[Mellanox] Add new thermal sensors for SN5600 (#12671)

- Why I did it
Add new thermal sensors for SN5600

- How I did it
Add new thermal sensors for SN5600: PCH and SODIMM

- How to verify it
Manual test

Author: Junchao-Mellanox

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

@@ -167,6 +167,12 @@
         }
     },
     'x86_64-nvidia_sn5600-r0': {
+        'thermal': {
+            "capability": {
+                "comex_amb": False,
+                "pch_temp": True
+            }
+        }
     }
 }
 
@@ -225,6 +231,11 @@ def get_gearbox_count(cls, sysfs_folder):
     def get_cpu_thermal_count(cls):
         return len(glob.glob('run/hw-management/thermal/cpu_core[!_]'))
 
+    @classmethod
+    @utils.read_only_cache()
+    def get_sodimm_thermal_count(cls):
+        return len(glob.glob('/run/hw-management/thermal/sodimm*_temp_input'))
+
     @classmethod
     @utils.read_only_cache()
     def get_minimum_table(cls):

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

@@ -119,6 +119,18 @@
             "name": "Ambient Switch Board Temp",
             "temperature": "swb_amb",
             "default_present": False
+        },
+        {
+            "name": "PCH Temp",
+            "temperature": "pch_temp",
+            "default_present": False
+        },
+        {
+            "name": "SODIMM {} Temp",
+            "temperature": "sodimm{}_temp_input",
+            "high_threshold": "sodimm{}_temp_max",
+            "high_critical_threshold": "sodimm{}_temp_crit",
+            "type": "indexable",
         }
     ],
     'linecard thermals': {
@@ -161,6 +173,8 @@ def initialize_chassis_thermals():
                 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
@@ -524,7 +538,7 @@ def monitor_asic_themal_zone(cls):
         else:
             cls.expect_cooling_state = None
 
-        
+
 class RemovableThermal(Thermal):
     def __init__(self, name, temp_file, high_th_file, high_crit_th_file, position, presence_cb):
         super(RemovableThermal, self).__init__(name, temp_file, high_th_file, high_crit_th_file, position)

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

@@ -38,6 +38,7 @@ 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.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
@@ -48,6 +49,7 @@ def test_chassis_thermal(self):
         thermal_dict = {thermal.get_name(): thermal for thermal in thermal_list}
         gearbox_thermal_rule = None
         cpu_thermal_rule = None
+        sodimm_thermal_rule = None
         for rule in THERMAL_NAMING_RULE['chassis thermals']:
             thermal_type = rule.get('type', 'single')
             if thermal_type == 'single':
@@ -69,9 +71,12 @@ def test_chassis_thermal(self):
                     gearbox_thermal_rule = rule
                 elif 'CPU Core' in rule['name']:
                     cpu_thermal_rule = rule
+                elif 'SODIMM' in rule['name']:
+                    sodimm_thermal_rule = rule
 
         gearbox_thermal_count = 0
         cpu_thermal_count = 0
+        sodimm_thermal_count = 0
         for thermal in thermal_list:
             if 'Gearbox' in thermal.get_name():
                 start_index = gearbox_thermal_rule.get('start_index', 1)
@@ -89,21 +94,32 @@ def test_chassis_thermal(self):
                 assert cpu_thermal_rule['high_threshold'].format(start_index) in thermal.high_threshold
                 assert cpu_thermal_rule['high_critical_threshold'].format(start_index) in thermal.high_critical_threshold
                 cpu_thermal_count += 1
+            elif 'SODIMM' in thermal.get_name():
+                start_index = sodimm_thermal_rule.get('start_index', 1)
+                start_index += sodimm_thermal_count
+                assert thermal.get_name() == sodimm_thermal_rule['name'].format(start_index)
+                assert sodimm_thermal_rule['temperature'].format(start_index) in thermal.temperature
+                assert sodimm_thermal_rule['high_threshold'].format(start_index) in thermal.high_threshold
+                assert sodimm_thermal_rule['high_critical_threshold'].format(start_index) in thermal.high_critical_threshold
+                sodimm_thermal_count += 1
 
         assert gearbox_thermal_count == 2
         assert cpu_thermal_count == 2
+        assert sodimm_thermal_count == 2
 
     @mock.patch('sonic_platform.device_data.DeviceDataManager.get_platform_name', mock.MagicMock(return_value='x86_64-nvidia_sn2201-r0'))
-    @mock.patch('sonic_platform.device_data.DeviceDataManager.get_thermal_capability', mock.MagicMock(return_value={'comex_amb': False, 'cpu_amb': True, 'swb_amb': True}))
-    def test_chassis_thermal_includes(self):
+    @mock.patch('sonic_platform.device_data.DeviceDataManager.get_thermal_capability')
+    def test_chassis_thermal_includes(self, mock_capability):
         from sonic_platform.thermal import THERMAL_NAMING_RULE
+        thermal_capability = {'comex_amb': False, 'cpu_amb': True, 'swb_amb': True}
+        mock_capability.return_value = thermal_capability
         chassis = Chassis()
         thermal_list = chassis.get_all_thermals()
         assert thermal_list
         thermal_dict = {thermal.get_name(): thermal for thermal in thermal_list}
         for rule in THERMAL_NAMING_RULE['chassis thermals']:
             default_present = rule.get('default_present', True)
-            if not default_present:
+            if not default_present and thermal_capability.get(rule['temperature']):
                 thermal_name = rule['name']
                 assert thermal_name in thermal_dict