diff options
| author | Raptor Engineering Development Team <support@raptorengineering.com> | 2018-01-22 00:51:36 -0600 |
|---|---|---|
| committer | Raptor Engineering Development Team <support@raptorengineering.com> | 2018-01-22 01:01:51 -0600 |
| commit | f0cfc21adc3b9ee7c4510df528b4a05dc77e3045 (patch) | |
| tree | 271c61062efb9f859139228991940249cc31119d | |
| parent | 5773eedf5b68897d3fb155d77de0fc9c8f168cb4 (diff) | |
| download | talos-openbmc-f0cfc21adc3b9ee7c4510df528b4a05dc77e3045.tar.gz talos-openbmc-f0cfc21adc3b9ee7c4510df528b4a05dc77e3045.zip | |
Recover from loss of either OCC
This fixes a corner case where the second OCC could drop off the bus undetected
| -rwxr-xr-x | meta-openbmc-machines/meta-openpower/common/recipes-phosphor/occ/openpower-occ-control/occ-active.sh | 42 |
1 files changed, 39 insertions, 3 deletions
diff --git a/meta-openbmc-machines/meta-openpower/common/recipes-phosphor/occ/openpower-occ-control/occ-active.sh b/meta-openbmc-machines/meta-openpower/common/recipes-phosphor/occ/openpower-occ-control/occ-active.sh index 9e4c4bec2..6a511c132 100755 --- a/meta-openbmc-machines/meta-openpower/common/recipes-phosphor/occ/openpower-occ-control/occ-active.sh +++ b/meta-openbmc-machines/meta-openpower/common/recipes-phosphor/occ/openpower-occ-control/occ-active.sh @@ -14,25 +14,59 @@ else exit -1 fi +# Get CPU count +CPU_COUNT=1 +STATUS_FLAGS=$(i2cget -y 12 0x31 0x7) +if [ $? != 0 ]; then + STATUS_FLAGS=$(i2cget -y 12 0x31 0x7) +fi +if [ $? != 0 ]; then + STATUS_FLAGS=$(i2cget -y 12 0x31 0x7) +fi +CPU_PRESENT_FLAG_N=$(( ${STATUS_FLAGS} & 0x20 )) +if [ $CPU_PRESENT_FLAG_N != 0 ]; then + CPU_COUNT=$(( ${CPU_COUNT} + 1 )) +fi +echo "Found $CPU_COUNT CPU(s)" + OCC_CONTROL_SERVICE="org.open_power.OCC.Control" if [ "$value" == "true" ]; then + # Disable existing OCC service + busctl tree $OCC_CONTROL_SERVICE --list | grep occ | xargs -r -n1 -I{} \ + busctl set-property $OCC_CONTROL_SERVICE {} org.open_power.OCC.Status \ + OccActive b 'false' + # Rescan bus to find OCCs openpower-proc-control scanFSI - # Wait 5 seconds before restarting fan controller - sleep 5 - systemctl restart phosphor-fan-control@0.service fi busctl tree $OCC_CONTROL_SERVICE --list | grep occ | xargs -r -n1 -I{} \ busctl set-property $OCC_CONTROL_SERVICE {} org.open_power.OCC.Status \ OccActive b $value +if [ "$value" == "true" ]; then + # Wait 5 seconds before restarting fan controller + sleep 5 + systemctl restart phosphor-fan-control@0.service +fi + if [ "$1" == "watchdog" ]; then # Start watchdog while [ 1 == 1 ]; do sleep 5 + + OCC_FAULT=0 if [ ! -e /sys/bus/platform/drivers/occ/sbefifo1-dev0/occ1-dev0/occ_status ]; then + OCC_FAULT=1 + fi + if [ $CPU_COUNT -gt 1 ]; then + if [ ! -e /sys/bus/platform/drivers/occ/sbefifo2-dev0/occ2-dev0/occ_status ]; then + OCC_FAULT=1 + fi + fi + + if [ $OCC_FAULT != 0 ]; then # OCC driver failed. Rescan bus to recover. # Disable existing OCC service busctl tree $OCC_CONTROL_SERVICE --list | grep occ | xargs -r -n1 -I{} \ @@ -45,6 +79,8 @@ if [ "$1" == "watchdog" ]; then busctl tree $OCC_CONTROL_SERVICE --list | grep occ | xargs -r -n1 -I{} \ busctl set-property $OCC_CONTROL_SERVICE {} org.open_power.OCC.Status \ OccActive b 'true' + # Give the OCC service time to start + sleep 5 # Restart fan controller systemctl restart phosphor-fan-control@0.service fi |
