Comment corrections, debug output and assert added

src/rocm_smi.cc

@@ -441,6 +441,12 @@ static rsmi_status_t get_frequencies(amd::smi::DevInfoTypes type,
 
   for (uint32_t i = 0; i < f->num_supported; ++i) {
     f->frequency[i] = freq_string_to_int(val_vec[i], &current);
+
+    // Our assumption is that frequencies are read in from lowest to highest.
+    // Check that that is true.
+    if (i > 0) {
+      assert(f->frequency[i-1] <= f->frequency[i]);
+    }
     if (current) {
       // Should only be 1 current frequency
       assert(f->current == RSMI_MAX_NUM_FREQUENCIES + 1);
@@ -511,8 +517,6 @@ static rsmi_status_t set_power_profile(uint32_t dv_ind,
   rsmi_status_t ret;
   rsmi_power_profile_status avail_profiles = {0, RSMI_PWR_PROF_PRST_INVALID, 0};
 
-  // TODO(cf): test if it is valid to OR profiles; if not the following is
-  // not necessary:
   // Determine if the provided profile is valid
   if (!is_power_of_2(profile)) {
     return RSMI_STATUS_INPUT_OUT_OF_BOUNDS;

src/rocm_smi_device.cc

@@ -98,7 +98,7 @@ static bool isRegularFile(std::string fname) {
   if (X) return X; \
 }
 
-Device::Device(std::string p) : path_(p) {
+Device::Device(std::string p, RocmSMI_env_vars const *e) : path_(p), env_(e) {
   monitor_ = nullptr;
 }
 
@@ -114,6 +114,7 @@ int Device::readDevInfoStr(DevInfoTypes type, std::string *retStr) {
   tempPath += "/device/";
   tempPath += kDevAttribNameMap.at(type);
 
+  DBG_FILE_ERROR(tempPath);
   if (!isRegularFile(tempPath)) {
     return EISDIR;
   }
@@ -121,6 +122,7 @@ int Device::readDevInfoStr(DevInfoTypes type, std::string *retStr) {
   std::ifstream fs;
   fs.open(tempPath);
 
+  DBG_FILE_ERROR(tempPath);
   if (!fs.is_open()) {
       return errno;
   }
@@ -139,12 +141,14 @@ int Device::writeDevInfoStr(DevInfoTypes type, std::string valStr) {
   std::ofstream fs;
   fs.open(tempPath);
 
+  DBG_FILE_ERROR(tempPath);
   if (!isRegularFile(tempPath)) {
     return EISDIR;
   }
 
+  DBG_FILE_ERROR(tempPath);
   if (!fs.is_open()) {
-      return errno;
+    return errno;
   }
 
   fs << valStr;
@@ -206,6 +210,7 @@ int Device::readDevInfoMultiLineStr(DevInfoTypes type,
   std::stringstream buffer;
 
 
+  DBG_FILE_ERROR(tempPath);
   if (!isRegularFile(tempPath)) {
     return EISDIR;
   }

src/rocm_smi_main.cc

@@ -48,6 +48,7 @@
 #include <string.h>
 #include <unistd.h>
 #include <sys/types.h>
+#include <stdlib.h>
 
 #include <string>
 #include <cstdint>
@@ -138,6 +139,8 @@ std::vector<std::shared_ptr<amd::smi::Device>> RocmSMI::s_monitor_devices;
 RocmSMI::RocmSMI(void) {
   auto i = 0;
 
+  GetEnvVariables();
+
   while (std::string(kAMDMonitorTypes[i]) != "") {
       amd_monitor_types_.insert(kAMDMonitorTypes[i]);
       ++i;
@@ -165,6 +168,20 @@ RocmSMI& RocmSMI::getInstance(void) {
   return singleton;
 }
 
+static int GetEnvVarInteger(const char *ev_str) {
+  ev_str = getenv(ev_str);
+
+  if (ev_str) {
+    return atoi(ev_str);
+  }
+  return 0;
+}
+
+// Get and store env. variables in this method
+void RocmSMI::GetEnvVariables(void) {
+  env_vars_.debug_output_bitfield = GetEnvVarInteger("RSMI_DEBUG_BITFIELD");
+}
+
 void
 RocmSMI::AddToDeviceList(std::string dev_name) {
   auto ret = 0;
@@ -173,7 +190,7 @@ RocmSMI::AddToDeviceList(std::string dev_name) {
   dev_path += "/";
   dev_path += dev_name;
 
-  auto dev = std::shared_ptr<Device>(new Device(dev_path));
+  auto dev = std::shared_ptr<Device>(new Device(dev_path, &env_vars_));
 
   auto m = monitors_.begin();
 
@@ -214,6 +231,8 @@ uint32_t RocmSMI::DiscoverDevices(void) {
   }
 
   auto drm_dir = opendir(kPathDRMRoot);
+  assert(drm_dir != nullptr);
+
   auto dentry = readdir(drm_dir);
 
   while (dentry != nullptr) {
@@ -261,7 +280,8 @@ uint32_t RocmSMI::DiscoverAMDMonitors(void) {
       fs.close();
 
       if (amd_monitor_types_.find(mon_type) != amd_monitor_types_.end()) {
-        monitors_.push_back(std::shared_ptr<Monitor>(new Monitor(mon_name)));
+        monitors_.push_back(std::shared_ptr<Monitor>(
+                                          new Monitor(mon_name, &env_vars_)));
       }
     }
     dentry = readdir(mon_dir);
@@ -311,7 +331,7 @@ uint32_t RocmSMI::DiscoverAMDPowerMonitors(bool force_update) {
 
     if (FileExists(tmp.c_str())) {
       std::shared_ptr<PowerMon> mon =
-                            std::shared_ptr<PowerMon>(new PowerMon(mon_name));
+                std::shared_ptr<PowerMon>(new PowerMon(mon_name, &env_vars_));
       power_mons_.push_back(mon);
       mon->set_dev_index(GetDeviceIndex(dentry->d_name));
     }

src/rocm_smi_monitor.cc

@@ -114,7 +114,8 @@ static const std::map<MonitorTypes, const char *> kMonitorNameMap = {
     {kMonTempHighest, kMonTempHighestName},
 };
 
-Monitor::Monitor(std::string path) : path_(path) {
+Monitor::Monitor(std::string path, RocmSMI_env_vars const *e) :
+                                                        path_(path), env_(e) {
 }
 Monitor::~Monitor(void) {
 }
@@ -135,6 +136,8 @@ Monitor::MakeMonitorPath(MonitorTypes type, int32_t sensor_id) {
 int Monitor::writeMonitor(MonitorTypes type, uint32_t sensor_id,
                                                             std::string val) {
   std::string sysfs_path = MakeMonitorPath(type, sensor_id);
+
+  DBG_FILE_ERROR(sysfs_path)
   return WriteSysfsStr(sysfs_path, val);
 }
 
@@ -146,6 +149,7 @@ int Monitor::readMonitor(MonitorTypes type, uint32_t sensor_id,
   std::string temp_str;
   std::string sysfs_path = MakeMonitorPath(type, sensor_id);
 
+  DBG_FILE_ERROR(sysfs_path)
   return ReadSysfsStr(sysfs_path, val);
 }
 

src/rocm_smi_power_mon.cc

@@ -55,6 +55,7 @@
 #include "rocm_smi/rocm_smi_main.h"
 #include "rocm_smi/rocm_smi_monitor.h"
 #include "rocm_smi/rocm_smi_utils.h"
+#include "rocm_smi/rocm_smi_common.h"
 
 namespace amd {
 namespace smi {
@@ -68,7 +69,8 @@ static const std::map<PowerMonTypes, const char *> kMonitorNameMap = {
     {kPowerAveGPUPower, kPowerMonPMName},
 };
 
-PowerMon::PowerMon(std::string path) : path_(path) {
+PowerMon::PowerMon(std::string path, RocmSMI_env_vars const *e) :
+                                                        path_(path), env_(e) {
 }
 PowerMon::~PowerMon(void) {
 }
@@ -139,6 +141,8 @@ int PowerMon::readPowerValue(PowerMonTypes type, uint64_t *power) {
 
   tempPath += "/";
   tempPath += kMonitorNameMap.at(type);
+
+  DBG_FILE_ERROR(tempPath)
   int ret = ReadSysfsStr(tempPath, &fstr);
 
   if (ret) {