[Accton] AS4630-54TE Support SystemHealthMonitor (#8183)

1. Implement FanDrawer-Fan hierarchy.
2. Enable thermalctld, disable pcied.
3. Implement SystemLED in Chassis.
4. Correct Fan direction
5. Implement require Fan APIs for SystemHealthMonitoring.
6. Handle non-ascii character while reading PSU model/serial num.

```
Check if System-health can pass the check and display the SystemLED correctly.


///////// booting, DIAG_LED = GREEN_BLINKING /////////

root@sonic:/tmp# show system-health detail 
System is currently booting...
root@sonic:/tmp# cat /sys/class/leds/diag/brightness 
5


///////// container_checker fail, DIAG_LED = AMBER /////////

root@sonic:/sys/bus/i2c/devices# show system-health detail
System status summary

  System status LED  STATUS_LED_COLOR_AMBER
  Services:
    Status: OK
  Hardware:
    Status: Not OK
    Reasons: container_checker is not Status ok

System services and devices monitor list

Name                        Status    Type
--------------------------  --------  ----------
container_checker           Not OK    Program
sonic                       OK        System
rsyslog                     OK        Process
root-overlay                OK        Filesystem
var-log                     OK        Filesystem
routeCheck                  OK        Program
diskCheck                   OK        Program
container_memory_telemetry  OK        Program
FAN-1F                      OK        Fan
FAN-1R                      OK        Fan
FAN-2F                      OK        Fan
FAN-2R                      OK        Fan
FAN-3F                      OK        Fan
FAN-3R                      OK        Fan
PSU-1 FAN-1                 OK        Fan
PSU-2 FAN-1                 OK        Fan
PSU 1                       OK        PSU
PSU 2                       OK        PSU

System services and devices ignore list

Name             Status    Type
---------------  --------  ------
asic             Ignored   Device
psu.temperature  Ignored   Device

///////// skip container_checker, DIAG_LED = GREEN /////////

root@sonic:/sys/bus/i2c/devices# vi /usr/share/sonic/device/x86_64-accton_as4630_54te-r0/system_health_monitoring_config.json 
root@sonic:/sys/bus/i2c/devices# 
root@sonic:/sys/bus/i2c/devices# 
root@sonic:/sys/bus/i2c/devices# show system-health detail
System status summary

  System status LED  STATUS_LED_COLOR_GREEN
  Services:
    Status: OK
  Hardware:
    Status: OK

System services and devices monitor list

Name                        Status    Type
--------------------------  --------  ----------
sonic                       OK        System
rsyslog                     OK        Process
root-overlay                OK        Filesystem
var-log                     OK        Filesystem
routeCheck                  OK        Program
diskCheck                   OK        Program
container_memory_telemetry  OK        Program
FAN-1F                      OK        Fan
FAN-1R                      OK        Fan
FAN-2F                      OK        Fan
FAN-2R                      OK        Fan
FAN-3F                      OK        Fan
FAN-3R                      OK        Fan
PSU-1 FAN-1                 OK        Fan
PSU-2 FAN-1                 OK        Fan
PSU 1                       OK        PSU
PSU 2                       OK        PSU

System services and devices ignore list

Name               Status    Type
-----------------  --------  -------
container_checker  Ignored   Service
psu.temperature    Ignored   Device
asic               Ignored   Device
```


Signed-off-by: Sean Wu <sean_wu@edge-core.com>

device/accton/x86_64-accton_as4630_54te-r0/pmon_daemon_control.json

@@ -1,5 +1,5 @@
 {
     "skip_ledd": true,
-    "skip_thermalctld": true
+    "skip_pcied": true
 }
 

device/accton/x86_64-accton_as4630_54te-r0/sonic_platform/__init__.py

@@ -1,2 +1,2 @@
-__all__ = ['chassis', 'eeprom', 'platform', 'psu', 'sfp', 'thermal', 'fan']
+__all__ = [ "platform", "chassis", "sfp", "eeprom", "component", "psu", "thermal", "fan", "fan_drawer" ]
 from . import platform 

device/accton/x86_64-accton_as4630_54te-r0/sonic_platform/chassis.py

@@ -27,6 +27,13 @@ PMON_REBOOT_CAUSE_PATH = "/usr/share/sonic/platform/api_files/reboot-cause/"
 REBOOT_CAUSE_FILE = "reboot-cause.txt"
 PREV_REBOOT_CAUSE_FILE = "previous-reboot-cause.txt"
 HOST_CHK_CMD = "docker > /dev/null 2>&1"
+SYSLED_FNODE = "/sys/class/leds/diag/brightness"
+SYSLED_MODES = {
+    "0" : "STATUS_LED_COLOR_OFF",
+    "1" : "STATUS_LED_COLOR_GREEN",
+    "2" : "STATUS_LED_COLOR_AMBER",
+    "5" : "STATUS_LED_COLOR_GREEN_BLINK"
+}
 
 
 class Chassis(ChassisBase):
@@ -55,12 +62,11 @@ class Chassis(ChassisBase):
         self.sfp_module_initialized = True
 
     def __initialize_fan(self):
-        from sonic_platform.fan import Fan
+        from sonic_platform.fan_drawer import FanDrawer
-        for fant_index in range(0, NUM_FAN_TRAY):
+        for fant_index in range(NUM_FAN_TRAY):
-            for fan_index in range(0, NUM_FAN):
+            fandrawer = FanDrawer(fant_index)
-                fan = Fan(fant_index, fan_index)
+            self._fan_drawer_list.append(fandrawer)
-                self._fan_list.append(fan)
+            self._fan_list.extend(fandrawer._fan_list)
-
     def __initialize_psu(self):
         from sonic_platform.psu import Psu
         for index in range(0, NUM_PSU):
@@ -192,3 +198,39 @@ class Chassis(ChassisBase):
             sys.stderr.write("SFP index {} out of range (1-{})\n".format(
                              index, len(self._sfp_list)))
         return sfp
+
+    def get_position_in_parent(self):
+        """
+        Retrieves 1-based relative physical position in parent device. If the agent cannot determine the parent-relative position
+        for some reason, or if the associated value of entPhysicalContainedIn is '0', then the value '-1' is returned
+        Returns:
+            integer: The 1-based relative physical position in parent device or -1 if cannot determine the position
+        """
+        return -1
+
+    def is_replaceable(self):
+        """
+        Indicate whether this device is replaceable.
+        Returns:
+            bool: True if it is replaceable.
+        """
+        return False
+
+
+    def initizalize_system_led(self):
+        return True
+
+    def get_status_led(self):
+        val = self._api_helper.read_txt_file(SYSLED_FNODE)
+        return SYSLED_MODES[val] if val in SYSLED_MODES else "UNKNOWN"
+
+    def set_status_led(self, color):
+        mode = None
+        for key, val in SYSLED_MODES.items():
+            if val == color:
+                mode = key
+                break
+        if mode is None:
+            return False
+        else:
+            return self._api_helper.write_txt_file(SYSLED_FNODE, mode)

device/accton/x86_64-accton_as4630_54te-r0/sonic_platform/fan.py

@@ -13,10 +13,10 @@ except ImportError as e:
     raise ImportError(str(e) + "- required module not found")
 
 PSU_FAN_MAX_RPM = 26688
-
+SPEED_TOLERANCE = 15
-CPLD_I2C_PATH = "/sys/bus/i2c/devices/3-0060/fan_"
+CPLD_FAN_I2C_PATH = "/sys/bus/i2c/devices/3-0060/fan_"
-PSU_HWMON_I2C_PATH ="/sys/bus/i2c/devices/{}-00{}/"
+I2C_PATH ="/sys/bus/i2c/devices/{}-00{}/"
-PSU_I2C_MAPPING = {
+PSU_HWMON_I2C_MAPPING = {
     0: {
         "num": 10,
         "addr": "58"
@@ -27,6 +27,20 @@ PSU_I2C_MAPPING = {
     },
 }
 
+PSU_CPLD_I2C_MAPPING = {
+    0: {
+        "num": 10,
+        "addr": "50"
+    },
+    1: {
+        "num": 11,
+        "addr": "51"
+    },
+}
+
+
+FAN_NAME_LIST = ["FAN-1F", "FAN-1R", "FAN-2F", "FAN-2R",
+                 "FAN-3F", "FAN-3R"]
 
 class Fan(FanBase):
     """Platform-specific Fan class"""
@@ -39,9 +53,14 @@ class Fan(FanBase):
 
         if self.is_psu_fan:
             self.psu_index = psu_index
-            self.psu_i2c_num = PSU_I2C_MAPPING[self.psu_index]['num']
+            self.psu_i2c_num = PSU_HWMON_I2C_MAPPING[self.psu_index]['num']
-            self.psu_i2c_addr = PSU_I2C_MAPPING[self.psu_index]['addr']
+            self.psu_i2c_addr = PSU_HWMON_I2C_MAPPING[self.psu_index]['addr']
-            self.psu_hwmon_path = PSU_HWMON_I2C_PATH.format(
+            self.psu_hwmon_path = I2C_PATH.format(
+                self.psu_i2c_num, self.psu_i2c_addr)
+
+            self.psu_i2c_num = PSU_CPLD_I2C_MAPPING[self.psu_index]['num']
+            self.psu_i2c_addr = PSU_CPLD_I2C_MAPPING[self.psu_index]['addr']
+            self.psu_cpld_path = I2C_PATH.format(
                 self.psu_i2c_num, self.psu_i2c_addr)
 
         FanBase.__init__(self)  
@@ -57,10 +76,10 @@ class Fan(FanBase):
 
 
         if not self.is_psu_fan:
-            dir_str = "{}{}{}".format(CPLD_I2C_PATH, 'direction_', self.fan_tray_index)
+            dir_str = "{}{}{}".format(CPLD_FAN_I2C_PATH, 'direction_', self.fan_tray_index+1)
             val=self._api_helper.read_txt_file(dir_str)
             if val is not None:
-                if val==0:#F2B
+                if int(val, 10)==0:#F2B
                     direction=self.FAN_DIRECTION_EXHAUST
                 else:
                     direction=self.FAN_DIRECTION_INTAKE
@@ -99,7 +118,7 @@ class Fan(FanBase):
             else:
                 return 0
         elif self.get_presence():            
-            speed_path = "{}{}".format(CPLD_I2C_PATH, 'duty_cycle_percentage')
+            speed_path = "{}{}".format(CPLD_FAN_I2C_PATH, 'duty_cycle_percentage')
             speed=self._api_helper.read_txt_file(speed_path)
             if speed is None:
                 return 0
@@ -116,7 +135,7 @@ class Fan(FanBase):
             0   : when PWM mode is use
             pwm : when pwm mode is not use
         """
-        return False #Not supported
+        return self.get_speed()
 
     def get_speed_tolerance(self):
         """
@@ -125,7 +144,7 @@ class Fan(FanBase):
             An integer, the percentage of variance from target speed which is
                  considered tolerable
         """
-        return False #Not supported
+        return SPEED_TOLERANCE
 
     def set_speed(self, speed):
         """
@@ -138,7 +157,7 @@ class Fan(FanBase):
         """
 
         if not self.is_psu_fan and self.get_presence():            
-            speed_path = "{}{}".format(CPLD_I2C_PATH, 'duty_cycle_percentage')
+            speed_path = "{}{}".format(CPLD_FAN_I2C_PATH, 'duty_cycle_percentage')
             return self._api_helper.write_txt_file(speed_path, int(speed))
 
         return False
@@ -154,18 +173,109 @@ class Fan(FanBase):
         """
         return False #Not supported
 
+    def get_status_led(self):
+        """
+        Gets the state of the fan status LED
+        Returns:
+            A string, one of the predefined STATUS_LED_COLOR_* strings above
+        """
+        status=self.get_presence()
+        if status is None:
+            return  self.STATUS_LED_COLOR_OFF
+
+        return {
+            1: self.STATUS_LED_COLOR_GREEN,
+            0: self.STATUS_LED_COLOR_RED
+        }.get(status, self.STATUS_LED_COLOR_OFF)
+
+    def get_name(self):
+        """
+        Retrieves the name of the device
+            Returns:
+            string: The name of the device
+        """
+        fan_name = FAN_NAME_LIST[self.fan_tray_index*2 + self.fan_index] \
+            if not self.is_psu_fan \
+            else "PSU-{} FAN-{}".format(self.psu_index+1, self.fan_index+1)
+
+        return fan_name
+
     def get_presence(self):
         """
         Retrieves the presence of the FAN
         Returns:
             bool: True if FAN is present, False if not
         """
-        present_path = "{}{}{}".format(CPLD_I2C_PATH, 'present_', self.fan_index+1)
+
+
+        if self.is_psu_fan:
+            present_path="{}{}".format(self.psu_cpld_path, 'psu_present')
+        else:
+            present_path = "{}{}{}".format(CPLD_FAN_I2C_PATH, 'present_', self.fan_tray_index+1)
+
         val=self._api_helper.read_txt_file(present_path)
-        if not self.is_psu_fan:
         if val is not None:
             return int(val, 10)==1
         else:
             return False
+
+    def get_status(self):
+        """
+        Retrieves the operational status of the device
+        Returns:
+            A boolean value, True if device is operating properly, False if not
+        """
+        if self.is_psu_fan:
+            psu_fan_path= "{}{}".format(self.psu_hwmon_path, 'psu_fan1_fault')
+            val=self._api_helper.read_txt_file(psu_fan_path)
+            if val is not None:
+                return int(val, 10)==0
             else:
-            return True
+                return False
+        else:
+            path = "{}{}{}".format(CPLD_FAN_I2C_PATH, 'fault_', self.fan_tray_index+1)
+            val=self._api_helper.read_txt_file(path)
+            if val is not None:
+                return int(val, 10)==0
+            else:
+                return False
+
+
+    def get_model(self):
+        """
+        Retrieves the model number (or part number) of the device
+        Returns:
+            string: Model/part number of device
+        """
+
+        return "N/A"
+
+    def get_serial(self):
+        """
+        Retrieves the serial number of the device
+        Returns:
+            string: Serial number of device
+        """
+        return "N/A"
+
+    def get_position_in_parent(self):
+        """
+        Retrieves 1-based relative physical position in parent device.
+        If the agent cannot determine the parent-relative position
+        for some reason, or if the associated value of
+        entPhysicalContainedIn is'0', then the value '-1' is returned
+        Returns:
+            integer: The 1-based relative physical position in parent device
+            or -1 if cannot determine the position
+        """
+        return (self.fan_tray_index+1) \
+            if not self.is_psu_fan else (self.psu_index+1)
+
+    def is_replaceable(self):
+        """
+        Indicate whether this device is replaceable.
+        Returns:
+            bool: True if it is replaceable.
+        """
+        return True if not self.is_psu_fan else False
+

device/accton/x86_64-accton_as4630_54te-r0/sonic_platform/fan_drawer.py

@@ -0,0 +1,90 @@
+########################################################################
+#
+# Module contains an implementation of SONiC Platform Base API and
+# provides the Fan-Drawers' information available in the platform.
+#
+########################################################################
+
+try:
+    from sonic_platform_base.fan_drawer_base import FanDrawerBase
+except ImportError as e:
+    raise ImportError(str(e) + "- required module not found")
+
+FANS_PER_FANTRAY = 2
+
+
+class FanDrawer(FanDrawerBase):
+    """Platform-specific Fan class"""
+
+    def __init__(self, fantray_index):
+
+        FanDrawerBase.__init__(self)
+        # FanTray is 0-based in platforms
+        self.fantrayindex = fantray_index
+        self.__initialize_fan_drawer()
+
+
+    def __initialize_fan_drawer(self):
+        from sonic_platform.fan import Fan
+        for i in range(FANS_PER_FANTRAY):
+            self._fan_list.append(Fan(self.fantrayindex, i))
+
+    def get_name(self):
+        """
+        Retrieves the fan drawer name
+        Returns:
+            string: The name of the device
+        """
+        return "FanTray{}".format(self.fantrayindex+1)
+
+    def get_presence(self):
+        """
+        Retrieves the presence of the device
+        Returns:
+            bool: True if device is present, False if not
+        """
+        return self._fan_list[0].get_presence()
+
+    def get_model(self):
+        """
+        Retrieves the model number (or part number) of the device
+        Returns:
+            string: Model/part number of device
+        """
+        return self._fan_list[0].get_model()
+
+    def get_serial(self):
+        """
+        Retrieves the serial number of the device
+        Returns:
+            string: Serial number of device
+        """
+        return self._fan_list[0].get_serial()
+
+    def get_status(self):
+        """
+        Retrieves the operational status of the device
+        Returns:
+            A boolean value, True if device is operating properly, False if not
+        """
+        return self._fan_list[0].get_status()
+
+    def get_position_in_parent(self):
+        """
+        Retrieves 1-based relative physical position in parent device.
+        If the agent cannot determine the parent-relative position
+        for some reason, or if the associated value of
+        entPhysicalContainedIn is'0', then the value '-1' is returned
+        Returns:
+            integer: The 1-based relative physical position in parent device
+            or -1 if cannot determine the position
+        """
+        return (self.fantrayindex+1)
+
+    def is_replaceable(self):
+        """
+        Indicate whether this device is replaceable.
+        Returns:
+            bool: True if it is replaceable.
+        """
+        return True

device/accton/x86_64-accton_as4630_54te-r0/sonic_platform/helper.py

@@ -51,7 +51,7 @@ class APIHelper():
 
     def read_txt_file(self, file_path):
         try:
-            with open(file_path, 'r') as fd:
+            with open(file_path, 'r', errors='replace') as fd:
                 data = fd.read()
                 return data.strip()
         except IOError:

device/accton/x86_64-accton_as4630_54te-r0/sonic_platform/psu.py

@@ -139,8 +139,15 @@ class Psu(PsuBase):
         Returns:
             A string, one of the predefined STATUS_LED_COLOR_* strings above
         """
+        status=self.get_status()
+        if status is None:
+            return  self.STATUS_LED_COLOR_OFF
+
+        return {
+            1: self.STATUS_LED_COLOR_GREEN,
+            0: self.STATUS_LED_COLOR_RED
+        }.get(status, self.STATUS_LED_COLOR_OFF)
 
-        return False  #Controlled by HW
 
     def get_temperature(self):
         """
@@ -226,3 +233,45 @@ class Psu(PsuBase):
             return int(val, 10) == 1
         else:
             return 0
+
+    def get_model(self):
+        """
+        Retrieves the model number (or part number) of the device
+        Returns:
+            string: Model/part number of device
+        """
+        model_path="{}{}".format(self.cpld_path, 'psu_model_name')
+        model=self._api_helper.read_txt_file(model_path)
+        if model is None:
+            return "N/A"
+
+        return model
+
+    def get_serial(self):
+        """
+        Retrieves the serial number of the device
+        Returns:
+            string: Serial number of device
+        """
+        serial_path="{}{}".format(self.cpld_path, 'psu_serial_number')
+        serial=self._api_helper.read_txt_file(serial_path)
+        if serial is None:
+            return "N/A"
+        return serial
+
+    def get_position_in_parent(self):
+        """
+        Retrieves 1-based relative physical position in parent device. If the agent cannot determine the parent-relative position
+        for some reason, or if the associated value of entPhysicalContainedIn is '0', then the value '-1' is returned
+        Returns:
+            integer: The 1-based relative physical position in parent device or -1 if cannot determine the position
+        """
+        return self.index+1
+
+    def is_replaceable(self):
+        """
+        Indicate whether this device is replaceable.
+        Returns:
+            bool: True if it is replaceable.
+        """
+        return True

device/accton/x86_64-accton_as4630_54te-r0/system_health_monitoring_config.json

@@ -0,0 +1,15 @@
+{
+    "services_to_ignore": [],
+    "devices_to_ignore": [
+        "asic",
+        "psu.temperature"
+
+    ],
+    "user_defined_checkers": [],
+    "polling_interval": 60,
+    "led_color": {
+        "fault": "STATUS_LED_COLOR_AMBER",
+        "normal": "STATUS_LED_COLOR_GREEN",
+        "booting": "STATUS_LED_COLOR_GREEN_BLINK"
+    }
+}