[201911] DellEMC S6100 SSD Monitor additional changes (sonic-net#7291)

Why I did it
Added soft-reboot plugin support.
Added SSD version s16425cq check
Added error message to display in console/SSH in case reboot is called in faulty/non-upgraded devices.

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

@@ -10,6 +10,7 @@ s6100/scripts/platform_reboot_override usr/share/sonic/device/x86_64-dell_s6100_
 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/soft-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/s6100_ssd_mon.sh usr/local/bin

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

@@ -1,10 +1,16 @@
 #!/bin/bash
 SSD_FW_UPGRADE="/host/ssd_fw_upgrade"
 
+_error_msg(){
+ echo "The SSD on this unit is faulty. Do not power-cycle/reboot this unit!"
+ echo "soft-/fast-/warm-reboot is allowed."
+ logger -p user.crit -t DELL_S6100_SSD_MON "The SSD on this unit is faulty. Do not power-cycle/reboot this unit!"
+ logger -p user.crit -t DELL_S6100_SSD_MON "soft-/fast-/warm-reboot is allowed."
+}
+
 # Check SSD Status
-if [ -e $SSD_FW_UPGRADE/GPIO7_low ] || [ -e $SSD_FW_UPGRADE/GPIO7_error ] || [ -e $SSD_FW_UPGRADE/GPIO_pending_upgrade ]; then
- logger -p user.crit -t DELL_S6100_SSD_MON "The SSD on this unit is faulty and does not support cold reboot."
- logger -p user.crit -t DELL_S6100_SSD_MON "Please perform a soft-/fast-/warm-reboot instead"
+if [ -e $SSD_FW_UPGRADE/GPIO7_low ] || [ -e $SSD_FW_UPGRADE/GPIO7_error ] || [ -e $SSD_FW_UPGRADE/GPIO7_pending_upgrade ]; then
+ _error_msg
  exit 1
 fi
 
@@ -19,8 +25,7 @@ if [ -e $SSD_FW_UPGRADE/GPIO7_high ]; then
  if [ $GPIO_STATUS == "0x01" ];then
  exit 0
  else
- logger -p user.crit -t DELL_S6100_SSD_MON "The SSD on this unit is faulty and does not support cold reboot."
- logger -p user.crit -t DELL_S6100_SSD_MON "Please perform a soft-/fast-/warm-reboot instead"
+ _error_msg
  exit 1
  fi
 fi

platform/broadcom/sonic-platform-modules-dell/s6100/scripts/s6100_ssd_mon.sh

@@ -10,8 +10,8 @@ if [ -e $SSD_FW_UPGRADE/GPIO7_high ]; then
  GPIO_STATUS=$(echo "$iSMART_CMD" | grep GPIO | awk '{print $NF}')
 
  if [ $GPIO_STATUS != "0x01" ];then
- logger -p user.crit -t DELL_S6100_SSD_MON "The SSD on this unit is faulty and does not support cold reboot."
- logger -p user.crit -t DELL_S6100_SSD_MON "If a reboot is required, please perform a soft-/fast-/warm-reboot."
+ logger -p user.crit -t DELL_S6100_SSD_MON "The SSD on this unit is faulty. Do not power-cycle/reboot this unit!"
+ logger -p user.crit -t DELL_S6100_SSD_MON "soft-/fast-/warm-reboot is allowed."
  rm -rf $SSD_FW_UPGRADE/GPIO7_*
  touch $SSD_FW_UPGRADE/GPIO7_low
  systemctl stop s6100-ssd-monitor.timer

platform/broadcom/sonic-platform-modules-dell/s6100/scripts/s6100_ssd_upgrade_status.sh

@@ -18,10 +18,15 @@ SSD_UPGRADE_LOG="$SSD_FW_UPGRADE/upgrade.log"
 SMART_CMD=`smartctl -a /dev/sda`
 
 SSD_FW_VERSION=$(echo "$SMART_CMD" | grep "Firmware Version" | awk '{print $NF}')
+SSD_FW_VERSION=${SSD_FW_VERSION,,}
 SSD_MODEL=$(echo "$SMART_CMD" | grep "Device Model" | awk '{print $NF}')
 
 if [ -e $SSD_FW_UPGRADE/GPIO7_pending_upgrade ]; then
- if [ $SSD_FW_VERSION == "S141002C" ] || [ $SSD_FW_VERSION == "S16425c1" ]; then
+ if [ $SSD_MODEL == "3IE" ] && [ $SSD_FW_VERSION == "s141002c" ]; then
+ # If SSD Firmware is not upgraded
+ exit 0
+ fi
+ if [ $SSD_FW_VERSION == "s16425c1" ] || [ $SSD_FW_VERSION == "s16425cq" ]; then
  # If SSD Firmware is not upgraded
  exit 0
  fi
@@ -51,47 +56,51 @@ elif [ $SSD_UPGRADE_STATUS2 == "2" ];then
 
  echo "$0 `date` Upgraded to unknown version after second mp_64 upgrade." >> $SSD_UPGRADE_LOG
 
-elif [ $SSD_FW_VERSION == "S141002G" ] || [ $SSD_FW_VERSION == "S16425cG" ]; then
+elif [ $SSD_FW_VERSION == "s141002g" ] || [ $SSD_FW_VERSION == "s16425cg" ]; then
  # If SSD Firmware is upgraded
  GPIO_STATUS=$(echo "$iSMART_CMD" | grep GPIO | awk '{print $NF}')
 
  if [ $GPIO_STATUS != "0x01" ];then
- logger -p user.crit -t DELL_S6100_SSD_MON "The SSD on this unit is faulty and does not support reboot."
- logger -p user.crit -t DELL_S6100_SSD_MON "If a reboot is required, please perform a soft-/fast-/warm-reboot."
+ logger -p user.crit -t DELL_S6100_SSD_MON "The SSD on this unit is faulty. Do not power-cycle/reboot this unit!"
+ logger -p user.crit -t DELL_S6100_SSD_MON "soft-/fast-/warm-reboot is allowed."
  rm -rf $SSD_FW_UPGRADE/GPIO7_*
  touch $SSD_FW_UPGRADE/GPIO7_low
- echo "$0 `date` The SSD on this unit is faulty and does not support cold reboot." >> $SSD_UPGRADE_LOG
- echo "$0 `date` If a reboot is required, please perform a soft-/fast-/warm-reboot." >> $SSD_UPGRADE_LOG
+ echo "$0 `date` The SSD on this unit is faulty. Do not power-cycle/reboot this unit!" >> $SSD_UPGRADE_LOG
+ echo "$0 `date` soft-/fast-/warm-reboot is allowed." >> $SSD_UPGRADE_LOG
 
  else
- rm -rf $SSD_FW_UPGRADE/GPIO7_*
- touch $SSD_FW_UPGRADE/GPIO7_high
- fi
-
- systemctl start --no-block s6100-ssd-monitor.timer
-
- if [ $SSD_UPGRADE_STATUS1 == "0" ]; then
- if [ $SSD_MODEL == "3IE" ];then
- echo "$0 `date` SSD FW upgraded from S141002C to S141002G in first mp_64." >> $SSD_UPGRADE_LOG
- else
- echo "$0 `date` SSD FW upgraded from S16425c1 to S16425cG in first mp_64." >> $SSD_UPGRADE_LOG
+ if [ $SSD_UPGRADE_STATUS1 == "0" ]; then
+ rm -rf $SSD_FW_UPGRADE/GPIO7_*
+ touch $SSD_FW_UPGRADE/GPIO7_high
+ systemctl start --no-block s6100-ssd-monitor.timer
+
+ if [ $SSD_MODEL == "3IE" ];then
+ echo "$0 `date` SSD FW upgraded from S141002C to S141002G in first mp_64." >> $SSD_UPGRADE_LOG
+ else
+ echo "$0 `date` SSD FW upgraded from S16425c1 to S16425cG in first mp_64." >> $SSD_UPGRADE_LOG
+ fi
+ elif [ $SSD_UPGRADE_STATUS2 == "1" ]; then
+ rm -rf $SSD_FW_UPGRADE/GPIO7_*
+ touch $SSD_FW_UPGRADE/GPIO7_low
+ logger -p user.crit -t DELL_S6100_SSD_MON "The SSD on this unit is faulty. Do not power-cycle/reboot this unit!"
+ logger -p user.crit -t DELL_S6100_SSD_MON "soft-/fast-/warm-reboot is allowed."
+
+ echo "$0 `date` SSD entered loader mode in first mp_64 and upgraded to latest version after second mp_64." >> $SSD_UPGRADE_LOG
  fi
- elif [ $SSD_UPGRADE_STATUS2 == "1" ]; then
- echo "$0 `date` SSD entered loader mode in first mp_64 and upgraded to latest version after second mp_64." >> $SSD_UPGRADE_LOG
  fi
 
 else
  if [ $SSD_UPGRADE_STATUS1 == "ff" ] && [ $SSD_UPGRADE_STATUS2 == "ff" ]; then
  rm -rf $SSD_FW_UPGRADE/GPIO7_*
  touch $SSD_FW_UPGRADE/GPIO7_pending_upgrade
 
- echo "$0 `date` SSD upgrade didn’t happened." >> $SSD_UPGRADE_LOG
+ echo "$0 `date` SSD upgrade didn’t happen." >> $SSD_UPGRADE_LOG
 
  elif [ $SSD_UPGRADE_STATUS1 == "1" ]; then
  rm -rf $SSD_FW_UPGRADE/GPIO7_*
  touch $SSD_FW_UPGRADE/GPIO7_low
- logger -p user.crit -t DELL_S6100_SSD_MON "The SSD on this unit is faulty and does not support reboot."
- logger -p user.crit -t DELL_S6100_SSD_MON "If a reboot is required, please perform a soft-/fast-/warm-reboot."
+ logger -p user.crit -t DELL_S6100_SSD_MON "The SSD on this unit is faulty. Do not power-cycle/reboot this unit!"
+ logger -p user.crit -t DELL_S6100_SSD_MON "soft-/fast-/warm-reboot is allowed."
 
  echo "$0 `date` SSD entered loader mode in first mp_64 upgrade." >> $SSD_UPGRADE_LOG
 

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

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