On Mon, 2002-09-16 at 15:01, Linus Torvalds wrote:
> Would it not be a lot better to just mask off PREEMPT_ACTIVE() instead of
> checking for it explicitly.
> The in_interrupt() etc stuff already effectively do this by masking off
> the HARDIRQ_MASK etc. I would prefer a patch to hardirq.h that just adds a
> #define to make preempt_count() not contain PREEMPT_ACTIVE - and make the
> PREEMPT_ACTIVE checks be a totally separate check (logic: it's not a
> count, so it shouldn't show up in preempt_count())
I liked this idea, and was working on implementing it when I ran into a
few roadblocks. Your ideas are welcome.
First, "preempt_count()" is used as an l-value in a lot of places, i.e.
look at all the "preempt_count() += foo" in the IRQ code. We cannot
mask things out of it.
Thus, I then looked into doing a separate function for the raw value,
say an "atomic_count()" ... the code just looked ugly mixing
"atomic_count()" and "preempt_count()" for no apparent reason.
Third, PREEMPT_ACTIVE actually _is_ part of the count. It helps assure
us a task is not preempted repeatedly. If we did not have it, we would
have to bump preempt_count on preemption. So we still need it in the
Simplest solution is to:
#define in_atomic() \
(preempt_count() & ~PREEMPT_ACTIVE) != kernel_locked())
although I still dislike the masking just to make the schedule()
Oh, and there is another problem: printk() from schedule() implicitly
calls wake_up(). My machine dies even with just a printk() and not a
BUG()... I suspect there may be some SMP issue in that whole mess too,
because setting oops_in_progress prior did not help.
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to email@example.com
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Mon Sep 23 2002 - 22:00:17 EST