From ab7c4ee66191ec260aede972074b5d13263ae0c2 Mon Sep 17 00:00:00 2001
From: Zhijian Li <zhijianli@microsoft.com>
Date: Mon, 14 Aug 2023 12:33:19 +0800
Subject: [PATCH] [Celestica-E1031] Enable CPU watchdog (#16083)

Enable CPU watchdog on Celestica-E1031.
---
 .../platform-modules-haliburton.install       |  2 +
 .../platform-modules-haliburton.postinst      |  5 +-
 .../haliburton/script/cpu_wdt                 | 89 +++++++++++++++++++
 .../haliburton/systemd/cpu_wdt.service        | 10 +++
 4 files changed, 105 insertions(+), 1 deletion(-)
 create mode 100644 platform/broadcom/sonic-platform-modules-cel/haliburton/script/cpu_wdt
 create mode 100644 platform/broadcom/sonic-platform-modules-cel/haliburton/systemd/cpu_wdt.service

diff --git a/platform/broadcom/sonic-platform-modules-cel/debian/platform-modules-haliburton.install b/platform/broadcom/sonic-platform-modules-cel/debian/platform-modules-haliburton.install
index 6c58afe6d5..c3d598623e 100644
--- a/platform/broadcom/sonic-platform-modules-cel/debian/platform-modules-haliburton.install
+++ b/platform/broadcom/sonic-platform-modules-cel/debian/platform-modules-haliburton.install
@@ -1,5 +1,6 @@
 haliburton/cfg/haliburton-modules.conf etc/modules-load.d
 haliburton/systemd/platform-modules-haliburton.service lib/systemd/system
+haliburton/systemd/cpu_wdt.service lib/systemd/system
 haliburton/script/fancontrol.sh etc/init.d
 haliburton/script/fancontrol.service lib/systemd/system
 services/fancontrol/fancontrol  usr/local/bin
@@ -8,4 +9,5 @@ services/platform_api/platform_api_mgnt.sh usr/local/bin
 haliburton/script/popmsg.sh usr/local/bin
 haliburton/script/udev_prefix.sh usr/local/bin
 haliburton/script/reload_udev.sh usr/local/bin
+haliburton/script/cpu_wdt usr/local/bin
 haliburton/script/50-ttyUSB-C0.rules etc/udev/rules.d
diff --git a/platform/broadcom/sonic-platform-modules-cel/debian/platform-modules-haliburton.postinst b/platform/broadcom/sonic-platform-modules-cel/debian/platform-modules-haliburton.postinst
index 57ac1be341..d427f17579 100644
--- a/platform/broadcom/sonic-platform-modules-cel/debian/platform-modules-haliburton.postinst
+++ b/platform/broadcom/sonic-platform-modules-cel/debian/platform-modules-haliburton.postinst
@@ -3,6 +3,7 @@ depmod -a
 sudo chmod +x /usr/local/bin/udev_prefix.sh
 sudo chmod +x /usr/local/bin/popmsg.sh
 sudo chmod +x /usr/local/bin/reload_udev.sh
+sudo chmod +x /usr/local/bin/cpu_wdt
 
 /usr/local/bin/platform_api_mgnt.sh install
 /etc/init.d/fancontrol.sh install
@@ -10,6 +11,8 @@ sudo chmod +x /usr/local/bin/reload_udev.sh
 
 systemctl enable platform-modules-haliburton.service
 systemctl enable fancontrol.service
+systemctl enable cpu_wdt.service
 
 systemctl start platform-modules-haliburton.service
-systemctl start fancontrol.service
\ No newline at end of file
+systemctl start fancontrol.service
+systemctl start cpu_wdt.service
diff --git a/platform/broadcom/sonic-platform-modules-cel/haliburton/script/cpu_wdt b/platform/broadcom/sonic-platform-modules-cel/haliburton/script/cpu_wdt
new file mode 100644
index 0000000000..8eed59013b
--- /dev/null
+++ b/platform/broadcom/sonic-platform-modules-cel/haliburton/script/cpu_wdt
@@ -0,0 +1,89 @@
+#!/bin/bash
+
+SYSLOG_IDENTIFIER="cpu_wdt"
+CPUWDT_MAIN_TASK_RUNNING_FLAG=true
+TIMEOUT=180
+KEEPALIVE=60
+
+function log_info()
+{
+    logger -p info -t ${SYSLOG_IDENTIFIER}  "$@"
+}
+
+function usage()
+{
+    echo "Usage: $0 ACTION [OPTIONS]..."
+    echo ""
+    echo "Actions:"
+    echo "  start    Start CPU WDT"
+    echo "  stop     Stop CPU WDT"
+    echo ""
+    echo "Options:"
+    echo "  -h              Show this help"
+    echo "  -t <timeout>    WDT timeout period: {30|60|180}, default 180"
+    echo "  -k <keepalive>  WDT keep alive period, {1..(timeout-5)}, default 60"
+    exit 1
+}
+
+function validate_action()
+{
+    if [[ "${ACTION}" != "start" && "${ACTION}" != "stop" ]]; then
+        echo -e "Invalid action: ${ACTION}\n"
+        usage
+    fi
+}
+
+function validate_options()
+{
+    if [[ ${TIMEOUT} != "30" && ${TIMEOUT} != "60" && ${TIMEOUT} != "180" ]]; then
+        echo -e "Invalid timeout value: ${TIMEOUT}\n"
+        usage
+    fi
+    if [[ ${KEEPALIVE} -le 0 || ${KEEPALIVE} -gt $((TIMEOUT - 5)) ]]; then
+        echo "Invalid keepalive value: ${KEEPALIVE}"
+        echo ""
+        usage
+    fi
+}
+
+trap 'log_info "Caught SIGHUP - ignoring..."' SIGHUP
+trap 'log_info "Caught SIGINT - exiting..."; CPUWDT_MAIN_TASK_RUNNING_FLAG=false' SIGINT
+trap 'log_info "Caught SIGTERM - exiting..."; CPUWDT_MAIN_TASK_RUNNING_FLAG=false' SIGTERM
+
+ACTION=$1
+shift
+validate_action
+
+while getopts "t:k:" OPTION; do
+    case $OPTION in
+    t)
+        TIMEOUT=${OPTARG}
+        ;;
+    k)
+        KEEPALIVE=${OPTARG}
+        ;;
+    *)
+        usage
+    esac
+done
+
+validate_options
+
+if [[ "${ACTION}" == "start" ]]; then
+    # enable
+    log_info "Enable CPU WDT.."
+    watchdogutil arm -s "${TIMEOUT}" > /dev/null
+    log_info "CPU WDT has been enabled with $TIMEOUT seconds timeout"
+
+    # keep alive
+    log_info "Enable keep alive messaging every $KEEPALIVE seconds"
+    while [[ ${CPUWDT_MAIN_TASK_RUNNING_FLAG} == "true" ]]; do
+        watchdogutil arm -s "${TIMEOUT}" > /dev/null
+        sleep "${KEEPALIVE}"
+    done
+    log_info "Keep alive messaging has been disabled"
+fi
+
+log_info "Disable CPU WDT.."
+watchdogutil disarm
+log_info "CPU WDT has been disabled!"
diff --git a/platform/broadcom/sonic-platform-modules-cel/haliburton/systemd/cpu_wdt.service b/platform/broadcom/sonic-platform-modules-cel/haliburton/systemd/cpu_wdt.service
new file mode 100644
index 0000000000..51cc537c20
--- /dev/null
+++ b/platform/broadcom/sonic-platform-modules-cel/haliburton/systemd/cpu_wdt.service
@@ -0,0 +1,10 @@
+[Unit]
+Description=CPU WDT
+After=platform-modules-haliburton.service
+Requires=platform-modules-haliburton.service
+
+[Service]
+ExecStart=-/usr/local/bin/cpu_wdt start
+
+[Install]
+WantedBy=multi-user.target