[SWDEV-529889] Fixed incorrect vendor_id reporting in amdsmi_get_gpu_asic_info

Signed-off-by: gabrpham_amdeng <Gabriel.Pham@amd.com>


[ROCm/amdsmi commit: 1ab57ce7dd]

projects/amdsmi/rocm_smi/src/rocm_smi.cc

@@ -1012,11 +1012,37 @@ rsmi_dev_subsystem_id_get(uint32_t dv_ind, uint16_t *id) {
 
 rsmi_status_t
 rsmi_dev_vendor_id_get(uint32_t dv_ind, uint16_t *id) {
+    // need to get this to fall back to kfd if sysfs doesn't work
   std::ostringstream ss;
   ss << __PRETTY_FUNCTION__ << "| ======= start =======";
   LOG_TRACE(ss);
   CHK_SUPPORT_NAME_ONLY(id)
-  return get_id(dv_ind, amd::smi::kDevVendorID, id);
+  rsmi_status_t status = get_id(dv_ind, amd::smi::kDevVendorID, id);
+  if (status != RSMI_STATUS_SUCCESS)
+  {
+    // /sys/class/kfd/kfd/topology/nodes/*/properties
+    GET_DEV_AND_KFDNODE_FROM_INDX
+    uint32_t node_id;
+    uint64_t kfd_vendor_id;
+    int ret_kfd = kfd_node->get_node_id(&node_id);
+    ret_kfd = amd::smi::read_node_properties(node_id, "vendor_id", &kfd_vendor_id);
+    if (ret_kfd == 0) {
+      *id = kfd_vendor_id;
+      status = RSMI_STATUS_SUCCESS;
+    } else {
+      *id = std::numeric_limits<uint16_t>::max();
+      status = RSMI_STATUS_NOT_SUPPORTED;
+    }
+    ss << __PRETTY_FUNCTION__
+       << " | Issue: Could not read device from sysfs, falling back to KFD" << "\n"
+       << " ; Device #: " << std::to_string(dv_ind) << "\n"
+       << " ; ret_kfd: " << std::to_string(ret_kfd) << "\n"
+       << " ; node: " << std::to_string(node_id) << "\n"
+       << " ; Data: vendor_id (from KFD)= " << std::to_string(*id) << "\n"
+       << " ; ret = " << getRSMIStatusString(status, false);
+    LOG_DEBUG(ss);
+  }
+  return status;
 }
 
 rsmi_status_t

projects/amdsmi/src/amd_smi/amd_smi.cc

@@ -1353,35 +1353,44 @@ amdsmi_get_gpu_asic_info(amdsmi_processor_handle processor_handle, amdsmi_asic_i
 
     amdsmi_status_t status;
     amd::smi::AMDSmiSystem::getInstance().init_drm();
-    if (gpu_device->check_if_drm_is_supported()) {
-        status = gpu_device->amdgpu_query_info(AMDGPU_INFO_DEV_INFO,
-            sizeof(struct drm_amdgpu_info_device), &dev_info);
-        ss << __PRETTY_FUNCTION__
-        << " | amdgpu_query_info(): "
-        << smi_amdgpu_get_status_string(status, true);
-        LOG_INFO(ss);
-        if (status != AMDSMI_STATUS_SUCCESS) {
-            amd::smi::AMDSmiSystem::getInstance().clean_up_drm();
-            return status;
-        }
-        SMIGPUDEVICE_MUTEX(gpu_device->get_mutex())
-        status = smi_amdgpu_get_market_name_from_dev_id(gpu_device, info->market_name);
-        if (status != AMDSMI_STATUS_SUCCESS) {
-            rsmi_wrapper(rsmi_dev_brand_get, processor_handle, 0,
-                         info->market_name, AMDSMI_256_LENGTH);
-        }
+    // removing drm check for now due to drm issues
+    // if (gpu_device->check_if_drm_is_supported()) {
+    //     status = gpu_device->amdgpu_query_info(AMDGPU_INFO_DEV_INFO,
+    //         sizeof(struct drm_amdgpu_info_device), &dev_info);
+    //     ss << __PRETTY_FUNCTION__
+    //     << " | amdgpu_query_info(): "
+    //     << smi_amdgpu_get_status_string(status, true);
+    //     LOG_INFO(ss);
+    //     if (status != AMDSMI_STATUS_SUCCESS) {
+    //         amd::smi::AMDSmiSystem::getInstance().clean_up_drm();
+    //         return status;
+    //     }
+    //     SMIGPUDEVICE_MUTEX(gpu_device->get_mutex())
+    //     status = smi_amdgpu_get_market_name_from_dev_id(gpu_device, info->market_name);
+    //     if (status != AMDSMI_STATUS_SUCCESS) {
+    //         rsmi_wrapper(rsmi_dev_brand_get, processor_handle, 0,
+    //                      info->market_name, AMDSMI_256_LENGTH);
+    //     }
 
-        info->device_id = dev_info.device_id;
-        info->rev_id = dev_info.pci_rev;
-        info->vendor_id = gpu_device->get_vendor_id();
-    } else {
-        status = rsmi_wrapper(rsmi_dev_brand_get, processor_handle, 0,
-                info->market_name, AMDSMI_256_LENGTH);
+    //     info->device_id = dev_info.device_id;
+    //     info->rev_id = dev_info.pci_rev;
+    //     info->vendor_id = gpu_device->get_vendor_id();
+    // } else {
+    uint16_t device_id = std::numeric_limits<uint16_t>::max();
+    status = rsmi_wrapper(rsmi_dev_id_get, processor_handle, 0, &device_id);
+    info->device_id = static_cast<uint64_t>(device_id);
 
-        status = rsmi_wrapper(rsmi_dev_vendor_id_get, processor_handle, 0,
-                              &vendor_id);
-        if (status == AMDSMI_STATUS_SUCCESS) info->vendor_id = vendor_id;
-    }
+    uint16_t rev_id = std::numeric_limits<uint16_t>::max();
+    status = rsmi_wrapper(rsmi_dev_revision_get, processor_handle, 0, &rev_id);
+    info->rev_id = static_cast<uint32_t>(rev_id);
+
+    status = rsmi_wrapper(rsmi_dev_brand_get, processor_handle, 0,
+            info->market_name, AMDSMI_256_LENGTH);
+
+    status = rsmi_wrapper(rsmi_dev_vendor_id_get, processor_handle, 0,
+                            &vendor_id);
+    if (status == AMDSMI_STATUS_SUCCESS) info->vendor_id = vendor_id;
+    // }
     // For other sysfs related information, get from rocm-smi
 
     // Ensure asic_serial defaults to an unsupported value