On Sun, Feb 25, 2007 at 08:30:11PM +0100, Frank van der Linden wrote:
> Cherry G. Mathew wrote:
> >rtk0 decided it wanted to share the SCI pin. That's an impossible
> >topology which indicates there's something wrong with the PCI
> >enumeration/irq allocation. I'll sniff around a bit more and see what
> >I can find.
> I need to look at your fix more, it doesn't seem quite right (though I 
> don't doubt that it works for the reported cases). The case of someone 
> wanting to share the SCI pin is something that has always looked dodgy 
> to me. It could be in the code that came from FreeBSD, 
> dev/acpi/acpi_pci_link.c:986:
> 
>        /*     
>         * If this is an ISA IRQ, try using the SCI if it is also an ISA
>         * interrupt as a fallback.
>         */
> 
> That always seemed strange to me. It seems like a wild, and probably 
> wrong, guess, but I did not have time to get to the bottom of that, so I 
> thought "hey, I guess it works in FreeBSD".

I also don't get "If this is an ISA IRQ" -- this source might *also*
have an ISA IRQ assigned, but if it's a PCI device on an ioapic system,
the ISA IRQ shouldn't be used for anything, right?  How can it be correct
to then use the ISA IRQ for routing purposes?