On Mon, Jul 18, 2011 at 09:54:57PM +1000, Benjamin Herrenschmidt wrote:
> On Mon, 2011-07-18 at 13:23 +0200, Peter Zijlstra wrote:
>
> > So how are we going to solve this? Naively I'd think that
> > current_thread_info() is short for task_thread_info(current), and thus
> > the platforms for where this isn't true are broken.
> >
> > I mean, what use is the thread_info not of a thread?
> >
> > Comments?
>
> Thomas just hit a bug in the platform code of said platform (powerpc
> heh ?) :-)
>
> We do it right for hard IRQs and for some reason never did it right for
> softirqs.
>
> The code is like this for the former:
>
> static inline void handle_one_irq(unsigned int irq)
> {
>
> .../...
>
> call_handle_irq(irq, desc, irqtp, desc->handle_irq);
> current->thread.ksp_limit = saved_sp_limit;
> irqtp->task = NULL;
>
> /* Set any flag that may have been set on the
> * alternate stack
> */
> if (irqtp->flags)
> set_bits(irqtp->flags, &curtp->flags);
> }
>
> So what we need, I suppose is to add those two last line to
> do_softirq_onstack() as well.
Hmmm... Would this explain preempt_count() inexplicably increasing by
three across a spin_unlock_irqrestore()? I ran into this situation when
testing on Power over the weekend.
Thanx, Paul
> Now indeed i386 needs a similar treatment on both hard and soft
> irqs (along with getting rid of that stupid duplication of
> call_on_stack in there, I don't think it's worth making the code
> horrible like that to save one clobber and PeterZ reckons we can
> probably avoid it using always_inline anyways).
>
> I'll let you guys sort i386 out tho, I'll look at fixing ppc tomorrow :-)
>
> Cheers,
> Ben.
>
>
--
To unsubscribe from this list: send the line "unsubscribe linux-arch" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
