Csm utils backport

csm-check-repair.sh

@@ -3,22 +3,26 @@
 # Copyright (c) Qualcomm Technologies, Inc. and/or its subsidiaries.
 # SPDX-License-Identifier: BSD-3-Clause-Clear
 #******************************************************************************/
+source csm-collect-conf.sh
 check-repair() {
-     local PATH="$1"
+     DEVICE_PATH="$1"
      ext4_images=(cache.img.raw persist.img.raw system.img.raw userdata.img.raw systemrw.img.raw)
      for img in ${ext4_images[@]}; do
-	 echo "checking for image : $PATH/$img"
-         /usr/sbin/e2fsck -n $PATH/$img
+	 echo "checking for image : $DEVICE_PATH/$img"
+         e2fsck -n $DEVICE_PATH/$img
          status=$?
          if [[ $status != 0 ]]; then
-             /usr/sbin/e2fsck -pf $PATH/$img
+             e2fsck -pf $DEVICE_PATH/$img
              status=$?
              if [[ $status -lt 3 ]]; then
-                     echo "$PATH/$img checked and repaired as needed..."
+                     echo "$DEVICE_PATH/$img checked and repaired as needed..."
              elif [[ $status -eq 4 || $status -eq 8 ]]; then
-                     echo "$PATH/$img could not be repaired, copying from flatimg..."
+                     echo "$DEVICE_PATH/$img could not be repaired, copying from flatimg..."
                      LASSEN_RAW_IMG_FOLDER=/lib/firmware/qcom/qdu100/flatimg/
-                     /usr/bin/rsync -av $LASSEN_RAW_IMG_FOLDER/$img $PATH/
+                     rsync -av $LASSEN_RAW_IMG_FOLDER/$img $DEVICE_PATH/
+                     if [ "$img" = "system.img.raw" ]; then
+                        restore-conf $DEVICE_PATH
+                     fi
              fi
          fi
      done

csm-collect-conf.sh

@@ -0,0 +1,61 @@
+#!/bin/bash
+#******************************************************************************
+# Copyright (c) Qualcomm Technologies, Inc. and/or its subsidiaries.
+# SPDX-License-Identifier: BSD-3-Clause-Clear
+#******************************************************************************/
+
+# device_collect_config - Receive indication from device to backup conf from systemrw
+# restore-conf - Restores the backed up conf on file corruption
+
+source /usr/bin/decimal-to-hex.sh
+backup_conf=(
+        "/etc/splane/internal/si5518_dload_image_cfg.conf"
+        "/etc/splane/internal/sync_timing_ptp2_boundary_clock_G_8275_1_multicast.conf"
+        "/etc/splane/internal/sync_timing_ptp2_slave_clock_G_8275_1_multicast.conf"
+        "/etc/splane/internal/sync_timing_ptp2_t_gm_G_8275_1_multicast.conf"
+        "/etc/splane/internal/sync_timing_ptp2_t_gm_only.conf"
+        "/etc/sync_timing_driver.conf"
+        "/etc/splane/internal/x100/splane_intern_cfg.conf"
+        "/etc/transceiver_dual_speed_cfg.conf"
+        "/etc/transceiver_fault_management_cfg.conf"
+        "/etc/transceiver_logger_cfg.conf"
+        "/etc/transceiver_perf_cfg.conf"
+        "/etc/fault_manager/public/Qualcomm_fault_configuration.conf"
+)
+
+device_collect_config() {
+    # mount system.img to view the device info
+    decimal_serial=$(cat /sys/bus/mhi/devices/mhi$1/serial_number | cut -d " " -f3)
+    serialno=$(convert_serial_to_hex "$decimal_serial")
+    DEVICE_FOLDER=/lib/firmware/qcom/qdu100/$serialno
+    mkdir -m 644 -p /tmp/tmp.$serialno
+    mount $DEVICE_FOLDER/system.img.raw /tmp/tmp.$serialno
+    DEVICE_CONF_PATH=$DEVICE_FOLDER/conf
+    if [ ! -d "$DEVICE_CONF_PATH" ];then
+        mkdir -m 644 -p $DEVICE_CONF_PATH
+    fi
+    echo "Backing up configs to backup path $DEVICE_CONF_PATH..."
+    for conf in ${backup_conf[@]}; do
+        rsync -av /tmp/tmp.$serialno$conf $DEVICE_CONF_PATH
+    done
+    umount /tmp/tmp.$serialno
+}
+
+restore-conf() {
+    DEVICE_CONF_PATH=$1/conf
+    serialno=$(basename $1)
+    if [ -d "$DEVICE_CONF_PATH" ];then
+        echo "Restoring configs from backup path $DEVICE_CONF_PATH..."
+        mkdir -m 644 -p /tmp/tmp.$serialno
+        mount $1/system.img.raw /tmp/tmp.$serialno
+        for conf_path in ${backup_conf[@]}; do
+            conf_name=$(basename $conf_path)
+            rsync -av $DEVICE_CONF_PATH/$conf_name /tmp/tmp.$serialno$conf_path
+        done
+        umount /tmp/tmp.$serialno
+    fi
+}
+
+if [[ "${BASH_SOURCE[0]}" == "${0}" ]]; then
+    device_collect_config $1
+fi

csm-collect-conf@.service

@@ -0,0 +1,9 @@
+[Unit]
+Description=Backup conf on Host
+
+[Service]
+Type=oneshot
+ExecStart=/usr/bin/csm-collect-conf.sh %i
+
+[Install]
+WantedBy=multi-user.target

csm-collect-conf@.timer

@@ -0,0 +1,9 @@
+[Unit]
+Description=Backup conf timer
+
+[Timer]
+OnCalendar=*:0/10
+Unit=csm-collect-conf@%i.service
+
+[Install]
+WantedBy=timers.target

csm-configure-ip.sh

@@ -37,14 +37,24 @@ config_interface()
 # @1:  MHI Channel no
 #
 configure_ipaddress() {
-    SWIP0_LOCAL_INTERFACE="mhi$1_IP_SW0"
+    if ! [[ "$1" =~ ^[0-9]+$ ]]; then
+        echo "Error: channel number must be a non-negative."
+        exit 2
+    fi
+
+    local channel=$1
+    local base=$(( channel * 2 ))
+    SWIP0_LOCAL_INTERFACE="mhi_swip${base}"
+    MPLANE_INTERFACE="mhi_swip$(( base + 1 ))"
+    echo "Host MHI interface name   --> $SWIP0_LOCAL_INTERFACE"
+    echo "Mplane MHI interface name --> $MPLANE_INTERFACE"
+
     SWIP0_LOCAL_ADDR="192.200.100.$1"
     SWIP0_REMOTE_ADDR="192.200.101.$1"
     echo "configure interface $SWIP0_LOCAL_INTERFACE $SWIP0_LOCAL_ADDR $SWIP0_REMOTE_ADDR"
     config_interface $SWIP0_LOCAL_INTERFACE $SWIP0_LOCAL_ADDR $SWIP0_REMOTE_ADDR
 
     # Configure mhi_swipe1 interface for QDU Mplane App - OEM OAM App
-    MPLANE_INTERFACE="mhi$1_IP_SW1"
     OAM_HOST_ADDR="192.200.102.$1"
     MPLANE_ADDR="192.200.103.$1"
     echo "configure interface $MPLANE_INTERFACE $OAM_HOST_ADDR $MPLANE_ADDR"

csm-device-health.sh

@@ -0,0 +1,67 @@
+#!/bin/bash
+#******************************************************************************
+# Copyright (c) Qualcomm Technologies, Inc. and/or its subsidiaries.
+# SPDX-License-Identifier: BSD-3-Clause-Clear
+#******************************************************************************/
+
+source decimal-to-hex.sh
+source csm-host-logger.sh
+
+BOOT_TIME_OUT=600
+INTERVAL=5            # Check every 5 seconds
+
+# Checks the boot status by polling for 
+# boot_flag file
+
+device_boot_check () {
+    LASSEN_FW_FOLDER=/lib/firmware/qcom/qdu100/
+    decimal_serial=$(cat /sys/bus/mhi/devices/mhi$1/serial_number | cut -d " " -f3)
+    serialno=$(convert_serial_to_hex "$decimal_serial")
+
+    mnt_dir=$(mktemp -d)
+    boot_status_file=$mnt_dir/post_boot/boot_flag
+    LASSEN_DEVICE_FOLDER=$LASSEN_FW_FOLDER$serialno
+    while [ "$BOOT_TIME_OUT" > 0 ]; do
+        if mount $LASSEN_DEVICE_FOLDER/userdata.img.raw $mnt_dir; then
+            if [[ -f $boot_status_file ]]; then
+                recv_msg=$(cat $boot_status_file)
+                echo "Received message from device $recv_msg"
+                boot_status=success
+                device_log_collection $1
+                rm -f $boot_status_file
+                umount $mnt_dir
+                return
+            fi
+        umount $mnt_dir
+        fi
+        sleep $INTERVAL
+        BOOT_TIME_OUT=$(( BOOT_TIME_OUT - INTERVAL ))
+    done
+
+    # Check exit status of timeout
+    echo "Boot up confirmation not received within BOOT_TIME_OUT"
+    echo "Invoke csm host debugger"
+    source csm-host-debugger.sh $1
+    boot_status="error"
+}
+# Udev rule triggers csm_nbdkit service with SAHARA channel as argument when Device detected on PCIe channel at boot
+# $1 - MHI#_CSM_CTRL
+channel="$1"
+echo "Device detected in $channel"
+# Retrieve channel no from the /dev/mhi#_CSM_CTRL string
+channelno=$(echo $channel | cut -d "_" -f1 | cut -d "i" -f2)
+echo "Channel number retrieved $channelno"
+# check for device boot
+boot_status=""
+device_boot_check $channelno
+if [ "$boot_status" == "error" ]; then
+    echo "Device boot failed"
+    exit
+fi
+
+#Start a timer to collect logs from device
+sudo systemctl start csm-host-logger@$channelno.timer
+sudo systemctl start csm-collect-conf@$channelno.timer
+
+echo "$(date) :Collect host log if X100 just crashed" >> /tmp/host_log_crash.log
+csm-host-log-crash.sh "$channelno" >> /tmp/host_log_crash.log 2>&1

csm-device-health@.service

@@ -0,0 +1,12 @@
+[Unit]
+Description=Run csm device health service
+After=csm-configure-ip@.service
+Requires=csm-configure-ip@.service
+
+[Service]
+Type=oneshot
+Group=root
+ExecStart=/usr/bin/csm-device-health.sh %I
+
+[Install]
+WantedBy=multi-user.target

csm-event-handler.sh

@@ -15,4 +15,7 @@ else
     echo "csm-event: $DEVICE: starting csm-configure-ip service for CSM_CTRL" > /dev/kmsg
     systemd-run --no-block systemctl restart csm-configure-ip@"$DEVICE".service &
     echo "csm-event: $DEVICE: csm-configure-ip service completed with status:$?" > /dev/kmsg
+    echo "csm-event: $DEVICE: starting csm-device-health service for CSM_CTRL" > /dev/kmsg
+    systemd-run --no-block systemctl restart csm-device-health@"$DEVICE".service &
+    echo "csm-event: $DEVICE: csm-device-health service completed with status:$?" > /dev/kmsg
 fi