meta-ampere: mtmitchell: Support always retrieve BERT when power cycl…

…e/reset/off

When BERT crash happens, depend on the Host UEFI firmware configuration,
the Host might not be rebooted immediately so no REBOOT_ACK from Host
for the BMC to check BERT crash. This might cause BERT crash is not
handled.
This commit changes to check BERT every time the Host is powered cycle,
reset or off to handle above case.

Signed-off-by: Dung Cao <[email protected]>

meta-ampere/meta-mitchell/recipes-ampere/host/ampere-hostctrl.bbappend

@@ -3,12 +3,24 @@ FILESEXTRAPATHS:append := "${THISDIR}/${PN}:"
 SRC_URI += " \
                 file://ampere_power_on_failure_check.sh \
                 file://ampere-host-on-host-check-override.conf \
+                file://ampere-bert-power-handle.service \
+                file://ampere-bert-power-handle.sh \
            "
 
-FILES:${PN} += "${systemd_system_unitdir}/[email protected]"
+FILES:${PN} += " \
+                 ${systemd_system_unitdir}/[email protected] \
+                 ${systemd_system_unitdir}/ampere-bert-power-handle.service \
+               "
+AMPERE_BERT_TMPL = "ampere-bert-power-handle.service"
+AMPERE_BERT_INSTMPL = "ampere-bert-power-handle.service"
+AMPERE_BERT_TGTFMT = "obmc-host-stop@{0}.target"
+AMPERE_BERT_FMT = "../${AMPERE_BERT_TMPL}:${AMPERE_BERT_TGTFMT}.requires/${AMPERE_BERT_INSTMPL}"
+SYSTEMD_LINK:${PN} += "${@compose_list_zip(d, 'AMPERE_BERT_FMT', 'OBMC_HOST_INSTANCES')}"
+SYSTEMD_SERVICE:${PN} += "${AMPERE_BERT_TMPL}"
 
 do_install:append() {
      install -m 0755 ${WORKDIR}/ampere_power_on_failure_check.sh ${D}/${sbindir}/
+     install -m 0755 ${WORKDIR}/ampere-bert-power-handle.sh ${D}/${sbindir}/
 
      install -d ${D}${systemd_system_unitdir}/ampere-host-on-host-check@0.service.d
      install -m 644 ${WORKDIR}/ampere-host-on-host-check-override.conf \

meta-ampere/meta-mitchell/recipes-ampere/host/ampere-hostctrl/ampere-bert-power-handle.service

@@ -0,0 +1,10 @@
+[Unit]
+Description=Ampere BERT handler for power action of the host
+After[email protected]
+Before[email protected]
+
+[Service]
+Restart=no
+ExecStart=/usr/sbin/ampere-bert-power-handle.sh
+SyslogIdentifier=ampere-bert-power-handle
+Type=oneshot

meta-ampere/meta-mitchell/recipes-ampere/host/ampere-hostctrl/ampere-bert-power-handle.sh

@@ -0,0 +1,43 @@
+#!/bin/bash
+
+# shellcheck disable=SC2046
+# shellcheck source=meta-ampere/meta-common/recipes-ampere/platform/ampere-utils/ampere_power_control_lock.sh
+source /usr/sbin/ampere_power_control_lock.sh
+
+wait_bert_complete() {
+	# Wait maximum 60 seconds for BERT completed
+	cnt=20
+	while [ $cnt -gt 0 ]
+	do
+		bert_done=$(busctl get-property com.ampere.CrashCapture.Trigger /com/ampere/crashcapture/trigger com.ampere.CrashCapture.Trigger TriggerActions | cut -d"." -f6)
+		if ! [ "$bert_done" == "Done\"" ]; then
+			sleep 3
+			cnt=$((cnt - 1))
+		else
+			break
+		fi
+	done
+	if [ "$cnt" -eq "0" ]; then
+		echo "Timeout 60 seconds, BERT is still not completed"
+		return 1
+	fi
+	return 0
+}
+
+
+echo "Notify Crash Capture to read BERT."
+busctl set-property com.ampere.CrashCapture.Trigger \
+       /com/ampere/crashcapture/trigger \
+       com.ampere.CrashCapture.Trigger \
+       TriggerProcess b true
+bert_timeout="0"
+# Wait until RAS BERT process completed
+wait_bert_complete
+bert_timeout=$?
+# If the crash capture process is crash or works unstable, it does
+# not unmask the power action. We should call unmask here to make sure
+# the power control is unmasked
+if [[ "${bert_timeout}" == "1" ]]; then
+    unmask_reboot_targets
+    unmask_off_targets
+fi

meta-ampere/meta-mitchell/recipes-ampere/platform/ampere-utils/ampere_power_util.sh

@@ -1,8 +1,6 @@
 #!/bin/bash
 
 # shellcheck disable=SC2046
-# shellcheck source=meta-ampere/meta-common/recipes-ampere/platform/ampere-utils/ampere_power_control_lock.sh
-source /usr/sbin/ampere_power_control_lock.sh
 
 # Usage of this utility
 function usage() {
@@ -85,44 +83,7 @@ force_reset() {
 	gpioset $(gpiofind host0-sysreset-n)=1
 }
 
-wait_bert_complete() {
-	# Wait maximum 60 seconds for BERT completed
-	cnt=20
-	while [ $cnt -gt 0 ]
-	do
-		bert_done=$(busctl get-property com.ampere.CrashCapture.Trigger /com/ampere/crashcapture/trigger com.ampere.CrashCapture.Trigger TriggerActions | cut -d"." -f6)
-		if ! [ "$bert_done" == "Done\"" ]; then
-			sleep 3
-			cnt=$((cnt - 1))
-		else
-			break
-		fi
-	done
-	if [ "$cnt" -eq "0" ]; then
-		echo "Timeout 60 seconds, BERT is still not completed"
-		return 1
-	fi
-	return 0
-}
-
 host_reboot_wa() {
-    echo "Notify Crash Capture reboot action from host."
-    busctl set-property com.ampere.CrashCapture.Trigger \
-       /com/ampere/crashcapture/trigger \
-       com.ampere.CrashCapture.Trigger \
-       TriggerProcess b true
-    bert_timeout="0"
-    # Wait until RAS BERT process completed
-    wait_bert_complete
-    bert_timeout=$?
-    # If the crash capture process is crash or works unstable, it does
-    # not unmask the power action. We should call unmask here to make sure
-    # the power control is unmasked
-    if [[ "${bert_timeout}" == "1" ]]; then
-        unmask_reboot_targets
-        unmask_off_targets
-    fi
-
     busctl set-property xyz.openbmc_project.State.Chassis \
         /xyz/openbmc_project/state/chassis0 xyz.openbmc_project.State.Chassis \
         RequestedPowerTransition s "xyz.openbmc_project.State.Chassis.Transition.Off"
@@ -136,12 +97,10 @@ host_reboot_wa() {
     done
     echo "The power is already Off."
 
-    # Keep the system off if BERT is timeout
-    if [[ "${bert_timeout}" == "0" ]]; then
-        busctl set-property xyz.openbmc_project.State.Host \
-            /xyz/openbmc_project/state/host0 xyz.openbmc_project.State.Host \
-            RequestedHostTransition s "xyz.openbmc_project.State.Host.Transition.On"
-    fi
+    busctl set-property xyz.openbmc_project.State.Host \
+        /xyz/openbmc_project/state/host0 xyz.openbmc_project.State.Host \
+        RequestedHostTransition s "xyz.openbmc_project.State.Host.Transition.On"
+
 }
 
 if [ ! -d "/run/openbmc/" ]; then