Adding functionality that will parse gpu_metrics sysfs file

Signed-off-by: Divya Shikre <DivyaUday.Shikre@amd.com>
Change-Id: I3a84870b83eb4cd0ed46f10bb19169c91f99fd8e

include/rocm_smi/rocm_smi.h

@@ -757,6 +757,72 @@ typedef struct {
 typedef rsmi_od_volt_freq_data_t rsmi_od_volt_freq_data;
 /// \endcond
 
+
+/**
+ * @brief The following structures hold the gpu metrics values for a device.
+ */
+
+struct metrics_table_header {
+  uint16_t      structure_size;
+  uint8_t       format_revision;
+  uint8_t       content_revision;
+};
+
+typedef struct {
+  struct metrics_table_header common_header;
+
+/* Driver attached timestamp (in ns) */
+  uint64_t      system_clock_counter;
+
+/* Temperature */
+  uint16_t      temperature_edge;
+  uint16_t      temperature_hotspot;
+  uint16_t      temperature_mem;
+  uint16_t      temperature_vrgfx;
+  uint16_t      temperature_vrsoc;
+  uint16_t      temperature_vrmem;
+
+/* Utilization */
+  uint16_t      average_gfx_activity;
+  uint16_t      average_umc_activity; // memory controller
+  uint16_t      average_mm_activity; // UVD or VCN
+
+/* Power/Energy */
+  uint16_t      average_socket_power;
+  uint32_t      energy_accumulator;
+
+/* Average clocks */
+  uint16_t      average_gfxclk_frequency;
+  uint16_t      average_socclk_frequency;
+  uint16_t      average_uclk_frequency;
+  uint16_t      average_vclk0_frequency;
+  uint16_t      average_dclk0_frequency;
+  uint16_t      average_vclk1_frequency;
+  uint16_t      average_dclk1_frequency;
+
+/* Current clocks */
+  uint16_t      current_gfxclk;
+  uint16_t      current_socclk;
+  uint16_t      current_uclk;
+  uint16_t      current_vclk0;
+  uint16_t      current_dclk0;
+  uint16_t      current_vclk1;
+  uint16_t      current_dclk1;
+
+/* Throttle status */
+  uint32_t      throttle_status;
+
+/* Fans */
+  uint16_t      current_fan_speed;
+
+/* Link width/speed */
+  uint8_t       pcie_link_width;
+  uint8_t       pcie_link_speed; // in 0.1 GT/s
+}rsmi_gpu_metrics_t;
+/// \cond Ignore in docs.
+typedef rsmi_gpu_metrics_t rsmi_gpu_metrics;
+/// \endcond
+
 /**
  * @brief This structure holds error counts.
  */
@@ -2014,6 +2080,29 @@ rsmi_status_t rsmi_dev_gpu_clk_freq_get(uint32_t dv_ind,
 rsmi_status_t rsmi_dev_od_volt_info_get(uint32_t dv_ind,
                                                rsmi_od_volt_freq_data_t *odv);
 
+/**
+ *  @brief This function retrieves the gpu metrics information
+ *
+ *  @details Given a device index @p dv_ind and a pointer to a
+ *  ::rsmi_gpu_metrics_t structure @p pgpu_metrics, this function will populate
+ *  @p pgpu_metrics. See ::rsmi_gpu_metrics_t for more details.
+ *
+ *  @param[in] dv_ind a device index
+ *
+ *  @param[inout] pgpu_metrics a pointer to an ::rsmi_gpu_metrics_t structure
+ *  If this parameter is nullptr, this function will return
+ *  ::RSMI_STATUS_INVALID_ARGS if the function is supported with the provided,
+ *  arguments and ::RSMI_STATUS_NOT_SUPPORTED if it is not supported with the
+ *  provided arguments.
+ *
+ *  @retval ::RSMI_STATUS_SUCCESS call was successful
+ *  @retval ::RSMI_STATUS_NOT_SUPPORTED installed software or hardware does not
+ *  support this function with the given arguments
+ *  @retval ::RSMI_STATUS_INVALID_ARGS the provided arguments are not valid
+ */
+rsmi_status_t rsmi_dev_gpu_metrics_info_get(uint32_t dv_ind,
+                                            rsmi_gpu_metrics_t *pgpu_metrics);
+
 /**
  *  @brief This function sets the clock frequency information
  *

include/rocm_smi/rocm_smi_device.h

@@ -152,7 +152,8 @@ enum DevInfoTypes {
   kDevFwVersionVcn,
   kDevSerialNumber,
   kDevMemPageBad,
-  kDevNumaNode
+  kDevNumaNode,
+  kDevGpuMetrics
 };
 
 typedef struct {
@@ -175,6 +176,7 @@ class Device {
     int readDevInfoLine(DevInfoTypes type, std::string *line);
     int readDevInfo(DevInfoTypes type, std::string *val);
     int readDevInfo(DevInfoTypes type, std::vector<std::string> *retVec);
+    int readDevInfo(DevInfoTypes type, std::vector<unsigned char> *retVec);
     int writeDevInfo(DevInfoTypes type, uint64_t val);
     int writeDevInfo(DevInfoTypes type, std::string val);
 
@@ -214,6 +216,8 @@ class Device {
     int readDevInfoStr(DevInfoTypes type, std::string *retStr);
     int readDevInfoMultiLineStr(DevInfoTypes type,
                                             std::vector<std::string> *retVec);
+    int readDevInfoBinary(DevInfoTypes type,
+                                            std::vector<unsigned char> *retVec);
     int writeDevInfoStr(DevInfoTypes type, std::string valStr);
     uint64_t bdfid_;
     uint64_t kfd_gpu_id_;

include/rocm_smi/rocm_smi_utils.h

@@ -80,6 +80,9 @@ rsmi_status_t handleException();
 rsmi_status_t
 GetDevValueVec(amd::smi::DevInfoTypes type,
                          uint32_t dv_ind, std::vector<std::string> *val_vec);
+rsmi_status_t
+GetDevBinaryVec(amd::smi::DevInfoTypes type,
+                         uint32_t dv_ind, std::vector<unsigned char> *val_vec);
 rsmi_status_t ErrnoToRsmiStatus(uint32_t err);
 
 struct pthread_wrap {

rocm_smi/example/rocm_smi_example.cc

@@ -408,6 +408,7 @@ int main() {
   rsmi_dev_perf_level_t pfl;
   rsmi_frequencies_t f;
   uint32_t num_monitor_devs = 0;
+  rsmi_gpu_metrics_t p;
 
   rsmi_num_monitor_devices(&num_monitor_devs);
   for (uint32_t i = 0; i< num_monitor_devs; ++i) {
@@ -415,6 +416,10 @@ int main() {
     CHK_RSMI_RET(ret)
     std::cout << "\t**Device ID: 0x" << std::hex << val_ui64 << std::endl;
 
+    ret = rsmi_dev_gpu_metrics_info_get(i, &p);
+    CHK_RSMI_RET(ret)
+    std::cout << "\t**GPU METRICS" << std::endl;
+
     ret = rsmi_dev_perf_level_get(i, &pfl);
     CHK_RSMI_RET(ret)
     std::cout << "\t**Performance Level:" <<

src/rocm_smi.cc

@@ -1033,7 +1033,6 @@ static rsmi_status_t get_od_clk_volt_info(uint32_t dv_ind,
   CATCH
 }
 
-
 rsmi_status_t rsmi_dev_od_clk_info_set(uint32_t dv_ind, rsmi_freq_ind_t level,
                                         uint64_t clkvalue,
                                         rsmi_clk_type_t clkType) {
@@ -2158,6 +2157,29 @@ rsmi_dev_od_volt_info_get(uint32_t dv_ind, rsmi_od_volt_freq_data_t *odv) {
   CATCH
 }
 
+rsmi_status_t
+rsmi_dev_gpu_metrics_info_get(uint32_t dv_ind, rsmi_gpu_metrics_t *smu) {
+  TRY
+  DEVICE_MUTEX
+  CHK_SUPPORT_NAME_ONLY(smu)
+
+  std::vector<unsigned char> val_vec;
+  rsmi_status_t ret = GetDevBinaryVec(amd::smi::kDevGpuMetrics, dv_ind,
+                                                                   &val_vec);
+  if (ret != RSMI_STATUS_SUCCESS) {
+    return ret;
+  }
+
+  if (val_vec.size() == 0) {
+    return RSMI_STATUS_NOT_YET_IMPLEMENTED;
+  }
+
+  smu = reinterpret_cast<rsmi_gpu_metrics_t *>(val_vec.data());
+
+  return ret;
+  CATCH
+}
+
 rsmi_status_t rsmi_dev_od_volt_curve_regions_get(uint32_t dv_ind,
                      uint32_t *num_regions, rsmi_freq_volt_region_t *buffer) {
   TRY

src/rocm_smi_device.cc

@@ -57,6 +57,7 @@
 #include <vector>
 #include <memory>
 #include <algorithm>
+#include <iterator>
 
 #include "rocm_smi/rocm_smi_main.h"
 #include "rocm_smi/rocm_smi_device.h"
@@ -106,6 +107,7 @@ static const char *kDevMemBusyPercentFName = "mem_busy_percent";
 static const char *kDevXGMIErrorFName = "xgmi_error";
 static const char *kDevSerialNumberFName = "serial_number";
 static const char *kDevNumaNodeFName = "numa_node";
+static const char *kDevGpuMetricsFName = "gpu_metrics";
 
 // Firmware version files
 static const char *kDevFwVersionAsdFName = "fw_version/asd_fw_version";
@@ -265,6 +267,7 @@ static const std::map<DevInfoTypes, const char *> kDevAttribNameMap = {
     {kDevSerialNumber, kDevSerialNumberFName},
     {kDevMemPageBad, kDevMemPageBadFName},
     {kDevNumaNode, kDevNumaNodeFName},
+    {kDevGpuMetrics, kDevGpuMetricsFName},
 };
 
 static const std::map<rsmi_dev_perf_level, const char *> kDevPerfLvlMap = {
@@ -375,6 +378,7 @@ static const std::map<const char *, dev_depends_t> kDevFuncDependsMap = {
   {"rsmi_dev_xgmi_error_reset",          {{kDevXGMIErrorFName}, {}}},
   {"rsmi_dev_memory_reserved_pages_get", {{kDevMemPageBadFName}, {}}},
   {"rsmi_topo_numa_affinity_get",        {{kDevNumaNodeFName}, {}}},
+  {"rsmi_dev_gpu_metrics_info_get",      {{kDevGpuMetricsFName}, {}}},
 
   // These functions with variants, but no sensors/units. (May or may not
   // have mandatory dependencies.)
@@ -624,6 +628,24 @@ int Device::readDevInfoLine(DevInfoTypes type, std::string *line) {
   return 0;
 }
 
+int Device::readDevInfoBinary(DevInfoTypes type,
+                                std::vector<unsigned char> *retVec) {
+  auto sysfs_path = path_;
+
+  sysfs_path += "/device/";
+  sysfs_path += kDevAttribNameMap.at(type);
+
+  std::ifstream fs(sysfs_path, std::ios::binary);
+  if (!fs.is_open()) {
+    return errno;
+  }
+  // copies all data into buffer
+  retVec->insert(retVec->begin(),
+               std::istreambuf_iterator<char>(fs),{});
+
+  return 0;
+}
+
 int Device::readDevInfoMultiLineStr(DevInfoTypes type,
                                            std::vector<std::string> *retVec) {
   std::string line;
@@ -754,6 +776,21 @@ int Device::readDevInfo(DevInfoTypes type, std::vector<std::string> *val) {
   return 0;
 }
 
+int Device::readDevInfo(DevInfoTypes type, std::vector<unsigned char> *val) {
+  assert(val != nullptr);
+
+  switch (type) {
+     case kDevGpuMetrics:
+      return readDevInfoBinary(type, val);
+      break;
+
+    default:
+      return EINVAL;
+  }
+
+  return 0;
+}
+
 int Device::readDevInfo(DevInfoTypes type, std::string *val) {
   assert(val != nullptr);
 

src/rocm_smi_utils.cc

@@ -204,6 +204,18 @@ rsmi_status_t GetDevValueVec(amd::smi::DevInfoTypes type,
   return ErrnoToRsmiStatus(ret);
 }
 
+rsmi_status_t GetDevBinaryVec(amd::smi::DevInfoTypes type,
+                         uint32_t dv_ind, std::vector<unsigned char> *val_vec) {
+  assert(val_vec != nullptr);
+  if (val_vec == nullptr) {
+    return RSMI_STATUS_INVALID_ARGS;
+  }
+  GET_DEV_FROM_INDX
+
+  int ret = dev->readDevInfo(type, val_vec);
+  return ErrnoToRsmiStatus(ret);
+}
+
 rsmi_status_t ErrnoToRsmiStatus(uint32_t err) {
   switch (err) {
     case 0:      return RSMI_STATUS_SUCCESS;