From 2244d946525e4acfaa689a4264a7423705d766b7 Mon Sep 17 00:00:00 2001
From: dgsudharsan <sudharsand@nvidia.com>
Date: Fri, 27 Sep 2024 06:22:09 +0000
Subject: [PATCH 1/3] [Mellanox] Fix retry logic on discovery of MST device

---
 platform/mellanox/mlnx-fw-upgrade.j2 | 16 ++++++++++------
 1 file changed, 10 insertions(+), 6 deletions(-)

diff --git a/platform/mellanox/mlnx-fw-upgrade.j2 b/platform/mellanox/mlnx-fw-upgrade.j2
index 6bfea7762070..ac97737f8a49 100755
--- a/platform/mellanox/mlnx-fw-upgrade.j2
+++ b/platform/mellanox/mlnx-fw-upgrade.j2
@@ -210,16 +210,20 @@ function WaitForDevice() {
     local -i QUERY_RETRY_COUNT_MAX="10"
     local -i QUERY_RETRY_COUNT="0"
     local -r DEVICE_TYPE=$(GetMstDeviceType)
+    local SPC_MST_DEV
+    local QUERY_RC=""
 
-    local SPC_MST_DEV=$(GetSPCMstDevice)
-
-    while [[ ("${QUERY_RETRY_COUNT}" -lt "${QUERY_RETRY_COUNT_MAX}") && ("${SPC_MST_DEV}" == "${UNKN_MST}") ]]; do
+    while : ; do
+        SPC_MST_DEV=$(GetSPCMstDevice)
+        ${QUERY_XML} -d ${SPC_MST_DEV} -o ${QUERY_FILE}
+        QUERY_RC="$?"
+        [[ ("${QUERY_RETRY_COUNT}" -lt "${QUERY_RETRY_COUNT_MAX}") && ("${QUERY_RC}" != "${EXIT_SUCCESS}") ]] || break
         sleep 1s
         ((QUERY_RETRY_COUNT++))
-        SPC_MST_DEV=$(GetSPCMstDevice)
+        LogInfo "Retrying MST device query ${QUERY_RETRY_COUNT}"
     done
 
-    if [[ "${SPC_MST_DEV}" == "${UNKN_MST}" ]]; then
+    if [[ "${QUERY_RC}" != "${EXIT_SUCCESS}" ]]; then
         # Couldn't Detect the Spectrum ASIC. Exit failure and print the detailed information
         output=$(${QUERY_CMD})
         failure_msg="${output#*Fail : }"
@@ -265,7 +269,7 @@ function GetSPCMstDevice() {
 
     if [[ ! -c "${_MST_DEVICE}" ]]; then
         echo "${UNKN_MST}"
-    else 
+    else
         echo "${_MST_DEVICE}"
     fi
 

From 580970ed4ceafa0297f4f00a90e5eba5580534a4 Mon Sep 17 00:00:00 2001
From: dgsudharsan <sudharsand@nvidia.com>
Date: Mon, 30 Sep 2024 16:02:59 +0000
Subject: [PATCH 2/3] Adding clear semaphore after device check

---
 files/scripts/syncd.sh               |  7 +------
 platform/mellanox/mlnx-fw-upgrade.j2 | 15 +++++++++++++++
 2 files changed, 16 insertions(+), 6 deletions(-)

diff --git a/files/scripts/syncd.sh b/files/scripts/syncd.sh
index 0930de72b87c..5b55b892a5b9 100755
--- a/files/scripts/syncd.sh
+++ b/files/scripts/syncd.sh
@@ -36,12 +36,7 @@ function startplatform() {
         debug "Starting Firmware update procedure"
         /usr/bin/mst start --with_i2cdev
 
-        local -r _MST_DEVICE="$(GetMstDevice)"
-        if [[ "${_MST_DEVICE}" != "${UNKN_MST}" ]]; then
-            /usr/bin/flint -d $_MST_DEVICE --clear_semaphore
-        fi
-
-        /usr/bin/mlnx-fw-upgrade.sh -v
+        /usr/bin/mlnx-fw-upgrade.sh -c -v
         if [[ "$?" -ne "${EXIT_SUCCESS}" ]]; then
             debug "Failed to upgrade fw. " "$?" "Restart syncd"
             exit 1
diff --git a/platform/mellanox/mlnx-fw-upgrade.j2 b/platform/mellanox/mlnx-fw-upgrade.j2
index ac97737f8a49..00796d54b096 100755
--- a/platform/mellanox/mlnx-fw-upgrade.j2
+++ b/platform/mellanox/mlnx-fw-upgrade.j2
@@ -77,6 +77,7 @@ function PrintHelp() {
     echo "  -s, --syslog   Use syslog logger (enabled when -u|--upgrade)"
     echo "  -v, --verbose  Verbose mode (enabled when -u|--upgrade)"
     echo "  -d, --dry-run  Compare the FW versions without installation. Return code "0" means the FW is up-to-date, return code "10" means an upgrade is required, otherwise an error is detected."
+    echo "  -c, --clear-semaphore   Clear hw resources before updating firmware"
     echo "  -h, --help     Print help"
     echo
     echo "Examples:"
@@ -103,6 +104,9 @@ function ParseArguments() {
             -d|--dry-run)
                 DRY_RUN="${YES_PARAM}"
             ;;
+            -c|--clear-semaphore)
+                CLEAR_SEMAPHORE="${YES_PARAM}"
+            ;;
             -h|--help)
                 PrintHelp
                 exit "${EXIT_SUCCESS}"
@@ -486,6 +490,15 @@ function Cleanup() {
     fi
 }
 
+function ClearSemaphore() {
+    if [[ "${CLEAR_SEMAPHORE}" == "${YES_PARAM}" ]]; then
+        local -r _MST_DEVICE="$(GetSPCMstDevice)"
+        if [[ "${_MST_DEVICE}" != "${UNKN_MST}" ]]; then
+            /usr/bin/flint -d $_MST_DEVICE --clear_semaphore
+        fi
+    fi
+}
+
 trap Cleanup EXIT
 
 ParseArguments "$@"
@@ -496,6 +509,8 @@ LockStateChange
 
 WaitForDevice
 
+ClearSemaphore
+
 if [ "${IMAGE_UPGRADE}" != "${YES_PARAM}" ]; then
     UpgradeFW
 else

From f9c29b1d8d69abc6b73b681be0353ab103704fbc Mon Sep 17 00:00:00 2001
From: dgsudharsan <sudharsand@nvidia.com>
Date: Mon, 30 Sep 2024 16:23:34 +0000
Subject: [PATCH 3/3] Clean up stale code in syncd.sh

---
 files/scripts/syncd.sh | 12 ------------
 1 file changed, 12 deletions(-)

diff --git a/files/scripts/syncd.sh b/files/scripts/syncd.sh
index 5b55b892a5b9..0c794ef5d9dc 100755
--- a/files/scripts/syncd.sh
+++ b/files/scripts/syncd.sh
@@ -2,18 +2,6 @@
 
 . /usr/local/bin/syncd_common.sh
 
-declare -r UNKN_MST="unknown"
-
-function GetMstDevice() {
-    local _MST_DEVICE="$(ls /dev/mst/*_pci_cr0 2>&1)"
-
-    if [[ ! -c "${_MST_DEVICE}" ]]; then
-        echo "${UNKN_MST}"
-    else
-        echo "${_MST_DEVICE}"
-    fi
-}
-
 function startplatform() {
 
     # platform specific tasks