[Mellanox] update asic and module temperature in a thread

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

@@ -79,6 +79,8 @@ class Chassis(ChassisBase):
     # System UID LED
     _led_uid = None
 
+    chassis_instance = None
+
     def __init__(self):
         super(Chassis, self).__init__()
 
@@ -124,6 +126,7 @@ def __init__(self):
         self._RJ45_port_inited = False
         self._RJ45_port_list = None
 
+        Chassis.chassis_instance = self
         logger.log_info("Chassis loaded successfully")
 
     def __del__(self):

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

@@ -227,3 +227,12 @@ def get_cpld_component_list(cls):
             # Currently, only fetching BIOS version is supported
             return ComponentCPLDSN2201.get_component_list()
         return ComponentCPLD.get_component_list()
+
+    @classmethod
+    @utils.read_only_cache()
+    def is_independent_mode(cls):
+        from sonic_py_common import device_info
+        _, hwsku_dir = device_info.get_paths_to_platform_and_hwsku_dirs()
+        sai_profile_file = os.path.join(hwsku_dir, 'sai.profile')
+        data = utils.read_key_value_file(sai_profile_file, delimeter='=')
+        return data.get('SAI_INDEPENDENT_MODULE_MODE') == '1'

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

@@ -31,6 +31,8 @@
     from . import utils
     from .device_data import DeviceDataManager
     from sonic_platform_base.sonic_xcvr.sfp_optoe_base import SfpOptoeBase
+    from sonic_platform_base.sonic_xcvr.fields import consts
+    from sonic_platform_base.sonic_xcvr.api.public import sff8636, sff8436
 
 except ImportError as e:
     raise ImportError (str(e) + "- required module not found")
@@ -135,6 +137,10 @@
 # SFP stderr
 SFP_EEPROM_NOT_AVAILABLE = 'Input/output error'
 
+SFP_DEFAULT_TEMP_WARNNING_THRESHOLD = 70.0
+SFP_DEFAULT_TEMP_CRITICAL_THRESHOLD = 80.0
+SFP_TEMPERATURE_SCALE = 8.0
+
 # SFP EEPROM limited bytes
 limited_eeprom = {
     SFP_TYPE_CMIS: {
@@ -406,6 +412,11 @@ def get_error_description(self):
             error_description = "Unknow SFP module status ({})".format(oper_status)
         return error_description
 
+    def get_error(self):
+        status_file_path = SFP_SDK_MODULE_SYSFS_ROOT_TEMPLATE.format(sdk_index) + SFP_SYSFS_STATUS
+        oper_state = utils.read_int_from_file(status_file_path)
+        return 1 if oper_state == SX_PORT_MODULE_STATUS_PLUGGED_WITH_ERROR else 0
+
     def _get_eeprom_path(self):
         return SFP_EEPROM_ROOT_TEMPLATE.format(self.sdk_index)
 
@@ -527,6 +538,71 @@ def get_tx_fault(self):
         api = self.get_xcvr_api()
         return [False] * api.NUM_CHANNELS if api else None
 
+    def get_temperature(self):
+        try:
+            if not self.is_sw_control():
+                temperature = utils.read_int_from_file(f'/sys/module/sx_core/asic0/module{self.sdk_index}/temperature/input',
+                                                       default=None)
+                return temperature / SFP_TEMPERATURE_SCALE if temperature is not None else 0.0
+        except:
+            return 0.0
+
+        temperature = super().get_temperature()
+        return temperature if temperature is not None else 0.0
+
+    def get_temperature_warning_threashold(self):
+        """Get temperature warning threshold
+
+        Returns:
+            int: temperature warning threshold
+        """
+        try:
+            if not self.is_sw_control():
+                emergency = utils.read_int_from_file(f'/sys/module/sx_core/asic0/module{self.sdk_index}/temperature/emergency',
+                                                     log_func=None,
+                                                     default=None)
+                return emergency / SFP_TEMPERATURE_SCALE if emergency is not None else SFP_DEFAULT_TEMP_WARNNING_THRESHOLD
+        except:
+            return SFP_DEFAULT_TEMP_WARNNING_THRESHOLD
+
+        thresh = self._get_temperature_threshold()
+        if thresh and consts.TEMP_HIGH_WARNING_FIELD in thresh:
+            return thresh[consts.TEMP_HIGH_WARNING_FIELD]
+        return SFP_DEFAULT_TEMP_WARNNING_THRESHOLD
+
+    def get_temperature_critical_threashold(self):
+        """Get temperature critical threshold
+
+        Returns:
+            int: temperature critical threshold
+        """
+        try:
+            if not self.is_sw_control():
+                critical = utils.read_int_from_file(f'/sys/module/sx_core/asic0/module{self.sdk_index}/temperature/critical',
+                                                    log_func=None,
+                                                    default=None)
+                return critical / SFP_TEMPERATURE_SCALE if critical is not None else SFP_DEFAULT_TEMP_CRITICAL_THRESHOLD
+        except:
+            return SFP_DEFAULT_TEMP_CRITICAL_THRESHOLD
+
+        thresh = self._get_temperature_threshold()
+        if thresh and consts.TEMP_HIGH_ALARM_FIELD in thresh:
+            return thresh[consts.TEMP_HIGH_ALARM_FIELD]
+        return SFP_DEFAULT_TEMP_CRITICAL_THRESHOLD
+
+    def _get_temperature_threshold(self):
+        api = self.get_xcvr_api()
+        if not api:
+            return None
+
+        thresh_support = api.get_transceiver_thresholds_support()
+        if thresh_support:
+            if isinstance(api, sff8636.Sff8636Api) or isinstance(api, sff8436.Sff8436Api):
+                return api.xcvr_eeprom.read(consts.TEMP_THRESHOLDS_FIELD)
+            return api.xcvr_eeprom.read(consts.THRESHOLDS_FIELD)
+        else:
+            return None
+
     def get_xcvr_api(self):
         """
         Retrieves the XcvrApi associated with this SFP
@@ -541,6 +617,17 @@ def get_xcvr_api(self):
                 self._xcvr_api.get_tx_fault = self.get_tx_fault
         return self._xcvr_api
 
+    def is_sw_control(self):
+        if not DeviceDataManager.is_independent_mode():
+            return False
+
+        db = utils.DbUtils.get_db_instance('STATE_DB')
+        control_type = db.get('STATE_DB', f'TRANSCEIVER_MODULES_MGMT|{self.sdk_index}', 'control_type')
+        if not control_type:
+            raise Exception(f'Module {self.sdk_index} is in initialization, please retry later')
+
+        return control_type == 'SW_CONTROL'
+
 
 class RJ45Port(NvidiaSFPCommon):
     """class derived from SFP, representing RJ45 ports"""

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

@@ -15,9 +15,36 @@
 # limitations under the License.
 #
 from sonic_platform_base.sonic_thermal_control.thermal_manager_base import ThermalManagerBase
+from . import thermal_updater
+from .device_data import DeviceDataManager
 
 
 class ThermalManager(ThermalManagerBase):
+    thermal_updater_task = None
+
     @classmethod
     def run_policy(cls, chassis):
         pass
+
+    @classmethod
+    def initialize(cls):
+        """
+        Initialize thermal manager, including register thermal condition types and thermal action types
+        and any other vendor specific initialization.
+        :return:
+        """
+        if DeviceDataManager.is_independent_mode():
+            from .chassis import Chassis
+            cls.thermal_updater_task = thermal_updater.ThermalUpdater(Chassis.chassis_instance.get_all_sfps())
+            cls.thermal_updater_task.start()
+
+
+    @classmethod
+    def deinitialize(cls):
+        """
+        Destroy thermal manager, including any vendor specific cleanup. The default behavior of this function
+        is a no-op.
+        :return:
+        """
+        if DeviceDataManager.is_independent_mode():
+            cls.thermal_updater_task.stop()