> > That's exactly the problem on ia64 - it does.
> > Could this also be a problem on i386 that we just haven't noticed yet?
>
> Unlikely. The IO-APIC on x86 is in that region, but it doesn't respond
> from external sources, it's not actually on the PCI bus and only visible
> from the CPU. And the CPU decodes that address internally and sends it on
> the APIC bus and thus PCI devices simply do not matter for it.
That's not true for the I/O APIC. That's only true for the local APIC.
The I/O APIC may be wired to the PCI-ISA bridge, specifically the APICCS#
(chip select) line may be driven by that bridge when a PCI cycle targets
the area assigned to the I/O APIC. This is certainly the case for
discrete implementations, like the i82371SB/AB (PCI-ISA bridge) + i82093AA
(I/O APIC) pair, possibly for others, including later I/O APICs integrated
into chipsets. Obviously cycles from CPUs travel to the PCI-ISA bridge
across the associated PCI bus.
-- + Maciej W. Rozycki, Technical University of Gdansk, Poland + +--------------------------------------------------------------+ + e-mail: macro@ds2.pg.gda.pl, PGP key available +- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/