OK, I've spent most of the day making my poor iBook compile kernels and I've tracked down the error. I can't seem to figure out why it would happen on my machine and no one elses, at least unless it's a compiler issue (I am using gcc 3.2.2).
Somewhere between 2.6.8-rc1 and 2.6.8-rc2 interrupts just stop getting delivered.
I tracked it down to this patch to linux/arch/ppc/syslib/open_pic.c :
/*
* Map an interrupt source to one or more CPUs
*/
-static void openpic_mapirq(u_int irq, u_int physmask, u_int keepmask)
+static void openpic_mapirq(u_int irq, cpumask_t physmask, cpumask_t keepmask)
{
if (ISR[irq] == 0)
return;
- if (keepmask != 0)
- physmask |= openpic_read(&ISR[irq]->Destination) & keepmask;
- openpic_write(&ISR[irq]->Destination, physmask);
+ if (!cpus_empty(keepmask)) {
+ cpumask_t irqdest = { .bits[0] = openpic_read(&ISR[irq]->Destination) };
+ cpus_and(irqdest, irqdest, keepmask);
+ cpus_or(physmask, physmask, irqdest);
+ }
+ openpic_write(&ISR[irq]->Destination, cpus_addr(physmask)[0]);
}
And this one: - openpic_mapirq(i, 1<<0, 0); + openpic_mapirq(i, CPU_MASK_CPU0, CPU_MASK_NONE);
Using printk's, I can see before the change I properly was writing "1" as the second argument to openpic_write, but afterwards it was 0.
It seems as though for some reason CPU_MASK_CPU0 is 0 on my kernel, rather than just a 1.
By patching the kernel to force it to write a 1 in openpic_write the kernel runs fine...
So, any ideas? I'll be glad to try out anything else if you'd like me to.
Vince
-- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
