Possible race condition in task_manager.c?

[garrulous]

Hello,

I believe I have found a possible race condition in experimental_task_manager / task_manager.c, task_events_wait().

The race occurs if a task calls task_events_wait(), while an interrupt calls task_events_set(), for the same event; the event is ignored and the waiting task waits indefinitely.

The problem appears fixed if I add 4 lines to task_events_wait(), indicated by the comments below:

uint32_t task_events_wait(uint32_t evt_mask)
{
    uint32_t current_events;

    ASSERT((evt_mask & ~TASK_FLAG_SIGNAL_MASK) == 0);

    for (;;)
    {
        current_events = s_task_state[s_current_task_id].flags & evt_mask;
        if (current_events != 0)
        {
            (void)nrf_atomic_u32_and(&s_task_state[s_current_task_id].flags, ~current_events);
            break;
        }

        TASK_STATE_SUSPENDED(s_current_task_id);
        
        // Race condition fix starts ----------------------------------------
        if (current_events != (s_task_state[s_current_task_id].flags & evt_mask))
        {
            continue;
        }
        // Race condition fix ends ------------------------------------------

        task_yield();
    }

    return current_events;
}

Appreciate if you could please confirm if this is indeed a race condition, and if this fix is appropriate.

Thanks!

[garrulous]

Revision to the above -- possibly better:

        // Race condition fix starts ----------------------------------------
        if (s_task_state[s_current_task_id].flags & evt_mask)
        {
            TASK_STATE_RUNNABLE(s_current_task_id);
            continue;
        }
        // Race condition fix ends ------------------------------------------