[ROCM] Enhance amd-smi node to display baseboard temp (#2943)

Added amdsmi_get_device_handle_from_node API to
retrieve device handle from a node handle.
Enhanced `amd-smi node` command to display baseboard
temperatures.

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

Author: bkanango

projects/amdsmi/CHANGELOG.md

@@ -8,6 +8,21 @@ Full documentation for amd_smi_lib is available at [https://rocm.docs.amd.com/pr
 
 ### Added
 
+- **Added `amdsmi_get_device_handle_from_node` API**.
+  - Added C API function to retrieve a device handle from a node handle.
+  - Provides inverse functionality to `amdsmi_get_node_handle`.
+  - Added python binding for the API and exported in `py-interface/__init__.py` for public API access.
+  - Returns `AMDSMI_STATUS_SUCCESS` on success, `AMDSMI_STATUS_NOT_FOUND` if no matching device found.
+
+- **Enhanced `amd-smi node` command to display baseboard temperatures**.
+  - Added `--base-board-temps` / `-b` option to display baseboard temperature sensors.
+  - Selective display: Use `-p` for NPM only, `-b` for Baseboard only.
+  - Default behavior (no flags): Shows both power management and baseboard temperatures.
+
+## amd_smi_lib for ROCm 7.11.0
+
+### Added
+
 - **Added `--hex` flag to `amd-smi bad-pages` command**.  
   - Added `--hex` option to display page addresses and sizes in hexadecimal format with `0x` prefix
 

projects/amdsmi/amdsmi_cli/amdsmi_commands.py

@@ -7489,65 +7489,114 @@ def ras(self, args, multiple_devices=False, gpu=None, cper=None, afid=None,
             time.sleep(1)
 
 
-    def node(self, args, multiple_devices=False, nodes=None, power_management=None):
+    def node(self, args, multiple_devices=False, nodes=None, power_management=None, base_board_temps=None):
         """List node informations
 
         Args:
             args (Namespace): Namespace containing the parsed CLI args
             multiple_devices (bool, optional): True if checking for multiple devices.
                 Defaults to False.
+            nodes (node_handle, optional): node_handle for target node. Defaults to None.
+            power_management (bool, optional): Value override for args.power_management. Defaults to None.
+            base_board_temps (bool, optional): Value override for args.base_board_temps. Defaults to None.
 
         Returns:
             None: Print output via AMDSMILogger to destination
         """
         # Set args.* to passed in arguments
         if nodes:
             args.nodes = nodes
-        if power_management:
-            args.power_management = power_management
+        # Store args that are applicable to the current platform
+        current_platform_args = ["power_management", "base_board_temps"]
+
+        # Check if any node-specific options were passed via command line
+        current_platform_values = []
+        if args.power_management:
+            current_platform_values += [args.power_management]
+        if args.base_board_temps:
+            current_platform_values += [args.base_board_temps]
+
+        # If no node options are passed, enable all by default
+        if not any(current_platform_values):
+            for arg in current_platform_args:
+                setattr(args, arg, True)
         if getattr(args, 'nodes', None) is None:
             args.nodes = self.node_handle
 
         if not self.group_check_printed:
             self.helpers.check_required_groups(check_render=True, check_video=False)
             self.group_check_printed = True
 
+        # Initialize variables for both power management and base board temps
+        npm_dict = {"limit": "N/A", "status": "N/A"}
+        power_unit = "W"
+        limit = "N/A"
+        base_board_temp_dict = {}
+
         # Get NPM info
-        if args.nodes is not None:
-            try:
-                npm_info = amdsmi_interface.amdsmi_get_npm_info(args.nodes)
-            except amdsmi_exception.AmdSmiLibraryException as e:
-                logging.debug("amdsmi_get_npm_info failed: %s", e.get_error_info())
+        if args.power_management:
+            if args.nodes is not None:
+                try:
+                    npm_info = amdsmi_interface.amdsmi_get_npm_info(args.nodes)
+                except amdsmi_exception.AmdSmiLibraryException as e:
+                    logging.debug("amdsmi_get_npm_info failed: %s", e.get_error_info())
+                    npm_info = "N/A"
+            else:
+                logging.debug('No node handle available to query NPM info')
                 npm_info = "N/A"
-        else:
-            logging.debug('No node handle available to query NPM info')
-            npm_info = "N/A"
 
-        # Log outputs
-        npm_dict = {"limit": "N/A", "status": "N/A"}
-        power_unit ="W"
+            if isinstance(npm_info, dict):
+                limit = npm_info.get('limit', "N/A")
+                status = npm_info.get('status', npm_info.get('current', "N/A"))
 
-        limit = "N/A"
-        if isinstance(npm_info, dict):
-            limit = npm_info.get('limit', "N/A")
-            status = npm_info.get('status', npm_info.get('current', "N/A"))
-
-            if limit !="N/A":
-                npm_dict['limit'] = limit
-            status = "DISABLED" if status == amdsmi_interface.amdsmi_wrapper.AMDSMI_NPM_STATUS_DISABLED else "ENABLED"
-            npm_dict.update({"status": status})
+                if limit !="N/A":
+                    npm_dict['limit'] = limit
+                status = "DISABLED" if status == amdsmi_interface.amdsmi_wrapper.AMDSMI_NPM_STATUS_DISABLED else "ENABLED"
+                npm_dict.update({"status": status})
+
+        # Get base board temperatures using node_handle
+        if args.base_board_temps:
+            if args.nodes is not None:
+                try:
+                    # Get device_handle from node_handle
+                    device_handle = amdsmi_interface.amdsmi_get_device_handle_from_node(args.nodes)
+                    gpu_id = self.helpers.get_gpu_id_from_device_handle(device_handle)
+                    base_board_temp_dict = self.helpers.get_base_board_temperatures(device_handle, gpu_id, self.logger)
+                except amdsmi_exception.AmdSmiLibraryException as e:
+                    logging.debug("Failed to get device handle from node: %s", e.get_error_info())
+                    base_board_temp_dict = {}
+
+        # Print output
         if self.logger.is_human_readable_format() and self.logger.destination == 'stdout':
-            print(f"NODE:\n    POWER_MANAGEMENT:\n        LIMIT: {npm_dict.get('limit', 'N/A')} {power_unit}\n        STATUS: {npm_dict.get('status', 'N/A')}")
+            node_output = ["NODE:"]
+            if args.power_management:
+                node_output.append("    POWER_MANAGEMENT:")
+                node_output.append(f"        LIMIT: {npm_dict.get('limit', 'N/A')} {power_unit}")
+                node_output.append(f"        STATUS: {npm_dict.get('status', 'N/A')}")
+            if args.base_board_temps and base_board_temp_dict:
+                node_output.append("    BASEBOARD:")
+                node_output.append("        TEMPERATURE:")
+                for temp_name, temp_value in base_board_temp_dict.items():
+                    node_output.append(f"            {temp_name.upper()}: {temp_value}")
+            print("\n".join(node_output))
         else:
             if self.logger.is_csv_format():
                 csv_dict = {}
-                csv_dict['limit'] = npm_dict.get('limit', "N/A")
-                csv_dict['status'] = npm_dict.get('status', "N/A")
+                if args.power_management:
+                    csv_dict['limit'] = npm_dict.get('limit', "N/A")
+                    csv_dict['status'] = npm_dict.get('status', "N/A")
+                if args.base_board_temps and base_board_temp_dict:
+                    csv_dict.update(base_board_temp_dict)
                 self.logger.output = csv_dict
             else:
                 # For JSON and human readable format with file output
-                npm_dict["limit"] = self.helpers.unit_format(self.logger, limit, power_unit)
-                self.logger.output = {'node': {'power_management': npm_dict}}
+                node_output = {}
+                if args.power_management:
+                    npm_dict["limit"] = self.helpers.unit_format(self.logger, limit, power_unit)
+                    node_output['power_management'] = npm_dict
+                if args.base_board_temps and base_board_temp_dict:
+                    node_output['base_board'] = {'temperature': base_board_temp_dict}
+                self.logger.output = {'node': node_output}
                 if multiple_devices:
                     self.logger.store_multiple_device_output()
                     return

projects/amdsmi/amdsmi_cli/amdsmi_parser.py

@@ -1690,13 +1690,14 @@ def _add_ras_parser(self, subparsers: argparse._SubParsersAction, func):
 
     def _add_node_parser(self, subparsers: argparse._SubParsersAction, func):
         # Subparser help text
-        node_help = "Gets power information for the node"
+        node_help = "Gets power and baseboard information for the node"
         node_subcommand_help = f"{self.description}\n\nReturns information for node 0 on the system.\
                                 \nIf no node argument is provided, all node information will be displayed."
         node_optionals_title = "Node arguments"
 
         # Help text for Node arguments
         power_management_help = "Displays power management information"
+        base_board_temps_help = "Displays baseboard temperatures"
 
         node_parser = subparsers.add_parser("node", help=node_help, description=node_subcommand_help)
         node_parser._optionals.title = node_optionals_title
@@ -1705,6 +1706,7 @@ def _add_node_parser(self, subparsers: argparse._SubParsersAction, func):
 
         # Optional Args
         node_parser.add_argument('-p', '--power-management', action='store_true', required=False, help=power_management_help)
+        node_parser.add_argument('-b', '--base-board-temps', action='store_true', required=False, help=base_board_temps_help)
 
         # Add Universal Arguments
         self._add_command_modifiers(node_parser)

projects/amdsmi/include/amd_smi/amdsmi.h

@@ -2820,6 +2820,27 @@ amdsmi_status_t amdsmi_get_processor_handles(amdsmi_socket_handle socket_handle,
  */
 amdsmi_status_t amdsmi_get_node_handle(amdsmi_processor_handle processor_handle, amdsmi_node_handle *node_handle);
 
+/**
+ *  @brief Get the processor (device) handle associated with a node handle.
+ *
+ *  @ingroup tagProcDiscovery
+ *
+ *  @platform{gpu_bm_linux}
+ *
+ *  @details This function retrieves the processor (device) handle from a node handle.
+ *  The @p node_handle must be provided for the node. This is the inverse operation
+ *  of amdsmi_get_node_handle API.
+ *  Currently, only AMD GPUs are supported.
+ *
+ *  @param[in] node_handle A pointer to a ::amdsmi_node_handle, this identifies
+ *  the node from which to retrieve the associated device handle.
+ *
+ *  @param[out] processor_handle A pointer to a block of memory where amdsmi_processor_handle
+ *  will be written.
+ *
+ *  @return ::amdsmi_status_t | ::AMDSMI_STATUS_SUCCESS on success, non-zero on fail
+ */
+amdsmi_status_t amdsmi_get_device_handle_from_node(amdsmi_node_handle node_handle, amdsmi_processor_handle *processor_handle);
 
 #ifdef ENABLE_ESMI_LIB
 /**

projects/amdsmi/py-interface/__init__.py

@@ -31,6 +31,9 @@
 from .amdsmi_interface import amdsmi_get_socket_info
 from .amdsmi_interface import amdsmi_get_processor_count_from_handles
 from .amdsmi_interface import amdsmi_get_processor_handles_by_type
+from .amdsmi_interface import amdsmi_get_node_handle
+from .amdsmi_interface import amdsmi_get_device_handle_from_node
+from .amdsmi_interface import amdsmi_get_npm_info
 
 # ESMI Dependent Functions
 try:

projects/amdsmi/py-interface/amdsmi_interface.py

@@ -4596,6 +4596,34 @@ def amdsmi_get_node_handle(processor_handle):
     return node_handle
 
 
+def amdsmi_get_device_handle_from_node(node_handle):
+    """
+    Get the processor (device) handle associated with a node handle.
+
+    This function retrieves the processor (device) handle from a node handle.
+    This is the inverse operation of amdsmi_get_node_handle.
+
+    Args:
+        node_handle: A node handle (amdsmi_node_handle) to get the device handle from.
+
+    Returns:
+        amdsmi_processor_handle: The processor handle associated with the node.
+
+    Raises:
+        AmdSmiParameterException: If node_handle is not the correct type.
+        AmdSmiLibraryException: If the library call fails.
+    """
+    if not isinstance(node_handle, amdsmi_wrapper.amdsmi_node_handle):
+        raise AmdSmiParameterException(node_handle, amdsmi_wrapper.amdsmi_node_handle)
+
+    processor_handle = amdsmi_wrapper.amdsmi_processor_handle()
+    _check_res(
+        amdsmi_wrapper.amdsmi_get_device_handle_from_node(node_handle, ctypes.byref(processor_handle))
+    )
+
+    return processor_handle
+
+
 def amdsmi_get_npm_info(node_handle: processor_handle_t) -> Dict[str, Any]:
     if not isinstance(node_handle, amdsmi_wrapper.amdsmi_node_handle):
         raise AmdSmiParameterException(node_handle, amdsmi_wrapper.amdsmi_node_handle)

projects/amdsmi/py-interface/amdsmi_wrapper.py

@@ -990,6 +990,21 @@ class struct_amdsmi_enumeration_info_t(Structure):
 class struct_amdsmi_pcie_info_t(Structure):
     pass
 
+class struct_pcie_static_(Structure):
+    pass
+
+struct_pcie_static_._pack_ = 1 # source:False
+struct_pcie_static_._fields_ = [
+    ('max_pcie_width', ctypes.c_uint16),
+    ('PADDING_0', ctypes.c_ubyte * 2),
+    ('max_pcie_speed', ctypes.c_uint32),
+    ('pcie_interface_version', ctypes.c_uint32),
+    ('slot_type', amdsmi_card_form_factor_t),
+    ('max_pcie_interface_version', ctypes.c_uint32),
+    ('PADDING_1', ctypes.c_ubyte * 4),
+    ('reserved', ctypes.c_uint64 * 9),
+]
+
 class struct_pcie_metric_(Structure):
     pass
 
@@ -1010,21 +1025,6 @@ class struct_pcie_metric_(Structure):
     ('reserved', ctypes.c_uint64 * 12),
 ]
 
-class struct_pcie_static_(Structure):
-    pass
-
-struct_pcie_static_._pack_ = 1 # source:False
-struct_pcie_static_._fields_ = [
-    ('max_pcie_width', ctypes.c_uint16),
-    ('PADDING_0', ctypes.c_ubyte * 2),
-    ('max_pcie_speed', ctypes.c_uint32),
-    ('pcie_interface_version', ctypes.c_uint32),
-    ('slot_type', amdsmi_card_form_factor_t),
-    ('max_pcie_interface_version', ctypes.c_uint32),
-    ('PADDING_1', ctypes.c_ubyte * 4),
-    ('reserved', ctypes.c_uint64 * 9),
-]
-
 struct_amdsmi_pcie_info_t._pack_ = 1 # source:False
 struct_amdsmi_pcie_info_t._fields_ = [
     ('pcie_static', struct_pcie_static_),
@@ -2617,6 +2617,9 @@ class struct_amdsmi_sock_info_t(Structure):
 amdsmi_get_node_handle = _libraries['libamd_smi.so'].amdsmi_get_node_handle
 amdsmi_get_node_handle.restype = amdsmi_status_t
 amdsmi_get_node_handle.argtypes = [amdsmi_processor_handle, ctypes.POINTER(ctypes.POINTER(None))]
+amdsmi_get_device_handle_from_node = _libraries['libamd_smi.so'].amdsmi_get_device_handle_from_node
+amdsmi_get_device_handle_from_node.restype = amdsmi_status_t
+amdsmi_get_device_handle_from_node.argtypes = [amdsmi_node_handle, ctypes.POINTER(ctypes.POINTER(None))]
 amdsmi_get_cpucore_handles = _libraries['libamd_smi.so'].amdsmi_get_cpucore_handles
 amdsmi_get_cpucore_handles.restype = amdsmi_status_t
 amdsmi_get_cpucore_handles.argtypes = [ctypes.POINTER(ctypes.c_uint32), ctypes.POINTER(ctypes.POINTER(None))]
@@ -3419,9 +3422,9 @@ class struct_amdsmi_cper_hdr_t(Structure):
     'AMDSMI_PROCESSOR_TYPE_NON_AMD_GPU',
     'AMDSMI_PROCESSOR_TYPE_UNKNOWN', 'AMDSMI_PTL_DATA_FORMAT_BF16',
     'AMDSMI_PTL_DATA_FORMAT_F16', 'AMDSMI_PTL_DATA_FORMAT_F32',
-    'AMDSMI_PTL_DATA_FORMAT_F64', 'AMDSMI_PTL_DATA_FORMAT_I8',
-    'AMDSMI_PTL_DATA_FORMAT_F8', 'AMDSMI_PTL_DATA_FORMAT_VECTOR',
-    'AMDSMI_PTL_DATA_FORMAT_INVALID',
+    'AMDSMI_PTL_DATA_FORMAT_F64', 'AMDSMI_PTL_DATA_FORMAT_F8',
+    'AMDSMI_PTL_DATA_FORMAT_I8', 'AMDSMI_PTL_DATA_FORMAT_INVALID',
+    'AMDSMI_PTL_DATA_FORMAT_VECTOR',
     'AMDSMI_PWR_PROF_PRST_3D_FULL_SCR_MASK',
     'AMDSMI_PWR_PROF_PRST_BOOTUP_DEFAULT',
     'AMDSMI_PWR_PROF_PRST_COMPUTE_MASK',
@@ -3604,7 +3607,8 @@ class struct_amdsmi_cper_hdr_t(Structure):
     'amdsmi_get_cpu_socket_power', 'amdsmi_get_cpu_socket_power_cap',
     'amdsmi_get_cpu_socket_power_cap_max',
     'amdsmi_get_cpu_socket_temperature', 'amdsmi_get_cpucore_handles',
-    'amdsmi_get_cpusocket_handles', 'amdsmi_get_dfc_ctrl',
+    'amdsmi_get_cpusocket_handles',
+    'amdsmi_get_device_handle_from_node', 'amdsmi_get_dfc_ctrl',
     'amdsmi_get_energy_count', 'amdsmi_get_esmi_err_msg',
     'amdsmi_get_fw_info',
     'amdsmi_get_gpu_accelerator_partition_profile',