Re: wait_on_irq, CPU1

Manfred Spraul (manfreds@colorfullife.com)
Wed, 29 Dec 1999 10:23:43 +0100

Andi Kleen wrote:
>
> On Tue, Dec 28, 1999 at 11:59:12PM +0100, Andrea Arcangeli wrote:
> > > I though about a tiny patch which calls
> > >smp_call_function() in wait_on_irq(): the IPI would call a stripped down
> > >version of show_registers().
> >
> > FYI: Andi just wrote the code to do that for the wait_on_bh case for other
> > needs, extending original's Andi's patch for wait_on_irq as well should be
> > trivial.
>
> It is part of the kbacktrace patch. It should give full backtrace for
> wait_on_bh and wait_on_irq. kbacktrace generally is intended as a replacement
> for the traditional *((int*)0)=0;
>

There are 2 buglets in the code:

* the kernel stack size is THREAD_SIZE (ie 8192), your patch assumes
that the stack size is 4096.

* your CPU enumeration (for_all_cpus) is wrong: the cpu number
(smp_processor_id()) are set by the motherboard, and they are not
guaranteed to be contiguous:

> +#define for_all_cpus(i) for((i)=0;(i)<smp_num_cpus;(i)++)

you must use "cpu_logical_map()", eg get_irq_list() in
linux/arch/i386/kernel/irq.c does that.

>
> + if (smp_call_function(getstack, sps, 1, 1) < 0)
> + printk("smp_call_function failed\n");
> + else {

I should really fix smp_call_function(): have you checked the actual
implementation:
it uses a semaphore, ie panic() waits for a semaphore before stopping
the kernel...

-
Manfred

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/