[Mellanox]New platform api -- chassis part (#3082)

* new platform api, chassis part

* Inject mlnx mlx libs to platform monitor

* address the review comments

* remove some confusing naming.

* Adjust the minor cause to a more human-readable way when rebooted by firmware

* address review comments

* expose host dir /host/reboot-cause to pmon docker so that the reboot causing by user command can be identified

* 1. Revert "expose host dir /host/reboot-cause to pmon docker so that the reboot causing by user command can be identified"
Since the only hardware-causing reboot should be handled by get_reboot_cause and the logic of handling reboot cause is about to move to the host side, no need to mount this dir to pmon docker.
This reverts commit 3feb96869d.
2. adjust log output by using sonic_daemon_base.daemon_base.Logger.
3. remove the logic of verifying /host/reboot-cause/ files.
4. fix typo.

* implement get_firmware_version and adjust the interfaces regarding components' version retrieving according to the Azure/sonic-platform-common#34

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

@@ -17,10 +17,14 @@ try:
     from sonic_platform.fan import FAN_PATH
     from sonic_platform.sfp import SFP
     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 io
     import re
     import subprocess
+    import syslog
 except ImportError as e:
     raise ImportError (str(e) + "- required module not found")
 
@@ -28,6 +32,41 @@ MLNX_NUM_PSU = 2
 
 GET_HWSKU_CMD = "sonic-cfggen -d -v DEVICE_METADATA.localhost.hwsku"
 
+EEPROM_CACHE_ROOT = '/var/cache/sonic/decode-syseeprom'
+EEPROM_CACHE_FILE = 'syseeprom_cache'
+
+HWMGMT_SYSTEM_ROOT = '/var/run/hw-management/system/'
+
+#reboot cause related definitions
+REBOOT_CAUSE_ROOT = HWMGMT_SYSTEM_ROOT
+
+REBOOT_CAUSE_POWER_LOSS_FILE = 'reset_main_pwr_fail'
+REBOOT_CAUSE_THERMAL_OVERLOAD_ASIC_FILE = 'reset_asic_thermal'
+REBOOT_CAUSE_WATCHDOG_FILE = 'reset_hotswap_or_wd'
+REBOOT_CAUSE_MLNX_FIRMWARE_RESET = 'reset_fw_reset'
+
+REBOOT_CAUSE_FILE_LENGTH = 1
+
+#version retrieving related definitions
+CPLD_VERSION_ROOT = HWMGMT_SYSTEM_ROOT
+
+CPLD1_VERSION_FILE = 'cpld1_version'
+CPLD2_VERSION_FILE = 'cpld2_version'
+CPLD_VERSION_MAX_LENGTH = 4
+
+FW_QUERY_VERSION_COMMAND = 'mlxfwmanager --query'
+BIOS_QUERY_VERSION_COMMAND = 'dmidecode -t 11'
+
+#components definitions
+COMPONENT_BIOS = "BIOS"
+COMPONENT_FIRMWARE = "ASIC-FIRMWARE"
+COMPONENT_CPLD1 = "CPLD1"
+COMPONENT_CPLD2 = "CPLD2"
+
+# Global logger class instance
+SYSLOG_IDENTIFIER = "mlnx-chassis"
+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}
@@ -37,7 +76,7 @@ class Chassis(ChassisBase):
     """Platform-specific Chassis class"""
 
     def __init__(self):
-        ChassisBase.__init__(self)
+        super(Chassis, self).__init__()
 
         # Initialize PSU list
         for index in range(MLNX_NUM_PSU):
@@ -67,12 +106,21 @@ class Chassis(ChassisBase):
         self.PORTS_IN_BLOCK = port_position_tuple[3]
 
         for index in range(self.PORT_START, self.PORT_END + 1):
-            if index in range(QSFP_PORT_START, self.PORTS_IN_BLOCK + 1):
+            if index in range(self.QSFP_PORT_START, self.PORTS_IN_BLOCK + 1):
                 sfp_module = SFP(index, 'QSFP')
             else:
                 sfp_module = SFP(index, 'SFP')
             self._sfp_list.append(sfp_module)
 
+        # Initialize EEPROM
+        self.eeprom = Eeprom()
+
+        # Initialize component list
+        self._component_name_list.append(COMPONENT_BIOS)
+        self._component_name_list.append(COMPONENT_FIRMWARE)
+        self._component_name_list.append(COMPONENT_CPLD1)
+        self._component_name_list.append(COMPONENT_CPLD2)
+
     def _extract_num_of_fans_and_fan_drawers(self):
         num_of_fan = 0
         num_of_drawer = 0
@@ -95,5 +143,178 @@ class Chassis(ChassisBase):
         position_tuple = port_position_tuple_list[hwsku_dict[out.rstrip('\n')]]
         return position_tuple
 
+    def get_base_mac(self):
+        """
+        Retrieves the base MAC address for the chassis
 
+        Returns:
+            A string containing the MAC address in the format
+            'XX:XX:XX:XX:XX:XX'
+        """
+        return self.eeprom.get_base_mac()
 
+    def get_serial_number(self):
+        """
+        Retrieves the hardware serial number for the chassis
+
+        Returns:
+            A string containing the hardware serial number for this chassis.
+        """
+        return self.eeprom.get_serial_number()
+
+    def get_system_eeprom_info(self):
+        """
+        Retrieves the full content of system EEPROM information for the chassis
+
+        Returns:
+            A dictionary where keys are the type code defined in
+            OCP ONIE TlvInfo EEPROM format and values are their corresponding
+            values.
+        """
+        return self.eeprom.get_system_eeprom_info()
+
+    def _read_generic_file(self, filename, len):
+        """
+        Read a generic file, returns the contents of the file
+        """
+        result = ''
+        try:
+            fileobj = io.open(filename)
+            result = fileobj.read(len)
+            fileobj.close()
+            return result
+        except:
+            logger.log_warning("Fail to read file {}, maybe it doesn't exist".format(filename))
+            return ''
+
+    def _verify_reboot_cause(self, filename):
+        '''
+        Open and read the reboot cause file in 
+        /var/run/hwmanagement/system (which is defined as REBOOT_CAUSE_ROOT)
+        If a reboot cause file doesn't exists, returns '0'.
+        '''
+        return bool(int(self._read_generic_file(join(REBOOT_CAUSE_ROOT, filename), REBOOT_CAUSE_FILE_LENGTH).rstrip('\n')))
+
+    def get_reboot_cause(self):
+        """
+        Retrieves the cause of the previous reboot
+
+        Returns:
+            A tuple (string, string) where the first element is a string
+            containing the cause of the previous reboot. This string must be
+            one of the predefined strings in this class. If the first string
+            is "REBOOT_CAUSE_HARDWARE_OTHER", the second string can be used
+            to pass a description of the reboot cause.
+        """
+        #read reboot causes files in the following order
+        minor_cause = ''
+        if self._verify_reboot_cause(REBOOT_CAUSE_POWER_LOSS_FILE):
+            major_cause = self.REBOOT_CAUSE_POWER_LOSS
+        elif self._verify_reboot_cause(REBOOT_CAUSE_THERMAL_OVERLOAD_ASIC_FILE):
+            major_cause = self.REBOOT_CAUSE_THERMAL_OVERLOAD_ASIC
+        elif self._verify_reboot_cause(REBOOT_CAUSE_WATCHDOG_FILE):
+            major_cause = self.REBOOT_CAUSE_WATCHDOG
+        else:
+            major_cause = self.REBOOT_CAUSE_HARDWARE_OTHER
+            if self._verify_reboot_cause(REBOOT_CAUSE_MLNX_FIRMWARE_RESET):
+                minor_cause = "Reset by ASIC firmware"
+            else:
+                major_cause = self.REBOOT_CAUSE_NON_HARDWARE
+
+        return major_cause, minor_cause
+
+    def _get_cpld_version(self, version_file):
+        cpld_version = self._read_generic_file(join(CPLD_VERSION_ROOT, version_file), CPLD_VERSION_MAX_LENGTH)
+        return cpld_version.rstrip('\n')
+
+    def _get_command_result(self, cmdline):
+        try:
+            proc = subprocess.Popen(cmdline, stdout=subprocess.PIPE, shell=True, stderr=subprocess.STDOUT)
+            stdout = proc.communicate()[0]
+            proc.wait()
+            result = stdout.rstrip('\n')
+
+        except OSError, e:
+            result = ''
+
+        return result
+
+    def _get_firmware_version(self):
+        """
+        firmware version is retrieved via command 'mlxfwmanager --query'
+        which should return result in the following convention
+            admin@mtbc-sonic-01-2410:~$ sudo mlxfwmanager --query
+            Querying Mellanox devices firmware ...
+
+            Device #1:
+            ----------
+
+            Device Type:      Spectrum
+            Part Number:      MSN2410-CxxxO_Ax_Bx
+            Description:      Spectrum based 25GbE/100GbE 1U Open Ethernet switch with ONIE; 48 SFP28 ports; 8 QSFP28 ports; x86 dual core; RoHS6
+            PSID:             MT_2860111033
+            PCI Device Name:  /dev/mst/mt52100_pci_cr0
+            Base MAC:         98039bf3f500
+            Versions:         Current        Available     
+                FW         ***13.2000.1140***N/A           
+
+            Status:           No matching image found
+
+        By using regular expression '(Versions:.*\n[\s]+FW[\s]+)([\S]+)',
+        we can extrace the version which is marked with *** in the above context
+        """
+        fw_ver_str = self._get_command_result(FW_QUERY_VERSION_COMMAND)
+        try: 
+            m = re.search('(Versions:.*\n[\s]+FW[\s]+)([\S]+)', fw_ver_str)
+            result = m.group(2)
+        except :
+            result = ''
+
+        return result
+
+    def _get_bios_version(self):
+        """
+        BIOS version is retrieved via command 'dmidecode -t 11'
+        which should return result in the following convention
+            # dmidecode 3.0
+            Getting SMBIOS data from sysfs.
+            SMBIOS 2.7 present.
+
+            Handle 0x0022, DMI type 11, 5 bytes
+            OEM Strings
+                    String 1:*0ABZS017_02.02.002*
+                    String 2: To Be Filled By O.E.M.
+
+        By using regular expression 'OEM[\s]*Strings\n[\s]*String[\s]*1:[\s]*([0-9a-zA-Z_\.]*)'
+        we can extrace the version string which is marked with * in the above context
+        """
+        bios_ver_str = self._get_command_result(BIOS_QUERY_VERSION_COMMAND)
+        try:
+            m = re.search('OEM[\s]*Strings\n[\s]*String[\s]*1:[\s]*([0-9a-zA-Z_\.]*)', bios_ver_str)
+            result = m.group(1)
+        except:
+            result = ''
+
+        return result
+
+    def get_firmware_version(self, component_name):
+        """
+        Retrieves platform-specific hardware/firmware versions for chassis
+        componenets such as BIOS, CPLD, FPGA, etc.
+        Args:
+            component_name: A string, the component name.
+
+        Returns:
+            A string containing platform-specific component versions
+        """
+        if component_name in self._component_name_list :
+            if component_name == COMPONENT_BIOS:
+                return self._get_bios_version()
+            elif component_name == COMPONENT_CPLD1:
+                return self._get_cpld_version(CPLD1_VERSION_FILE)
+            elif component_name == COMPONENT_CPLD2:
+                return self._get_cpld_version(CPLD2_VERSION_FILE)
+            elif component_name == COMPONENT_FIRMWARE:
+                return self._get_firmware_version()
+
+        return None

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

@@ -0,0 +1,149 @@
+#!/usr/bin/env python
+
+#############################################################################
+# Mellanox
+#
+# Module contains an implementation of SONiC Platform Base API and
+# provides the eeprom information which are available in the platform
+#
+#############################################################################
+import exceptions
+import os
+import sys
+import re
+from cStringIO import StringIO
+
+try:
+    from sonic_platform_base.sonic_eeprom import eeprom_tlvinfo
+except ImportError as e:
+    raise ImportError (str(e) + "- required module not found")
+
+#
+# CACHE_XXX stuffs are supposted to be moved to the base classes
+# since they are common for all vendors
+# they are defined in decode-syseeprom which might be removed in the future
+# currently we just copy them here
+#
+CACHE_ROOT = '/var/cache/sonic/decode-syseeprom'
+CACHE_FILE = 'syseeprom_cache'
+
+#
+# this is mlnx-specific
+# should this be moved to chass.py or here, which better?
+#
+EEPROM_SYMLINK = "/var/run/hw-management/eeprom/vpd_info"
+
+class Eeprom(eeprom_tlvinfo.TlvInfoDecoder):
+    RETRIES = 3
+    EEPROM_DECODE_HEADLINES = 6
+    EEPROM_DECODE_MAXITEM = 3
+    EEPROM_DECODE_OFFSET = 0
+    EEPROM_DECODE_CONTENT = 2
+
+    def __init__(self):
+        for attempt in range(self.RETRIES):
+            if not os.path.islink(EEPROM_SYMLINK):
+                time.sleep(1)
+            else:
+                break  
+
+        if not (os.path.exists(EEPROM_SYMLINK) \
+                or os.path.isfile(os.path.join(CACHE_ROOT, CACHE_FILE))):
+            log_error("Nowhere to read syseeprom from! No symlink or cache file found")
+            raise RuntimeError("No syseeprom symlink or cache file found")
+
+        self.eeprom_path = EEPROM_SYMLINK
+        super(Eeprom, self).__init__(self.eeprom_path, 0, '', True)
+        self._eeprom_loaded = False
+        self._load_eeprom()
+        self._eeprom_loaded = True
+
+    def _load_eeprom(self):
+        if not os.path.exists(CACHE_ROOT):
+            try:
+                os.makedirs(CACHE_ROOT)
+            except:
+                pass
+
+        try:
+            self.set_cache_name(os.path.join(CACHE_ROOT, CACHE_FILE))
+        except:
+            pass
+
+        eeprom = self.read_eeprom()
+        if eeprom is None :
+            return 0
+
+        try:
+            self.update_cache(eeprom)
+        except:
+            pass
+
+        self._base_mac = self.mgmtaddrstr(eeprom)
+        if self._base_mac == None:
+            self._base_mac = "Undefined."
+
+        self._serial_str = self.serial_number_str(eeprom)
+        if self._serial_str == None:
+            self._serial_str = "Undefined."
+
+        original_stdout = sys.stdout
+        sys.stdout = StringIO()
+        self.decode_eeprom(eeprom)
+        decode_output = sys.stdout.getvalue()
+        sys.stdout = original_stdout
+
+        #parse decode_output into a dictionary
+        decode_output.replace('\0', '')
+        lines = decode_output.split('\n')
+        lines = lines[self.EEPROM_DECODE_HEADLINES:]
+        self._eeprom_info_dict = dict()
+
+        for line in lines:
+            try:
+                match = re.search('(0x[0-9a-fA-F]{2})([\s]+[\S]+[\s]+)([\S]+)', line)
+                if match is not None:
+                    idx = match.group(1)
+                    value = match.group(3).rstrip('\0')
+
+                self._eeprom_info_dict[idx] = value
+            except:
+                pass
+
+        return 0
+
+    def get_base_mac(self):
+        """
+        Retrieves the base MAC address for the chassis
+
+        Returns:
+            A string containing the MAC address in the format
+            'XX:XX:XX:XX:XX:XX'
+        """
+        if not self._eeprom_loaded:
+            self._load_eeprom()
+        return self._base_mac
+
+    def get_serial_number(self):
+        """
+        Retrieves the hardware serial number for the chassis
+
+        Returns:
+            A string containing the hardware serial number for this chassis.
+        """
+        if not self._eeprom_loaded:
+            self._load_eeprom()
+        return self._serial_str
+
+    def get_system_eeprom_info(self):
+        """
+        Retrieves the full content of system EEPROM information for the chassis
+
+        Returns:
+            A dictionary where keys are the type code defined in
+            OCP ONIE TlvInfo EEPROM format and values are their corresponding
+            values.
+        """
+        if not self._eeprom_loaded:
+            self._load_eeprom()
+        return self._eeprom_info_dict

platform/mellanox/rules.mk

@@ -25,3 +25,6 @@ $(SYNCD)_RDEPENDS += $(MLNX_SAI)
 
 # Inject mlnx sdk libs to platform monitor
 $(DOCKER_PLATFORM_MONITOR)_DEPENDS += $(APPLIBS) $(SX_COMPLIB) $(SXD_LIBS) $(SX_GEN_UTILS) $(PYTHON_SDK_API) $(APPLIBS_DEV) $(SX_COMPLIB_DEV) $(SXD_LIBS_DEV) $(SX_GEN_UTILS_DEV)
+
+# Inject mlnx mlx libs to platform monitor
+$(DOCKER_PLATFORM_MONITOR)_DEPENDS += $(MFT)