DellEMC S6100 SSD Monitor Porting Changes in master (#7289)

Why I did it
To monitor the SSD health condition in DellEMC S6100 platform post upgrade.

A daemon is introduced to monitor the SSD every one hour.

To check for SSD status at boot time and at the time of cold-reboot.

All these changes are supported only for newer SSD firmware.

Porting changes from 201911 branch

Added a platform_reboot_pre_check script to prevent cold-reboot based on SSD status.
Depends on Azure/sonic-utilities#1556
DO NOT MERGE UNTIL ABOVE PR IS MERGED

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

@@ -10,20 +10,28 @@ 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/reboot_plugin usr/share/sonic/device/x86_64-dell_s6100_c2538-r0
 s6100/scripts/ssd-fw-upgrade 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
+s6100/scripts/s6100_ssd_upgrade_status.sh usr/local/bin
 common/actions.sh usr/share/sonic/device/x86_64-dell_s6100_c2538-r0
 s6100/scripts/platform_sensors.py usr/local/bin
+s6100/scripts/platform_reboot_pre_check usr/share/sonic/device/x86_64-dell_s6100_c2538-r0
 s6100/scripts/hw-management-generate-dump.sh usr/bin
 s6100/modules/sonic_platform-1.0-py2-none-any.whl usr/share/sonic/device/x86_64-dell_s6100_c2538-r0
 s6100/modules/sonic_platform-1.0-py3-none-any.whl usr/share/sonic/device/x86_64-dell_s6100_c2538-r0
 s6100/scripts/platform_watchdog_enable.sh usr/local/bin
 s6100/scripts/platform_watchdog_disable.sh usr/local/bin
 s6100/scripts/sensors usr/bin
+s6100/scripts/iSMART_64 usr/local/bin
 s6100/systemd/platform-modules-s6100.service etc/systemd/system
 s6100/systemd/s6100-lpc-monitor.service etc/systemd/system
+s6100/systemd/s6100-ssd-monitor.service etc/systemd/system
+s6100/systemd/s6100-ssd-monitor.timer etc/systemd/system
+s6100/systemd/s6100-ssd-upgrade-status.service etc/systemd/system
 s6100/systemd/s6100-reboot-cause.service etc/systemd/system
 s6100/systemd/s6100-i2c-enumerate.service etc/systemd/system
 s6100/scripts/s6100_serial_getty_monitor etc/monit/conf.d

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

@@ -0,0 +1,38 @@
+#!/bin/bash
+SSD_FW_UPGRADE="/host/ssd_fw_upgrade"
+
+_error_msg(){
+    echo "The SSD on this unit is $1. 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 $1. 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_pending_upgrade ]; then
+    _error_msg "running older firmware"
+    exit 1
+fi
+
+if [ -e $SSD_FW_UPGRADE/GPIO7_low ] || [ -e $SSD_FW_UPGRADE/GPIO7_error ]; then
+    _error_msg "faulty"
+    exit 1
+fi
+
+if [ -e $SSD_FW_UPGRADE/GPIO7_high ]; then
+    iSMART="/usr/local/bin/iSMART_64"
+    iSMART_OPTIONS="-d /dev/sda"
+
+    iSMART_CMD=`$iSMART $iSMART_OPTIONS`
+
+    GPIO_STATUS=$(echo "$iSMART_CMD" | grep GPIO | awk '{print $NF}')
+
+    if [ $GPIO_STATUS == "0x01" ];then
+	exit 0
+    else
+        _error_msg "faulty"
+        exit 1
+    fi
+fi
+
+exit 1

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

@@ -43,6 +43,8 @@ if [[ "$1" == "init" ]]; then
         /usr/local/bin/platform_watchdog_disable.sh
     fi
 
+    systemctl start --no-block s6100-ssd-upgrade-status.service
+
     is_fast_warm=$(cat /proc/cmdline | grep SONIC_BOOT_TYPE | wc -l)
 
     if [[ "$is_fast_warm" == "1" ]]; then

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

@@ -0,0 +1,21 @@
+#!/bin/bash
+
+SSD_FW_UPGRADE="/host/ssd_fw_upgrade"
+
+if [ -e $SSD_FW_UPGRADE/GPIO7_high ]; then
+    iSMART="/usr/local/bin/iSMART_64"
+    iSMART_OPTIONS="-d /dev/sda"
+
+    iSMART_CMD=`$iSMART $iSMART_OPTIONS`
+    GPIO_STATUS=$(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. 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
+    fi
+else
+    systemctl stop s6100-ssd-monitor.timer
+fi

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

@@ -0,0 +1,118 @@
+#!/bin/bash
+
+SSD_FW_UPGRADE="/host/ssd_fw_upgrade"
+
+if [ -e $SSD_FW_UPGRADE/GPIO7_high ]; then
+    systemctl start --no-block s6100-ssd-monitor.timer
+    exit 0
+fi
+
+if [ -e $SSD_FW_UPGRADE/GPIO7_low ] || [ -e $SSD_FW_UPGRADE/GPIO7_error ]; then
+    exit 0
+fi
+
+[ ! -d $SSD_FW_UPGRADE ] && mkdir $SSD_FW_UPGRADE
+
+SSD_UPGRADE_LOG="$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_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
+fi
+
+echo "$0 `date` SSD FW upgrade logs post reboot." >> $SSD_UPGRADE_LOG
+
+iSMART="/usr/local/bin/iSMART_64"
+iSMART_OPTIONS="-d /dev/sda"
+iSMART_CMD=`$iSMART $iSMART_OPTIONS`
+
+SSD_UPGRADE_STATUS1=`io_rd_wr.py --set --val 06 --offset 210; io_rd_wr.py --set --val 09 --offset 211; io_rd_wr.py --get --offset 212`
+SSD_UPGRADE_STATUS1=$(echo "$SSD_UPGRADE_STATUS1" | awk '{print $NF}')
+
+SSD_UPGRADE_STATUS2=`io_rd_wr.py --set --val 06 --offset 210; io_rd_wr.py --set --val 0A --offset 211; io_rd_wr.py --get --offset 212`
+SSD_UPGRADE_STATUS2=$(echo "$SSD_UPGRADE_STATUS2" | awk '{print $NF}')
+
+if [ $SSD_UPGRADE_STATUS1 == "2" ]; then
+    rm -rf $SSD_FW_UPGRADE/GPIO7_*
+    touch $SSD_FW_UPGRADE/GPIO7_error
+
+    echo "$0 `date` Upgraded to unknown version after first mp_64 upgrade." >> $SSD_UPGRADE_LOG
+
+elif [ $SSD_MODEL == "3IE3" ] && [ $SSD_UPGRADE_STATUS2 == "2" ];then
+    rm -rf $SSD_FW_UPGRADE/GPIO7_*
+    touch $SSD_FW_UPGRADE/GPIO7_error
+
+    echo "$0 `date` Upgraded to unknown version after second mp_64 upgrade." >> $SSD_UPGRADE_LOG
+
+elif [ $SSD_FW_VERSION == "s210506g" ] || [ $SSD_FW_VERSION == "s16425cg" ]; then
+    # If SSD Firmware is upgraded
+    GPIO_STATUS=$(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. 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. Do not power-cycle/reboot this unit!" >> $SSD_UPGRADE_LOG
+        echo "$0 `date` soft-/fast-/warm-reboot is allowed." >> $SSD_UPGRADE_LOG
+
+    else
+        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 S210506G 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_MODEL == "3IE3" ] && [ $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
+    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 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. 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
+
+        if [ $SSD_MODEL == "3IE3" ] && [ $SSD_UPGRADE_STATUS2 == "0" ]; then
+            echo "$0 `date` SSD entered loader mode in first mp_64 and recovered back to older version in second mp_64." >> $SSD_UPGRADE_LOG
+        fi
+    fi
+
+fi
+
+echo "$0 `date` SMF Register 1 = $SSD_UPGRADE_STATUS1" >> $SSD_UPGRADE_LOG
+echo "$0 `date` SMF Register 2 = $SSD_UPGRADE_STATUS2" >> $SSD_UPGRADE_LOG
+echo "$SMART_CMD" >> $SSD_UPGRADE_LOG
+echo "$iSMART_CMD" >> $SSD_UPGRADE_LOG
+sync

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

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

platform/broadcom/sonic-platform-modules-dell/s6100/systemd/s6100-ssd-monitor.service

@@ -0,0 +1,12 @@
+[Unit]
+Description=Dell S6100 SSD monitoring poller
+DefaultDependencies=no
+
+[Service]
+User=root
+ExecStart=/usr/local/bin/s6100_ssd_mon.sh
+RemainAfterExit=no
+
+[Install]
+WantedBy=multi-user.target
+

platform/broadcom/sonic-platform-modules-dell/s6100/systemd/s6100-ssd-monitor.timer

@@ -0,0 +1,12 @@
+[Unit]
+Description=Dell S6100 SSD monitoring poller timer
+DefaultDependencies=no
+After=pmon.service
+
+[Timer]
+OnBootSec=5min
+OnUnitActiveSec=60min
+
+[Install]
+WantedBy=timers.target
+

platform/broadcom/sonic-platform-modules-dell/s6100/systemd/s6100-ssd-upgrade-status.service

@@ -0,0 +1,14 @@
+[Unit]
+Description= Checking Dell S6100 SSD upgrade status
+After=pmon.service
+DefaultDependencies=no
+
+[Service]
+User=root
+Type=oneshot
+ExecStart=/usr/local/bin/s6100_ssd_upgrade_status.sh
+RemainAfterExit=no
+
+[Install]
+WantedBy=multi-user.target
+