Re: [PATCH] 3 patches against linuxppc_dev-2.4.18-pre7

["David Müller (ELSOFT AG)" <d.mueller@elsoft.ch>]

Hi

Armin Kuster wrote:

> "David Müller (ELSOFT AG)" wrote:
>
>>Hello
>>
>>The attached patch fixes the following issues:
>>
>>    - arch/ppc/kernel/ppc4xx_pic.c
>>       fixed constant in ppc405_pic_get_irq()
>>       (should be NR_AIC_IRQS as in ppc403_pic_get_irq())
>>
>
> David,
>
> include/asm-ppc/irq.h
> NR_IRQS = (NR_AIC_IRQS + NR_BOARD_IRQS) and was defined to allow
> cascading UIC's.
> NR_BOARD_IRQS is defined in some of the board files and its default is
> in include/asm-ppc/ibm4xx.h
>
> So i don't see the need to patch that file but the others look fine.
>
>

Exactly, NR_IRQS is the the total number of irqs possible in the system
as this number of irqs is reserved in the irq_desc table.

For example, we have a system with a PPC405GP and an additional PC like 8259 PIC
pair, we have NR_AIC_IRQS (32) + NR_BOARD_IRQS (16) = NR_IRQS (48).

In ppc405_pic_get_irq() the following code is executed:

	irq = 32 - ffs(bits);

	if (irq == NR_IRQS)
		irq = -1;

	return(irq);

with NR_IRQS equal 48 the condition is never true !

This results in an endless while-loop in do_IRQ() in arch/ppc/kernel/irq.c

Either this or my understanding how things are working is totally wrong. ;-)
Additionally, i don't see me why ppc403_pic_get_irq() is using NC_AIC_IRQS while
ppc405_pic_get_irq() is using NR_IRQS. Shouldn't they be "symmetric".


Dave


** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/