21.08.2013, 21:24, "Sebastian Andrzej Siewior" <bigeasy@xxxxxxxxxxxxx>:
> Now that I looked at it for a while, would this fix your trouble?
>
> diff --git a/kernel/rtmutex.c b/kernel/rtmutex.c
> --- a/kernel/rtmutex.c
> +++ b/kernel/rtmutex.c
> @@ -724,6 +724,7 @@ static void noinline __sched rt_spin_lock_slowlock(struct rt_mutex *lock)
> struct task_struct *lock_owner, *self = current;
> struct rt_mutex_waiter waiter, *top_waiter;
> int ret;
> + int new_state;
>
> rt_mutex_init_waiter(&waiter, true);
>
> @@ -744,8 +745,11 @@ static void noinline __sched rt_spin_lock_slowlock(struct rt_mutex *lock)
> * try_to_wake_up().
> */
> pi_lock(&self->pi_lock);
> + new_state = TASK_UNINTERRUPTIBLE;
> + if (task_is_traced(self))
> + new_state |= __TASK_TRACED;
> self->saved_state = self->state;
> - __set_current_state(TASK_UNINTERRUPTIBLE);
> + __set_current_state(new_state);
> pi_unlock(&self->pi_lock);
>
> ret = task_blocks_on_rt_mutex(lock, &waiter, self, 0);
>
> This should avoid that the trace state is lost while waiting on that
> mutex and the checks for "is traced" may remain the same.
This was the first thing I tried but it did not work. ptrace_check_attach() passes TASK_TRACED to wait_task_inactive() which tests for equality:
unsigned long wait_task_inactive(struct task_struct *p, long match_state)
< ... >
if (match_state && unlikely(p->state != match_state)
&& unlikely(p->saved_state != match_state)) {
But this test will fail because we've added TASK_UNINTERRUPTIBLE and possibly removed some other bits from p->state. So I decided to add pi_lock locking instead - it is slower but it works with 100% guarantee.
Also adding __TASK_TRACED does not help with all the other places where saved_state is checked and more races might still be hidden.
--
To unsubscribe from this list: send the line "unsubscribe linux-rt-users" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
