Show more info in stderr when rsmi_init() fails

Some rsmi apps fail without much explanation when
rsmi_init() fails. This patch hopes to provide some clues to
the reason for the failure.

Change-Id: Id51308dc327b9871d537dd3e709b677db4ef10bc


[ROCm/rocm_smi_lib commit: 6377e0258d]

projects/rocm-smi-lib/src/rocm_smi.cc

@@ -435,6 +435,7 @@ rsmi_init(uint64_t flags) {
       smi.Initialize(flags);
     } catch(...) {
       smi.Cleanup();
+      std::cerr << "rsmi_init() failed" << std::endl;
       throw amd::smi::rsmi_exception(RSMI_STATUS_INIT_ERROR, __FUNCTION__);
     }
   }

projects/rocm-smi-lib/src/rocm_smi_io_link.cc

@@ -177,7 +177,14 @@ int DiscoverIOLinks(std::map<std::pair<uint32_t, uint32_t>,
   links->clear();
 
   auto kfd_node_dir = opendir(kKFDNodesPathRoot);
-  assert(kfd_node_dir != nullptr);
+
+  if (kfd_node_dir == nullptr) {
+    std::string err_msg = "Failed to open KFD nodes directory ";
+    err_msg += kKFDNodesPathRoot;
+    err_msg += ".";
+    perror(err_msg.c_str());
+    return 1;
+  }
 
   auto dentry_kfd = readdir(kfd_node_dir);
   while (dentry_kfd != nullptr) {
@@ -222,6 +229,10 @@ int DiscoverIOLinks(std::map<std::pair<uint32_t, uint32_t>,
     }
 
     if (closedir(io_link_dir)) {
+      std::string err_msg = "Failed to close KFD nodes directory ";
+      err_msg += kKFDNodesPathRoot;
+      err_msg += ".";
+      perror(err_msg.c_str());
       return 1;
     }
 

projects/rocm-smi-lib/src/rocm_smi_kfd.cc

@@ -559,12 +559,16 @@ int DiscoverKFDNodes(std::map<uint64_t, std::shared_ptr<KFDNode>> *nodes) {
       node->get_property_value(kKFDNodePropLOCATION_IDStr,
                                                         &kfd_gpu_node_bus_fn);
     if (ret != 0) {
+      std:: cerr << "Failed to open properties file for kfd node " <<
+                                       node->node_index() << "." << std::endl;
       closedir(kfd_node_dir);
       return ret;
     }
     ret =
         node->get_property_value(kKFDNodePropDOMAINStr, &kfd_gpu_node_domain);
     if (ret != 0) {
+      std::cerr << "Failed to get \"domain\" properity from properties "
+              "files for kfd node " << node->node_index() << "." << std::endl;
       closedir(kfd_node_dir);
       return ret;
     }
@@ -577,6 +581,10 @@ int DiscoverKFDNodes(std::map<uint64_t, std::shared_ptr<KFDNode>> *nodes) {
   }
 
   if (closedir(kfd_node_dir)) {
+    std::string err_str = "Failed to close KFD node directory ";
+    err_str += kKFDNodesPathRoot;
+    err_str += ".";
+    perror(err_str.c_str());
     return 1;
   }
   return 0;

projects/rocm-smi-lib/src/rocm_smi_main.cc

@@ -56,6 +56,7 @@
 #include <functional>
 #include <cerrno>
 #include <unordered_map>
+#include <iostream>
 
 #include "rocm_smi/rocm_smi.h"
 #include "rocm_smi/rocm_smi_device.h"
@@ -230,6 +231,7 @@ static uint32_t GetMonitorDevices(const std::shared_ptr<amd::smi::Device> &d,
   if (d->monitor() != nullptr) {
     // Calculate BDFID and set for this device
     if (ConstructBDFID(d->path(), &bdfid) != 0) {
+      std::cerr << "Failed to construct BDFID." << std::endl;
       return 1;
     }
     d->set_bdfid(bdfid);
@@ -458,6 +460,7 @@ static bool isAMDGPU(std::string dev_path) {
 
 uint32_t RocmSMI::DiscoverAmdgpuDevices(void) {
   uint32_t ret = 0;
+  std::string err_msg;
 
   // If this gets called more than once, clear previous findings.
   devices_.clear();
@@ -471,6 +474,10 @@ uint32_t RocmSMI::DiscoverAmdgpuDevices(void) {
 
   auto drm_dir = opendir(kPathDRMRoot);
   if (drm_dir == nullptr) {
+    err_msg = "Failed to open drm root directory ";
+    err_msg += kPathDRMRoot;
+    err_msg += ".";
+    perror(err_msg.c_str());
     return 1;
   }
 
@@ -492,6 +499,10 @@ uint32_t RocmSMI::DiscoverAmdgpuDevices(void) {
   }
 
   if (closedir(drm_dir)) {
+    err_msg = "Failed to close drm root directory ";
+    err_msg += kPathDRMRoot;
+    err_msg += ".";
+    perror(err_msg.c_str());
     return 1;
   }
   return 0;
@@ -504,6 +515,7 @@ uint32_t RocmSMI::DiscoverAMDMonitors(void) {
 
   std::string mon_name;
   std::string tmp;
+  std::string err_msg;
 
   while (dentry != nullptr) {
     if (dentry->d_name[0] == '.') {
@@ -521,7 +533,11 @@ uint32_t RocmSMI::DiscoverAMDMonitors(void) {
       fs.open(tmp);
 
       if (!fs.is_open()) {
-          return 1;
+        err_msg = "Failed to open monitor file ";
+        err_msg += tmp;
+        err_msg += ".";
+        perror(err_msg.c_str());
+        return 1;
       }
       std::string mon_type;
       fs >> mon_type;
@@ -539,6 +555,10 @@ uint32_t RocmSMI::DiscoverAMDMonitors(void) {
   }
 
   if (closedir(mon_dir)) {
+    err_msg = "Failed to close monitor directory ";
+    err_msg += kPathHWMonRoot;
+    err_msg += ".";
+    perror(err_msg.c_str());
     return 1;
   }
   return 0;