[Mellanox]Fix issues found in thermal control

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

@@ -38,17 +38,22 @@ def __init__(self, fan_index, drawer_index = 1, psu_fan = False):
             self.fan_speed_set_path = "fan{}_speed_set".format(self.index)
             self.fan_presence_path = "fan{}_status".format(self.drawer_index)
             self.fan_max_speed_path = "fan{}_max".format(self.index)
+            self._name = "fan{}".format(fan_index + 1)
         else:
             self.fan_speed_get_path = "psu{}_fan1_speed_get".format(self.index)
             self.fan_presence_path = "psu{}_fan1_speed_get".format(self.index)
-            self.fan_max_speed_path = "psu{}_max".format(self.index)
+            self._name = "fan(PSU{})".format(fan_index)
+            self.fan_max_speed_path = None
         self.fan_status_path = "fan{}_fault".format(self.index)
         self.fan_green_led_path = "led_fan{}_green".format(self.drawer_index)
         self.fan_red_led_path = "led_fan{}_red".format(self.drawer_index)
         self.fan_orange_led_path = "led_fan{}_orange".format(self.drawer_index)
         self.fan_pwm_path = "pwm1"
         self.fan_led_cap_path = "led_fan{}_capability".format(self.drawer_index)
 
+    def get_name(self):
+        return self._name
+
     def get_status(self):
         """
         Retrieves the operational status of fan
@@ -123,7 +128,11 @@ def get_speed(self):
                 speed_in_rpm = int(fan_curr_speed.read())
         except (ValueError, IOError):
             speed_in_rpm = 0
-
+
+        if self.fan_max_speed_path is None:
+            # in case of max speed unsupported, we just return speed in unit of RPM.
+            return speed_in_rpm
+
         max_speed_in_rpm = self._get_max_speed_in_rpm()
         speed = 100*speed_in_rpm/max_speed_in_rpm
 
@@ -136,11 +145,10 @@ def get_target_speed(self):
         Returns:
             int: percentage of the max fan speed
         """
-        speed = 0
-
         if self.is_psu_fan:
             # Not like system fan, psu fan speed can not be modified, so target speed is N/A 
-            return speed
+            return self.get_speed()
+
         try:
             with open(os.path.join(FAN_PATH, self.fan_speed_set_path), 'r') as fan_pwm:
                 pwm = int(fan_pwm.read())
@@ -243,4 +251,4 @@ def get_speed_tolerance(self):
                  considered tolerable
         """
         # The tolerance value is fixed as 20% for all the Mellanox platform
-        return 20
+        return 20

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

@@ -24,6 +24,7 @@ def __init__(self):
             self._chassis.initialize_psu()
             self._chassis.initialize_fan()
             self._chassis.initialize_eeprom()
+            self._chassis.initialize_thermals()
 
     def _is_host(self):
         """

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

@@ -60,6 +60,7 @@ def __init__(self, psu_index, sku):
         psu_oper_status = "thermal/psu{}_pwr_status".format(self.index)
         #psu_oper_status should always be present for all SKUs
         self.psu_oper_status = os.path.join(self.psu_path, psu_oper_status)
+        self._name = "PSU{}".format(psu_index + 1)
 
         if sku in hwsku_dict_psu:
             filemap = psu_profile_list[hwsku_dict_psu[sku]]
@@ -92,7 +93,10 @@ def __init__(self, psu_index, sku):
 
         fan = Fan(psu_index, psu_index, True)
         if fan.get_presence():
-            self._fan = fan
+            self._fan_list.append(fan)
+
+    def get_name(self):
+        return self._name
 
     def _read_generic_file(self, filename, len):
         """

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

@@ -36,28 +36,36 @@
 
 THERMAL_API_GET_TEMPERATURE = "get_temperature"
 THERMAL_API_GET_HIGH_THRESHOLD = "get_high_threshold"
+THERMAL_API_GET_HIGH_CRITICAL_THRESHOLD = "get_high_critical_threshold"
+
+THERMAL_API_INVALID_HIGH_THRESHOLD = 0.0
 
 HW_MGMT_THERMAL_ROOT = "/var/run/hw-management/thermal/"
 
 thermal_api_handler_cpu_core = {
     THERMAL_API_GET_TEMPERATURE:"cpu_core{}",
-    THERMAL_API_GET_HIGH_THRESHOLD:"cpu_core{}_max"
+    THERMAL_API_GET_HIGH_THRESHOLD:"cpu_core{}_max",
+    THERMAL_API_GET_HIGH_CRITICAL_THRESHOLD:"cpu_core{}_crit"
 }
 thermal_api_handler_cpu_pack = {
     THERMAL_API_GET_TEMPERATURE:"cpu_pack",
-    THERMAL_API_GET_HIGH_THRESHOLD:"cpu_pack_max"
+    THERMAL_API_GET_HIGH_THRESHOLD:"cpu_pack_max",
+    THERMAL_API_GET_HIGH_CRITICAL_THRESHOLD:"cpu_pack_crit"
 }
 thermal_api_handler_module = {
     THERMAL_API_GET_TEMPERATURE:"module{}_temp_input",
-    THERMAL_API_GET_HIGH_THRESHOLD:"module{}_temp_crit"
+    THERMAL_API_GET_HIGH_THRESHOLD:"module{}_temp_crit",
+    THERMAL_API_GET_HIGH_CRITICAL_THRESHOLD:"module{}_temp_emergency"
 }
 thermal_api_handler_psu = {
     THERMAL_API_GET_TEMPERATURE:"psu{}_temp",
-    THERMAL_API_GET_HIGH_THRESHOLD:"psu{}_temp_max"
+    THERMAL_API_GET_HIGH_THRESHOLD:"psu{}_temp_max",
+    THERMAL_API_GET_HIGH_CRITICAL_THRESHOLD:None
 }
 thermal_api_handler_gearbox = {
     THERMAL_API_GET_TEMPERATURE:"temp_input_gearbox{}",
-    THERMAL_API_GET_HIGH_THRESHOLD:None
+    THERMAL_API_GET_HIGH_THRESHOLD:None,
+    THERMAL_API_GET_HIGH_CRITICAL_THRESHOLD:None
 }
 thermal_ambient_apis = {
     THERMAL_DEV_ASIC_AMBIENT : "asic",
@@ -233,6 +241,7 @@
     },
 ]
 
+
 def initialize_thermals(sku, thermal_list, psu_list):
     # create thermal objects for all categories of sensors
     tp_index = hwsku_dict_thermal[sku]
@@ -262,6 +271,8 @@ def initialize_thermals(sku, thermal_list, psu_list):
                         thermal = Thermal(category, start + index, True)
                         thermal_list.append(thermal)
 
+
+
 class Thermal(ThermalBase):
     def __init__(self, category, index, has_index, dependency = None, hint = None):
         """
@@ -280,9 +291,11 @@ def __init__(self, category, index, has_index, dependency = None, hint = None):
         self.category = category
         self.temperature = self._get_file_from_api(THERMAL_API_GET_TEMPERATURE)
         self.high_threshold = self._get_file_from_api(THERMAL_API_GET_HIGH_THRESHOLD)
+        self.high_critical_threshold = self._get_file_from_api(THERMAL_API_GET_HIGH_CRITICAL_THRESHOLD)
         self.dependency = dependency
         self.dependent_hint = hint
 
+
     def get_name(self):
         """
         Retrieves the name of the device
@@ -292,6 +305,7 @@ def get_name(self):
         """
         return self.name
 
+
     def _read_generic_file(self, filename, len):
         """
         Read a generic file, returns the contents of the file
@@ -304,6 +318,7 @@ def _read_generic_file(self, filename, len):
             logger.log_info("Fail to read file {} due to {}".format(filename, repr(e)))
         return result
 
+
     def _get_file_from_api(self, api_name):
         if self.category == THERMAL_DEV_CATEGORY_AMBIENT:
             if api_name == THERMAL_API_GET_TEMPERATURE:
@@ -315,9 +330,13 @@ def _get_file_from_api(self, api_name):
             if self.category in thermal_device_categories_singleton:
                 filename = handler
             else:
-                filename = handler.format(self.index)
+                if handler:
+                    filename = handler.format(self.index)
+                else:
+                    return None
         return join(HW_MGMT_THERMAL_ROOT, filename)
 
+
     def get_temperature(self):
         """
         Retrieves current temperature reading from thermal
@@ -337,8 +356,11 @@ def get_temperature(self):
         if value_str is None:
             return None
         value_float = float(value_str)
+        if self.category == THERMAL_DEV_CATEGORY_MODULE and value_float == THERMAL_API_INVALID_HIGH_THRESHOLD:
+            return None
         return value_float / 1000.0
 
+
     def get_high_threshold(self):
         """
         Retrieves the high threshold temperature of thermal
@@ -353,4 +375,25 @@ def get_high_threshold(self):
         if value_str is None:
             return None
         value_float = float(value_str)
+        if self.category == THERMAL_DEV_CATEGORY_MODULE and value_float == THERMAL_API_INVALID_HIGH_THRESHOLD:
+            return None
+        return value_float / 1000.0
+
+
+    def get_high_critical_threshold(self):
+        """
+        Retrieves the high critical threshold temperature of thermal
+
+        Returns:
+            A float number, the high critical threshold temperature of thermal in Celsius
+            up to nearest thousandth of one degree Celsius, e.g. 30.125
+        """
+        if self.high_critical_threshold is None:
+            return None
+        value_str = self._read_generic_file(self.high_critical_threshold, 0)
+        if value_str is None:
+            return None
+        value_float = float(value_str)
+        if self.category == THERMAL_DEV_CATEGORY_MODULE and value_float == THERMAL_API_INVALID_HIGH_THRESHOLD:
+            return None
         return value_float / 1000.0