Re: [PATCH RFC 09/13] exit: Use for_each_thread() in do_wait()

[Kirill Tkhai <tkhai@yandex.ru>]

27.05.2015, 12:34, "Kirill Tkhai" <tkhai@yandex.ru>:
> 26.05.2015, 22:47, "Oleg Nesterov" <oleg@redhat.com>:
>>  On 05/25, Kirill Tkhai wrote:
>>>   Refactoring, no functionality change.
>>  Hmm. unless I missed something this change is wrong.
>>>   --- a/kernel/exit.c
>>>   +++ b/kernel/exit.c
>>>   @@ -1538,8 +1538,7 @@ static long do_wait(struct wait_opts *wo)
>>>
>>>            set_current_state(TASK_INTERRUPTIBLE);
>>>            read_lock(&tasklist_lock);
>>>   - tsk = current;
>>>   - do {
>>>   + for_each_thread(current, tsk) {
>>>                    retval = do_wait_thread(wo, tsk);
>>>                    if (retval)
>>>                            goto end;
>>>   @@ -1550,7 +1549,7 @@ static long do_wait(struct wait_opts *wo)
>>>
>>>                    if (wo->wo_flags & __WNOTHREAD)
>>>                            break;
>>>   - } while_each_thread(current, tsk);
>>>   + }
>>  Please note the __WNOTHREAD check. This is the rare case when we
>>  actually want while_each_thread() (although it should die anyway).
>>
>>  for_each_thread() always starts from ->group_leader, but we need
>>  to start from "current" first.
>
> Sure, this must be like below. Thanks!
> I won't resend the whole series with only this one patch changed to
> do not bomb mail boxes. Waiting for the review.
>
> diff --git a/kernel/exit.c b/kernel/exit.c
> index a268093..e4963d3 100644
> --- a/kernel/exit.c
> +++ b/kernel/exit.c
> @@ -1538,8 +1538,10 @@ static long do_wait(struct wait_opts *wo)
>
>          set_current_state(TASK_INTERRUPTIBLE);
>          read_lock(&tasklist_lock);
> - tsk = current;
> - do {
> + for_each_thread(current, tsk) {
> + if (wo->wo_flags & __WNOTHREAD)
> + tsk = current;
> +
>                  retval = do_wait_thread(wo, tsk);
>                  if (retval)
>                          goto end;
> @@ -1550,7 +1552,7 @@ static long do_wait(struct wait_opts *wo)
>
>                  if (wo->wo_flags & __WNOTHREAD)
>                          break;
> - } while_each_thread(current, tsk);
> + }
>          read_unlock(&tasklist_lock);
>
>  notask:

Hm. Once again. Is the problem in __WNOTHREAD only?
Should we firstly reap our own children in common case?