Handle and report NMI/hardfaults

board/early_init.h

@@ -63,3 +63,19 @@ void early_initialization(void) {
     jump_to_bootloader();
   }
 }
+
+// we might hit NMI / HardFault close to brown-out.
+// we don't want to infinitely loop, so let's set a logging cookie and reset.
+#define NMI_COOKIE 0x4E4D4921U
+#define HARDFAULT_COOKIE 0x48464C54U
+extern uint32_t fault_cookie;
+
+void NMI_Handler(void) {
+  fault_cookie = NMI_COOKIE;
+  NVIC_SystemReset();
+}
+
+void HardFault_Handler(void) {
+  fault_cookie = HARDFAULT_COOKIE;
+  NVIC_SystemReset();
+}

board/health.h

@@ -7,6 +7,8 @@
 #define HEALTH_FLAG_HEARTBEAT_LOST           (1U << 4)
 #define HEALTH_FLAG_SAFETY_RX_CHECKS_INVALID (1U << 5)
 #define HEALTH_FLAG_SOM_RESET_TRIGGERED      (1U << 6)
+#define HEALTH_FLAG_NMI_RESET                (1U << 7)
+#define HEALTH_FLAG_HARDFAULT_RESET          ((uint16_t)1U << 8)
 
 struct __attribute__((packed)) health_t {
   uint32_t uptime_pkt;

board/main.c

@@ -316,6 +316,14 @@ int main(void) {
   led_set(LED_GREEN, false);
   led_set(LED_BLUE, false);
 
+  if (fault_cookie == NMI_COOKIE) {
+    reset_flags |= HEALTH_FLAG_NMI_RESET;
+  }
+  if (fault_cookie == HARDFAULT_COOKIE) {
+    reset_flags |= HEALTH_FLAG_HARDFAULT_RESET;
+  }
+  fault_cookie = 0U;
+
   print("**** INTERRUPTS ON ****\n");
   enable_interrupts();
 

board/main_comms.h

@@ -22,6 +22,7 @@ static int get_health_pkt(void *dat) {
   health->flags_pkt |= heartbeat_lost ? HEALTH_FLAG_HEARTBEAT_LOST : 0U;
   health->flags_pkt |= safety_rx_checks_invalid ? HEALTH_FLAG_SAFETY_RX_CHECKS_INVALID : 0U;
   health->flags_pkt |= bootkick_reset_triggered ? HEALTH_FLAG_SOM_RESET_TRIGGERED : 0U;
+  health->flags_pkt |= reset_flags;
 
   health->safety_tx_blocked_pkt = safety_tx_blocked;
   health->safety_rx_invalid_pkt = safety_rx_invalid;

board/main_declarations.h

@@ -12,6 +12,7 @@ void pwm_set(TIM_TypeDef *TIM, uint8_t channel, uint8_t percentage);
 extern uint8_t hw_type;
 extern board *current_board;
 extern uint32_t uptime_cnt;
+extern uint16_t reset_flags;
 
 // heartbeat state
 extern uint32_t heartbeat_counter;

board/main_definitions.h

@@ -6,6 +6,7 @@
 uint8_t hw_type = 0;
 board *current_board;
 uint32_t uptime_cnt = 0;
+uint16_t reset_flags = 0U;
 
 // heartbeat state
 uint32_t heartbeat_counter = 0;

board/stm32h7/stm32h7x5_flash.ld

@@ -54,6 +54,8 @@ ENTRY(Reset_Handler)
 
 /* Highest address of the user mode stack */
 enter_bootloader_mode = 0x38001FFC;
+fault_cookie = 0x38003FFC;
+
 _estack = 0x20020000;    /* end of RAM */
 _app_start = 0x08020000; /* Reserve Sector 0(128K) for bootloader */
 

python/__init__.py

@@ -146,6 +146,8 @@ class Panda:
   HEALTH_FLAG_HEARTBEAT_LOST = 1 << 4
   HEALTH_FLAG_SAFETY_RX_CHECKS_INVALID = 1 << 5
   HEALTH_FLAG_SOM_RESET_TRIGGERED = 1 << 6
+  HEALTH_FLAG_NMI_RESET = 1 << 7
+  HEALTH_FLAG_HARDFAULT_RESET = 1 << 8
 
   H7_DEVICES = [HW_TYPE_RED_PANDA, HW_TYPE_TRES, HW_TYPE_CUATRO, HW_TYPE_BODY]
   SUPPORTED_DEVICES = H7_DEVICES
@@ -549,6 +551,8 @@ def health(self):
       "sbu1_voltage_mV": a[17],
       "sbu2_voltage_mV": a[18],
       "som_reset_triggered": bool(flags & self.HEALTH_FLAG_SOM_RESET_TRIGGERED),
+      "nmi_reset": bool(flags & self.HEALTH_FLAG_NMI_RESET),
+      "hardfault_reset": bool(flags & self.HEALTH_FLAG_HARDFAULT_RESET),
       "sound_output_level": a[19],
       "temperature": a[20] - 40.0,
     }