From a7e25d9feb97d748a9bc67718fbaa400a6893ccc Mon Sep 17 00:00:00 2001
From: Nazarii Hnydyn <nazariig@nvidia.com>
Date: Mon, 10 Jun 2024 14:00:02 +0300
Subject: [PATCH] [Mellanox] Improve FW upgrade: add locking mechanism.
 (#18925)

User implications:
On cold/warm/fast reboot if there is a syncd service start ongoing, the target procedure will stay blocked until the original call is done. This may delay the action in time.

- Why I did it
Improve the upgrade fw script avoiding errors in the logs when cold reboot is executed during service restart
Work item tracking

- How I did it
Added locking mechanism for Mellanox FW upgrade

- How to verify it
Run cold reboot after DUT first install

Signed-off-by: Nazarii Hnydyn <nazariig@nvidia.com>
---
 platform/mellanox/mlnx-fw-upgrade.j2 | 26 ++++++++++++++++++++++++++
 1 file changed, 26 insertions(+)

diff --git a/platform/mellanox/mlnx-fw-upgrade.j2 b/platform/mellanox/mlnx-fw-upgrade.j2
index eb288624982c..e4c567c1b76e 100755
--- a/platform/mellanox/mlnx-fw-upgrade.j2
+++ b/platform/mellanox/mlnx-fw-upgrade.j2
@@ -20,6 +20,8 @@ declare -r SCRIPT_NAME="$(basename "$0")"
 declare -r SCRIPT_PATH="$(readlink -f "$0")"
 declare -r SCRIPT_DIR="$(dirname "$SCRIPT_PATH")"
 
+declare -r LOCKFILE="/tmp/mlxfwmanager-lock"
+
 declare -r YES_PARAM="yes"
 declare -r NO_PARAM="no"
 
@@ -163,6 +165,20 @@ function ExitSuccess() {
     exit "${EXIT_SUCCESS}"
 }
 
+function LockStateChange() {
+    LogInfo "Locking ${LOCKFILE} from ${SCRIPT_NAME}"
+
+    exec {LOCKFD}>${LOCKFILE}
+    /usr/bin/flock -x ${LOCKFD}
+
+    LogInfo "Locked ${LOCKFILE} (${LOCKFD}) from ${SCRIPT_NAME}"
+}
+
+function UnlockStateChange() {
+    LogInfo "Unlocking ${LOCKFILE} (${LOCKFD}) from ${SCRIPT_NAME}"
+    /usr/bin/flock -u ${LOCKFD}
+}
+
 function WaitForDevice() {
     local -i QUERY_RETRY_COUNT_MAX="10"
     local -i QUERY_RETRY_COUNT="0"
@@ -372,10 +388,20 @@ function ExitIfQEMU() {
     fi
 }
 
+function Cleanup() {
+    if [[ -n "${LOCKFD}" ]]; then
+        UnlockStateChange
+    fi
+}
+
+trap Cleanup EXIT
+
 ParseArguments "$@"
 
 ExitIfQEMU
 
+LockStateChange
+
 WaitForDevice
 
 if [ "${IMAGE_UPGRADE}" != "${YES_PARAM}" ]; then