Merge pull request #110 from iranzo/mellanoxcheckfix

fix: checks on mellanox failed on some hosts, switch to ethtool

Author: dcritch

checks/mellanox-firmware-version

@@ -20,15 +20,15 @@ if oc auth can-i debug node >/dev/null 2>&1; then
   for node in $(oc get nodes -o go-template='{{range .items}}{{$node := .}}{{range .status.conditions}}{{if eq .type "Ready"}}{{if eq .status "True"}}node/{{$node.metadata.name}}{{"\n"}}{{end}}{{end}}{{end}}{{end}}'); do
     # shellcheck disable=SC1083
     ocdebugorwait # Pause for no OC debug running
-    if ! FIRMWAREVERS=$(oc debug --image="${OCDEBUGIMAGE}" "${node}" -- chroot /host sh -c "for id in ${IDS}; do for device in \$(lspci -D -d "\${id}" | awk '{ print \$1 }'); do echo -n \"\${device},\"; grep -aoP '(?<=FFV)[0-9,.]{8}' /sys/bus/pci/devices/\${device}/vpd;done;done" 2>/dev/null); then
+    if ! FIRMWAREVERS=$(oc debug --image="${OCDEBUGIMAGE}" "${node}" -- chroot /host sh -c "for id in ${IDS}; do for device in \$(lspci -D -d "\${id}" | awk '{ print \$1 }'); do echo -n \"\${device},\" ; ethtool -i \$(ls /sys/bus/pci/devices/\${device}/net/)|grep firmware-version|cut -d: -f2-|xargs echo|awk '{ print \$1 }';done;done" 2>/dev/null); then
       msg "${ORANGE}Error running oc debug in ${node}${NOCOLOR}"
     else
       if [ -n "${FIRMWAREVERS}" ]; then
         for result in ${FIRMWAREVERS}; do
           dev=$(echo ${result} | awk -F, '{print $1}')
           fw=$(echo ${result} | awk -F, '{print $2}' | awk -F. '{print $1"."$2}')
           if [[ $(expr ${fw} \< ${MIN_VERS}) -eq 1 ]]; then
-            msg "Firmware for Mellanox card ${RED}${dev}${NOCOLOR} on ${RED}${node}${NOCOLOR} is below the minimum recommended version. Please upgrade to at least ${GREEN}${MIN_VERS}${NOCOLOR}."
+            msg "Firmware for Mellanox card ${RED}${dev}${NOCOLOR} (${fw}) on ${RED}${node}${NOCOLOR} is below the minimum recommended version. Please upgrade to at least ${GREEN}${MIN_VERS}${NOCOLOR}."
             errors=$(("${errors}" + 1))
             fw_errors=$(("${fw_errors}" + 1))
             if [ ! -z "${ERRORFILE}" ]; then
@@ -49,4 +49,4 @@ else
   msg "Couldn't debug nodes, check permissions"
   exit ${OCSKIP}
 fi
-exit ${OCUNKNOWNN}
+exit ${OCUNKNOWN}