Merge "Return the driver loading status" into amd-dev

include/amd_smi/amdsmi.h

@@ -178,6 +178,7 @@ typedef enum {
     AMDSMI_STATUS_NOT_FOUND = 31,  //!< Device Not found
     AMDSMI_STATUS_NOT_INIT = 32,  //!< Device not initialized
     AMDSMI_STATUS_NO_SLOT = 33,  //!< No more free slot
+    AMDSMI_STATUS_DRIVER_NOT_LOADED = 34,  //!< Processor driver not loaded
     // Data and size errors
     AMDSMI_STATUS_NO_DATA = 40,  //!< No data was found for a given input
     AMDSMI_STATUS_INSUFFICIENT_SIZE = 41,  //!< Not enough resources were available for the operation

rocm_smi/include/rocm_smi/rocm_smi.h

@@ -150,6 +150,20 @@ typedef enum {
   RSMI_INIT_FLAG_RESRV_TEST1 = 0x800000000000000,  //!< Reserved for test
 } rsmi_init_flags_t;
 
+/**
+ * @brief Driver loading status
+ *
+ * The driver loading status from initState sysfs
+ */
+
+typedef enum {
+  RSMI_DRIVER_NOT_FOUND   = 0,             //!< Cannot find the driver
+  RSMI_DRIVER_MODULE_STATE_LIVE,           //!< Driver loaded and live
+  RSMI_DRIVER_MODULE_STATE_LOADING,        //!< Driver is loading(coming)
+  RSMI_DRIVER_MODULE_STATE_UNLOADING,      //!< Driver is unloading(going)
+  RSMI_DRIVER_MODULE_STATE_UNKNOWN,        //!< Driver state unknown
+} rsmi_driver_state_t;
+
 /**
  * @brief PowerPlay performance levels
  */
@@ -1038,6 +1052,13 @@ rsmi_status_t rsmi_init(uint64_t init_flags);
  */
 rsmi_status_t rsmi_shut_down(void);
 
+/**
+ *  @brief Get driver loading status
+ *
+ *  @details The status could be not found, live, loading, unloading.
+ */
+rsmi_status_t rsmi_driver_status(rsmi_driver_state_t* state);
+
 /** @} */  // end of InitShut
 
 /*****************************************************************************/

rocm_smi/src/rocm_smi.cc

@@ -508,6 +508,33 @@ rsmi_shut_down(void) {
   CATCH
 }
 
+
+rsmi_status_t rsmi_driver_status(rsmi_driver_state_t* state) {
+  TRY
+  if (state == nullptr) {
+    return RSMI_STATUS_INVALID_ARGS;
+  }
+
+  // live, coming, going
+  static const char *kDevInitStateID = "/sys/module/amdgpu/initstate";
+  std::ifstream infile(kDevInitStateID);
+  if (!infile) {
+    *state = RSMI_DRIVER_NOT_FOUND;
+    return RSMI_STATUS_SUCCESS;
+  }
+
+  std::string stat_str;
+  infile >> stat_str;
+
+  *state = RSMI_DRIVER_MODULE_STATE_UNKNOWN;
+  if (stat_str == "live") *state = RSMI_DRIVER_MODULE_STATE_LIVE;
+  if (stat_str == "coming") *state = RSMI_DRIVER_MODULE_STATE_LOADING;
+  if (stat_str == "going") *state = RSMI_DRIVER_MODULE_STATE_UNLOADING;
+
+  return RSMI_STATUS_SUCCESS;
+  CATCH
+}
+
 rsmi_status_t
 rsmi_num_monitor_devices(uint32_t *num_devices) {
   TRY

src/amd_smi/amd_smi_system.cc

@@ -208,8 +208,13 @@ amdsmi_status_t AMDSmiSystem::get_cpu_model(uint32_t cpu_model) {
 
 amdsmi_status_t AMDSmiSystem::populate_amd_gpu_devices() {
     // init rsmi
+    rsmi_driver_state_t state;
     rsmi_status_t ret = rsmi_init(0);
     if (ret != RSMI_STATUS_SUCCESS) {
+        if (rsmi_driver_status(&state) == RSMI_STATUS_SUCCESS &&
+                state != RSMI_DRIVER_MODULE_STATE_LIVE) {
+            return AMDSMI_STATUS_DRIVER_NOT_LOADED;
+        }
         return amd::smi::rsmi_to_amdsmi_status(ret);
     }