Add thermal control support for SONiC (#3949)

device/mellanox/x86_64-mlnx_lssn2700-r0/thermal_policy.json

@@ -0,0 +1 @@
+../x86_64-mlnx_msn2700-r0/thermal_policy.json

device/mellanox/x86_64-mlnx_msn2010-r0/thermal_policy.json

@@ -0,0 +1 @@
+../x86_64-mlnx_msn2700-r0/thermal_policy.json

device/mellanox/x86_64-mlnx_msn2100-r0/thermal_policy.json

@@ -0,0 +1 @@
+../x86_64-mlnx_msn2700-r0/thermal_policy.json

device/mellanox/x86_64-mlnx_msn2410-r0/thermal_policy.json

@@ -0,0 +1 @@
+../x86_64-mlnx_msn2700-r0/thermal_policy.json

device/mellanox/x86_64-mlnx_msn2700-r0/thermal_policy.json

@@ -0,0 +1,72 @@
+{   
+    "thermal_control_algorithm": {
+        "run_at_boot_up": "false",
+        "fan_speed_when_suspend": "60"
+    },
+    "info_types": [
+        {
+            "type": "fan_info"
+        },
+        {
+            "type": "psu_info"
+        },
+        {
+            "type": "chassis_info"
+        }
+    ],
+    "policies": [
+        {
+            "name": "any fan absence",
+            "conditions": [
+                {
+                    "type": "fan.any.absence"
+                }
+            ],
+            "actions": [
+                {
+                    "type": "thermal_control.control",
+                    "status": "false"
+                },
+                {
+                    "type": "fan.all.set_speed",
+                    "speed": "100"
+                }
+            ]
+        },
+        {
+            "name": "any psu absence",
+            "conditions": [
+                {
+                    "type": "psu.any.absence"
+                }
+            ],
+            "actions": [
+                {
+                    "type": "thermal_control.control",
+                    "status": "false"
+                },
+                {
+                    "type": "fan.all.set_speed",
+                    "speed": "100"
+                }
+            ]
+        },
+        {
+            "name": "all fan and psu presence",
+            "conditions": [
+                {
+                    "type": "fan.all.presence"
+                },
+                {
+                    "type": "psu.all.presence"
+                }
+            ],
+            "actions": [
+                {
+                    "type": "fan.all.set_speed",
+                    "speed": "60"
+                }
+            ]
+        }
+    ]
+}

device/mellanox/x86_64-mlnx_msn2700_simx-r0/thermal_policy.json

@@ -0,0 +1 @@
+../x86_64-mlnx_msn2700-r0/thermal_policy.json

device/mellanox/x86_64-mlnx_msn2740-r0/thermal_policy.json

@@ -0,0 +1 @@
+../x86_64-mlnx_msn2700-r0/thermal_policy.json

device/mellanox/x86_64-mlnx_msn3700-r0/thermal_policy.json

@@ -0,0 +1 @@
+../x86_64-mlnx_msn2700-r0/thermal_policy.json

device/mellanox/x86_64-mlnx_msn3700_simx-r0/thermal_policy.json

@@ -0,0 +1 @@
+../x86_64-mlnx_msn2700-r0/thermal_policy.json

device/mellanox/x86_64-mlnx_msn3700c-r0/thermal_policy.json

@@ -0,0 +1 @@
+../x86_64-mlnx_msn2700-r0/thermal_policy.json

device/mellanox/x86_64-mlnx_msn3800-r0/thermal_policy.json

@@ -0,0 +1 @@
+../x86_64-mlnx_msn2700-r0/thermal_policy.json

dockers/docker-platform-monitor/docker-pmon.supervisord.conf.j2

@@ -91,3 +91,14 @@ stdout_logfile=syslog
 stderr_logfile=syslog
 startsecs=0
 {% endif %}
+
+{% if not skip_thermalctld %}
+[program:thermalctld]
+command=/usr/bin/thermalctld
+priority=9
+autostart=false
+autorestart=true
+stdout_logfile=syslog
+stderr_logfile=syslog
+startsecs=0
+{% endif %}

dockers/docker-platform-monitor/start.sh.j2

@@ -75,3 +75,7 @@ supervisorctl start psud
 supervisorctl start syseepromd
 {% endif %}
 
+{% if not skip_thermalctld %}
+supervisorctl start thermalctld
+{% endif %}
+

platform/mellanox/hw-management/0002-hw-management.sh-Disable-thermal-policy-running-in-h.patch

@@ -0,0 +1,31 @@
+From 76b02916794be2e2558fcff1d11609a594f633d7 Mon Sep 17 00:00:00 2001
+From: Stephen Sun <stephens@mellanox.com>
+Date: Fri, 14 Feb 2020 13:48:00 +0800
+Subject: [PATCH] Disable thermal policy running in hw-mgmt service SONiC
+ thermal control algorithm has been supported.
+
+Signed-off-by: Stephen Sun <stephens@mellanox.com>
+---
+ usr/usr/bin/hw-management.sh | 6 +++++-
+ 1 file changed, 5 insertions(+), 1 deletion(-)
+
+diff --git a/usr/usr/bin/hw-management.sh b/usr/usr/bin/hw-management.sh
+index 2cdbfb2..48b41d5 100755
+--- a/usr/usr/bin/hw-management.sh
++++ b/usr/usr/bin/hw-management.sh
+@@ -799,7 +799,11 @@ do_start()
+ 	#disabled for leopard chipless bringup.
+ 	echo 1 > $config_path/suspend
+ 
+-	$THERMAL_CONTROL $thermal_type $max_tachos $max_psus&
++#
++#	Disable thermal control algorithm in hw-management service
++#	because there has already been that in SONiC
++#
++#	$THERMAL_CONTROL $thermal_type $max_tachos $max_psus&
+ }
+ 
+ do_stop()
+-- 
+1.9.1
+

platform/mellanox/mlnx-platform-api.mk

@@ -3,6 +3,7 @@
 SONIC_PLATFORM_API_PY2 = mlnx_platform_api-1.0-py2-none-any.whl
 $(SONIC_PLATFORM_API_PY2)_SRC_PATH = $(PLATFORM_PATH)/mlnx-platform-api
 $(SONIC_PLATFORM_API_PY2)_PYTHON_VERSION = 2
+$(SONIC_PLATFORM_API_PY2)_DEPENDS = $(SONIC_PLATFORM_COMMON_PY2) $(SONIC_DAEMON_BASE_PY2) $(SONIC_CONFIG_ENGINE)
 SONIC_PYTHON_WHEELS += $(SONIC_PLATFORM_API_PY2)
 
 export mlnx_platform_api_py2_wheel_path="$(addprefix $(PYTHON_WHEELS_PATH)/,$(SONIC_PLATFORM_API_PY2))"

platform/mellanox/mlnx-platform-api/.gitignore

@@ -0,0 +1,2 @@
+*.pyc
+.cache/

platform/mellanox/mlnx-platform-api/pytest.ini

@@ -0,0 +1,3 @@
+[pytest]
+filterwarnings =
+    ignore::DeprecationWarning

platform/mellanox/mlnx-platform-api/setup.cfg

@@ -0,0 +1,2 @@
+[aliases]
+test=pytest

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

@@ -12,6 +12,14 @@ setup(
     maintainer_email='kevinw@mellanox.com',
     packages=[
         'sonic_platform',
+        'tests'
+    ],
+    setup_requires= [
+        'pytest-runner'
+    ],
+    tests_require = [
+        'pytest',
+        'mock>=2.0.0'
     ],
     classifiers=[
         'Development Status :: 3 - Alpha',
@@ -26,5 +34,6 @@ setup(
         'Topic :: Utilities',
     ],
     keywords='sonic SONiC platform PLATFORM',
+    test_suite='setup.get_test_suite'
 )
 

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

@@ -1,2 +1,2 @@
 __all__ = ["platform", "chassis"]
-from sonic_platform import *
+from sonic_platform import *

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

@@ -442,3 +442,8 @@ class Chassis(ChassisBase):
             return True, {'sfp':port_dict}
         else:
             return True, {'sfp':{}}
+
+    def get_thermal_manager(self):
+        from .thermal_manager import ThermalManager
+        return ThermalManager
+

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

@@ -38,10 +38,12 @@ class Fan(FanBase):
             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 = 'psu_{}_fan_{}'.format(self.index, 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)
@@ -49,6 +51,9 @@ class Fan(FanBase):
         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 @@ class Fan(FanBase):
                 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 @@ class Fan(FanBase):
         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())

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

@@ -24,6 +24,7 @@ class Platform(PlatformBase):
             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 @@ class Psu(PsuBase):
         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 @@ class Psu(PsuBase):
 
         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

@@ -36,24 +36,31 @@ THERMAL_DEV_BOARD_AMBIENT = "board_amb"
 
 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{}",
@@ -281,6 +288,7 @@ thermal_profile_list = [
     }
 ]
 
+
 def initialize_thermals(sku, thermal_list, psu_list):
     # create thermal objects for all categories of sensors
     tp_index = hwsku_dict_thermal[sku]
@@ -310,6 +318,8 @@ 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):
         """
@@ -328,9 +338,11 @@ class Thermal(ThermalBase):
         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
@@ -340,6 +352,7 @@ class Thermal(ThermalBase):
         """
         return self.name
 
+
     def _read_generic_file(self, filename, len):
         """
         Read a generic file, returns the contents of the file
@@ -352,6 +365,7 @@ class Thermal(ThermalBase):
             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:
@@ -363,9 +377,13 @@ class Thermal(ThermalBase):
             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
@@ -385,8 +403,11 @@ class Thermal(ThermalBase):
         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
@@ -401,4 +422,25 @@ class Thermal(ThermalBase):
         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

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

@@ -0,0 +1,108 @@
+from sonic_platform_base.sonic_thermal_control.thermal_action_base import ThermalPolicyActionBase
+from sonic_platform_base.sonic_thermal_control.thermal_json_object import thermal_json_object
+
+
+class SetFanSpeedAction(ThermalPolicyActionBase):
+    """
+    Base thermal action class to set speed for fans
+    """
+    # JSON field definition
+    JSON_FIELD_SPEED = 'speed'
+
+    def __init__(self):
+        """
+        Constructor of SetFanSpeedAction which actually do nothing.
+        """
+        self.speed = None
+
+    def load_from_json(self, json_obj):
+        """
+        Construct SetFanSpeedAction via JSON. JSON example:
+            {
+                "type": "fan.all.set_speed"
+                "speed": "100"
+            }
+        :param json_obj: A JSON object representing a SetFanSpeedAction action.
+        :return:
+        """
+        if SetFanSpeedAction.JSON_FIELD_SPEED in json_obj:
+            speed = float(json_obj[SetFanSpeedAction.JSON_FIELD_SPEED])
+            if speed < 0 or speed > 100:
+                raise ValueError('SetFanSpeedAction invalid speed value {} in JSON policy file, valid value should be [0, 100]'.
+                                 format(speed))
+            self.speed = float(json_obj[SetFanSpeedAction.JSON_FIELD_SPEED])
+        else:
+            raise ValueError('SetFanSpeedAction missing mandatory field {} in JSON policy file'.
+                             format(SetFanSpeedAction.JSON_FIELD_SPEED))
+
+
+@thermal_json_object('fan.all.set_speed')
+class SetAllFanSpeedAction(SetFanSpeedAction):
+    """
+    Action to set speed for all fans
+    """
+    def execute(self, thermal_info_dict):
+        """
+        Set speed for all fans
+        :param thermal_info_dict: A dictionary stores all thermal information.
+        :return:
+        """
+        from .thermal_infos import FanInfo
+        if FanInfo.INFO_NAME in thermal_info_dict and isinstance(thermal_info_dict[FanInfo.INFO_NAME], FanInfo):
+            fan_info_obj = thermal_info_dict[FanInfo.INFO_NAME]
+            for fan in fan_info_obj.get_presence_fans():
+                fan.set_speed(self.speed)
+
+
+@thermal_json_object('thermal_control.control')
+class ControlThermalAlgoAction(ThermalPolicyActionBase):
+    """
+    Action to control the thermal control algorithm
+    """
+    # JSON field definition
+    JSON_FIELD_STATUS = 'status'
+
+    def __init__(self):
+        self.status = True
+
+    def load_from_json(self, json_obj):
+        """
+        Construct ControlThermalAlgoAction via JSON. JSON example:
+            {
+                "type": "thermal_control.control"
+                "status": "true"
+            }
+        :param json_obj: A JSON object representing a ControlThermalAlgoAction action.
+        :return:
+        """
+        if ControlThermalAlgoAction.JSON_FIELD_STATUS in json_obj:
+            status_str = json_obj[ControlThermalAlgoAction.JSON_FIELD_STATUS].lower()
+            if status_str == 'true':
+                self.status = True
+            elif status_str == 'false':
+                self.status = False
+            else:
+                raise ValueError('Invalid {} field value, please specify true of false'.
+                                 format(ControlThermalAlgoAction.JSON_FIELD_STATUS))
+        else:
+            raise ValueError('ControlThermalAlgoAction '
+                             'missing mandatory field {} in JSON policy file'.
+                             format(ControlThermalAlgoAction.JSON_FIELD_STATUS))
+
+    def execute(self, thermal_info_dict):
+        """
+        Disable thermal control algorithm
+        :param thermal_info_dict: A dictionary stores all thermal information.
+        :return:
+        """
+        from .thermal_infos import ChassisInfo
+        if ChassisInfo.INFO_NAME in thermal_info_dict:
+            chassis_info_obj = thermal_info_dict[ChassisInfo.INFO_NAME]
+            chassis = chassis_info_obj.get_chassis()
+            thermal_manager = chassis.get_thermal_manager()
+            if self.status:
+                thermal_manager.start_thermal_control_algorithm()
+            else:
+                thermal_manager.stop_thermal_control_algorithm()
+
+

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

@@ -0,0 +1,63 @@
+from sonic_platform_base.sonic_thermal_control.thermal_condition_base import ThermalPolicyConditionBase
+from sonic_platform_base.sonic_thermal_control.thermal_json_object import thermal_json_object
+
+
+class FanCondition(ThermalPolicyConditionBase):
+    def get_fan_info(self, thermal_info_dict):
+        from .thermal_infos import FanInfo
+        if FanInfo.INFO_NAME in thermal_info_dict and isinstance(thermal_info_dict[FanInfo.INFO_NAME], FanInfo):
+            return thermal_info_dict[FanInfo.INFO_NAME]
+        else:
+            return None
+
+
+@thermal_json_object('fan.any.absence')
+class AnyFanAbsenceCondition(FanCondition):
+    def is_match(self, thermal_info_dict):
+        fan_info_obj = self.get_fan_info(thermal_info_dict)
+        return len(fan_info_obj.get_absence_fans()) > 0 if fan_info_obj else False
+
+
+@thermal_json_object('fan.all.absence')
+class AllFanAbsenceCondition(FanCondition):
+    def is_match(self, thermal_info_dict):
+        fan_info_obj = self.get_fan_info(thermal_info_dict)
+        return len(fan_info_obj.get_presence_fans()) == 0 if fan_info_obj else False
+
+
+@thermal_json_object('fan.all.presence')
+class AllFanPresenceCondition(FanCondition):
+    def is_match(self, thermal_info_dict):
+        fan_info_obj = self.get_fan_info(thermal_info_dict)
+        return len(fan_info_obj.get_absence_fans()) == 0 if fan_info_obj else False
+
+
+class PsuCondition(ThermalPolicyConditionBase):
+    def get_psu_info(self, thermal_info_dict):
+        from .thermal_infos import PsuInfo
+        if PsuInfo.INFO_NAME in thermal_info_dict and isinstance(thermal_info_dict[PsuInfo.INFO_NAME], PsuInfo):
+            return thermal_info_dict[PsuInfo.INFO_NAME]
+        else:
+            return None
+
+
+@thermal_json_object('psu.any.absence')
+class AnyPsuAbsenceCondition(PsuCondition):
+    def is_match(self, thermal_info_dict):
+        psu_info_obj = self.get_psu_info(thermal_info_dict)
+        return len(psu_info_obj.get_absence_psus()) > 0 if psu_info_obj else False
+
+
+@thermal_json_object('psu.all.absence')
+class AllPsuAbsenceCondition(PsuCondition):
+    def is_match(self, thermal_info_dict):
+        psu_info_obj = self.get_psu_info(thermal_info_dict)
+        return len(psu_info_obj.get_presence_psus()) == 0 if psu_info_obj else False
+
+
+@thermal_json_object('psu.all.presence')
+class AllPsuPresenceCondition(PsuCondition):
+    def is_match(self, thermal_info_dict):
+        psu_info_obj = self.get_psu_info(thermal_info_dict)
+        return len(psu_info_obj.get_absence_psus()) == 0 if psu_info_obj else False
+

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

@@ -0,0 +1,136 @@
+from sonic_platform_base.sonic_thermal_control.thermal_info_base import ThermalPolicyInfoBase
+from sonic_platform_base.sonic_thermal_control.thermal_json_object import thermal_json_object
+
+
+@thermal_json_object('fan_info')
+class FanInfo(ThermalPolicyInfoBase):
+    """
+    Fan information needed by thermal policy
+    """
+
+    # Fan information name
+    INFO_NAME = 'fan_info'
+
+    def __init__(self):
+        self._absence_fans = set()
+        self._presence_fans = set()
+        self._status_changed = False
+
+    def collect(self, chassis):
+        """
+        Collect absence and presence fans.
+        :param chassis: The chassis object
+        :return:
+        """
+        self._status_changed = False
+        for fan in chassis.get_all_fans():
+            if fan.get_presence() and fan not in self._presence_fans:
+                self._presence_fans.add(fan)
+                self._status_changed = True
+                if fan in self._absence_fans:
+                    self._absence_fans.remove(fan)
+            elif not fan.get_presence() and fan not in self._absence_fans:
+                self._absence_fans.add(fan)
+                self._status_changed = True
+                if fan in self._presence_fans:
+                    self._presence_fans.remove(fan)
+
+    def get_absence_fans(self):
+        """
+        Retrieves absence fans
+        :return: A set of absence fans
+        """
+        return self._absence_fans
+
+    def get_presence_fans(self):
+        """
+        Retrieves presence fans
+        :return: A set of presence fans
+        """
+        return self._presence_fans
+
+    def is_status_changed(self):
+        """
+        Retrieves if the status of fan information changed
+        :return: True if status changed else False
+        """
+        return self._status_changed
+
+
+@thermal_json_object('psu_info')
+class PsuInfo(ThermalPolicyInfoBase):
+    """
+    PSU information needed by thermal policy
+    """
+    INFO_NAME = 'psu_info'
+
+    def __init__(self):
+        self._absence_psus = set()
+        self._presence_psus = set()
+        self._status_changed = False
+
+    def collect(self, chassis):
+        """
+        Collect absence and presence PSUs.
+        :param chassis: The chassis object
+        :return:
+        """
+        self._status_changed = False
+        for psu in chassis.get_all_psus():
+            if psu.get_presence() and psu not in self._presence_psus:
+                self._presence_psus.add(psu)
+                self._status_changed = True
+                if psu in self._absence_psus:
+                    self._absence_psus.remove(psu)
+            elif not psu.get_presence() and psu not in self._absence_psus:
+                self._absence_psus.add(psu)
+                self._status_changed = True
+                if psu in self._presence_psus:
+                    self._presence_psus.remove(psu)
+
+    def get_absence_psus(self):
+        """
+        Retrieves presence PSUs
+        :return: A set of absence PSUs
+        """
+        return self._absence_psus
+
+    def get_presence_psus(self):
+        """
+        Retrieves presence PSUs
+        :return: A set of presence fans
+        """
+        return self._presence_psus
+
+    def is_status_changed(self):
+        """
+        Retrieves if the status of PSU information changed
+        :return: True if status changed else False
+        """
+        return self._status_changed
+
+
+@thermal_json_object('chassis_info')
+class ChassisInfo(ThermalPolicyInfoBase):
+    """
+    Chassis information needed by thermal policy
+    """
+    INFO_NAME = 'chassis_info'
+
+    def __init__(self):
+        self._chassis = None
+
+    def collect(self, chassis):
+        """
+        Collect platform chassis.
+        :param chassis: The chassis object
+        :return:
+        """
+        self._chassis = chassis
+
+    def get_chassis(self):
+        """
+        Retrieves platform chassis object
+        :return: A platform chassis object.
+        """
+        return self._chassis

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

@@ -0,0 +1,50 @@
+import os
+from sonic_platform_base.sonic_thermal_control.thermal_manager_base import ThermalManagerBase
+from .thermal_actions import *
+from .thermal_conditions import *
+from .thermal_infos import *
+
+
+class ThermalManager(ThermalManagerBase):
+    THERMAL_ALGORITHM_CONTROL_PATH = '/var/run/hw-management/config/suspend'
+
+    @classmethod
+    def start_thermal_control_algorithm(cls):
+        """
+        Start thermal control algorithm
+
+        Returns:
+            bool: True if set success, False if fail. 
+        """
+        cls._control_thermal_control_algorithm(False)
+
+    @classmethod
+    def stop_thermal_control_algorithm(cls):
+        """
+        Stop thermal control algorithm
+
+        Returns:
+            bool: True if set success, False if fail. 
+        """
+        cls._control_thermal_control_algorithm(True)
+
+    @classmethod
+    def _control_thermal_control_algorithm(cls, suspend):
+        """
+        Control thermal control algorithm
+
+        Args:
+            suspend: Bool, indicate suspend the algorithm or not
+
+        Returns:
+            bool: True if set success, False if fail. 
+        """
+        status = True
+        write_value = 1 if suspend else 0
+        try:
+            with open(cls.THERMAL_ALGORITHM_CONTROL_PATH, 'w') as control_file:
+                control_file.write(str(write_value))
+        except (ValueError, IOError):
+            status = False
+
+        return status

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

@@ -0,0 +1,44 @@
+class MockFan:
+    def __init__(self):
+        self.presence = True
+        self.speed = 60
+
+    def get_presence(self):
+        return self.presence
+
+    def set_speed(self, speed):
+        self.speed = speed
+
+
+class MockPsu:
+    def __init__(self):
+        self.presence = True
+
+    def get_presence(self):
+        return self.presence
+
+
+class MockChassis:
+    def __init__(self):
+        self.fan_list = []
+        self.psu_list = []
+
+    def get_all_psus(self):
+        return self.psu_list
+
+    def get_all_fans(self):
+        return self.fan_list
+
+    def get_thermal_manager(self):
+        from sonic_platform.thermal_manager import ThermalManager
+        return ThermalManager
+
+    def make_fan_absence(self):
+        fan = MockFan()
+        fan.presence = False
+        self.fan_list.append(fan)
+
+    def make_psu_absence(self):
+        psu = MockPsu()
+        psu.presence = False
+        self.psu_list.append(psu)

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

@@ -0,0 +1,272 @@
+import os
+import sys
+import pytest
+import json
+from mock import MagicMock
+from .mock_platform import MockChassis, MockFan, MockPsu
+
+test_path = os.path.dirname(os.path.abspath(__file__))
+modules_path = os.path.dirname(test_path)
+sys.path.insert(0, modules_path)
+
+from sonic_platform.thermal_manager import ThermalManager
+from sonic_platform.thermal_infos import FanInfo, PsuInfo
+
+
+@pytest.fixture(scope='session', autouse=True)
+def thermal_manager():
+    policy_file = os.path.join(test_path, 'thermal_policy.json')
+    ThermalManager.load(policy_file)
+    return ThermalManager
+
+
+def test_load_policy(thermal_manager):
+    assert 'psu_info' in thermal_manager._thermal_info_dict
+    assert 'fan_info' in thermal_manager._thermal_info_dict
+    assert 'chassis_info' in thermal_manager._thermal_info_dict
+
+    assert 'any fan absence' in thermal_manager._policy_dict
+    assert 'any psu absence' in thermal_manager._policy_dict
+    assert 'all fan and psu presence' in thermal_manager._policy_dict
+
+    assert thermal_manager._fan_speed_when_suspend == 60
+    assert thermal_manager._run_thermal_algorithm_at_boot_up == False
+
+
+def test_fan_info():
+    chassis = MockChassis()
+    chassis.make_fan_absence()
+    fan_info = FanInfo()
+    fan_info.collect(chassis)
+    assert len(fan_info.get_absence_fans()) == 1
+    assert len(fan_info.get_presence_fans()) == 0
+    assert fan_info.is_status_changed()
+
+    fan_list = chassis.get_all_fans()
+    fan_list[0].presence = True
+    fan_info.collect(chassis)
+    assert len(fan_info.get_absence_fans()) == 0
+    assert len(fan_info.get_presence_fans()) == 1
+    assert fan_info.is_status_changed()
+
+
+def test_psu_info():
+    chassis = MockChassis()
+    chassis.make_psu_absence()
+    psu_info = PsuInfo()
+    psu_info.collect(chassis)
+    assert len(psu_info.get_absence_psus()) == 1
+    assert len(psu_info.get_presence_psus()) == 0
+    assert psu_info.is_status_changed()
+
+    psu_list = chassis.get_all_psus()
+    psu_list[0].presence = True
+    psu_info.collect(chassis)
+    assert len(psu_info.get_absence_psus()) == 0
+    assert len(psu_info.get_presence_psus()) == 1
+    assert psu_info.is_status_changed()
+
+
+def test_fan_policy(thermal_manager):
+    chassis = MockChassis()
+    chassis.make_fan_absence()
+    chassis.fan_list.append(MockFan())
+    thermal_manager.start_thermal_control_algorithm = MagicMock()
+    thermal_manager.stop_thermal_control_algorithm = MagicMock()
+    thermal_manager.run_policy(chassis)
+
+    fan_list = chassis.get_all_fans()
+    assert fan_list[1].speed == 100
+    thermal_manager.stop_thermal_control_algorithm.assert_called_once()
+
+    fan_list[0].presence = True
+    thermal_manager.run_policy(chassis)
+    thermal_manager.start_thermal_control_algorithm.assert_called_once()
+
+
+def test_psu_policy(thermal_manager):
+    chassis = MockChassis()
+    chassis.make_psu_absence()
+    chassis.fan_list.append(MockFan())
+    thermal_manager.start_thermal_control_algorithm = MagicMock()
+    thermal_manager.stop_thermal_control_algorithm = MagicMock()
+    thermal_manager.run_policy(chassis)
+
+    fan_list = chassis.get_all_fans()
+    assert fan_list[0].speed == 100
+    thermal_manager.stop_thermal_control_algorithm.assert_called_once()
+
+    psu_list = chassis.get_all_psus()
+    psu_list[0].presence = True
+    thermal_manager.run_policy(chassis)
+    thermal_manager.start_thermal_control_algorithm.assert_called_once()
+
+
+def test_any_fan_absence_condition():
+    chassis = MockChassis()
+    chassis.make_fan_absence()
+    fan_info = FanInfo()
+    fan_info.collect(chassis)
+
+    from sonic_platform.thermal_conditions import AnyFanAbsenceCondition
+    condition = AnyFanAbsenceCondition()
+    assert condition.is_match({'fan_info': fan_info})
+
+    fan = chassis.get_all_fans()[0]
+    fan.presence = True
+    fan_info.collect(chassis)
+    assert not condition.is_match({'fan_info': fan_info})
+
+
+def test_all_fan_absence_condition():
+    chassis = MockChassis()
+    chassis.make_fan_absence()
+    fan = MockFan()
+    fan_list = chassis.get_all_fans()
+    fan_list.append(fan)
+    fan_info = FanInfo()
+    fan_info.collect(chassis)
+
+    from sonic_platform.thermal_conditions import AllFanAbsenceCondition
+    condition = AllFanAbsenceCondition()
+    assert not condition.is_match({'fan_info': fan_info})
+
+    fan.presence = False
+    fan_info.collect(chassis)
+    assert condition.is_match({'fan_info': fan_info})
+
+
+def test_all_fan_presence_condition():
+    chassis = MockChassis()
+    chassis.make_fan_absence()
+    fan = MockFan()
+    fan_list = chassis.get_all_fans()
+    fan_list.append(fan)
+    fan_info = FanInfo()
+    fan_info.collect(chassis)
+
+    from sonic_platform.thermal_conditions import AllFanPresenceCondition
+    condition = AllFanPresenceCondition()
+    assert not condition.is_match({'fan_info': fan_info})
+
+    fan_list[0].presence = True
+    fan_info.collect(chassis)
+    assert condition.is_match({'fan_info': fan_info})
+
+
+def test_any_psu_absence_condition():
+    chassis = MockChassis()
+    chassis.make_psu_absence()
+    psu_info = PsuInfo()
+    psu_info.collect(chassis)
+
+    from sonic_platform.thermal_conditions import AnyPsuAbsenceCondition
+    condition = AnyPsuAbsenceCondition()
+    assert condition.is_match({'psu_info': psu_info})
+
+    psu = chassis.get_all_psus()[0]
+    psu.presence = True
+    psu_info.collect(chassis)
+    assert not condition.is_match({'psu_info': psu_info})
+
+
+def test_all_psu_absence_condition():
+    chassis = MockChassis()
+    chassis.make_psu_absence()
+    psu = MockPsu()
+    psu_list = chassis.get_all_psus()
+    psu_list.append(psu)
+    psu_info = PsuInfo()
+    psu_info.collect(chassis)
+
+    from sonic_platform.thermal_conditions import AllPsuAbsenceCondition
+    condition = AllPsuAbsenceCondition()
+    assert not condition.is_match({'psu_info': psu_info})
+
+    psu.presence = False
+    psu_info.collect(chassis)
+    assert condition.is_match({'psu_info': psu_info})
+
+
+def test_all_fan_presence_condition():
+    chassis = MockChassis()
+    chassis.make_psu_absence()
+    psu = MockPsu()
+    psu_list = chassis.get_all_psus()
+    psu_list.append(psu)
+    psu_info = PsuInfo()
+    psu_info.collect(chassis)
+
+    from sonic_platform.thermal_conditions import AllPsuPresenceCondition
+    condition = AllPsuPresenceCondition()
+    assert not condition.is_match({'psu_info': psu_info})
+
+    psu_list[0].presence = True
+    psu_info.collect(chassis)
+    assert condition.is_match({'psu_info': psu_info})
+
+
+def test_load_set_fan_speed_action():
+    from sonic_platform.thermal_actions import SetAllFanSpeedAction
+    action = SetAllFanSpeedAction()
+    json_str = '{\"speed\": \"50\"}'
+    json_obj = json.loads(json_str)
+    action.load_from_json(json_obj)
+    assert action.speed == 50
+
+    json_str = '{\"speed\": \"-1\"}'
+    json_obj = json.loads(json_str)
+    with pytest.raises(ValueError):
+        action.load_from_json(json_obj)
+
+    json_str = '{\"speed\": \"101\"}'
+    json_obj = json.loads(json_str)
+    with pytest.raises(ValueError):
+        action.load_from_json(json_obj)
+
+    json_str = '{\"invalid\": \"101\"}'
+    json_obj = json.loads(json_str)
+    with pytest.raises(ValueError):
+        action.load_from_json(json_obj)
+
+
+def test_execute_set_fan_speed_action():
+    chassis = MockChassis()
+    fan_list = chassis.get_all_fans()
+    fan_list.append(MockFan())
+    fan_list.append(MockFan())
+    fan_info = FanInfo()
+    fan_info.collect(chassis)
+
+    from sonic_platform.thermal_actions import SetAllFanSpeedAction
+    action = SetAllFanSpeedAction()
+    action.speed = 99
+    action.execute({'fan_info': fan_info})
+    assert fan_list[0].speed == 99
+    assert fan_list[1].speed == 99
+
+
+def test_load_control_thermal_algo_action():
+    from sonic_platform.thermal_actions import ControlThermalAlgoAction
+    action = ControlThermalAlgoAction()
+    json_str = '{\"status\": \"false\"}'
+    json_obj = json.loads(json_str)
+    action.load_from_json(json_obj)
+    assert not action.status 
+
+    json_str = '{\"status\": \"true\"}'
+    json_obj = json.loads(json_str)
+    action.load_from_json(json_obj)
+    assert action.status 
+
+    json_str = '{\"status\": \"invalid\"}'
+    json_obj = json.loads(json_str)
+    with pytest.raises(ValueError):
+        action.load_from_json(json_obj)
+
+    json_str = '{\"invalid\": \"true\"}'
+    json_obj = json.loads(json_str)
+    with pytest.raises(ValueError):
+        action.load_from_json(json_obj)
+
+

platform/mellanox/mlnx-platform-api/tests/thermal_policy.json

@@ -0,0 +1,72 @@
+{
+    "thermal_control_algorithm": {
+        "run_at_boot_up": "false",
+        "fan_speed_when_suspend": "60"
+    },
+    "info_types": [
+        {
+            "type": "fan_info"
+        },
+        {
+            "type": "psu_info"
+        },
+        {
+            "type": "chassis_info"
+        }
+    ],
+    "policies": [
+        {
+            "name": "any fan absence",
+            "conditions": [
+                {
+                    "type": "fan.any.absence"
+                }
+            ],
+            "actions": [
+                {
+                    "type": "thermal_control.control",
+                    "status": "false"
+                },
+                {
+                    "type": "fan.all.set_speed",
+                    "speed": "100"
+                }
+            ]
+        },
+        {
+            "name": "any psu absence",
+            "conditions": [
+                {
+                    "type": "psu.any.absence"
+                }
+            ],
+            "actions": [
+                {
+                    "type": "thermal_control.control",
+                    "status": "false"
+                },
+                {
+                    "type": "fan.all.set_speed",
+                    "speed": "100"
+                }
+            ]
+        },
+        {
+            "name": "all fan and psu presence",
+            "conditions": [
+                {
+                    "type": "fan.all.presence"
+                },
+                {
+                    "type": "psu.all.presence"
+                }
+            ],
+            "actions": [
+                {
+                    "type": "thermal_control.control",
+                    "status": "true"
+                }
+            ]
+        }
+    ]
+}

rules/docker-platform-monitor.mk

@@ -10,7 +10,7 @@ $(DOCKER_PLATFORM_MONITOR)_DEPENDS += $(LIBSENSORS) $(LM_SENSORS) $(FANCONTROL)
 ifeq ($(CONFIGURED_PLATFORM),barefoot)
 $(DOCKER_PLATFORM_MONITOR)_DEPENDS += $(PYTHON_THRIFT)
 endif
-$(DOCKER_PLATFORM_MONITOR)_PYTHON_DEBS += $(SONIC_LEDD) $(SONIC_XCVRD) $(SONIC_PSUD) $(SONIC_SYSEEPROMD)
+$(DOCKER_PLATFORM_MONITOR)_PYTHON_DEBS += $(SONIC_LEDD) $(SONIC_XCVRD) $(SONIC_PSUD) $(SONIC_SYSEEPROMD) $(SONIC_THERMALCTLD)
 $(DOCKER_PLATFORM_MONITOR)_PYTHON_WHEELS += $(SONIC_PLATFORM_COMMON_PY2)
 $(DOCKER_PLATFORM_MONITOR)_PYTHON_WHEELS += $(SWSSSDK_PY2)
 $(DOCKER_PLATFORM_MONITOR)_PYTHON_WHEELS += $(SONIC_PLATFORM_API_PY2)

rules/sonic-thermalctld.mk

@@ -0,0 +1,6 @@
+# sonic-thermalctld (SONiC Thermal control daemon) Debian package
+
+SONIC_THERMALCTLD = python-sonic-thermalctld_1.0-1_all.deb
+$(SONIC_THERMALCTLD)_SRC_PATH = $(SRC_PATH)/sonic-platform-daemons/sonic-thermalctld
+$(SONIC_THERMALCTLD)_WHEEL_DEPENDS = $(SONIC_DAEMON_BASE_PY2)
+SONIC_PYTHON_STDEB_DEBS += $(SONIC_THERMALCTLD)

src/sonic-daemon-base/sonic_daemon_base/task_base.py

@@ -0,0 +1,50 @@
+import multiprocessing
+import os
+import signal
+import threading
+
+
+#
+# ProcessTaskBase =====================================================================
+#
+class ProcessTaskBase(object):  # TODO: put this class to swss-platform-common
+    def __init__(self):
+        self.task_process = None
+        self.task_stopping_event = multiprocessing.Event()
+
+    def task_worker(self):
+        pass
+
+    def task_run(self):
+        if self.task_stopping_event.is_set():
+            return
+
+        self.task_process = multiprocessing.Process(target=self.task_worker)
+        self.task_process.start()
+
+    def task_stop(self):
+        self.task_stopping_event.set()
+        os.kill(self.task_process.pid, signal.SIGKILL)
+
+
+#
+# ThreadTaskBase =====================================================================
+#
+class ThreadTaskBase(object):  # TODO: put this class to swss-platform-common;
+    def __init__(self):
+        self.task_thread = None
+        self.task_stopping_event = threading.Event()
+
+    def task_worker(self):
+        pass
+
+    def task_run(self):
+        if self.task_stopping_event.is_set():
+            return
+
+        self.task_thread = threading.Thread(target=self.task_worker)
+        self.task_thread.start()
+
+    def task_stop(self):
+        self.task_stopping_event.set()
+        self.task_thread.join()