[SWDEV-560235] Add gpu and base board temperatures to monitor

Added gpu_board and base_board temperatures non-default
monitor temperature.

Signed-off-by: Bindhiya Kanangot Balakrishnan <[email protected]>

Author: bkanango

projects/amdsmi/CHANGELOG.md

@@ -8,6 +8,19 @@ Full documentation for amd_smi_lib is available at [https://rocm.docs.amd.com/pr
 
 ### Added
 
+- **Added GPU and base board temperature `amd-smi monitor` CLI support**.  
+  - Added `--gpu-board-temps` option to `amd-smi monitor` command for GPU board temperature sensors
+  - Added `--base-board-temps` option to `amd-smi monitor` command for base board temperature sensors
+
+- **Added Node Power Management (NPM) support**.  
+  - Added new Node Power Management APIs and CLI for node monitoring
+  - Added C API functions:
+    - `amdsmi_get_node_handle()`: Get handle for node devices
+    - `amdsmi_get_npm_info()`: Retrieve Node Power Management information
+  - Added Python API wrappers for new node device functions
+  - Added `amd-smi node` CLI command for Node Power Management operations
+  - Currently supported for OAM_ID 0 only.
+
 - **Added the following C API's to amdsmi_interface.py**.  
   - amdsmi_get_cpu_handle()
   - amdsmi_get_esmi_err_msg()

projects/amdsmi/amdsmi_cli/amdsmi_commands.py

@@ -5607,8 +5607,9 @@ def reset(self, args, multiple_devices=False, gpu=None, gpureset=None,
 
     def monitor(self, args, multiple_devices=False, watching_output=False, gpu=None,
                     watch=None, watch_time=None, iterations=None, power_usage=None,
-                    temperature=None, gfx_util=None, mem_util=None, encoder=None,
-                    decoder=None, ecc=None, vram_usage=None, pcie=None, process=None,
+                    temperature=None, base_board_temps=None, gpu_board_temps=None,
+                    gfx_util=None, mem_util=None, encoder=None, decoder=None,
+                    ecc=None, vram_usage=None, pcie=None, process=None,
                     violation=None):
         """ Populate a table with each GPU as an index to rows of targeted data
 
@@ -5621,6 +5622,8 @@ def monitor(self, args, multiple_devices=False, watching_output=False, gpu=None,
             iterations (int, optional): Value override for args.iterations. Defaults to None.
             power_usage (bool, optional): Value override for args.power_usage. Defaults to None.
             temperature (bool, optional): Value override for args.temperature. Defaults to None.
+            base_board_temps (bool, optional): Value override for args.base_board_temps. Defaults to None.
+            gpu_board_temps (bool, optional): Value override for args.gpu_board_temps. Defaults to None.
             gfx (bool, optional): Value override for args.gfx. Defaults to None.
             mem_util (bool, optional): Value override for args.mem. Defaults to None.
             encoder (bool, optional): Value override for args.encoder. Defaults to None.
@@ -5653,6 +5656,10 @@ def monitor(self, args, multiple_devices=False, watching_output=False, gpu=None,
             args.power_usage = power_usage
         if temperature:
             args.temperature = temperature
+        if base_board_temps:
+            args.base_board_temps = base_board_temps
+        if gpu_board_temps:
+            args.gpu_board_temps = gpu_board_temps
         if gfx_util:
             args.gfx = gfx_util
         if mem_util:
@@ -5685,9 +5692,10 @@ def monitor(self, args, multiple_devices=False, watching_output=False, gpu=None,
 
         # If all arguments are False, the print all values
         # Don't include process in this logic as it's an optional edge case
-        if not any([args.power_usage, args.temperature, args.gfx, args.mem,
-                    args.encoder, args.decoder, args.ecc, args.vram_usage,
-                    args.pcie, args.violation]):
+        if not any([args.power_usage, args.temperature, args.base_board_temps,
+                    args.gpu_board_temps, args.gfx, args.mem, args.encoder,
+                    args.decoder, args.ecc, args.vram_usage, args.pcie,
+                    args.violation]):
             args.power_usage = args.temperature = args.gfx = args.mem = \
                 args.encoder = args.decoder = args.vram_usage = True
             # set extra args for default output filtering
@@ -5869,6 +5877,41 @@ def monitor(self, args, multiple_devices=False, watching_output=False, gpu=None,
             self.logger.table_header += 'GPU_T'.rjust(8)
             self.logger.table_header += 'MEM_T'.rjust(8)
 
+
+        if args.gpu_board_temps:
+            try:
+                gpu_board_temp_dict = self.helpers.get_gpu_board_temperatures(args.gpu, gpu_id, self.logger)
+
+                temp_unit_json = 'C'
+                # Add GPU board sensor headers
+                if gpu_board_temp_dict:
+                    for temp_sensor in sorted(gpu_board_temp_dict.keys()):
+                        self.logger.table_header += f"{temp_sensor}".rjust(max(len(temp_sensor)+2, 7))
+                    for temp_type, temp_value in gpu_board_temp_dict.items():
+                        if self.logger.is_json_format() and isinstance(temp_value, dict):
+                            temp_value['unit'] = temp_unit_json
+                        monitor_values[temp_type] = temp_value
+            except Exception as e:
+                logging.debug("Failed to get GPU board temperatures on gpu %s | %s", gpu_id, e)
+
+
+        if args.base_board_temps:
+            try:
+                base_board_temp_dict = self.helpers.get_base_board_temperatures(args.gpu, gpu_id, self.logger)
+
+                temp_unit_json = 'C'
+                # Add base board sensor headers
+                if base_board_temp_dict:
+                    for temp_sensor in sorted(base_board_temp_dict.keys()):
+                        self.logger.table_header += f"{temp_sensor}".rjust(max(len(temp_sensor)+2, 7))
+                    for temp_type, temp_value in base_board_temp_dict.items():
+                        if self.logger.is_json_format() and isinstance(temp_value, dict):
+                            temp_value['unit'] = temp_unit_json
+                        monitor_values[temp_type] = temp_value
+            except Exception as e:
+                logging.debug("Failed to get base board temperatures on gpu %s | %s", gpu_id, e)
+
+
         if args.gfx:
             try:
                 gfx_clk = gpu_metrics_info['current_gfxclk']

projects/amdsmi/amdsmi_cli/amdsmi_logger.py

@@ -263,6 +263,9 @@ def _convert_json_to_tabular(self, json_object: Dict[str, any], dynamic=False):
                 # Remove excess two values after a new line in table_values
                 table_values = table_values[:table_values.rfind('\n')]
                 table_values += '\n'
+            # Board temperature key patterns
+            elif any(pattern in key for pattern in ['IBC', 'OAM', 'RETIMER', 'UBB', 'HSC', 'VR', 'VDDCR', 'NODE', 'VDD', 'HBM']):
+                table_values += string_value.rjust(max((len(key)+2), 7))
             # Default spacing
             else:
                 table_values += string_value.rjust(10)

projects/amdsmi/amdsmi_cli/amdsmi_parser.py

@@ -1412,6 +1412,8 @@ def _add_monitor_parser(self, subparsers: argparse._SubParsersAction, func):
         # Help text for Arguments only on Guest and BM platforms
         power_usage_help = "Monitor power usage and power cap in Watts"
         temperature_help = "Monitor temperature in Celsius"
+        base_board_temps_help = "Monitor base board temperatures in Celsius"
+        gpu_board_temps_help = "Monitor GPU board temperatures in Celsius"
         gfx_util_help = "Monitor graphics utilization (%%) and clock (MHz)"
         mem_util_help = "Monitor memory utilization (%%) and clock (MHz)"
         encoder_util_help = "Monitor encoder utilization (%%) and clock (MHz)"
@@ -1431,6 +1433,8 @@ def _add_monitor_parser(self, subparsers: argparse._SubParsersAction, func):
         # Add monitor arguments
         monitor_parser.add_argument('-p', '--power-usage', action='store_true', required=False, help=power_usage_help)
         monitor_parser.add_argument('-t', '--temperature', action='store_true', required=False, help=temperature_help)
+        monitor_parser.add_argument('-b', '--base-board-temps', action='store_true', required=False, help=base_board_temps_help)
+        monitor_parser.add_argument('-o', '--gpu-board-temps', action='store_true', required=False, help=gpu_board_temps_help)
         monitor_parser.add_argument('-u', '--gfx', action='store_true', required=False, help=gfx_util_help)
         monitor_parser.add_argument('-m', '--mem', action='store_true', required=False, help=mem_util_help)
         monitor_parser.add_argument('-n', '--encoder', action='store_true', required=False, help=encoder_util_help)