Move customized OCC watchdog to RCS overlay

2 files changed, 153 insertions, 0 deletions

diff --git a/meta-rcs/meta-openpower/recipes-phosphor/occ/openpower-occ-control/occ-active.sh b/meta-rcs/meta-openpower/recipes-phosphor/occ/openpower-occ-control/occ-active.sh
new file mode 100755
index 000000000..fa3a3ea3c
--- /dev/null
+++ b/meta-rcs/meta-openpower/recipes-phosphor/occ/openpower-occ-control/occ-active.sh
@@ -0,0 +1,152 @@
+#!/bin/sh
+# set and unset occ active for all occ's in system
+
+if [ "$1" == "disable" ]; then
+    value='false'
+elif [ "$1" == "enable" ]; then
+    value='true'
+elif [ "$1" == "watchdog" ]; then
+    value='true'
+else
+    echo "Usage: occ-active.sh [argument]"
+    echo "  enable  - set occ's to active state"
+    echo "  disable - set occ's to inactive state"
+    exit -1
+fi
+
+if [ "$value" == "true" ]; then
+	# Verify host IPL is complete
+	while [ ! -e /run/openbmc/host@0-ipl-complete ]; do
+		echo "Waiting for host IPL to complete before engaging OCC interface..."
+		sleep 1
+	done
+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
+
+	# Bind OCCs
+	echo occ-hwmon.1 > /sys/bus/platform/drivers/occ-hwmon/bind
+	if [ $CPU_COUNT -gt 1 ]; then
+		echo occ-hwmon.2 > /sys/bus/platform/drivers/occ-hwmon/bind
+	fi
+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
+else
+        # Unbind all OCC drivers to avoid bus contention on next IPL
+	echo occ-hwmon.1 > /sys/bus/platform/drivers/occ-hwmon/unbind || true
+	echo occ-hwmon.2 > /sys/bus/platform/drivers/occ-hwmon/unbind || true
+	echo sbefifo1-dev0 > /sys/bus/platform/drivers/occ/unbind || true
+	echo sbefifo2-dev0 > /sys/bus/platform/drivers/occ/unbind || true
+fi
+
+if [ "$1" == "watchdog" ]; then
+	# Start watchdog
+	while [ 1 == 1 ]; do
+		sleep 5
+
+		if [ ! -e /run/openbmc/host@0-ipl-complete ]; then
+			echo "Exiting due to missing IPL complete flag file"
+			exit 0
+		fi
+
+		OCC_FAULT=0
+		if [ ! -e /sys/bus/platform/drivers/occ/sbefifo1-dev0/occ-hwmon.1/occ_error ]; then
+			OCC_FAULT=1
+		else
+			STATUS=$(cat /sys/bus/platform/drivers/occ/sbefifo1-dev0/occ-hwmon.1/occ_error)
+			if [ $STATUS != 0 ]; then
+				OCC_FAULT=1
+			fi
+		fi
+		if [ $CPU_COUNT -gt 1 ]; then
+			if [ ! -e /sys/bus/platform/drivers/occ/sbefifo2-dev0/occ-hwmon.2/occ_error ]; then
+				OCC_FAULT=1
+			else
+				STATUS=$(cat /sys/bus/platform/drivers/occ/sbefifo2-dev0/occ-hwmon.2/occ_error)
+				if [ $STATUS != 0 ]; then
+					OCC_FAULT=1
+				fi
+			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{} \
+				busctl set-property $OCC_CONTROL_SERVICE {} org.open_power.OCC.Status \
+				OccActive b 'false'
+
+			# Give enough time for the OCC control service to attempt unload before forcible unbind
+			sleep 5
+
+			# Unbind all OCC drivers
+			echo occ-hwmon.1 > /sys/bus/platform/drivers/occ-hwmon/unbind || true
+			echo occ-hwmon.2 > /sys/bus/platform/drivers/occ-hwmon/unbind || true
+			echo sbefifo1-dev0 > /sys/bus/platform/drivers/occ/unbind || true
+			echo sbefifo2-dev0 > /sys/bus/platform/drivers/occ/unbind || true
+
+			# Wait 10 seconds for OCC to figure out something has happened / bus contention to stop
+			sleep 10
+
+			# Make sure system hasn't gone down in the interim
+			if [ ! -e /run/openbmc/host@0-ipl-complete ]; then
+				echo "Exiting due to missing IPL complete flag file"
+				exit 0
+			fi
+
+			# Rescan bus to find OCCs
+			openpower-proc-control scanFSI
+
+			# Bind OCCs
+			echo occ-hwmon.1 > /sys/bus/platform/drivers/occ-hwmon/bind
+			if [ $CPU_COUNT -gt 1 ]; then
+				echo occ-hwmon.2 > /sys/bus/platform/drivers/occ-hwmon/bind
+			fi
+
+			# Re-enable OCC service
+			busctl tree $OCC_CONTROL_SERVICE --list | grep occ | xargs -r -n1 -I{} \
+				busctl set-property $OCC_CONTROL_SERVICE {} org.open_power.OCC.Status \
+				OccActive b 'true'
+
+			# Give the OCC service time to start
+			sleep 5
+
+			# Restart fan controller
+			systemctl restart phosphor-fan-control@0.service
+		fi
+	done
+fi
+
+exit 0
diff --git a/meta-rcs/meta-openpower/recipes-phosphor/occ/openpower-occ-control_git.bbappend b/meta-rcs/meta-openpower/recipes-phosphor/occ/openpower-occ-control_git.bbappend
new file mode 100644
index 000000000..fbeb943ba
--- /dev/null
+++ b/meta-rcs/meta-openpower/recipes-phosphor/occ/openpower-occ-control_git.bbappend
@@ -0,0 +1 @@
+FILESEXTRAPATHS_prepend := "${THISDIR}/openpower-occ-control:"