Watchdogs

Overview

The ESP-IDF has support for two types of watchdogs: The Interrupt Watchdog Timer and the Task Watchdog Timer (TWDT). The Interrupt Watchdog Timer and the TWDT can both be enabled using Project Configuration Menu, however the TWDT can also be enabled during runtime. The Interrupt Watchdog is responsible for detecting instances where FreeRTOS task switching is blocked for a prolonged period of time. The TWDT is responsible for detecting instances of tasks running without yielding for a prolonged period.

Interrupt watchdog

The interrupt watchdog makes sure the FreeRTOS task switching interrupt isn’t blocked for a long time. This is bad because no other tasks, including potentially important ones like the WiFi task and the idle task, can’t get any CPU runtime. A blocked task switching interrupt can happen because a program runs into an infinite loop with interrupts disabled or hangs in an interrupt.

The default action of the interrupt watchdog is to invoke the panic handler. causing a register dump and an opportunity for the programmer to find out, using either OpenOCD or gdbstub, what bit of code is stuck with interrupts disabled. Depending on the configuration of the panic handler, it can also blindly reset the CPU, which may be preferred in a production environment.

The interrupt watchdog is built around the hardware watchdog in timer group 1. If this watchdog for some reason cannot execute the NMI handler that invokes the panic handler (e.g. because IRAM is overwritten by garbage), it will hard-reset the SOC. If the panic handler executes, it will display the panic reason as “Interrupt wdt timeout on CPU0” or “Interrupt wdt timeout on CPU1” (as applicable).

Configuration

The interrupt watchdog is enabled by default via the CONFIG_ESP_INT_WDT configuration flag. The timeout is configured by setting CONFIG_ESP_INT_WDT_TIMEOUT_MS. The default timeout is higher if PSRAM support is enabled, as a critical section or interrupt routine that accesses a large amount of PSRAM will take longer to complete in some circumstances. The INT WDT timeout should always be longer than the period between FreeRTOS ticks (see CONFIG_FREERTOS_HZ).

Tuning

If you find the Interrupt watchdog timeout is triggering because an interrupt or critical section is running longer than the timeout period, consider rewriting the code: critical sections should be made as short as possible, with non-critical computation happening outside the critical section. Interrupt handlers should also perform the minimum possible amount of computation, consider pushing data into a queue from the ISR and processing it in a task instead. Neither critical sections or interrupt handlers should ever block waiting for another event to occur.

If changing the code to reduce the processing time is not possible or desirable, it’s possible to increase the CONFIG_ESP_INT_WDT_TIMEOUT_MS setting instead.

Task Watchdog Timer

The Task Watchdog Timer (TWDT) is responsible for detecting instances of tasks running for a prolonged period of time without yielding. This is a symptom of CPU starvation and is usually caused by a higher priority task looping without yielding to a lower-priority task thus starving the lower priority task from CPU time. This can be an indicator of poorly written code that spinloops on a peripheral, or a task that is stuck in an infinite loop.

By default the TWDT will watch the Idle Tasks of each CPU, however any task can subscribe to be watched by the TWDT. Each watched task must ‘reset’ the TWDT periodically to indicate that they have been allocated CPU time. If a task does not reset within the TWDT timeout period, a warning will be printed with information about which tasks failed to reset the TWDT in time and which tasks are currently running.

It is also possible to redefine the function esp_task_wdt_isr_user_handler in the user code, in order to receive the timeout event and handle it differently.

The TWDT is built around the Hardware Watchdog Timer in Timer Group 0. The TWDT can be initialized by calling esp_task_wdt_init() which will configure the hardware timer. A task can then subscribe to the TWDT using esp_task_wdt_add() in order to be watched. Each subscribed task must periodically call esp_task_wdt_reset() to reset the TWDT. Failure by any subscribed tasks to periodically call esp_task_wdt_reset() indicates that one or more tasks have been starved of CPU time or are stuck in a loop somewhere.

A watched task can be unsubscribed from the TWDT using esp_task_wdt_delete(). A task that has been unsubscribed should no longer call esp_task_wdt_reset(). Once all tasks have unsubscribed form the TWDT, the TWDT can be deinitialized by calling esp_task_wdt_deinit().

The default timeout period for the TWDT is set using config item CONFIG_ESP_TASK_WDT_TIMEOUT_S. This should be set to at least as long as you expect any single task will need to monopolise the CPU (for example, if you expect the app will do a long intensive calculation and should not yield to other tasks). It is also possible to change this timeout at runtime by calling esp_task_wdt_init().

The following config options control TWDT configuration at startup. They are all enabled by default:

JTAG and watchdogs

While debugging using OpenOCD, the CPUs will be halted every time a breakpoint is reached. However if the watchdog timers continue to run when a breakpoint is encountered, they will eventually trigger a reset making it very difficult to debug code. Therefore OpenOCD will disable the hardware timers of both the interrupt and task watchdogs at every breakpoint. Moreover, OpenOCD will not reenable them upon leaving the breakpoint. This means that interrupt watchdog and task watchdog functionality will essentially be disabled. No warnings or panics from either watchdogs will be generated when the ESP32 is connected to OpenOCD via JTAG.

Interrupt Watchdog API Reference

Functions

void esp_int_wdt_init(void)

Initialize the non-CPU-specific parts of interrupt watchdog. This is called in the init code if the interrupt watchdog is enabled in menuconfig.

Task Watchdog API Reference

A full example using the Task Watchdog is available in esp-idf: system/task_watchdog

Functions

esp_err_t esp_task_wdt_init(uint32_t timeout, bool panic)

Initialize the Task Watchdog Timer (TWDT)

This function configures and initializes the TWDT. If the TWDT is already initialized when this function is called, this function will update the TWDT’s timeout period and panic configurations instead. After initializing the TWDT, any task can elect to be watched by the TWDT by subscribing to it using esp_task_wdt_add().

Return

  • ESP_OK: Initialization was successful

  • ESP_ERR_NO_MEM: Initialization failed due to lack of memory

Note

esp_task_wdt_init() must only be called after the scheduler started

Parameters

  • [in] timeout: Timeout period of TWDT in seconds

  • [in] panic: Flag that controls whether the panic handler will be executed when the TWDT times out

esp_err_t esp_task_wdt_deinit(void)

Deinitialize the Task Watchdog Timer (TWDT)

This function will deinitialize the TWDT. Calling this function whilst tasks are still subscribed to the TWDT, or when the TWDT is already deinitialized, will result in an error code being returned.

Return

  • ESP_OK: TWDT successfully deinitialized

  • ESP_ERR_INVALID_STATE: Error, tasks are still subscribed to the TWDT

  • ESP_ERR_NOT_FOUND: Error, TWDT has already been deinitialized

esp_err_t esp_task_wdt_add(TaskHandle_t handle)

Subscribe a task to the Task Watchdog Timer (TWDT)

This function subscribes a task to the TWDT. Each subscribed task must periodically call esp_task_wdt_reset() to prevent the TWDT from elapsing its timeout period. Failure to do so will result in a TWDT timeout. If the task being subscribed is one of the Idle Tasks, this function will automatically enable esp_task_wdt_reset() to called from the Idle Hook of the Idle Task. Calling this function whilst the TWDT is uninitialized or attempting to subscribe an already subscribed task will result in an error code being returned.

Return

  • ESP_OK: Successfully subscribed the task to the TWDT

  • ESP_ERR_INVALID_ARG: Error, the task is already subscribed

  • ESP_ERR_NO_MEM: Error, could not subscribe the task due to lack of memory

  • ESP_ERR_INVALID_STATE: Error, the TWDT has not been initialized yet

Parameters

  • [in] handle: Handle of the task. Input NULL to subscribe the current running task to the TWDT

esp_err_t esp_task_wdt_reset(void)

Reset the Task Watchdog Timer (TWDT) on behalf of the currently running task.

This function will reset the TWDT on behalf of the currently running task. Each subscribed task must periodically call this function to prevent the TWDT from timing out. If one or more subscribed tasks fail to reset the TWDT on their own behalf, a TWDT timeout will occur. If the IDLE tasks have been subscribed to the TWDT, they will automatically call this function from their idle hooks. Calling this function from a task that has not subscribed to the TWDT, or when the TWDT is uninitialized will result in an error code being returned.

Return

  • ESP_OK: Successfully reset the TWDT on behalf of the currently running task

  • ESP_ERR_NOT_FOUND: Error, the current running task has not subscribed to the TWDT

  • ESP_ERR_INVALID_STATE: Error, the TWDT has not been initialized yet

esp_err_t esp_task_wdt_delete(TaskHandle_t handle)

Unsubscribes a task from the Task Watchdog Timer (TWDT)

This function will unsubscribe a task from the TWDT. After being unsubscribed, the task should no longer call esp_task_wdt_reset(). If the task is an IDLE task, this function will automatically disable the calling of esp_task_wdt_reset() from the Idle Hook. Calling this function whilst the TWDT is uninitialized or attempting to unsubscribe an already unsubscribed task from the TWDT will result in an error code being returned.

Return

  • ESP_OK: Successfully unsubscribed the task from the TWDT

  • ESP_ERR_INVALID_ARG: Error, the task is already unsubscribed

  • ESP_ERR_INVALID_STATE: Error, the TWDT has not been initialized yet

Parameters

  • [in] handle: Handle of the task. Input NULL to unsubscribe the current running task.

esp_err_t esp_task_wdt_status(TaskHandle_t handle)

Query whether a task is subscribed to the Task Watchdog Timer (TWDT)

This function will query whether a task is currently subscribed to the TWDT, or whether the TWDT is initialized.

Return

:

  • ESP_OK: The task is currently subscribed to the TWDT

  • ESP_ERR_NOT_FOUND: The task is currently not subscribed to the TWDT

  • ESP_ERR_INVALID_STATE: The TWDT is not initialized, therefore no tasks can be subscribed

Parameters

  • [in] handle: Handle of the task. Input NULL to query the current running task.

Provide feedback about this document