Add errors for existing but empty dev files

Change-Id: Iad9febc50f9b8e6085f8b605249ee884d2f134d6
Signed-off-by: Galantsev, Dmitrii <dmitrii.galantsev@amd.com>


[ROCm/amdsmi commit: a4b470fe71]

projects/amdsmi/python_smi_tools/rocm_smi.py

@@ -1791,15 +1791,19 @@ def showClocks(deviceList):
         for clk_type in sorted(rsmi_clk_names_dict):
             if rocmsmi.rsmi_dev_gpu_clk_freq_get(device, rsmi_clk_names_dict[clk_type], None) == 1:
                 ret = rocmsmi.rsmi_dev_gpu_clk_freq_get(device, rsmi_clk_names_dict[clk_type], byref(freq))
-                if rsmi_ret_ok(ret, device, 'get_clk_freq_' + clk_type, True):
-                    printLog(device, 'Supported %s frequencies on GPU%s' % (clk_type, str(device)), None)
-                    for x in range(freq.num_supported):
-                        fr = '{:>.0f}Mhz'.format(freq.frequency[x] / 1000000)
-                        if x == freq.current:
-                            printLog(device, str(x), str(fr) + ' *')
-                        else:
-                            printLog(device, str(x), str(fr))
-                    printLog(device, '', None)
+                if ret == rsmi_status_t.RSMI_STATUS_UNEXPECTED_DATA:
+                    printLog(device, 'Clock [%s] on device [%s] exists but EMPTY! Likely driver error!' % (clk_type, str(device)))
+                    continue
+                if not rsmi_ret_ok(ret, device, 'get_clk_freq_' + clk_type, True):
+                    continue
+                printLog(device, 'Supported %s frequencies on GPU%s' % (clk_type, str(device)), None)
+                for x in range(freq.num_supported):
+                    fr = '{:>.0f}Mhz'.format(freq.frequency[x] / 1000000)
+                    if x == freq.current:
+                        printLog(device, str(x), str(fr) + ' *')
+                    else:
+                        printLog(device, str(x), str(fr))
+                printLog(device, '', None)
             else:
                 logging.debug('{} frequency is unsupported on device[{}]'.format(clk_type, device))
                 printLog(device, '', None)

projects/amdsmi/src/rocm_smi.cc

@@ -3740,6 +3740,10 @@ rsmi_dev_memory_reserved_pages_get(uint32_t dv_ind, uint32_t *num_pages,
 
   ret = GetDevValueVec(amd::smi::kDevMemPageBad, dv_ind, &val_vec);
 
+  // file is empty, which is valid for no errors
+  if (ret == RSMI_STATUS_UNEXPECTED_DATA) {
+    ret = RSMI_STATUS_SUCCESS;
+  }
   if (ret == RSMI_STATUS_FILE_ERROR) {
     return RSMI_STATUS_NOT_SUPPORTED;
   }

projects/amdsmi/src/rocm_smi_device.cc

@@ -858,8 +858,8 @@ int Device::readDevInfoMultiLineStr(DevInfoTypes type,
     ss << "Read devInfoMultiLineStr for DevInfoType ("
        << RocmSMI::devInfoTypesStrings.at(type) << ")"
        << ", but contained no string lines";
-    LOG_INFO(ss);
-    return 0;
+    LOG_ERROR(ss);
+    return ENXIO;
   }
   // Remove any *trailing* empty (whitespace) lines
   while (!retVec->empty() &&
@@ -882,6 +882,7 @@ int Device::readDevInfoMultiLineStr(DevInfoTypes type,
        << RocmSMI::devInfoTypesStrings.at(type) << ")"
        << ", but lines were empty";
     LOG_INFO(ss);
+    return ENXIO;
   }
   return 0;
 }

projects/amdsmi/tests/rocm_smi_test/functional/frequencies_read.cc

@@ -123,16 +123,22 @@ void TestFrequenciesRead::Run(void) {
           // Verify api support checking functionality is working
           err = rsmi_dev_gpu_clk_freq_get(i, t, nullptr);
           ASSERT_EQ(err, RSMI_STATUS_NOT_SUPPORTED);
-        } else {
-            CHK_ERR_ASRT(err)
-            IF_VERB(STANDARD) {
-              std::cout << "\t**Supported " << name << " clock frequencies: ";
-              std::cout << f.num_supported << std::endl;
-              print_frequencies(&f);
-              // Verify api support checking functionality is working
-              err = rsmi_dev_gpu_clk_freq_get(i, t, nullptr);
-              ASSERT_EQ(err, RSMI_STATUS_INVALID_ARGS);
-            }
+        }
+
+        // special driver issue, shouldn't normally occur
+        if (err == RSMI_STATUS_UNEXPECTED_DATA) {
+          std::cerr << "WARN: Clock file [" << FreqEnumToStr(t) << "] exists on device [" << i << "] but empty!" << std::endl;
+          std::cerr << "      Likely a driver issue!" << std::endl;
+        }
+
+        CHK_ERR_ASRT(err)
+        IF_VERB(STANDARD) {
+          std::cout << "\t**Supported " << name << " clock frequencies: ";
+          std::cout << f.num_supported << std::endl;
+          print_frequencies(&f);
+          // Verify api support checking functionality is working
+          err = rsmi_dev_gpu_clk_freq_get(i, t, nullptr);
+          ASSERT_EQ(err, RSMI_STATUS_INVALID_ARGS);
         }
       };
 

projects/amdsmi/tests/rocm_smi_test/functional/frequencies_read_write.cc

@@ -114,14 +114,20 @@ void TestFrequenciesReadWrite::Run(void) {
           std::cout << "\t**Set " << FreqEnumToStr(rsmi_clk) <<
                                ": Not supported on this machine" << std::endl;
           return false;
-        } else {
-         //   CHK_ERR_ASRT(ret)
-          IF_VERB(STANDARD) {
-            std::cout << "Initial frequency for clock " <<
-                FreqEnumToStr(rsmi_clk) << " is " << f.current << std::endl;
-          }
-          return true;
         }
+
+        // special driver issue, shouldn't normally occur
+        if (ret == RSMI_STATUS_UNEXPECTED_DATA) {
+          std::cerr << "WARN: Clock file [" << FreqEnumToStr(rsmi_clk) << "] exists on device [" << dv_ind << "] but empty!" << std::endl;
+          std::cerr << "      Likely a driver issue!" << std::endl;
+        }
+
+        // CHK_ERR_ASRT(ret)
+        IF_VERB(STANDARD) {
+          std::cout << "Initial frequency for clock " <<
+              FreqEnumToStr(rsmi_clk) << " is " << f.current << std::endl;
+        }
+        return true;
       };
 
       auto freq_write = [&]() {
@@ -177,44 +183,6 @@ void TestFrequenciesReadWrite::Run(void) {
       }
       freq_write();
       CHK_ERR_ASRT(ret)
-#if 0
-      ret = rsmi_dev_gpu_clk_freq_get(dv_ind, rsmi_clk, &f);
-      CHK_ERR_ASRT(ret)
-
-      IF_VERB(STANDARD) {
-        std::cout << "Initial frequency for clock " << rsmi_clk << " is " <<
-                                                        f.current << std::endl;
-      }
-      // Set clocks to something other than the usual default of the lowest
-      // frequency.
-      freq_bitmask = 0b01100;  // Try the 3rd and 4th clocks
-
-      std::string freq_bm_str =
-              std::bitset<RSMI_MAX_NUM_FREQUENCIES>(freq_bitmask).to_string();
-
-      freq_bm_str.erase(0, std::min(freq_bm_str.find_first_not_of('0'),
-                                                       freq_bm_str.size()-1));
-
-      IF_VERB(STANDARD) {
-      std::cout << "Setting frequency mask for clock " << rsmi_clk <<
-          " to 0b" << freq_bm_str << " ..." << std::endl;
-      }
-      ret = rsmi_dev_gpu_clk_freq_set(dv_ind, rsmi_clk, freq_bitmask);
-      CHK_ERR_ASRT(ret)
-
-      ret = rsmi_dev_gpu_clk_freq_get(dv_ind, rsmi_clk, &f);
-      CHK_ERR_ASRT(ret)
-
-      IF_VERB(STANDARD) {
-        std::cout << "Frequency is now index " << f.current << std::endl;
-        std::cout << "Resetting mask to all frequencies." << std::endl;
-      }
-      ret = rsmi_dev_gpu_clk_freq_set(dv_ind, rsmi_clk, 0xFFFFFFFF);
-      CHK_ERR_ASRT(ret)
-
-      ret = rsmi_dev_perf_level_set(dv_ind, RSMI_DEV_PERF_LEVEL_AUTO);
-      CHK_ERR_ASRT(ret)
-#endif
     }
   }
 }