PCI vector and ACPI

PCI vector and ACPI

[Mika Penttilä]

There seems to be some bad interaction between PCI VECTOR 
(use_pci_vector()) and ACPI PRTs. Namely, both mp_parse_prt() and 
pcibios_fixup_irqs() do the IO_APIC_VECTOR() translation on irq, the 
result being  IO_APIC_VECTOR(IO_APIC_VECTOR(irq)). I wonder if this is 
the cause of some bug reports with acpi and pci vector? Maybe 
mp_parse_prt() should not do the irq->vector translation?

--Mika

RE: Update MSI Patches

[Nguyen, Tom L]

Sent: Friday, August 15, 2003, Zwane Mwaikambo wrote:
> Ok i tested it on the following configurations;

> UP i8259 w/ CONFIG_PCI_USE_VECTOR is ok
> UP IOAPIC/MP1.4 w/ or w/o CONFIG_PCI_USE_VECTOR is ok
> UP IOAPIC/ACPI w/ or w/o CONFIG_PCI_USE_VECTOR is ok
> 8x SMP/MP1.4 w/ or w/o CONFIG_PCI_USE_VECTOR is ok
> 8x SMP/ACPI w/ or w/o CONFIG_PCI_USE_VECTOR is ok

> I also just looked again and it does look like the additional interrupt 
> controller startup/ack/mask etc member functions are too much duplicated 
> code (even if it wont all be in the final image). Perhaps just keep them 
> like this;

> static unsigned int startup_edge_ioapic(unsigned int index)
> {
>	int irq = vector_to_irq(index);
>	...
>
>	if (platform_legacy_irq(irq)) {
>		...
>	}
>
>	...
>	return
>}
Thank you for testing it through different configurations!
That is what we thought initially by renaming irq to something like offset or
index to avoid some duplicate functions. However, I think if keeping them like
above perhaps may raise a confusion again of why not name vector since the 
function vector_to_irq(...) converts a vector to irq. Please provide us final 
thoughts.


> Also there is a warning whilst compiling which looks like it must be 
> fixed.

> Index: linux-2.6.0-test3-msi/arch/i386/kernel/mpparse.c
> ===================================================================
> RCS file: /build/cvsroot/linux-2.6.0-test3/arch/i386/kernel/mpparse.c,v
> retrieving revision 1.2
> diff -u -p -B -r1.2 mpparse.c
> --- linux-2.6.0-test3-msi/arch/i386/kernel/mpparse.c	15 Aug 2003 07:38:39 -0000	1.2
> +++ linux-2.6.0-test3-msi/arch/i386/kernel/mpparse.c	15 Aug 2003 08:02:28 -0000
> @@ -1133,11 +1133,12 @@ void __init mp_parse_prt (void)
> 
> 		mp_ioapic_routing[ioapic].pin_programmed[idx] |= (1<<bit);
> 
> -		if (!io_apic_set_pci_routing(ioapic, ioapic_pin, irq))
> +		if (!io_apic_set_pci_routing(ioapic, ioapic_pin, irq)) {
> 			if (use_pci_vector() && !platform_legacy_irq(irq))
> 				entry->irq = IO_APIC_VECTOR(irq);
> 			else
> 				entry->irq = irq;
> +		}
> 
> 		printk(KERN_DEBUG "%02x:%02x:%02x[%c] -> %d-%d -> IRQ %d\n",
>  			entry->id.segment, entry->id.bus, 
> -
Thanks! Will do ...

Thanks,
Long

PCI vector and ACPI

[Mika Penttilä]

There seems to be some bad interaction between PCI VECTOR 
(use_pci_vector()) and ACPI PRTs. Namely, both mp_parse_prt() and 
pcibios_fixup_irqs() do the IO_APIC_VECTOR() translation on irq, the 
result being  IO_APIC_VECTOR(IO_APIC_VECTOR(irq)). I wonder if this is 
the cause of some bug reports with acpi and pci vector? Maybe 
mp_parse_prt() should not do the irq->vector translation?

--Mika