13_0_6_caps {
SMU_CAP(ALL),
};
+#define SMU_13_0_6_NUM_XGMI_LINKS 8
+#define SMU_13_0_6_MAX_GFX_CLKS 8
+#define SMU_13_0_6_MAX_CLKS 4
+#define SMU_13_0_6_MAX_XCC 8
+#define SMU_13_0_6_MAX_VCN 4
+#define SMU_13_0_6_MAX_JPEG 40
+
extern void smu_v13_0_6_set_ppt_funcs(struct smu_context *smu);
extern void smu_v13_0_6_set_temp_funcs(struct smu_context *smu);
bool smu_v13_0_6_cap_supported(struct smu_context *smu, enum smu_v13_0_6_caps
cap);
@@ -95,4 +102,115 @@ void smu_v13_0_12_tables_fini(struct smu_context *smu);
extern const struct cmn2asic_mapping smu_v13_0_12_feature_mask_map[];
extern const struct cmn2asic_msg_mapping smu_v13_0_12_message_map[];
extern const struct smu_temp_funcs smu_v13_0_12_temp_funcs;
+
+#if defined(SWSMU_CODE_LAYER_L2)
+#include "smu_cmn.h"
+
+/* SMUv 13.0.6 GPU metrics*/
+#define SMU_13_0_6_METRICS_FIELDS(SMU_SCALAR, SMU_ARRAY)
\
+ SMU_SCALAR(SMU_MATTR(TEMPERATURE_HOTSPOT), SMU_MUNIT(TEMP_1), \
+ SMU_MTYPE(U16), temperature_hotspot) \
+ SMU_SCALAR(SMU_MATTR(TEMPERATURE_MEM), SMU_MUNIT(TEMP_1), \
+ SMU_MTYPE(U16), temperature_mem) \
+ SMU_SCALAR(SMU_MATTR(TEMPERATURE_VRSOC), SMU_MUNIT(TEMP_1), \
+ SMU_MTYPE(U16), temperature_vrsoc) \
+ SMU_SCALAR(SMU_MATTR(CURR_SOCKET_POWER), SMU_MUNIT(POWER_1), \
+ SMU_MTYPE(U16), curr_socket_power) \
+ SMU_SCALAR(SMU_MATTR(AVERAGE_GFX_ACTIVITY), SMU_MUNIT(PERCENT), \
+ SMU_MTYPE(U16), average_gfx_activity) \
+ SMU_SCALAR(SMU_MATTR(AVERAGE_UMC_ACTIVITY), SMU_MUNIT(PERCENT), \
+ SMU_MTYPE(U16), average_umc_activity) \
+ SMU_SCALAR(SMU_MATTR(MEM_MAX_BANDWIDTH), SMU_MUNIT(BW_1), \
+ SMU_MTYPE(U64), mem_max_bandwidth) \
+ SMU_SCALAR(SMU_MATTR(ENERGY_ACCUMULATOR), SMU_MUNIT(NONE), \
+ SMU_MTYPE(U64), energy_accumulator) \
+ SMU_SCALAR(SMU_MATTR(SYSTEM_CLOCK_COUNTER), SMU_MUNIT(TIME_1), \
+ SMU_MTYPE(U64), system_clock_counter) \
+ SMU_SCALAR(SMU_MATTR(ACCUMULATION_COUNTER), SMU_MUNIT(NONE), \
+ SMU_MTYPE(U32), accumulation_counter) \
+ SMU_SCALAR(SMU_MATTR(PROCHOT_RESIDENCY_ACC), SMU_MUNIT(NONE), \
+ SMU_MTYPE(U32), prochot_residency_acc) \
+ SMU_SCALAR(SMU_MATTR(PPT_RESIDENCY_ACC), SMU_MUNIT(NONE), \
+ SMU_MTYPE(U32), ppt_residency_acc) \
+ SMU_SCALAR(SMU_MATTR(SOCKET_THM_RESIDENCY_ACC), SMU_MUNIT(NONE), \
+ SMU_MTYPE(U32), socket_thm_residency_acc) \
+ SMU_SCALAR(SMU_MATTR(VR_THM_RESIDENCY_ACC), SMU_MUNIT(NONE), \
+ SMU_MTYPE(U32), vr_thm_residency_acc) \
+ SMU_SCALAR(SMU_MATTR(HBM_THM_RESIDENCY_ACC), SMU_MUNIT(NONE), \
+ SMU_MTYPE(U32), hbm_thm_residency_acc) \
+ SMU_SCALAR(SMU_MATTR(GFXCLK_LOCK_STATUS), SMU_MUNIT(NONE), \
+ SMU_MTYPE(U32), gfxclk_lock_status) \
+ SMU_SCALAR(SMU_MATTR(PCIE_LINK_WIDTH), SMU_MUNIT(NONE), \
+ SMU_MTYPE(U16), pcie_link_width) \
+ SMU_SCALAR(SMU_MATTR(PCIE_LINK_SPEED), SMU_MUNIT(SPEED_2), \
+ SMU_MTYPE(U16), pcie_link_speed) \
+ SMU_SCALAR(SMU_MATTR(XGMI_LINK_WIDTH), SMU_MUNIT(NONE), \
+ SMU_MTYPE(U16), xgmi_link_width) \
+ SMU_SCALAR(SMU_MATTR(XGMI_LINK_SPEED), SMU_MUNIT(SPEED_1), \
+ SMU_MTYPE(U16), xgmi_link_speed) \
+ SMU_SCALAR(SMU_MATTR(GFX_ACTIVITY_ACC), SMU_MUNIT(PERCENT), \
+ SMU_MTYPE(U32), gfx_activity_acc) \
+ SMU_SCALAR(SMU_MATTR(MEM_ACTIVITY_ACC), SMU_MUNIT(PERCENT), \
+ SMU_MTYPE(U32), mem_activity_acc) \
+ SMU_SCALAR(SMU_MATTR(PCIE_BANDWIDTH_ACC), SMU_MUNIT(PERCENT), \
+ SMU_MTYPE(U64), pcie_bandwidth_acc) \
+ SMU_SCALAR(SMU_MATTR(PCIE_BANDWIDTH_INST), SMU_MUNIT(BW_1), \
+ SMU_MTYPE(U64), pcie_bandwidth_inst) \
+ SMU_SCALAR(SMU_MATTR(PCIE_L0_TO_RECOV_COUNT_ACC), SMU_MUNIT(NONE), \
+ SMU_MTYPE(U64), pcie_l0_to_recov_count_acc) \
+ SMU_SCALAR(SMU_MATTR(PCIE_REPLAY_COUNT_ACC), SMU_MUNIT(NONE), \
+ SMU_MTYPE(U64), pcie_replay_count_acc) \
+ SMU_SCALAR(SMU_MATTR(PCIE_REPLAY_ROVER_COUNT_ACC), SMU_MUNIT(NONE), \
+ SMU_MTYPE(U64), pcie_replay_rover_count_acc) \
+ SMU_SCALAR(SMU_MATTR(PCIE_NAK_SENT_COUNT_ACC), SMU_MUNIT(NONE), \
+ SMU_MTYPE(U32), pcie_nak_sent_count_acc) \
+ SMU_SCALAR(SMU_MATTR(PCIE_NAK_RCVD_COUNT_ACC), SMU_MUNIT(NONE), \
+ SMU_MTYPE(U32), pcie_nak_rcvd_count_acc) \
+ SMU_ARRAY(SMU_MATTR(XGMI_READ_DATA_ACC), SMU_MUNIT(DATA_1), \
+ SMU_MTYPE(U64), xgmi_read_data_acc, \
+ SMU_13_0_6_NUM_XGMI_LINKS) \
+ SMU_ARRAY(SMU_MATTR(XGMI_WRITE_DATA_ACC), SMU_MUNIT(DATA_1), \
+ SMU_MTYPE(U64), xgmi_write_data_acc, \
+ SMU_13_0_6_NUM_XGMI_LINKS) \
+ SMU_ARRAY(SMU_MATTR(XGMI_LINK_STATUS), SMU_MUNIT(NONE), \
+ SMU_MTYPE(U16), xgmi_link_status, SMU_13_0_6_NUM_XGMI_LINKS) \
+ SMU_SCALAR(SMU_MATTR(FIRMWARE_TIMESTAMP), SMU_MUNIT(TIME_2), \
+ SMU_MTYPE(U64), firmware_timestamp) \
+ SMU_ARRAY(SMU_MATTR(CURRENT_GFXCLK), SMU_MUNIT(CLOCK_1), \
+ SMU_MTYPE(U16), current_gfxclk, SMU_13_0_6_MAX_GFX_CLKS) \
+ SMU_ARRAY(SMU_MATTR(CURRENT_SOCCLK), SMU_MUNIT(CLOCK_1), \
+ SMU_MTYPE(U16), current_socclk, SMU_13_0_6_MAX_CLKS) \
+ SMU_ARRAY(SMU_MATTR(CURRENT_VCLK0), SMU_MUNIT(CLOCK_1), \
+ SMU_MTYPE(U16), current_vclk0, SMU_13_0_6_MAX_CLKS) \
+ SMU_ARRAY(SMU_MATTR(CURRENT_DCLK0), SMU_MUNIT(CLOCK_1), \
+ SMU_MTYPE(U16), current_dclk0, SMU_13_0_6_MAX_CLKS) \
+ SMU_SCALAR(SMU_MATTR(CURRENT_UCLK), SMU_MUNIT(CLOCK_1), \
+ SMU_MTYPE(U16), current_uclk) \
+ SMU_SCALAR(SMU_MATTR(PCIE_LC_PERF_OTHER_END_RECOVERY), \
+ SMU_MUNIT(NONE), SMU_MTYPE(U32), \
+ pcie_lc_perf_other_end_recovery) \
+ SMU_ARRAY(SMU_MATTR(GFX_BUSY_INST), SMU_MUNIT(PERCENT), \
+ SMU_MTYPE(U32), gfx_busy_inst, SMU_13_0_6_MAX_XCC) \
+ SMU_ARRAY(SMU_MATTR(JPEG_BUSY), SMU_MUNIT(PERCENT), SMU_MTYPE(U16), \
+ jpeg_busy, SMU_13_0_6_MAX_JPEG) \
+ SMU_ARRAY(SMU_MATTR(VCN_BUSY), SMU_MUNIT(PERCENT), SMU_MTYPE(U16), \
+ vcn_busy, SMU_13_0_6_MAX_VCN) \
+ SMU_ARRAY(SMU_MATTR(GFX_BUSY_ACC), SMU_MUNIT(PERCENT), SMU_MTYPE(U64), \
+ gfx_busy_acc, SMU_13_0_6_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_13_0_6_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_13_0_6_MAX_XCC) \
+ SMU_ARRAY(SMU_MATTR(GFX_LOW_UTILIZATION_ACC), SMU_MUNIT(NONE), \
+ SMU_MTYPE(U64), gfx_low_utilization_acc, SMU_13_0_6_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_13_0_6_MAX_XCC)
+
+DECLARE_SMU_METRICS_CLASS(smu_v13_0_6_gpu_metrics, SMU_13_0_6_METRICS_FIELDS);
+
+#endif /* SWSMU_CODE_LAYER_L2 */
+
#endif
--
2.49.0
