Added --decode argument to amd-smi so we can parse cper files directly

amdsmi_cli/amdsmi_commands.py

@@ -7190,7 +7190,7 @@ def partition(self, args, multiple_devices=False, gpu=None, current=None, memory
                         output_file.write(legend_output + '\n')
 
 
-    def ras(self, args, multiple_devices=False, gpu=None, cper=None, afid=None,
+    def ras(self, args, multiple_devices=False, gpu=None, cper=None, afid=None, decode=None,
             severity=None, folder=None, file_limit=None, cper_file=None, follow=None):
         """
         Retrieve and process CPER (RAS) entries for a target GPU.
@@ -7210,6 +7210,8 @@ def ras(self, args, multiple_devices=False, gpu=None, cper=None, afid=None,
             args.cper = cper
         if afid:
             args.afid = afid
+        if decode:
+            args.decode = decode
         if severity:
             args.severity = severity
         if folder:
@@ -7225,7 +7227,11 @@ def ras(self, args, multiple_devices=False, gpu=None, cper=None, afid=None,
 
         if args.afid:
             if args.cper_file:
-                afids = self.helpers.pvtDumpAfids(args.cper_file)
+                if args.decode:
+                    args.cursor = [0]
+                    self.helpers.ras_cper(args, None, self.logger, 0)
+                    return
+                afids = self.helpers.cper_dump_afids(args.cper_file)
                 print(' '.join(map(str, afids)))
                 return
             else:

amdsmi_cli/amdsmi_helpers.py

@@ -34,6 +34,7 @@
 import pwd
 import stat
 from typing import Tuple, Optional, Union
+import tempfile
 
 from enum import Enum
 from pathlib import Path
@@ -1471,7 +1472,9 @@ def display_cper_files_generated(self, entries, device_handle, folder):
         for entry_index, entry in enumerate(entries.values()):
             # Assume 'entry' is a dictionary with keys: "error_severity" and "notify_type".
             timestamp = entry.get("timestamp", "unknown")
-            gpu_id = self.get_gpu_id_from_device_handle(device_handle)
+            gpu_id = '-'
+            if not isinstance(device_handle, Path):
+                gpu_id = self.get_gpu_id_from_device_handle(device_handle)
             prefix = self._severity_as_string(entry.get("error_severity", "Unknown"),
                                               entry.get("notify_type", "Unknown"),
                                               False)
@@ -1481,7 +1484,7 @@ def display_cper_files_generated(self, entries, device_handle, folder):
                                                 entry.get("notify_type", "Unknown"),
                                                 True)
                 cper_data_file = f"{prefix}_{self.get_cper_count() + 1}.cper"
-                afids = self.pvtDumpAfids(cper_data_file)
+                afids = self.cper_dump_afids(cper_data_file)
                 afids_str = ' '.join(map(str, afids))
                 output += f" {cper_data_file:<17} {afids_str}"
 
@@ -1494,7 +1497,8 @@ def _print_header(self, folder):
             print(f" {'file_name':<17} {'list of afids'}", end="")
         print("")
 
-    def dump_cper_entries(self, folder, entries, cper_data, device_handle, file_limit=None):
+
+    def dump_cper_entries(self, folder, entries, cper_data, device_handle, file_limit=None, cper_file=None):
         """
         Dump CPER entries to files in the specified folder. Handles batch deletion if file limit is exceeded.
 
@@ -1504,6 +1508,7 @@ def dump_cper_entries(self, folder, entries, cper_data, device_handle, file_limi
         cper_data (list): List of CPER data objects with 'bytes' and 'size' keys.
         device_handle: Device handle for GPU identification.
         file_limit (int, optional): Maximum number of files to retain in the folder.
+        cper_file (str, optional): cper file name to use when saving to folder
         """
         # Initialize header display
         if not getattr(self, "_cper_display_initialized", False):
@@ -1524,7 +1529,10 @@ def dump_cper_entries(self, folder, entries, cper_data, device_handle, file_limi
 
                 # Generate filenames
                 count = self.get_cper_count() + 1
-                cper_name = f"{prefix}-{count}.cper"
+                if cper_file:
+                    cper_name = cper_file
+                else:
+                   cper_name = f"{prefix}-{count}.cper"
                 json_name = f"{prefix}-{count}.json"
                 cper_path = folder / cper_name
                 json_path = folder / json_name
@@ -1553,7 +1561,9 @@ def dump_cper_entries(self, folder, entries, cper_data, device_handle, file_limi
 
                 # Collect data for printing
                 timestamp = entry.get("timestamp", "unknown")
-                gpu_id = self.get_gpu_id_from_device_handle(device_handle)
+                gpu_id = '-'
+                if not isinstance(device_handle, Path):
+                    gpu_id = self.get_gpu_id_from_device_handle(device_handle)
                 severity = self._severity_as_string(error_severity, notify_type, False)
                 output_rows[cper_path] = [timestamp, gpu_id, severity, cper_name]
                 self.increment_cper_count()
@@ -1576,7 +1586,7 @@ def dump_cper_entries(self, folder, entries, cper_data, device_handle, file_limi
             for cper_path, row in output_rows.items():
                 timestamp, gpu_id, severity, fname = row
                 try:
-                    afids = self.pvtDumpAfids(cper_path)
+                    afids = self.cper_dump_afids(cper_path)
                     afids_str = ' '.join(map(str, afids))
                 except Exception as e:
                     afids_str = "Error fetching AFIDs"
@@ -1593,6 +1603,26 @@ def dump_cper_entries(self, folder, entries, cper_data, device_handle, file_limi
                 ))
             except Exception as e:
                 logging.debug(f"Failed to dump entries as JSON: {e}")
+
+    def dump_cper_entries_as_json(self, entries, _cper_data, _device_handle):
+        """
+        Return the CPER entries as a formatted JSON string and print it.
+        Parameters largely mirror dump_cper_entries so that callers can reuse the same argument list.
+        Unused arguments (_cper_data, _device_handle) are retained for API symmetry.
+        Returns:
+        str: The JSON representation of the CPER entries, or an empty string on failure.
+        """
+        try:
+            entries_json = json.dumps(
+                entries,
+                indent=2,
+                default=lambda o: o.decode("utf-8") if isinstance(o, bytes) else o,
+            )
+            print(entries_json)
+            return entries_json
+        except Exception as e:
+            logging.debug(f"Failed to serialize CPER entries as JSON: {e}")
+            return ""
 
     def write_binary(self, data, size, filepath):
         """
@@ -1652,7 +1682,7 @@ def binary_to_hexdump_string(self, data: Union[bytes, List[int]]) -> str:
 
         return "\n".join(lines)
 
-    def pvtDumpAfids(self, cper_file):
+    def cper_dump_afids(self, cper_file):
         # 1) Fetch the CPER “file” and ensure we have raw bytes
         raw_data = cper_file
         if hasattr(raw_data, "read"):
@@ -1743,14 +1773,16 @@ def ras_cper(self, args, device_handle, logger, gpu_idx):
 
         buffer_size = 1048576
 
-        gpu_id = self.get_gpu_id_from_device_handle(device_handle)
-        if args.follow and not getattr(self, "_cper_follow_prompted", False):
-            print("Press CTRL + C to stop.")
-            self._cper_follow_prompted = True
-
-        primary_partition = self.is_primary_partition(device_handle, gpu_id)
-        if not primary_partition:
-            return
+        if args.decode and args.cper_file:
+            device_handle = args.cper_file
+        else:
+            gpu_id = self.get_gpu_id_from_device_handle(device_handle)
+            if args.follow and not getattr(self, "_cper_follow_prompted", False):
+                print("Press CTRL + C to stop.")
+                self._cper_follow_prompted = True
+            primary_partition = self.is_primary_partition(device_handle, gpu_id)
+            if not primary_partition:
+                return
 
         if args.folder and not getattr(self, "_cper_folder_prompted", False):
             self._cper_folder_prompted = True
@@ -1759,6 +1791,7 @@ def ras_cper(self, args, device_handle, logger, gpu_idx):
         self.stop = False
 
         num_entries = 0
+        entries = {}
         while True:
             try:
                 entries, new_cursor, cper_data, status_code = amdsmi_interface.amdsmi_get_gpu_cper_entries(
@@ -1779,7 +1812,15 @@ def ras_cper(self, args, device_handle, logger, gpu_idx):
             args.cursor[gpu_idx] = new_cursor
             if len(entries) == 0:
                 break
-            if args.folder:
+            if args.decode and args.cper_file:
+                if args.json:
+                    self.dump_cper_entries_as_json(entries, cper_data, device_handle)
+                elif args.folder:
+                    self.dump_cper_entries(args.folder, entries, cper_data, device_handle, args.file_limit)
+                else:
+                     with tempfile.TemporaryDirectory() as tmp_dir:
+                        self.dump_cper_entries(tmp_dir, entries, cper_data, device_handle, args.file_limit, os.path.basename(args.cper_file))
+            elif args.folder:
                 self.dump_cper_entries(args.folder, entries, cper_data, device_handle, args.file_limit)
             else:
                 self.display_cper_files_generated(entries, device_handle, args.folder)

amdsmi_cli/amdsmi_parser.py

@@ -1534,6 +1534,7 @@ def _add_ras_parser(self, subparsers: argparse._SubParsersAction, func):
         # Help text for RAS arguments
         cper_help = "Trigger current CPER data retrieval"
         afid_help = "Generate an AFID (AMD Field ID) given a CPER record file"
+        decode_help = "Decode out-of-band CPER files captured by or collected from other systems"
         severity_choices = ["nonfatal-uncorrected", "fatal", "nonfatal-corrected", "all"]
         severity_choices_str = ", ".join(severity_choices)
         severity_help = f"Set the SEVERITY filters from the following:\n    {severity_choices_str}"
@@ -1562,6 +1563,7 @@ def _add_ras_parser(self, subparsers: argparse._SubParsersAction, func):
         # AFID Arguments
         afid_group = ras_parser.add_argument_group("AFID Arguments")
         afid_group.add_argument("--cper-file", action=self._check_cper_file_path(), metavar="CPER_FILE", help=cper_file_help)
+        afid_group.add_argument("--decode", action="store_true", help=decode_help)
 
         # Add common modifiers and device selection arguments.
         self._add_device_arguments(ras_parser, required=False)

cmake.sh

@@ -0,0 +1,5 @@
+#!/bin/bash -xe
+  set -xe
+  cmake   -DCMAKE_BUILD_TYPE=Debug   -DCMAKE_INSTALL_PREFIX=/opt/rocm-6.5.0   -DBUILD_TESTS=ON   ..
+  #-DBUILD_WRAPPER=ON   #
+

py-interface/amdsmi_interface.py

@@ -2713,15 +2713,20 @@ def notifyTypeToString(notify_type_b):
     return "".join(guid[::-1])
 
 def amdsmi_get_gpu_cper_entries(
-    processor_handle: processor_handle_t,
+    device_handle: amdsmi_wrapper.amdsmi_processor_handle | Path,
+    # processor_handle: Union[amdsmi_wrapper.amdsmi_processor_handle, str],
     severity_mask: int,
     buffer_size: int = 4 * 1048576,
     cursor: int = 0
 ) -> Tuple[Dict[str, Any], int, List[Dict[str, Any]], int]:
 
-    if not isinstance(processor_handle, amdsmi_wrapper.amdsmi_processor_handle):
+    if isinstance(device_handle, Path):
+        if not os.path.isfile(device_handle):
+            raise AmdSmiParameterException(device_handle, str)
+        device_handle = ctypes.c_char_p(str(device_handle).encode("utf-8"))
+    elif not isinstance(device_handle, amdsmi_wrapper.amdsmi_processor_handle):
         raise AmdSmiParameterException(
-            processor_handle, amdsmi_wrapper.amdsmi_processor_handle
+            device_handle, amdsmi_wrapper.amdsmi_processor_handle
         )
 
     # Allocate a buffer for CPER data.
@@ -2737,7 +2742,7 @@ def amdsmi_get_gpu_cper_entries(
 
     # Call the underlying AMD-SMI API.
     status_code = amdsmi_wrapper.amdsmi_get_gpu_cper_entries(
-        processor_handle,
+        device_handle,
         ctypes.c_uint32(severity_mask),
         buf,
         ctypes.byref(buf_size),

src/amd_smi/amd_smi.cc

@@ -4404,6 +4404,12 @@ amdsmi_get_gpu_cper_entries(
     uint64_t *entry_count,
     uint64_t *cursor) {
 
+    std::string path;
+    if(amd::smi::FileExists(static_cast<char const *>(processor_handle))) {
+        path = std::string(static_cast<char const *>(processor_handle));
+    }
+    else {
+
     AMDSMI_CHECK_INIT();
     if (!amd::smi::is_sudo_user()) {
         return AMDSMI_STATUS_NO_PERM;
@@ -4414,10 +4420,10 @@ amdsmi_get_gpu_cper_entries(
     if (status != AMDSMI_STATUS_SUCCESS) {
         return status;
     }
-
-    std::string path = std::string("/sys/kernel/debug/dri/") +
+    path = std::string("/sys/kernel/debug/dri/") +
         std::to_string(gpu_device->get_card_id()) +
         "/amdgpu_ring_cper";
+    }
 
     return amdsmi_get_gpu_cper_entries_by_path(
         path.c_str(),