[SWDEV-554697] CPER not properly displaying warnings for non-zero partition id's (#687)

* Get primary gpu_id for non-primary partitions.

Signed-off-by: Bindhiya Kanangot Balakrishnan <Bindhiya.KanangotBalakrishnan@amd.com>

* corrected partitions warning print logic

Signed-off-by: Maisam Arif <Maisam.Arif@amd.com>
Change-Id: I08be6c78ddd46e5316dc9d538de4908b65b21d43

* Updated patch with latest changes and modified
xgmi partition_id check.

Signed-off-by: Bindhiya Kanangot Balakrishnan <Bindhiya.KanangotBalakrishnan@amd.com>

* Typo correction

Signed-off-by: Bindhiya Kanangot Balakrishnan <Bindhiya.KanangotBalakrishnan@amd.com>

* adjusted logging

Signed-off-by: Maisam Arif <Maisam.Arif@amd.com>
Change-Id: I6d425102d8583aabbcd4d7f55c9c733428524d59

---------

Signed-off-by: Bindhiya Kanangot Balakrishnan <Bindhiya.KanangotBalakrishnan@amd.com>
Signed-off-by: Maisam Arif <Maisam.Arif@amd.com>
Co-authored-by: Oosman Saeed <oossaeed@amd.com>
Co-authored-by: Bindhiya Kanangot Balakrishnan <Bindhiya.KanangotBalakrishnan@amd.com>
Co-authored-by: Maisam Arif <Maisam.Arif@amd.com>

[ROCm/amdsmi commit: 5398eaa6b3]

projects/amdsmi/amdsmi_cli/amdsmi_commands.py

@@ -6429,15 +6429,9 @@ class AMDSMICommands():
         # Populate the possible gpus and their bdfs
         xgmi_values = []
         for gpu in args.gpu:
-            partition_id = -1
+            primary_partition = self.helpers.is_primary_partition(gpu)
-            try:
+            if not primary_partition:
-                kfd_info = amdsmi_interface.amdsmi_get_gpu_kfd_info(gpu)
+                logging.debug(f"Skipping xgmi command due to non zero partition {gpu}")
-                partition_id = kfd_info['current_partition_id']
-            except amdsmi_exception.AmdSmiLibraryException as e:
-                logging.debug("Failed to get kfd info for gpu %s | %s", gpu, e.get_error_info())
-
-            if partition_id != 0:
-                logging.debug(f"Skipping xgmi command due to non zero partition {gpu} - {partition_id}")
                 continue
 
             logging.debug("check1 device_handle: %s", gpu)
@@ -6497,14 +6491,8 @@ class AMDSMICommands():
 
                 # Populate link metrics
                 for dest_gpu in args.gpu:
-                    partition_id = -1
+                    primary_partition = self.helpers.is_primary_partition(dest_gpu)
-                    try:
+                    if not primary_partition:
-                        kfd_info = amdsmi_interface.amdsmi_get_gpu_kfd_info(dest_gpu)
-                        partition_id = kfd_info['current_partition_id']
-                    except amdsmi_exception.AmdSmiLibraryException as e:
-                        logging.debug("Failed to get kfd info for gpu %s | %s", dest_gpu, e.get_error_info())
-
-                    if partition_id != 0:
                         continue
 
                     dest_gpu_id = self.helpers.get_gpu_id_from_device_handle(dest_gpu)
@@ -7077,6 +7065,36 @@ class AMDSMICommands():
             args.gpu = [args.gpu]
 
         args.cursor = [0] * len(args.gpu)
+
+        # Using all the devices given in args.gpu
+        # Populate a list of all the primary partition GPU ids (GPU 0, GPU 1, etc)
+        partition_warning_flag = True
+        primary_partition_gpu_ids = set() # set of all primary partition GPU ids from arg.gpu
+        for device_handle in args.gpu:
+            # First get the partition
+            partition_id = self.helpers.get_partition_id(device_handle)
+            # If there is a single primary partition within args.gpu then we don't need to print the warning
+            if partition_id == 0:
+                partition_warning_flag = False
+                break
+            # Then attempt to get the primary GPU id for that partition
+            primary_partition_gpu_id = self.helpers.get_primary_partition_gpu_id(device_handle)
+            # Add to the set if it's a non-primary partition and we found a valid primary GPU id
+            if partition_id != 0 and primary_partition_gpu_id is not None:
+                primary_partition_gpu_ids.add(primary_partition_gpu_id)
+
+        if partition_warning_flag:
+            # Create a list of the primary partitions
+            primary_partitions_str = " ".join(f"GPU{gpu_id}" for gpu_id in primary_partition_gpu_ids)
+
+            print("WARNING: CPER files are only available on primary partitions")
+            if len(primary_partition_gpu_ids) > 1:
+                print(f"Try with primary partitions {primary_partitions_str}",end="")
+            else:
+                print(f"Try with primary partition {primary_partitions_str}",end="")
+
+            print()
+
         while True:
             for idx, device_handle in enumerate(args.gpu):
                 self.helpers.ras_cper(args, device_handle, self.logger, idx)

projects/amdsmi/amdsmi_cli/amdsmi_helpers.py

@@ -1188,7 +1188,7 @@ class AMDSMIHelpers():
         if not getattr(self, "_cper_display_initialized", False):
             # Warning if no folder was specified elsewhere
             if not getattr(self, "_cper_warning_printed", False):
-               print(f"WARNING:No CPER files will be dumped unless --folder=<folder_name> is specified and cper entries exist.")
+               print(f"WARNING: No CPER files will be dumped unless --folder=<folder_name> is specified and cper entries exist.")
                self._cper_warning_printed = True
 
             self._print_header(folder)
@@ -1410,6 +1410,43 @@ class AMDSMIHelpers():
             else:
                 raise ValueError("Unexpected Error getting afids from CPER file") from e
 
+    def get_partition_id(self, device_handle, gpu_id = None) -> int:
+        partition_id = -1
+        try:
+            kfd_info = amdsmi_interface.amdsmi_get_gpu_kfd_info(device_handle)
+            partition_id = kfd_info['current_partition_id']
+        except amdsmi_exception.AmdSmiLibraryException as e:
+            logging.debug("Failed to get kfd info for gpu %s | %s", gpu_id, e.get_error_info())
+        return partition_id
+
+    def get_primary_partition_gpu_id(self, device_handle) -> Union[int, None]:
+        try:
+            bdf = amdsmi_interface.amdsmi_get_gpu_device_bdf(device_handle)
+            if bdf is None:
+                logging.debug("Failed to get device BDF: BDF is None")
+                return None
+            # Construct primary partition BDF (base + ".0" for function 0)
+            primary_bdf = bdf[:10] + ".0"
+            try:
+                primary_device_handle = amdsmi_interface.amdsmi_get_processor_handle_from_bdf(primary_bdf)
+                partition_id = self.get_partition_id(primary_device_handle)
+                if partition_id == 0:
+                    return self.get_gpu_id_from_device_handle(primary_device_handle)
+                return None
+            except amdsmi_exception.AmdSmiLibraryException as e:
+                logging.debug("Failed to get primary partition device handle with BDF %s: %s", primary_bdf, e.get_error_info())
+                return None
+        except amdsmi_exception.AmdSmiLibraryException as e:
+            logging.debug("Failed to get partition device BDF: %s", e.get_error_info())
+            return None
+
+    def is_primary_partition(self, device_handle, gpu_id = None) -> bool:
+        partition_id = self.get_partition_id(device_handle, gpu_id)
+        if partition_id != 0:
+            logging.debug(f"Skipping gpu {gpu_id} on non zero partition {partition_id}")
+            return False
+        return True
+
     def ras_cper(self, args, device_handle, logger, gpu_idx):
         # Parse severity mask dynamically from the --severity option.
         severity_mask = 0
@@ -1437,15 +1474,8 @@ class AMDSMIHelpers():
             print("Press CTRL + C to stop.")
             self._cper_follow_prompted = True
 
-        partition_id = -1
+        primary_partition = self.is_primary_partition(device_handle, gpu_id)
-        try:
+        if not primary_partition:
-            kfd_info = amdsmi_interface.amdsmi_get_gpu_kfd_info(device_handle)
-            partition_id = kfd_info['current_partition_id']
-        except amdsmi_exception.AmdSmiLibraryException as e:
-            logging.debug("Failed to get kfd info for gpu %s | %s", gpu_id, e.get_error_info())
-
-        if partition_id != 0:
-            logging.debug(f"Skipping gpu {gpu_id} on non zero partition {partition_id}")
             return
 
         if args.folder and not getattr(self, "_cper_folder_prompted", False):