ECC_check.c

/*
  create a thread that checks for ECC errors in flash. We disable
  fault IRQs in the thread and monitor it from the main thread. If it
  doesn't complete in 1s then erase flash This is the only way I've
  found to check for ECC errors without causing a bus fault or
  unconditionally erasing flash
 */
struct ecccheck {
    const uint32_t *start;
    uint32_t size;
    uint32_t sum;
    bool done;
};

static void ecccheck_thread(void *arg)
{
    auto *m = (struct ecccheck *)arg;
    const uint32_t *p = m->start;
    __disable_fault_irq();
    for (uint32_t i=0; i<m->size/4; i++) {
        /*
          when we have an ECC double error the ldr in the following
          line keeps repeating forever. The parent thread should then
          timeout waiting for this thread to complete and will erase
          all of flash, allowing the user to load a new firmware.
          Unfortunately it doesn't work as ChibiOS never schedules the
          parent thread again while this lower priority thread is
          stuck executing this ldr, so it doesn't work
         */
        m->sum += *p++;
    }
    __enable_fault_irq();
    m->done = true;
}

void check_ecc_errors(void)
{
    struct ecccheck m {};
    m.start = (const uint32_t *)FLASH_BASE;
    m.size = BOARD_FLASH_SIZE*1024;
    auto *thd = chThdCreateFromHeap(NULL, 1024, "ECC_check", 2, ecccheck_thread, &m);
    for (uint32_t i=0; i<1000; i++) {
        thread_sleep_ms(1);
        if (m.done) {
            break;
        }
    }
    if (!m.done) {
        // we must have ECC errors in flash blocking the read. Erase
        // flash
        flash_set_keep_unlocked(true);
        uint32_t page = 0;
        while (stm32_flash_erasepage(flash_base_page+(page++))) ;
        flash_set_keep_unlocked(false);
        thread_sleep_ms(100);
    }
    chThdRelease(thd);
}