[DellEMC] S6100 Last Reboot Reason Thermal Support (#3767)

files/image_config/platform/rc.local

@@ -240,6 +240,12 @@ if [ -f $FIRST_BOOT_FILE ]; then
     # Notify firstboot to Platform, to use it for reboot-cause
     touch /tmp/notify_firstboot_to_platform
 
+    # Create /host/reboot-cause/platform/ directory
+    # can be used to track last reboot reason by some platforms
+    if [ ! -d /host/reboot-cause/platform ]; then
+        mkdir -p /host/reboot-cause/platform
+    fi
+
     if [ -d /host/image-$SONIC_VERSION/platform/$platform ]; then
         dpkg -i /host/image-$SONIC_VERSION/platform/$platform/*.deb
     fi

platform/broadcom/sonic-platform-modules-dell/common/nvram_rd_wr.py

@@ -0,0 +1,97 @@
+#!/usr/bin/python
+#Script to read/write the nvram
+
+import sys
+import os
+import getopt
+import struct
+
+nvram_resource='/dev/nvram'
+
+def usage():
+    ''' This is the Usage Method '''
+
+    print 'Utility for NVRAM read/write'
+    print '\t\t nvram_rd_wr.py --get --offset <offset>'
+    print '\t\t nvram_rd_wr.py --set --val <val>  --offset <offset>'
+    sys.exit(1)
+
+def nvram_reg_read(nvram_resource,offset):
+    fd=os.open(nvram_resource, os.O_RDONLY)
+    if(fd<0):
+        print 'file open failed %s"%nvram_resource'
+        return
+    if(os.lseek(fd, offset, os.SEEK_SET) != offset):
+        print 'lseek failed on %s'%nvram_resource
+        return
+    buf=os.read(fd,1)
+    reg_val1=ord(buf)
+    print 'value %x'%reg_val1
+    os.close(fd)
+
+def nvram_reg_write(nvram_resource,offset,val):
+    fd=os.open(nvram_resource,os.O_RDWR)
+    if(fd<0):
+        print 'file open failed %s"%nvram_resource'
+        return
+    if(os.lseek(fd, offset, os.SEEK_SET) != offset):
+        print 'lseek failed on %s'%nvram_resource
+        return
+    ret=os.write(fd,struct.pack('B',val))
+    if(ret != 1):
+        print 'write failed %d'%ret
+        return
+    os.close(fd)
+
+def main(argv):
+
+    ''' The main function will read the user input from the
+    command line argument and  process the request  '''
+
+    opts = ''
+    val = ''
+    choice = ''
+    resouce = ''
+    offset = ''
+
+    try:
+        opts, args = getopt.getopt(argv, "hgs:" , \
+        ["val=","offset=","help", "get", "set"])
+
+    except getopt.GetoptError:
+        usage()
+
+    if not os.path.exists(nvram_resource):
+        print 'NVRAM is not initialized'
+        sys.exit(1)
+
+    for opt,arg in opts:
+
+        if opt in ('-h','--help'):
+            choice = 'help'
+
+        elif opt in ('-g', '--get'):
+            choice = 'get'
+
+        elif opt in ('-s', '--set'):
+            choice = 'set'
+
+        elif opt ==  '--offset':
+            offset = int(arg,16) - 0xE
+
+        elif opt ==  '--val':
+            val = int(arg,16)
+
+    if choice == 'get' and offset != '':
+        nvram_reg_read(nvram_resource,offset)
+
+    elif choice == 'set' and offset != '' and val != '':
+        nvram_reg_write(nvram_resource,offset,val)
+
+    else:
+        usage()
+
+#Calling the main method
+if __name__ == "__main__":
+    main(sys.argv[1:])
+

platform/broadcom/sonic-platform-modules-dell/debian/platform-modules-s6100.install

@@ -2,7 +2,11 @@ s6100/scripts/iom_power_*.sh usr/local/bin
 s6100/scripts/s6100_platform.sh usr/local/bin
 common/dell_i2c_utils.sh usr/local/bin
 common/io_rd_wr.py usr/local/bin
+common/nvram_rd_wr.py usr/local/bin
 s6100/scripts/platform_reboot_override usr/share/sonic/device/x86_64-dell_s6100_c2538-r0
+s6100/scripts/fast-reboot_plugin usr/share/sonic/device/x86_64-dell_s6100_c2538-r0
+s6100/scripts/track_reboot_reason.sh usr/share/sonic/device/x86_64-dell_s6100_c2538-r0
+s6100/scripts/warm-reboot_plugin usr/share/sonic/device/x86_64-dell_s6100_c2538-r0
 s6100/scripts/override.conf /etc/systemd/system/systemd-reboot.service.d
 common/dell_lpc_mon.sh usr/local/bin
 s6100/scripts/platform_sensors.py usr/local/bin
@@ -12,6 +16,7 @@ s6100/scripts/platform_watchdog_disable.sh usr/local/bin
 s6100/scripts/sensors usr/bin
 s6100/systemd/platform-modules-s6100.service etc/systemd/system
 s6100/systemd/s6100-lpc-monitor.service etc/systemd/system
+s6100/systemd/s6100-reboot-cause.service etc/systemd/system
 tools/flashrom/flashrom usr/local/bin/
 common/fw-updater usr/local/bin
 common/onie_mode_set usr/local/bin

platform/broadcom/sonic-platform-modules-dell/s6100/scripts/fast-reboot_plugin

@@ -0,0 +1,6 @@
+#!/bin/bash
+
+if [[ -d /sys/devices/platform/SMF.512/hwmon/ ]]; then
+    cd /sys/devices/platform/SMF.512/hwmon/*
+    echo 0xcc > mb_poweron_reason
+fi

platform/broadcom/sonic-platform-modules-dell/s6100/scripts/platform_reboot_override

@@ -1,11 +1,17 @@
 #!/usr/bin/python
 import sys
 import os
+import subprocess
 import struct
 
 PORT_RES = '/dev/port'
 
 
+def log_software_reboot():
+    # Run plugin script which will track the cli triggered reboot, fastboot, warmboot
+    res = subprocess.check_output(['/usr/share/sonic/device/x86_64-dell_s6100_c2538-r0/fast-reboot_plugin'])
+    return
+
 def portio_reg_write(resource, offset, val):
     fd = os.open(resource, os.O_RDWR)
     if(fd < 0):
@@ -21,5 +27,6 @@ def portio_reg_write(resource, offset, val):
     os.close(fd)
 
 if __name__ == "__main__":
+    log_software_reboot()
     portio_reg_write(PORT_RES, 0xcf9, 0xe)
 

platform/broadcom/sonic-platform-modules-dell/s6100/scripts/s6100_platform.sh

@@ -237,24 +237,6 @@ reset_muxes() {
     io_rd_wr.py --set --val 0xff --offset 0x20b
 }
 
-track_reboot_reason() {
-    if [[ -d /sys/devices/platform/SMF.512/hwmon/ ]]; then
-        rv=$(cd /sys/devices/platform/SMF.512/hwmon/*; cat mb_poweron_reason)
-        reason=$(echo $rv | cut -d 'x' -f2)
-        if [ $reason == "ff" ]; then
-            cd /sys/devices/platform/SMF.512/hwmon/*
-            if [[ -e /tmp/notify_firstboot_to_platform ]]; then
-                echo 0x01 > mb_poweron_reason
-            else
-                echo 0xbb > mb_poweron_reason
-            fi
-        elif [ $reason == "bb" ] || [ $reason == "1" ]; then
-            cd /sys/devices/platform/SMF.512/hwmon/*
-            echo 0xaa > mb_poweron_reason
-        fi
-    fi
-}
-
 install_python_api_package() {
     device="/usr/share/sonic/device"
     platform=$(/usr/local/bin/sonic-cfggen -H -v DEVICE_METADATA.localhost.platform)
@@ -277,7 +259,8 @@ if [[ "$1" == "init" ]]; then
     modprobe dell_ich
     modprobe dell_s6100_iom_cpld
     modprobe dell_s6100_lpc
-    track_reboot_reason
+    modprobe nvram
+    systemctl start s6100-reboot-cause.service
 
     # Disable Watchdog Timer
     if [[ -e /usr/local/bin/platform_watchdog_disable.sh ]]; then
@@ -313,6 +296,7 @@ elif [[ "$1" == "deinit" ]]; then
     modprobe -r i2c-mux-pca954x
     modprobe -r i2c-dev
     modprobe -r dell_ich
+    modprobe -r nvram
     remove_python_api_package
 else
      echo "s6100_platform : Invalid option !"

platform/broadcom/sonic-platform-modules-dell/s6100/scripts/track_reboot_reason.sh

@@ -0,0 +1,168 @@
+#!/bin/bash
+
+reboot_dir_found=false
+reboot_file_found=false
+smf_dir_missing=0
+nvram_missing=0
+
+REBOOT_CAUSE_FILE=/host/reboot-cause/reboot-cause.txt
+REBOOT_REASON_FILE=/host/reboot-cause/platform/reboot_reason
+BIOS_VERSION_FILE=/host/reboot-cause/platform/bios_minor_version
+SMF_POWERON_REASON=/sys/devices/platform/SMF.512/hwmon/*/smf_poweron_reason
+SMF_RESET_REASON=/sys/devices/platform/SMF.512/hwmon/*/smf_reset_reason
+MAILBOX_POWERON_REASON=/sys/devices/platform/SMF.512/hwmon/*/mb_poweron_reason
+NVRAM_DEVICE_FILE=/dev/nvram
+RESET_REASON_FILE=/host/reboot-cause/platform/reset_reason
+SMF_DIR=/sys/devices/platform/SMF.512/hwmon/
+
+while [[ ! -d $SMF_DIR ]]
+do
+    sleep 0.5
+    let smf_dir_missing=$smf_dir_missing+1
+    if [[ "$smf_dir_missing" = "5" ]]; then
+        echo "SMF is not initialized"
+        smf_dir_missing=0
+    fi
+done
+
+SMF_RESET=$(cat $SMF_RESET_REASON)
+
+if [[ -d /host/reboot-cause/platform ]]; then
+    reboot_dir_found=true
+    if [[ -f $REBOOT_REASON_FILE ]]; then
+        reboot_file_found=true
+    fi
+fi
+
+SMF_BIOS_REG=$(io_rd_wr.py --get --offset 0x203 | cut -d " " -f 3)
+SMF_BIOS_REG=$((16#$SMF_BIOS_REG))
+bios_secondary_boot=$(($SMF_BIOS_REG & 1))
+
+_get_smf_reset_register(){
+    BIOS_VERSION=$(/usr/sbin/dmidecode -s system-version)
+    BIOS_VERSION_MINOR=$(echo $BIOS_VERSION | cut -d'-' -f 2)
+
+    if [[ $BIOS_VERSION_MINOR -gt 7 ]]; then
+        echo $BIOS_VERSION > $BIOS_VERSION_FILE
+    elif [[ "$bios_secondary_boot" = "0" ]]; then
+        # For Primary BIOS with older version
+        if [[ -e $BIOS_VERSION_FILE ]]; then
+            rm $BIOS_VERSION_FILE
+        fi
+    fi
+
+    if [[ -e $BIOS_VERSION_FILE ]]; then
+        while [[ ! -e $NVRAM_DEVICE_FILE ]]
+        do
+            sleep 1
+            let nvram_missing=$nvram_missing+1
+            if [[ "$nvram_missing" = "5" ]]; then
+                echo "NVRAM is not initialized"
+                nvram_missing=0
+            fi
+        done
+        first_reset=$(nvram_rd_wr.py --get --offset 0x5c | cut -d " " -f 2)
+        second_reset=$(nvram_rd_wr.py --get --offset 0x5d | cut -d " " -f 2)
+        third_reset=$(nvram_rd_wr.py --get --offset 0x5e | cut -d " " -f 2)
+        fourth_reset=$(nvram_rd_wr.py --get --offset 0x5f | cut -d " " -f 2)
+
+        if [[ "$first_reset" != "ee" ]]; then
+            SMF_RESET=$first_reset
+        fi
+
+        # Saving NVRAM values for future debugging
+        if [[ $reboot_dir_found = true ]]; then
+            echo "First reset - $first_reset" > $RESET_REASON_FILE
+            echo "Second reset - $second_reset" >> $RESET_REASON_FILE
+            echo "Third reset - $third_reset" >> $RESET_REASON_FILE
+            echo "Fourth reset - $fourth_reset" >> $RESET_REASON_FILE
+        fi
+
+        # Clearing NVRAM values to holding next reset values
+        nvram_rd_wr.py --set --val 0xee --offset 0x58
+        nvram_rd_wr.py --set --val 0xee --offset 0x5c
+        nvram_rd_wr.py --set --val 0xee --offset 0x5d
+        nvram_rd_wr.py --set --val 0xee --offset 0x5e
+        nvram_rd_wr.py --set --val 0xee --offset 0x5f
+    fi
+}
+
+_is_thermal_reset() {
+    prev_thermal=$(cat $REBOOT_REASON_FILE)
+    curr_poweron_reason=$(cat $SMF_POWERON_REASON)
+    if [[ $curr_poweron_reason = "11" ]]; then
+        echo 0
+        return
+    fi
+    if [[ $prev_thermal = $curr_poweron_reason ]]; then
+        echo 2
+        return
+    else
+        echo "$curr_poweron_reason" > $REBOOT_REASON_FILE
+        echo 1
+        return
+    fi
+
+    echo 0
+    return
+}
+
+_is_watchdog_reset(){
+    curr_reset_reason=$SMF_RESET
+    if [[ $curr_reset_reason = "33" ]]; then
+        echo 1
+        return
+    fi
+
+    echo 0
+    return
+}
+
+_is_unknown_reset(){
+    if [[ -f $REBOOT_CAUSE_FILE ]]; then
+        if [[ $1 = 0 ]]; then
+            echo "Unknown software reboot" > $REBOOT_CAUSE_FILE
+            return
+        fi
+        curr_poweron_reason=$(cat $SMF_POWERON_REASON)
+        curr_reset_reason=$SMF_RESET
+        mb_poweron_reason=$(cat $MAILBOX_POWERON_REASON)
+        echo "Unknown POR: $curr_poweron_reason RST: $curr_reset_reason MBR: $mb_poweron_reason" > $REBOOT_CAUSE_FILE
+    fi
+    
+}
+
+update_mailbox_register(){
+    if [[ "$bios_secondary_boot" = "1" ]]; then
+        echo "Secondary BIOS booted"
+    fi
+
+    if [[ $reboot_file_found = false ]]; then
+        echo "None" > $REBOOT_REASON_FILE
+    fi
+
+    _get_smf_reset_register
+    if [[ -d /sys/devices/platform/SMF.512/hwmon/ ]]; then
+        is_thermal_reboot=$(_is_thermal_reset)
+
+        is_wd_reboot=$(_is_watchdog_reset)
+
+        mbr=$(cat $MAILBOX_POWERON_REASON)
+        reason=$(echo $mbr | cut -d 'x' -f2)
+        if [[ $reason = "ff" ]]; then
+            echo "None" > $REBOOT_REASON_FILE
+            echo 0xbb > $MAILBOX_POWERON_REASON
+        elif [[ $is_thermal_reboot = 1 ]]; then
+            echo 0xee > $MAILBOX_POWERON_REASON
+        elif [[ $is_wd_reboot = 1 ]] && [[ $reason != "cc" ]]; then
+            echo 0xdd > $MAILBOX_POWERON_REASON
+        elif [[ $reason = "cc" ]]; then
+            echo 0xaa > $MAILBOX_POWERON_REASON
+        else
+            _is_unknown_reset $is_thermal_reboot
+            echo 0x99 > $MAILBOX_POWERON_REASON
+        fi
+    fi
+}
+
+update_mailbox_register

platform/broadcom/sonic-platform-modules-dell/s6100/scripts/warm-reboot_plugin

@@ -0,0 +1 @@
+fast-reboot_plugin

platform/broadcom/sonic-platform-modules-dell/s6100/sonic_platform/chassis.py

@@ -18,7 +18,9 @@ try:
     from sonic_platform.module import Module
     from sonic_platform.thermal import Thermal
     from sonic_platform.component import Component
+    from sonic_platform.watchdog import Watchdog
     from eeprom import Eeprom
+    import time
 except ImportError as e:
     raise ImportError(str(e) + "- required module not found")
 
@@ -43,6 +45,8 @@ class Chassis(ChassisBase):
     reset_reason_dict[33] = ChassisBase.REBOOT_CAUSE_WATCHDOG
     reset_reason_dict[44] = ChassisBase.REBOOT_CAUSE_NON_HARDWARE
     reset_reason_dict[55] = ChassisBase.REBOOT_CAUSE_NON_HARDWARE
+    reset_reason_dict[66] = ChassisBase.REBOOT_CAUSE_HARDWARE_OTHER
+    reset_reason_dict[77] = ChassisBase.REBOOT_CAUSE_HARDWARE_OTHER
 
     power_reason_dict = {}
     power_reason_dict[11] = ChassisBase.REBOOT_CAUSE_POWER_LOSS
@@ -76,10 +80,15 @@ class Chassis(ChassisBase):
             component = Component(i)
             self._component_list.append(component)
 
+        self._watchdog = Watchdog()
+
     def _get_reboot_reason_smf_register(self):
-        # Returns 0xAA on software reload
+        # In S6100, mb_poweron_reason register will
-        # Returns 0xFF on power-cycle
+        # Returns 0xaa or 0xcc on software reload
-        # Returns 0x01 on first-boot
+        # Returns 0xff or 0xbb on power-cycle
+        # Returns 0xdd on Watchdog
+        # Returns 0xee on Thermal Shutdown
+        # Returns 0x99 on Unknown reset
         smf_mb_reg_reason = self._get_pmc_register('mb_poweron_reason')
         return int(smf_mb_reg_reason, 16)
 
@@ -188,27 +197,21 @@ class Chassis(ChassisBase):
         power_reason = int(self._get_pmc_register('smf_poweron_reason'))
         smf_mb_reg_reason = self._get_reboot_reason_smf_register()
 
-        if ((smf_mb_reg_reason == 0x01) and (power_reason == 0x11)):
-            return (ChassisBase.REBOOT_CAUSE_NON_HARDWARE, None)
-
-        # Reset_Reason = 11 ==> PowerLoss
-        # So return the reboot reason from Last Power_Reason Dictionary
-        # If Reset_Reason is not 11 return from Reset_Reason dictionary
-        # Also check if power_reason, reset_reason are valid values by
-        # checking key presence in dictionary else return
-        # REBOOT_CAUSE_HARDWARE_OTHER as the Power_Reason and Reset_Reason
-        # registers returned invalid data
-
-        # In S6100, if Reset_Reason is not 11 and smf_mb_reg_reason
-        # is ff or bb, then it is PowerLoss
-        if (reset_reason == 11):
-            if (power_reason in self.power_reason_dict):
-                return (self.power_reason_dict[power_reason], None)
-        else:
         if ((smf_mb_reg_reason == 0xbb) or (smf_mb_reg_reason == 0xff)):
             return (ChassisBase.REBOOT_CAUSE_POWER_LOSS, None)
-
+        elif ((smf_mb_reg_reason == 0xaa) or (smf_mb_reg_reason == 0xcc)):
-            if (reset_reason in self.reset_reason_dict):
+            return (ChassisBase.REBOOT_CAUSE_NON_HARDWARE, None)
-                return (self.reset_reason_dict[reset_reason], None)
+        elif (smf_mb_reg_reason == 0xdd):
+            return (ChassisBase.REBOOT_CAUSE_WATCHDOG, None)
+        elif (smf_mb_reg_reason == 0xee):
+            return (self.power_reason_dict[power_reason], None)
+        elif (reset_reason == 66):
+            return (ChassisBase.REBOOT_CAUSE_HARDWARE_OTHER,
+                    "Emulated Cold Reset")
+        elif (reset_reason == 77):
+            return (ChassisBase.REBOOT_CAUSE_HARDWARE_OTHER,
+                    "Emulated Warm Reset")
+        else:
+            return (ChassisBase.REBOOT_CAUSE_NON_HARDWARE, None)
 
         return (ChassisBase.REBOOT_CAUSE_HARDWARE_OTHER, "Invalid Reason")

platform/broadcom/sonic-platform-modules-dell/s6100/sonic_platform/eeprom.py

@@ -35,9 +35,13 @@ class Eeprom(eeprom_tlvinfo.TlvInfoDecoder):
             else:
                 self.eeprom_data = self.read_eeprom()
         except:
-            self.eeprom_data = "N/A"
             if not self.is_module:
+                try:
+                    self.eeprom_data = self.read_eeprom()
+                except:
                     raise RuntimeError("Eeprom is not Programmed")
+            else:
+                self.eeprom_data = "N/A"
         else:
             eeprom = self.eeprom_data
 

platform/broadcom/sonic-platform-modules-dell/s6100/systemd/s6100-reboot-cause.service

@@ -0,0 +1,12 @@
+[Unit]
+Description=Read Dell S6100 reboot cause
+Before=process-reboot-cause.service
+DefaultDependencies=no
+
+[Service]
+Type=oneshot
+ExecStart=/usr/share/sonic/device/x86_64-dell_s6100_c2538-r0/track_reboot_reason.sh
+RemainAfterExit=no
+
+[Install]
+WantedBy=multi-user.target