Thang Q. Nguyen | 98de8b9 | 2023-05-05 15:33:27 +0700 | [diff] [blame] | 1 | #!/bin/bash |
| 2 | |
| 3 | # This script monitors S0/S1 fault GPIO and detects errors or warnings from CPUs |
| 4 | # |
| 5 | # According to OpenBMC_Software_Funcional_Specification, section 3.16, |
| 6 | # |
| 7 | # When the BMC detects the GPIO_FAULT signal indicating an SCP booting failure: |
| 8 | # • If a non-critical error/warning from the SCP occurs, the BMC blinks the Fault LED once. |
| 9 | # • If a critical error from the SCP occurs, the BMC turns on the Fault LED. |
| 10 | # The BMC monitors the GPIO_FAULT signal from the SCP during SCP booting to determine whether |
| 11 | # the error is non-critical or critical. A fatal error is indicated when the signal is On and then Off |
| 12 | # continuously, followed by a “quiet” period of about three seconds, and this pattern repeats. If the “quiet” |
| 13 | # period is longer than three seconds, the error is non-fatal. The BMC must set up appropriate debounce |
| 14 | # times to detect such errors. The BMC is expected to turn on the Fault LED forever for fatal errors, or to |
| 15 | # turn on the Fault LED and turn it off when the fault clears for non-fatal errors. |
| 16 | # |
| 17 | # Usage: <app_name> <socket 0/1> |
| 18 | |
| 19 | # shellcheck source=/dev/null |
| 20 | source /usr/sbin/gpio-lib.sh |
| 21 | |
| 22 | # global variables |
| 23 | error_flag='/tmp/fault_err' |
| 24 | warning_flag='/tmp/fault_warning' |
| 25 | |
| 26 | duty_cycle=250000 |
| 27 | scan_pulse=100000 |
| 28 | blank_num=8 |
| 29 | |
| 30 | curr_pattern=0 |
| 31 | prev_pattern=0 |
| 32 | |
| 33 | gpio_status=0 |
| 34 | repeat=0 |
| 35 | |
| 36 | socket=$1 |
| 37 | |
| 38 | socket1_present=15 |
| 39 | socket1_status=1 |
| 40 | |
| 41 | S0_fault_gpio=73 |
| 42 | S1_fault_gpio=201 |
| 43 | |
| 44 | map_event_name() { |
| 45 | case $curr_pattern in |
| 46 | 1) |
| 47 | event_name="RAS_GPIO_INVALID_LCS" |
| 48 | ;; |
| 49 | 2) |
| 50 | event_name="RAS_GPIO_FILE_HDR_INVALID" |
| 51 | ;; |
| 52 | 3) |
| 53 | event_name="RAS_GPIO_FILE_INTEGRITY_INVALID" |
| 54 | ;; |
| 55 | 4) |
| 56 | event_name="RAS_GPIO_KEY_CERT_AUTH_ERR" |
| 57 | ;; |
| 58 | 5) |
| 59 | event_name="RAS_GPIO_CNT_CERT_AUTH_ERR" |
| 60 | ;; |
| 61 | 6) |
| 62 | event_name="RAS_GPIO_I2C_HARDWARE_ERR" |
| 63 | ;; |
| 64 | 7) |
| 65 | event_name="RAS_GPIO_CRYPTO_ENGINE_ERR" |
| 66 | ;; |
| 67 | 8) |
| 68 | event_name="RAS_GPIO_ROTPK_EFUSE_INVALID" |
| 69 | ;; |
| 70 | 9) |
| 71 | event_name="RAS_GPIO_SEED_EFUSE_INVALID" |
| 72 | ;; |
| 73 | 10) |
| 74 | event_name="RAS_GPIO_LCS_FROM_EFUSE_INVALID" |
| 75 | ;; |
| 76 | 11) |
| 77 | event_name="RAS_GPIO_PRIM_ROLLBACK_EFUSE_INVALID" |
| 78 | ;; |
| 79 | 12) |
| 80 | event_name="RAS_GPIO_SEC_ROLLBACK_EFUSE_INVALID" |
| 81 | ;; |
| 82 | 13) |
| 83 | event_name="RAS_GPIO_HUK_EFUSE_INVALID" |
| 84 | ;; |
| 85 | 14) |
| 86 | event_name="RAS_GPIO_CERT_DATA_INVALID" |
| 87 | ;; |
| 88 | 15) |
| 89 | event_name="RAS_GPIO_INTERNAL_HW_ERR" |
| 90 | ;; |
| 91 | *) |
| 92 | event_name="NOT_SUPPORT" |
| 93 | ;; |
| 94 | esac |
| 95 | } |
| 96 | |
| 97 | detect_patern_repeat() { |
| 98 | local prev=0 |
| 99 | local curr=0 |
| 100 | local cnt=13 |
| 101 | |
| 102 | while true |
| 103 | do |
| 104 | usleep $scan_pulse |
| 105 | gpio_status=$(cat /sys/class/gpio/gpio"$gpio_Id"/value) |
| 106 | prev=$curr |
| 107 | curr=$gpio_status |
| 108 | if [ "$prev" == 0 ] && [ "$curr" == 1 ]; then |
| 109 | # patern start repeating, check if previous and current pattern are the same |
| 110 | repeat=1 |
| 111 | break |
| 112 | fi |
| 113 | if [ "$cnt" == 0 ]; then |
| 114 | map_event_name |
| 115 | echo "detected a warning from fault GPIO #$fault_gpio $socket, event $event_name" |
| 116 | # pattern not repeat, this is a warning, turn on warning flag |
| 117 | touch $warning_flag |
| 118 | break |
| 119 | fi |
| 120 | cnt=$(( cnt - 1 )) |
| 121 | done |
| 122 | } |
| 123 | |
| 124 | detect_pattern() { |
| 125 | local cnt_falling_edge=0 |
| 126 | local cnt_blank=0 |
| 127 | |
| 128 | local prev=0 |
| 129 | local curr=0 |
| 130 | |
| 131 | while true |
| 132 | do |
| 133 | prev=$curr |
| 134 | curr=$gpio_status |
| 135 | # count the falling edges, if they appear, just reset cnt_blank |
| 136 | if [ "$prev" == 1 ] && [ "$curr" == 0 ]; then |
| 137 | cnt_falling_edge=$(( cnt_falling_edge + 1 )) |
| 138 | cnt_blank=0 |
| 139 | continue |
| 140 | # check if we are in the quite gap |
| 141 | elif [ "$prev" == 0 ] && [ "$curr" == 0 ]; then |
| 142 | cnt_blank=$(( cnt_blank + 1 )) |
| 143 | if [ "$cnt_blank" == "$blank_num" ]; then |
| 144 | # echo "pattern number falling_edge=$cnt_falling_edge blank=$cnt_blank" |
| 145 | curr_pattern=$cnt_falling_edge |
| 146 | # after count all falling edges, now check if patern repeat after 3s |
| 147 | detect_patern_repeat |
| 148 | break |
| 149 | fi |
| 150 | fi |
| 151 | usleep $scan_pulse |
| 152 | gpio_status=$(cat /sys/class/gpio/gpio"$gpio_Id"/value) |
| 153 | done |
| 154 | } |
| 155 | |
| 156 | gpio_config_input() { |
| 157 | echo "$gpio_Id" > /sys/class/gpio/export |
| 158 | echo "in" > /sys/class/gpio/gpio"${gpio_Id}"/direction |
| 159 | } |
| 160 | |
| 161 | init_sysfs_fault_gpio() { |
| 162 | gpio_Id=$(gpio_number "$fault_gpio") |
| 163 | if [ -d /sys/class/gpio/gpio"$gpio_Id" ]; then |
| 164 | return |
| 165 | fi |
| 166 | gpio_config_input "$fault_gpio" |
| 167 | } |
| 168 | |
| 169 | # init |
| 170 | if [ "$socket" == "0" ]; then |
| 171 | fault_gpio=$S0_fault_gpio |
| 172 | else |
| 173 | socket1_status=$(gpioget 0 "$socket1_present") |
| 174 | if [ "$socket1_status" == 1 ]; then |
| 175 | echo "socket 1 not present" |
| 176 | exit 1 |
| 177 | fi |
| 178 | fault_gpio=$S1_fault_gpio |
| 179 | fi |
| 180 | |
| 181 | init_sysfs_fault_gpio |
| 182 | |
| 183 | # daemon start |
| 184 | while true |
| 185 | do |
| 186 | # detect when pattern starts |
| 187 | if [ "$gpio_status" == 1 ]; then |
| 188 | # now, there is something on gpio, check if that is a pattern |
| 189 | detect_pattern |
| 190 | if [ "$repeat" == 1 ] && [ "$prev_pattern" == "$curr_pattern" ]; then |
| 191 | map_event_name |
| 192 | echo "detected an error from fault GPIO #$fault_gpio $socket, event#$curr_pattern $event_name" |
| 193 | touch $error_flag |
| 194 | repeat=0 |
| 195 | fi |
| 196 | prev_pattern=$curr_pattern |
| 197 | curr_pattern=0 |
| 198 | continue |
| 199 | fi |
| 200 | usleep $duty_cycle |
| 201 | gpio_status=$(cat /sys/class/gpio/gpio"$gpio_Id"/value) |
| 202 | |
| 203 | done |
| 204 | |
| 205 | exit 1 |