Re: Linux 3.19-rc5

[Oleg Nesterov <oleg@redhat.com>]

On 02/01, Linus Torvalds wrote:
>
> On Sun, Feb 1, 2015 at 11:43 AM, Oleg Nesterov <oleg@redhat.com> wrote:
> >
> > And personally I agree. sched_annotate_sleep() looks self-documented, it
> > is clear that it is used to suppress the warning.
>
> But *that's not the problem*.
>
> If it was just silencing the warning, things would be fine.
>
> But it is actively screwing task state up, and actually changing
> behavior of the kernel (in a very subtle part of the code too), and
> doing so in ways that potentially introduce WORSE BUGS THAN THE WHOLE
> DAMN DEBUG SUPPORT WAS SUPPOSED TO FIND IN THE FIRST PLACE.

I understand, that is why I suggested to change it.

> I like your patch, but I'm going to combine it with mine that actually
> fixes a real bug,

Sure, I agree.

> because what you don't see in that patch of yours:
> ...
>
>
> is that the whole "if (WARN_ONCE()" remains horribly buggy, because of
> the line that follows it:
>
>                 __set_current_state(TASK_RUNNING);
>
> in the if-statement.

Ah. I just forgot to mention that this change should be rediffed on top
of your patch, of course it is not enough.

> I'll just combine it with yours to avoid extra noise in this area, and
> mark you as the author, fixing *both* of the incorrect state changes.
> Ok?

Please combine them, but don't mark me as an author, I do not want to
take the false credits ;)

Oleg.