diff --git a/esphome/components/esp32/__init__.py b/esphome/components/esp32/__init__.py index 7e13156b4d..9f3d7f1dc9 100644 --- a/esphome/components/esp32/__init__.py +++ b/esphome/components/esp32/__init__.py @@ -3108,9 +3108,10 @@ def _parse_register(config, regex, line): STACKTRACE_ESP32_PC_RE = re.compile(r".*PC\s*:\s*(?:0x)?(4[0-9a-fA-F]{7}).*") -STACKTRACE_ESP32_EXCVADDR_RE = re.compile(r"EXCVADDR\s*:\s*(?:0x)?(4[0-9a-fA-F]{7})") +STACKTRACE_ESP32_EXCVADDR_RE = re.compile(r".*EXCVADDR\s*:\s*(?:0x)?(4[0-9a-fA-F]{7})") STACKTRACE_ESP32_C3_PC_RE = re.compile(r"MEPC\s*:\s*(?:0x)?(4[0-9a-fA-F]{7})") STACKTRACE_ESP32_C3_RA_RE = re.compile(r"RA\s*:\s*(?:0x)?(4[0-9a-fA-F]{7})") +STACKTRACE_ESP32_C3_MTVAL_RE = re.compile(r".*MTVAL\s*:\s*(?:0x)?(4[0-9a-fA-F]{7})") STACKTRACE_BAD_ALLOC_RE = re.compile( r"^last failed alloc call: (4[0-9a-fA-F]{7})\((\d+)\)$" ) @@ -3128,9 +3129,10 @@ def process_stacktrace(config, line, backtrace_state): # ESP32 PC/EXCVADDR _parse_register(config, STACKTRACE_ESP32_PC_RE, line) _parse_register(config, STACKTRACE_ESP32_EXCVADDR_RE, line) - # ESP32-C3 PC/RA + # ESP32-C3 PC/RA/MTVAL _parse_register(config, STACKTRACE_ESP32_C3_PC_RE, line) _parse_register(config, STACKTRACE_ESP32_C3_RA_RE, line) + _parse_register(config, STACKTRACE_ESP32_C3_MTVAL_RE, line) # bad alloc match = re.match(STACKTRACE_BAD_ALLOC_RE, line) diff --git a/esphome/components/esp32/crash_handler.cpp b/esphome/components/esp32/crash_handler.cpp index a7de48a6ee..4c0f430daf 100644 --- a/esphome/components/esp32/crash_handler.cpp +++ b/esphome/components/esp32/crash_handler.cpp @@ -122,7 +122,7 @@ static uint8_t IRAM_ATTR capture_riscv_backtrace(RvExcFrame *frame, uint32_t *ou // Magic is second to validate the data. Remaining fields can change between versions. // Version is uint32_t because it would be padded to 4 bytes anyway before the next // uint32_t field, so we use the full width rather than wasting 3 bytes of padding. -static constexpr uint32_t CRASH_DATA_VERSION = 2; +static constexpr uint32_t CRASH_DATA_VERSION = 3; struct RawCrashData { uint32_t version; uint32_t magic; @@ -132,7 +132,8 @@ struct RawCrashData { uint8_t exception; // panic_exception_t enum (FAULT/ABORT/IWDT/TWDT/DEBUG) uint8_t pseudo_excause; // Whether cause is a pseudo exception (Xtensa SoC-level panic) uint32_t backtrace[MAX_BACKTRACE]; - uint32_t cause; // Architecture-specific: exccause (Xtensa) or mcause (RISC-V) + uint32_t cause; // Architecture-specific: exccause (Xtensa) or mcause (RISC-V) + uint32_t fault_addr; // Faulting memory address: excvaddr (Xtensa) or mtval (RISC-V) uint8_t crashed_core; #if SOC_CPU_CORES_NUM > 1 static_assert(SOC_CPU_CORES_NUM == 2, "Dual-core logic assumes exactly 2 cores"); @@ -240,6 +241,16 @@ static const char *get_exception_reason() { nullptr, "LoadProhibited", "StoreProhibited", + nullptr, + nullptr, + "Cp0Dis", + "Cp1Dis", + "Cp2Dis", + "Cp3Dis", + "Cp4Dis", + "Cp5Dis", + "Cp6Dis", + "Cp7Dis", }; uint32_t cause = s_raw_crash_data.cause; if (cause < sizeof(REASON) / sizeof(REASON[0]) && REASON[cause] != nullptr) @@ -332,12 +343,24 @@ void crash_handler_log() { ESP_LOGE(TAG, "*** CRASH DETECTED ON PREVIOUS BOOT ***"); const char *reason = get_exception_reason(); if (reason != nullptr) { - ESP_LOGE(TAG, " Reason: %s - %s", get_exception_type(), reason); + ESP_LOGE(TAG, " Reason: %s - %s (cause %" PRIu32 ")", get_exception_type(), reason, s_raw_crash_data.cause); } else { ESP_LOGE(TAG, " Reason: %s", get_exception_type()); } ESP_LOGE(TAG, " Crashed core: %d", s_raw_crash_data.crashed_core); ESP_LOGE(TAG, " PC: 0x%08" PRIX32 " (fault location)", s_raw_crash_data.pc); + // Faulting memory address — only meaningful for real CPU faults, not + // aborts/watchdogs or SoC-level pseudo exceptions. Uses the same register + // name as ESP-IDF's live register dump for the architecture (EXCVADDR on + // Xtensa, MTVAL on RISC-V) so the CLI decodes it when it happens to be a + // code address. + if (s_raw_crash_data.exception == PANIC_EXCEPTION_FAULT && !s_raw_crash_data.pseudo_excause) { +#if CONFIG_IDF_TARGET_ARCH_XTENSA + ESP_LOGE(TAG, " EXCVADDR: 0x%08" PRIX32 " (faulting address)", s_raw_crash_data.fault_addr); +#elif CONFIG_IDF_TARGET_ARCH_RISCV + ESP_LOGE(TAG, " MTVAL: 0x%08" PRIX32 " (faulting address)", s_raw_crash_data.fault_addr); +#endif + } log_backtrace(s_raw_crash_data.backtrace, s_raw_crash_data.backtrace_count, s_raw_crash_data.reg_frame_count); #if SOC_CPU_CORES_NUM > 1 @@ -382,6 +405,9 @@ void IRAM_ATTR __wrap_esp_panic_handler(panic_info_t *info) { s_raw_crash_data.exception = (uint8_t) info->exception; s_raw_crash_data.pseudo_excause = info->pseudo_excause ? 1 : 0; s_raw_crash_data.crashed_core = (uint8_t) info->core; + // Zero unconditionally so a null frame doesn't leave stale .noinit data from a previous boot + s_raw_crash_data.cause = 0; + s_raw_crash_data.fault_addr = 0; #if SOC_CPU_CORES_NUM > 1 s_raw_crash_data.other_backtrace_count = 0; s_raw_crash_data.other_reg_frame_count = 0; @@ -392,6 +418,7 @@ void IRAM_ATTR __wrap_esp_panic_handler(panic_info_t *info) { if (info->frame != nullptr) { auto *xt_frame = (XtExcFrame *) info->frame; s_raw_crash_data.cause = xt_frame->exccause; + s_raw_crash_data.fault_addr = xt_frame->excvaddr; s_raw_crash_data.backtrace_count = walk_xtensa_backtrace(xt_frame, s_raw_crash_data.backtrace, MAX_BACKTRACE); } @@ -414,6 +441,7 @@ void IRAM_ATTR __wrap_esp_panic_handler(panic_info_t *info) { if (info->frame != nullptr) { auto *rv_frame = (RvExcFrame *) info->frame; s_raw_crash_data.cause = rv_frame->mcause; + s_raw_crash_data.fault_addr = rv_frame->mtval; s_raw_crash_data.backtrace_count = capture_riscv_backtrace(rv_frame, s_raw_crash_data.backtrace, MAX_BACKTRACE, &s_raw_crash_data.reg_frame_count); } diff --git a/tests/unit_tests/components/test_esp_stacktrace.py b/tests/unit_tests/components/test_esp_stacktrace.py index f231ac5fb7..eb7e63fc4d 100644 --- a/tests/unit_tests/components/test_esp_stacktrace.py +++ b/tests/unit_tests/components/test_esp_stacktrace.py @@ -137,3 +137,45 @@ def test_process_stacktrace_esp32_crash_handler( state = process_stacktrace(config, line_bt1, False) mock_esp32_decode_pc.assert_called_once_with(config, "42005ABC") assert state is False + + mock_esp32_decode_pc.reset_mock() + + # Reason line carries no address, must not trigger a decode + line_reason = "[E][esp32.crash:079]: Reason: Fault - LoadProhibited (cause 28)" + state = process_stacktrace(config, line_reason, False) + mock_esp32_decode_pc.assert_not_called() + assert state is False + + mock_esp32_decode_pc.reset_mock() + + # EXCVADDR pointing at code (e.g. jumping through a corrupted pointer) decodes + line_excvaddr = "[E][esp32.crash:081]: EXCVADDR: 0x400D9ABC (faulting address)" + state = process_stacktrace(config, line_excvaddr, False) + mock_esp32_decode_pc.assert_called_once_with(config, "400D9ABC") + assert state is False + + mock_esp32_decode_pc.reset_mock() + + # EXCVADDR pointing at data (heap/null) is not a code address, must be ignored + line_excvaddr_data = ( + "[E][esp32.crash:081]: EXCVADDR: 0x0000001C (faulting address)" + ) + state = process_stacktrace(config, line_excvaddr_data, False) + mock_esp32_decode_pc.assert_not_called() + assert state is False + + mock_esp32_decode_pc.reset_mock() + + # RISC-V MTVAL pointing at code decodes + line_mtval = "[E][esp32.crash:081]: MTVAL: 0x42001234 (faulting address)" + state = process_stacktrace(config, line_mtval, False) + mock_esp32_decode_pc.assert_called_once_with(config, "42001234") + assert state is False + + mock_esp32_decode_pc.reset_mock() + + # RISC-V MTVAL pointing at data must be ignored + line_mtval_data = "[E][esp32.crash:081]: MTVAL: 0x3FC80123 (faulting address)" + state = process_stacktrace(config, line_mtval_data, False) + mock_esp32_decode_pc.assert_not_called() + assert state is False