RTOS/middleware/safety/watchdog_manager.c

/**
 * @file watchdog_manager.c
 * @brief Watchdog Manager for safety-critical applications
 */

#include "kernel.h"
#include "task.h"
#include <string.h>

/* Watchdog Configuration */
#define WATCHDOG_MAX_TASKS          16
#define WATCHDOG_DEFAULT_TIMEOUT    100  /* ms */
#define WATCHDOG_MAX_ALIVE_COUNT    5

/* Watchdog Task Status */
typedef enum {
    WATCHDOG_TASK_ALIVE = 0,
    WATCHDOG_TASK_TIMEOUT = 1,
    WATCHDOG_TASK_SUSPENDED = 2
} WatchdogTaskStatus_t;

/* Watchdog Task Entry */
typedef struct {
    TaskHandle_t task;
    char task_name[16];
    uint32_t timeout_ms;
    uint32_t last_alive_time;
    uint32_t alive_count;
    WatchdogTaskStatus_t status;
    bool is_supervised;
} WatchdogTaskEntry_t;

/* Watchdog Manager State */
typedef struct {
    bool initialized;
    bool enabled;
    WatchdogTaskEntry_t tasks[WATCHDOG_MAX_TASKS];
    uint8_t task_count;
    uint32_t global_timeout_ms;
    uint32_t last_service_time;
    Mutex_t mutex;
    void (*system_reset_callback)(void);
    void (*task_timeout_callback)(TaskHandle_t task);
} WatchdogManagerState_t;

static WatchdogManagerState_t watchdog_manager;

/* Initialize Watchdog Manager */
KernelStatus_t watchdog_manager_init(uint32_t global_timeout_ms) {
    if (watchdog_manager.initialized) {
        return KERNEL_ERROR;
    }

    memset(&watchdog_manager, 0, sizeof(WatchdogManagerState_t));
    watchdog_manager.global_timeout_ms = global_timeout_ms;
    watchdog_manager.enabled = false;
    watchdog_manager.last_service_time = kernel_get_tick_count();

    mutex_create(&watchdog_manager.mutex, false);

    watchdog_manager.initialized = true;
    return KERNEL_OK;
}

/* Register Task for Supervision */
KernelStatus_t watchdog_register_task(TaskHandle_t task, const char* name,
                                      uint32_t timeout_ms) {
    if (!watchdog_manager.initialized || task == NULL) {
        return KERNEL_ERROR;
    }

    if (watchdog_manager.task_count >= WATCHDOG_MAX_TASKS) {
        return KERNEL_OUT_OF_MEMORY;
    }

    mutex_lock(&watchdog_manager.mutex, 100);

    WatchdogTaskEntry_t* entry = &watchdog_manager.tasks[watchdog_manager.task_count];
    entry->task = task;
    strncpy(entry->task_name, name, sizeof(entry->task_name) - 1);
    entry->timeout_ms = timeout_ms;
    entry->last_alive_time = kernel_get_tick_count();
    entry->alive_count = 0;
    entry->status = WATCHDOG_TASK_ALIVE;
    entry->is_supervised = true;

    watchdog_manager.task_count++;

    mutex_unlock(&watchdog_manager.mutex);

    return KERNEL_OK;
}

/* Task Alive Indication */
KernelStatus_t watchdog_task_alive(TaskHandle_t task) {
    if (!watchdog_manager.initialized || task == NULL) {
        return KERNEL_ERROR;
    }

    mutex_lock(&watchdog_manager.mutex, 100);

    for (uint8_t i = 0; i < watchdog_manager.task_count; i++) {
        if (watchdog_manager.tasks[i].task == task) {
            watchdog_manager.tasks[i].last_alive_time = kernel_get_tick_count();
            watchdog_manager.tasks[i].alive_count++;
            watchdog_manager.tasks[i].status = WATCHDOG_TASK_ALIVE;
            break;
        }
    }

    mutex_unlock(&watchdog_manager.mutex);

    return KERNEL_OK;
}

/* Enable Watchdog */
KernelStatus_t watchdog_enable(void) {
    if (!watchdog_manager.initialized) {
        return KERNEL_ERROR;
    }

    watchdog_manager.enabled = true;
    watchdog_manager.last_service_time = kernel_get_tick_count();

    /* Enable hardware watchdog */
    hal_watchdog_enable(watchdog_manager.global_timeout_ms);

    return KERNEL_OK;
}

/* Disable Watchdog */
KernelStatus_t watchdog_disable(void) {
    if (!watchdog_manager.initialized) {
        return KERNEL_ERROR;
    }

    watchdog_manager.enabled = false;

    /* Disable hardware watchdog */
    hal_watchdog_disable();

    return KERNEL_OK;
}

/* Service Watchdog */
KernelStatus_t watchdog_service(void) {
    if (!watchdog_manager.initialized || !watchdog_manager.enabled) {
        return KERNEL_ERROR;
    }

    /* Check all supervised tasks */
    uint32_t current_time = kernel_get_tick_count();
    bool all_tasks_alive = true;

    mutex_lock(&watchdog_manager.mutex, 100);

    for (uint8_t i = 0; i < watchdog_manager.task_count; i++) {
        WatchdogTaskEntry_t* entry = &watchdog_manager.tasks[i];

        if (!entry->is_supervised) {
            continue;
        }

        /* Check if task is alive */
        if ((current_time - entry->last_alive_time) > entry->timeout_ms) {
            entry->status = WATCHDOG_TASK_TIMEOUT;
            all_tasks_alive = false;

            /* Call task timeout callback */
            if (watchdog_manager.task_timeout_callback != NULL) {
                watchdog_manager.task_timeout_callback(entry->task);
            }
        }
    }

    mutex_unlock(&watchdog_manager.mutex);

    if (all_tasks_alive) {
        /* Service hardware watchdog */
        hal_watchdog_service();
        watchdog_manager.last_service_time = current_time;
        return KERNEL_OK;
    } else {
        /* Don't service watchdog - will trigger reset */
        return KERNEL_ERROR;
    }
}

/* Register Callbacks */
KernelStatus_t watchdog_register_callbacks(
    void (*system_reset_callback)(void),
    void (*task_timeout_callback)(TaskHandle_t task)) {

    if (!watchdog_manager.initialized) {
        return KERNEL_ERROR;
    }

    watchdog_manager.system_reset_callback = system_reset_callback;
    watchdog_manager.task_timeout_callback = task_timeout_callback;

    return KERNEL_OK;
}

/* Watchdog Main Function */
void watchdog_manager_main_function(void) {
    if (!watchdog_manager.initialized || !watchdog_manager.enabled) {
        return;
    }

    /* Service watchdog periodically */
    watchdog_service();
}