Add support for reading GPU's unique ID

Add support and testing for reading the Unique ID associated with a
specific GPU. This ID will persist across reboots, even if the GPU is
moved to a different machine. Note that this is per-GPU, not per-card,
as some cards have multiple GPUs, and each GPU will get a unique
identifier

Change-Id: Idce50c6febc2ceb1a4c1200d2489ec8b9d8fe174

include/rocm_smi/rocm_smi.h

@@ -726,6 +726,22 @@ rsmi_dev_subsystem_name_get(uint32_t dv_ind, char *name, size_t len);
  */
 rsmi_status_t rsmi_dev_subsystem_vendor_id_get(uint32_t dv_ind, uint16_t *id);
 
+/**
+ *  @brief Get Unique ID
+ *
+ *  @details Given a device index @p dv_ind and a pointer to a uint64_t @p
+ *  id, this function will write the unique ID of the GPU pointed to @p
+ *  id.
+ *
+ *  @param[in] dv_ind a device index
+ *
+ *  @param[inout] id a pointer to uint64_t to which the unique ID of the GPU
+ *  is written
+ *
+ *  @retval ::RSMI_STATUS_SUCCESS is returned upon successful call.
+ */
+rsmi_status_t rsmi_dev_unique_id_get(uint32_t dv_ind, uint64_t *id);
+
 /** @} */  // end of IDQuer
 
 /*****************************************************************************/

include/rocm_smi/rocm_smi_device.h

@@ -91,6 +91,7 @@ enum DevInfoTypes {
   kDevMemUsedVisVRAM,
   kDevMemUsedVRAM,
   kDevPCIEReplayCount,
+  kDevUniqueId,
 };
 
 class Device {

src/rocm_smi.cc

@@ -2107,3 +2107,14 @@ rsmi_dev_pci_replay_counter_get(uint32_t dv_ind, uint64_t *counter) {
   CATCH
 }
 
+rsmi_status_t
+rsmi_dev_unique_id_get(uint32_t dv_ind, uint64_t *unique_id) {
+  TRY
+  DEVICE_MUTEX
+  rsmi_status_t ret;
+
+  ret = get_dev_value_int(amd::smi::kDevUniqueId, dv_ind, unique_id);
+  return ret;
+
+  CATCH
+}

src/rocm_smi_device.cc

@@ -97,6 +97,7 @@ static const char *kDevMemUsedGTTFName = "mem_info_gtt_used";
 static const char *kDevMemUsedVisVRAMFName = "mem_info_vis_vram_used";
 static const char *kDevMemUsedVRAMFName = "mem_info_vram_used";
 static const char *kDevPCIEReplayCountFName = "pcie_replay_count";
+static const char *kDevUniqueIdFName = "unique_id";
 
 // Strings that are found within sysfs files
 static const char *kDevPerfLevelAutoStr = "auto";
@@ -138,6 +139,7 @@ static const std::map<DevInfoTypes, const char *> kDevAttribNameMap = {
     {kDevMemUsedVisVRAM, kDevMemUsedVisVRAMFName},
     {kDevMemUsedVRAM, kDevMemUsedVRAMFName},
     {kDevPCIEReplayCount, kDevPCIEReplayCountFName},
+    {kDevUniqueId, kDevUniqueIdFName},
 };
 
 static const std::map<rsmi_dev_perf_level, const char *> kDevPerfLvlMap = {
@@ -374,6 +376,11 @@ int Device::readDevInfo(DevInfoTypes type, uint64_t *val) {
       RET_IF_NONZERO(ret);
       *val = std::stoul(tempStr, 0);
       break;
+    case kDevUniqueId:
+      ret = readDevInfoStr(type, &tempStr);
+      RET_IF_NONZERO(ret);
+      *val = std::stoul(tempStr, 0, 16);
+      break;
 
     default:
       return -1;

tests/rocm_smi_test/functional/sys_info_read.cc

@@ -120,6 +120,19 @@ void TestSysInfoRead::Run(void) {
       std::cout << " (" << std::dec << val_ui64 << ")" << std::endl;
     }
 
+    err = rsmi_dev_unique_id_get(i, &val_ui64);
+    if (err == RSMI_STATUS_NOT_SUPPORTED) {
+        std::cout <<
+            "\t**rsmi_dev_unique_id() is not supported"
+            " on this machine" << std::endl;
+    } else {
+        CHK_ERR_ASRT(err)
+        IF_VERB(STANDARD) {
+            std::cout << "\t**GPU Unique ID : " << std::hex << val_ui64 <<
+            std::endl;
+        }
+    }
+
     err = rsmi_version_get(&ver);
     CHK_ERR_ASRT(err)