[Mellanox] support new platform api, thermal and psu part (#3175)

* support new platform api, thermal and psu part
for psu, all APIs are supported.
for thermal, we support
  get_temperature,
  get_high_threshold
for the thermal sensors of cpu core, cpu pack, psu and sfp module
and get_temperature for the ambient thermal sensors around the asic, port, fan, comex and board.

* 1. address review comments
2. improve the handling of PSU inserting/removal
3. tolerance diverse psu thermal sensor file name conventions

* 1. adjust thermal code according to the latest version of hw-management
2. check power_good_status rather than whether file existing ahead of reading voltage, current and power of PSU

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

@@ -8,19 +8,19 @@
 #
 #############################################################################
 
-import sys
-
 try:
     from sonic_platform_base.chassis_base import ChassisBase
     from sonic_platform.psu import Psu
     from sonic_platform.fan import Fan
     from sonic_platform.fan import FAN_PATH
     from sonic_platform.sfp import SFP
+    from sonic_platform.thermal import Thermal, initialize_thermals
     from sonic_platform.watchdog import get_watchdog
     from sonic_daemon_base.daemon_base import Logger
     from eeprom import Eeprom
     from os import listdir
     from os.path import isfile, join
+    import sys
     import io
     import re
     import subprocess
@@ -64,12 +64,12 @@
 COMPONENT_CPLD2 = "CPLD2"
 
 # Global logger class instance
-SYSLOG_IDENTIFIER = "mlnx-chassis"
+SYSLOG_IDENTIFIER = "mlnx-chassis-api"
 logger = Logger(SYSLOG_IDENTIFIER)
 
 # magic code defnition for port number, qsfp port position of each hwsku
 # port_position_tuple = (PORT_START, QSFP_PORT_START, PORT_END, PORT_IN_BLOCK, EEPROM_OFFSET)
-hwsku_dict = {'ACS-MSN2700': 0, "LS-SN2700":0, 'ACS-MSN2740': 0, 'ACS-MSN2100': 1, 'ACS-MSN2410': 2, 'ACS-MSN2010': 3, 'ACS-MSN3700': 0, 'ACS-MSN3700C': 0, 'Mellanox-SN2700': 0, 'Mellanox-SN2700-D48C8': 0}
+hwsku_dict_port = {'ACS-MSN2700': 0, "LS-SN2700":0, 'ACS-MSN2740': 0, 'ACS-MSN2100': 1, 'ACS-MSN2410': 2, 'ACS-MSN2010': 3, 'ACS-MSN3700': 0, 'ACS-MSN3700C': 0, 'Mellanox-SN2700': 0, 'Mellanox-SN2700-D48C8': 0}
 port_position_tuple_list = [(0, 0, 31, 32, 1), (0, 0, 15, 16, 1), (0, 48, 55, 56, 1),(0, 18, 21, 22, 1)]
 
 class Chassis(ChassisBase):
@@ -78,9 +78,12 @@ class Chassis(ChassisBase):
     def __init__(self):
         super(Chassis, self).__init__()
 
+        # Initialize SKU name
+        self.sku_name = self._get_sku_name()
+
         # Initialize PSU list
         for index in range(MLNX_NUM_PSU):
-            psu = Psu(index)
+            psu = Psu(index, self.sku_name)
             self._psu_list.append(psu)
 
         # Initialize watchdog
@@ -112,6 +115,9 @@ def __init__(self):
                 sfp_module = SFP(index, 'SFP')
             self._sfp_list.append(sfp_module)
 
+        # Initialize thermals
+        initialize_thermals(self.sku_name, self._thermal_list, self._psu_list)
+
         # Initialize EEPROM
         self.eeprom = Eeprom()
 
@@ -137,10 +143,13 @@ def _extract_num_of_fans_and_fan_drawers(self):
 
         return num_of_fan, num_of_drawer
 
-    def _get_port_position_tuple_by_sku_name(self):
+    def _get_sku_name(self):
         p = subprocess.Popen(GET_HWSKU_CMD, shell=True, stdout=subprocess.PIPE)
         out, err = p.communicate()
-        position_tuple = port_position_tuple_list[hwsku_dict[out.rstrip('\n')]]
+        return out.rstrip('\n')
+
+    def _get_port_position_tuple_by_sku_name(self):
+        position_tuple = port_position_tuple_list[hwsku_dict_port[self.sku_name]]
         return position_tuple
 
     def get_base_mac(self):
@@ -183,8 +192,8 @@ def _read_generic_file(self, filename, len):
             result = fileobj.read(len)
             fileobj.close()
             return result
-        except:
-            logger.log_warning("Fail to read file {}, maybe it doesn't exist".format(filename))
+        except Exception as e:
+            logger.log_info("Fail to read file {} due to {}".format(filename, repr(e)))
             return ''
 
     def _verify_reboot_cause(self, filename):

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

@@ -8,48 +8,113 @@
 #
 #############################################################################
 
-import os.path
-
 try:
+    import os.path
     from sonic_platform_base.psu_base import PsuBase
+    from sonic_daemon_base.daemon_base import Logger
     from sonic_platform.fan import Fan
 except ImportError as e:
     raise ImportError (str(e) + "- required module not found")
 
+# Global logger class instance
+SYSLOG_IDENTIFIER = "mlnx-psu-api"
+logger = Logger(SYSLOG_IDENTIFIER)
+
 psu_list = []
 
+PSU_CURRENT = "current"
+PSU_VOLTAGE = "voltage"
+PSU_POWER = "power"
+
+# SKUs with unplugable PSUs:
+# 1. don't have psuX_status and should be treated as always present
+# 2. don't have voltage, current and power values
+hwsku_dict_with_unplugable_psu = ['ACS-MSN2010', 'ACS-MSN2100']
+
+# in most SKUs the file psuX_curr, psuX_volt and psuX_power contain current, voltage and power data respectively. 
+# but there are exceptions which will be handled by the following dictionary
+hwsku_dict_psu = {'ACS-MSN3700': 1, 'ACS-MSN3700C': 1, 'ACS-MSN3800': 1}
+psu_profile_list = [
+    # default filename convention
+    {
+        PSU_CURRENT : "power/psu{}_curr",
+        PSU_VOLTAGE : "power/psu{}_volt",
+        PSU_POWER : "power/psu{}_power"
+    },
+    # for 3700, 3700c, 3800
+    {
+        PSU_CURRENT : "power/psu{}_curr",
+        PSU_VOLTAGE : "power/psu{}_volt_out2",
+        PSU_POWER : "power/psu{}_power"
+    }
+]
+
 class Psu(PsuBase):
     """Platform-specific Psu class"""
-    def __init__(self, psu_index):
+    def __init__(self, psu_index, sku):
         global psu_list
         PsuBase.__init__(self)
         # PSU is 1-based on Mellanox platform
         self.index = psu_index + 1
         psu_list.append(self.index)
-        self.psu_path = "/var/run/hw-management/thermal/"
-        self.psu_oper_status = "psu{}_pwr_status".format(self.index)
-        self.psu_presence = "psu{}_status".format(self.index)
-        if os.path.exists(os.path.join(self.psu_path, self.psu_presence)):
-            self.presence_file_exists = True
+        self.psu_path = "/var/run/hw-management/"
+        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)
+
+        if sku in hwsku_dict_psu:
+            filemap = psu_profile_list[hwsku_dict_psu[sku]]
+        else:
+            filemap = psu_profile_list[0]
+
+        if sku in hwsku_dict_with_unplugable_psu:
+            self.always_presence = True
+            self.psu_voltage = None
+            self.psu_current = None
+            self.psu_power = None
+            self.psu_presence = None
         else:
-            self.presence_file_exists = False
+            self.always_presence = False
+            psu_voltage = filemap[PSU_VOLTAGE].format(self.index)
+            psu_voltage = os.path.join(self.psu_path, psu_voltage)
+            self.psu_voltage = psu_voltage
+
+            psu_current = filemap[PSU_CURRENT].format(self.index)
+            psu_current = os.path.join(self.psu_path, psu_current)
+            self.psu_current = psu_current
+
+            psu_power = filemap[PSU_POWER].format(self.index)
+            psu_power = os.path.join(self.psu_path, psu_power)
+            self.psu_power = psu_power
+
+            psu_presence = "thermal/psu{}_status".format(self.index)
+            psu_presence = os.path.join(self.psu_path, psu_presence)
+            self.psu_presence = psu_presence
+
         fan = Fan(psu_index, psu_index, True)
         if fan.get_presence():
             self._fan = fan
 
-    def get_status(self):
+    def _read_generic_file(self, filename, len):
+        """
+        Read a generic file, returns the contents of the file
+        """
+        result = 0
+        try:
+            with open(filename, 'r') as fileobj:
+                result = int(fileobj.read())
+        except Exception as e:
+            logger.log_info("Fail to read file {} due to {}".format(filename, repr(e)))
+        return result
+
+    def get_powergood_status(self):
         """
         Retrieves the operational status of power supply unit (PSU) defined
 
         Returns:
             bool: True if PSU is operating properly, False if not
         """
-        status = 0
-        try:
-            with open(os.path.join(self.psu_path, self.psu_oper_status), 'r') as power_status:
-                status = int(power_status.read())
-        except (ValueError, IOError):
-            status = 0
+        status = self._read_generic_file(os.path.join(self.psu_path, self.psu_oper_status), 0)
 
         return status == 1
 
@@ -60,15 +125,48 @@ def get_presence(self):
         Returns:
             bool: True if PSU is present, False if not
         """
-        status = 0
-        if self.presence_file_exists:
-            try:
-                with open(os.path.join(self.psu_path, self.psu_presence), 'r') as presence_status:
-                    status = int(presence_status.read())
-            except (ValueError, IOError):
-                status = 0
+        if self.always_presence:
+            return self.always_presence
         else:
-            status = self.index in psu_list
+            status = self._read_generic_file(self.psu_presence, 0)
+            return status == 1
 
-        return status == 1
+    def get_voltage(self):
+        """
+        Retrieves current PSU voltage output
 
+        Returns:
+            A float number, the output voltage in volts, 
+            e.g. 12.1 
+        """
+        if self.psu_voltage is not None and self.get_powergood_status():
+            voltage = self._read_generic_file(self.psu_voltage, 0)
+            return float(voltage) / 1000
+        else:
+            return None
+
+    def get_current(self):
+        """
+        Retrieves present electric current supplied by PSU
+
+        Returns:
+            A float number, the electric current in amperes, e.g 15.4
+        """
+        if self.psu_current is not None and self.get_powergood_status():
+            amperes = self._read_generic_file(self.psu_current, 0)
+            return float(amperes) / 1000
+        else:
+            return None
+
+    def get_power(self):
+        """
+        Retrieves current energy supplied by PSU
+
+        Returns:
+            A float number, the power in watts, e.g. 302.6
+        """
+        if self.psu_power is not None and self.get_powergood_status():
+            power = self._read_generic_file(self.psu_power, 0)
+            return float(power) / 1000000
+        else:
+            return None