[Mellanox] support thermal sensor which has discrete index

Author: Junchao-Mellanox

platform/mellanox/mlnx-platform-api/sonic_platform/thermal.py

@@ -1,5 +1,6 @@
 #
-# Copyright (c) 2019-2024 NVIDIA CORPORATION & AFFILIATES.
+# SPDX-FileCopyrightText: NVIDIA CORPORATION & AFFILIATES
+# Copyright (c) 2019-2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 # Apache-2.0
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
@@ -26,7 +27,9 @@
     from sonic_platform_base.thermal_base import ThermalBase
     from sonic_py_common.logger import Logger
     import copy
+    import glob
     import os
+    import re
 
     from .device_data import DeviceDataManager
     from . import utils
@@ -133,7 +136,9 @@
             "temperature": "sodimm{}_temp_input",
             "high_threshold": "sodimm{}_temp_max",
             "high_critical_threshold": "sodimm{}_temp_crit",
-            "type": "indexable",
+            "search_pattern": '/run/hw-management/thermal/sodimm*_temp_input',
+            'index_pattern': r'sodimm(\d+)_temp_input',
+            "type": "discrete",
         }
     ],
     'linecard thermals': {
@@ -153,21 +158,22 @@ def initialize_chassis_thermals():
     rules = THERMAL_NAMING_RULE['chassis thermals']
     position = 1
     for rule in rules:
-        if 'type' in rule and rule['type'] == 'indexable':
+        thermal_type = rule.get('type')
+        if thermal_type == 'indexable':
             count = 0
             if 'Gearbox' in rule['name']:
                 count = DeviceDataManager.get_gearbox_count('/run/hw-management/config')
             elif 'CPU Core' in rule['name']:
                 count = DeviceDataManager.get_cpu_thermal_count()
-            elif 'SODIMM' in rule['name']:
-                count = DeviceDataManager.get_sodimm_thermal_count()
             if count == 0:
                 logger.log_debug('Failed to get thermal object count for {}'.format(rule['name']))
                 continue
 
             for index in range(count):
                 thermal_list.append(create_indexable_thermal(rule, index, CHASSIS_THERMAL_SYSFS_FOLDER, position))
                 position += 1
+        elif thermal_type == 'discrete':
+            thermal_list.extend(create_discrete_thermal(rule))
         else:
             thermal_object = create_single_thermal(rule, CHASSIS_THERMAL_SYSFS_FOLDER, position)
             if thermal_object:
@@ -274,6 +280,23 @@ def create_single_thermal(rule, sysfs_folder, position, presence_cb=None):
         return RemovableThermal(name, temp_file, high_th_file, high_crit_th_file, high_th_default, high_crit_th_default, scale, position, presence_cb)
 
 
+def create_discrete_thermal(rule):
+    search_pattern = rule.get('search_pattern')
+    index_pattern = rule.get('index_pattern')
+    position = 1
+    thermal_list = []
+    for file_path in glob.iglob(search_pattern):
+        file_name = os.path.basename(file_path)
+        match = re.search(index_pattern, file_name)
+        if not match:
+            logger.log_error(f'Failed to extract index from {file_name} using pattern {index_pattern}')
+            continue
+        index = int(match.group(1))
+        thermal_list.append(create_indexable_thermal(rule, index - 1, CHASSIS_THERMAL_SYSFS_FOLDER, position))
+        position += 1
+    return thermal_list
+
+
 def _check_thermal_sysfs_existence(file_path, presence_cb):
     if presence_cb:
         status, _ = presence_cb()

platform/mellanox/mlnx-platform-api/tests/test_thermal.py

@@ -1,5 +1,6 @@
 #
-# Copyright (c) 2021-2023 NVIDIA CORPORATION & AFFILIATES.
+# SPDX-FileCopyrightText: NVIDIA CORPORATION & AFFILIATES
+# Copyright (c) 2021-2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 # Apache-2.0
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
@@ -39,7 +40,9 @@ class TestThermal:
     @mock.patch('os.path.exists', mock.MagicMock(return_value=True))
     @mock.patch('sonic_platform.device_data.DeviceDataManager.get_gearbox_count', mock.MagicMock(return_value=2))
     @mock.patch('sonic_platform.device_data.DeviceDataManager.get_cpu_thermal_count', mock.MagicMock(return_value=2))
-    @mock.patch('sonic_platform.device_data.DeviceDataManager.get_sodimm_thermal_count', mock.MagicMock(return_value=2))
+    @mock.patch('sonic_platform.thermal.glob.iglob', mock.MagicMock(
+        return_value=['/run/hw-management/thermal/sodimm1_temp_input',
+                      '/run/hw-management/thermal/sodimm2_temp_input']))
     @mock.patch('sonic_platform.device_data.DeviceDataManager.get_platform_name', mock.MagicMock(return_value='x86_64-mlnx_msn2700-r0'))
     def test_chassis_thermal(self):
         from sonic_platform.thermal import THERMAL_NAMING_RULE