[device/celestica]: Add thermalctld support on DX010 platform APIs (#…

…6089)

**- Why I did it**
- The thermalctld daemon on the Pmon docker requires support from the thermal manager API.

**- How I did it**
- Removed the old function for detecting a faulty fan.
- Removed the old function for detecting excess temperature.
- Implement thermal_manager APIs based on ThermalManagerBase
- Implement thermal_conditions APIs based on ThermalPolicyConditionBase
- Implement thermal_actions APIs based on ThermalPolicyActionBase
- Implement thermal_info APIs based on ThermalPolicyInfoBase
- Add thermal_policy.json

device/celestica/x86_64-cel_seastone-r0/fancontrol-B2F

@@ -9,5 +9,4 @@ MINSTOP=13-002e/pwm1=89 13-002e/pwm2=89 13-002e/pwm3=89 13-002e/pwm4=89 13-002e/
 MINPWM=13-002e/pwm1=89 13-002e/pwm2=89 13-002e/pwm3=89 13-002e/pwm4=89 13-002e/pwm5=89 13-004d/pwm1=89 13-004d/pwm2=89 13-004d/pwm3=89 13-004d/pwm4=89 13-004d/pwm5=89
 MAXPWM=13-002e/pwm1=255 13-002e/pwm2=255 13-002e/pwm3=255 13-002e/pwm4=255 13-002e/pwm5=255 13-004d/pwm1=255 13-004d/pwm2=255 13-004d/pwm3=255 13-004d/pwm4=255 13-004d/pwm5=255
 THYST=13-002e/pwm1=3 13-002e/pwm2=3 13-002e/pwm3=3 13-002e/pwm4=3 13-002e/pwm5=3 13-004d/pwm1=3 13-004d/pwm2=3 13-004d/pwm3=3 13-004d/pwm4=3 13-004d/pwm5=3
-MAXTEMPCRIT=/sys/bus/i2c/devices/7-004a/hwmon/hwmon*/temp1_input=65 /sys/bus/i2c/devices/14-0048/hwmon/hwmon*/temp1_input=75
-MAXTEMPTYPE=/sys/bus/i2c/devices/7-004a/hwmon/hwmon*/temp1_input=ASIC /sys/bus/i2c/devices/14-0048/hwmon/hwmon*/temp1_input=CPU
+

device/celestica/x86_64-cel_seastone-r0/fancontrol-F2B

@@ -9,5 +9,4 @@ MINSTOP=13-002e/pwm1=89 13-002e/pwm2=89 13-002e/pwm3=89 13-002e/pwm4=89 13-002e/
 MINPWM=13-002e/pwm1=89 13-002e/pwm2=89 13-002e/pwm3=89 13-002e/pwm4=89 13-002e/pwm5=89 13-004d/pwm1=89 13-004d/pwm2=89 13-004d/pwm3=89 13-004d/pwm4=89 13-004d/pwm5=89
 MAXPWM=13-002e/pwm1=255 13-002e/pwm2=255 13-002e/pwm3=255 13-002e/pwm4=255 13-002e/pwm5=255 13-004d/pwm1=255 13-004d/pwm2=255 13-004d/pwm3=255 13-004d/pwm4=255 13-004d/pwm5=255
 THYST=13-002e/pwm1=3 13-002e/pwm2=3 13-002e/pwm3=3 13-002e/pwm4=3 13-002e/pwm5=3 13-004d/pwm1=3 13-004d/pwm2=3 13-004d/pwm3=3 13-004d/pwm4=3 13-004d/pwm5=3
-MAXTEMPCRIT=/sys/bus/i2c/devices/7-004a/hwmon/hwmon*/temp1_input=75 /sys/bus/i2c/devices/14-0048/hwmon/hwmon*/temp1_input=75
-MAXTEMPTYPE=/sys/bus/i2c/devices/7-004a/hwmon/hwmon*/temp1_input=ASIC /sys/bus/i2c/devices/14-0048/hwmon/hwmon*/temp1_input=CPU
+

device/celestica/x86_64-cel_seastone-r0/sonic_platform/chassis.py

@@ -68,8 +68,9 @@ def __initialize_fan(self):
 
  def __initialize_thermals(self):
  from sonic_platform.thermal import Thermal
+ airflow = self.__get_air_flow()
  for index in range(0, NUM_THERMAL):
- thermal = Thermal(index)
+ thermal = Thermal(index, airflow)
  self._thermal_list.append(thermal)
 
  def __initialize_eeprom(self):
@@ -82,6 +83,11 @@ def __initialize_components(self):
  component = Component(index)
  self._component_list.append(component)
 
+ def __get_air_flow(self):
+ air_flow_path = '/usr/share/sonic/device/{}/fan_airflow'.format(self._api_helper.platform) if self.is_host else '/usr/share/sonic/platform/fan_airflow'
+ air_flow = self._api_helper.read_one_line_file(air_flow_path)
+ return air_flow or 'B2F'
+
  def get_base_mac(self):
  """
  Retrieves the base MAC address for the chassis
@@ -261,3 +267,7 @@ def get_status(self):
  A boolean value, True if device is operating properly, False if not
  """
  return True
+
+ def get_thermal_manager(self):
+ from .thermal_manager import ThermalManager
+ return ThermalManager

device/celestica/x86_64-cel_seastone-r0/sonic_platform/fan.py

@@ -306,4 +306,20 @@ def get_status(self):
  Returns:
  A boolean value, True if device is operating properly, False if not
  """
- return self.get_presence() and self.get_speed() > 0
+ status = 1
+ if self.is_psu_fan:
+ fan_fault_sysfs_name = "fan1_fault"
+ fan_fault_sysfs_path = self.__search_file_by_name(
+ self.psu_hwmon_path, fan_fault_sysfs_name)
+ status = self._api_helper.read_one_line_file(fan_fault_sysfs_path)
+
+ elif self.get_presence():
+ chip = self.emc2305_chip_mapping[self.fan_index]
+ device = chip['device']
+ fan_index = chip['index_map']
+ sysfs_path = "%s%s/%s" % (
+ EMC2305_PATH, device, 'fan{}_fault')
+ sysfs_path = sysfs_path.format(fan_index[self.fan_tray_index])
+ status = self._api_helper.read_one_line_file(sysfs_path)
+
+ return False if int(status) != 0 else True

device/celestica/x86_64-cel_seastone-r0/sonic_platform/thermal.py

@@ -18,46 +18,75 @@
 except ImportError as e:
  raise ImportError(str(e) + "- required module not found")
 
+THERMAL_INFO = {
+ 0: {
+ "F2B_max": 50,
+ "B2F_max": 55,
+ "postion": "asic",
+ "name": "Front-panel temp sensor 1",
+ "i2c_path": "i2c-5/5-0048/hwmon/hwmon1", # u4 system-inlet
+ },
+ 1: {
+ "F2B_max": 50,
+ "B2F_max": 55,
+ "postion": "asic",
+ "name": "Front-panel temp sensor 2",
+ "i2c_path": "i2c-6/6-0049/hwmon/hwmon2", # u2 system-inlet
+ },
+ 2: {
+ "F2B_max": 70,
+ "F2B_max_crit": 75,
+ "B2F_max": 60,
+ "B2F_max_crit": 65,
+ "postion": "asic",
+ "name": "ASIC temp sensor",
+ "i2c_path": "i2c-7/7-004a/hwmon/hwmon3", # u44 bmc56960-on-board
+ },
+ 3: {
+ "F2B_max": 70,
+ "F2B_max_crit": 75,
+ "B2F_max": 70,
+ "B2F_max_crit": 75,
+ "postion": "cpu",
+ "name": "Rear-panel temp sensor 1",
+ "i2c_path": "i2c-14/14-0048/hwmon/hwmon4", # u9200 cpu-on-board
+ },
+ 4: {
+ "F2B_max": 70,
+ "B2F_max": 55,
+ "postion": "cpu",
+ "name": "Rear-panel temp sensor 2",
+ "i2c_path": "i2c-15/15-004e/hwmon/hwmon5" # u9201 system-outlet
+ }
+}
+NULL_VAL = "N/A"
+I2C_ADAPTER_PATH = "/sys/class/i2c-adapter"
+
 
 class Thermal(ThermalBase):
  """Platform-specific Thermal class"""
 
- THERMAL_NAME_LIST = []
- I2C_ADAPTER_PATH = "/sys/class/i2c-adapter"
  SS_CONFIG_PATH = "/usr/share/sonic/device/x86_64-cel_seastone-r0/sensors.conf"
 
- def __init__(self, thermal_index):
+ def __init__(self, thermal_index, airflow):
  self.index = thermal_index
  self._api_helper = APIHelper()
-
- # Add thermal name
- self.THERMAL_NAME_LIST.append("Front-panel temp sensor 1")
- self.THERMAL_NAME_LIST.append("Front-panel temp sensor 2")
- self.THERMAL_NAME_LIST.append("ASIC temp sensor")
- self.THERMAL_NAME_LIST.append("Rear-panel temp sensor 1")
- self.THERMAL_NAME_LIST.append("Rear-panel temp sensor 2")
-
- # Set hwmon path
- i2c_path = {
- 0: "i2c-5/5-0048/hwmon/hwmon1", # u4 system-inlet
- 1: "i2c-6/6-0049/hwmon/hwmon2", # u2 system-inlet
- 2: "i2c-7/7-004a/hwmon/hwmon3", # u44 bmc56960-on-board
- 3: "i2c-14/14-0048/hwmon/hwmon4", # u9200 cpu-on-board
- 4: "i2c-15/15-004e/hwmon/hwmon5" # u9201 system-outlet
- }.get(self.index, None)
-
- self.hwmon_path = "{}/{}".format(self.I2C_ADAPTER_PATH, i2c_path)
- self.ss_key = self.THERMAL_NAME_LIST[self.index]
+ self._airflow = airflow
+ self._thermal_info = THERMAL_INFO[self.index]
+ self._hwmon_path = "{}/{}".format(I2C_ADAPTER_PATH,
+ self._thermal_info["i2c_path"])
+ self.name = self.get_name()
+ self.postion = self._thermal_info["postion"]
  self.ss_index = 1
 
  def __get_temp(self, temp_file):
- temp_file_path = os.path.join(self.hwmon_path, temp_file)
+ temp_file_path = os.path.join(self._hwmon_path, temp_file)
  raw_temp = self._api_helper.read_txt_file(temp_file_path)
  temp = float(raw_temp)/1000
  return float("{:.3f}".format(temp))
 
  def __set_threshold(self, file_name, temperature):
- temp_file_path = os.path.join(self.hwmon_path, file_name)
+ temp_file_path = os.path.join(self._hwmon_path, file_name)
  try:
  with open(temp_file_path, 'w') as fd:
  fd.write(str(temperature))
@@ -82,8 +111,17 @@ def get_high_threshold(self):
  A float number, the high threshold temperature of thermal in Celsius
  up to nearest thousandth of one degree Celsius, e.g. 30.125
  """
- temp_file = "temp{}_max".format(self.ss_index)
- return self.__get_temp(temp_file)
+ max_crit_key = '{}_max'.format(self._airflow)
+ return self._thermal_info.get(max_crit_key, None)
+
+ def get_low_threshold(self):
+ """
+ Retrieves the low threshold temperature of thermal
+ Returns:
+ A float number, the low threshold temperature of thermal in Celsius
+ up to nearest thousandth of one degree Celsius, e.g. 30.125
+ """
+ return 0.0
 
  def set_high_threshold(self, temperature):
  """
@@ -104,7 +142,7 @@ def set_high_threshold(self, temperature):
  f.seek(0)
  ss_found = False
  for idx, val in enumerate(content):
- if self.ss_key in val:
+ if self.name in val:
  ss_found = True
  elif ss_found and temp_file in val:
  content[idx] = " set {} {}\n".format(
@@ -117,13 +155,43 @@ def set_high_threshold(self, temperature):
 
  return is_set & file_set
 
+ def set_low_threshold(self, temperature):
+ """
+ Sets the low threshold temperature of thermal
+ Args : 
+ temperature: A float number up to nearest thousandth of one degree Celsius,
+ e.g. 30.125
+ Returns:
+ A boolean, True if threshold is set successfully, False if not
+ """
+ return False
+
+ 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
+ """
+ max_crit_key = '{}_max_crit'.format(self._airflow)
+ return self._thermal_info.get(max_crit_key, None)
+
+ def get_low_critical_threshold(self):
+ """
+ Retrieves the low critical threshold temperature of thermal
+ Returns:
+ A float number, the low critical threshold temperature of thermal in Celsius
+ up to nearest thousandth of one degree Celsius, e.g. 30.125
+ """
+ return 0.0
+
  def get_name(self):
  """
  Retrieves the name of the thermal device
  Returns:
  string: The name of the thermal device
  """
- return self.THERMAL_NAME_LIST[self.index]
+ return self._thermal_info["name"]
 
  def get_presence(self):
  """
@@ -132,9 +200,25 @@ def get_presence(self):
  bool: True if PSU is present, False if not
  """
  temp_file = "temp{}_input".format(self.ss_index)
- temp_file_path = os.path.join(self.hwmon_path, temp_file)
+ temp_file_path = os.path.join(self._hwmon_path, temp_file)
  return os.path.isfile(temp_file_path)
 
+ def get_model(self):
+ """
+ Retrieves the model number (or part number) of the device
+ Returns:
+ string: Model/part number of device
+ """
+ return NULL_VAL
+
+ def get_serial(self):
+ """
+ Retrieves the serial number of the device
+ Returns:
+ string: Serial number of device
+ """
+ return NULL_VAL
+
  def get_status(self):
  """
  Retrieves the operational status of the device
@@ -145,7 +229,7 @@ def get_status(self):
  return False
 
  fault_file = "temp{}_fault".format(self.ss_index)
- fault_file_path = os.path.join(self.hwmon_path, fault_file)
+ fault_file_path = os.path.join(self._hwmon_path, fault_file)
  if not os.path.isfile(fault_file_path):
  return True
 

device/celestica/x86_64-cel_seastone-r0/sonic_platform/thermal_actions.py

@@ -0,0 +1,78 @@
+
+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
+from .thermal_infos import ChassisInfo
+from .helper import APIHelper
+
+
+@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:
+ """
+ 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()
+
+
+@thermal_json_object('switch.power_cycling')
+class SwitchPolicyAction(ThermalPolicyActionBase):
+ """
+ Base class for thermal action. Once all thermal conditions in a thermal policy are matched,
+ all predefined thermal action will be executed.
+ """
+
+ def execute(self, thermal_info_dict):
+ """
+ Take action when thermal condition matches. For example, power cycle the switch.
+ :param thermal_info_dict: A dictionary stores all thermal information.
+ :return:
+ """
+ thermal_overload_position_path = '/tmp/thermal_overload_position'
+ thermal_overload_position = APIHelper().read_one_line_file(
+ thermal_overload_position_path)
+
+ cmd = 'bash /usr/share/sonic/platform/thermal_overload_control.sh {}'.format(
+ thermal_overload_position)
+ APIHelper().run_command(cmd)