clear_IO_APIC() et al

clear_IO_APIC() et al

[Jan Beulich]

In c/s 20650:b543acc1aaad you made clear_IO_APIC() call both
clear_IO_APIC_pin() and clear_IO_APIC_raw(), and I don't really
follow why the former would be necessary here - the only code
paths we care about are boot and shutdown/crash, and hence
all we want is clearing the actual IO-APIC RTEs (without regard
to interrupt re-mapping).

Furthermore, even clear_IO_APIC_pin() is only called during early
boot, so it would seem to me that we should only need the "raw"
variant.

The context of the question is the need to add some extra code
here to clear the remoteIRR bit if still set (e.g. during a crash), as
that bit remaining set can otherwise lead to later confusion (during
crash handling in the secondary kernel, and when booted through
kexec in Xen itself).

Jan

Re: clear_IO_APIC() et al

[Keir Fraser]

On 16/08/2011 08:46, "Jan Beulich" <JBeulich@novell.com> wrote:

> In c/s 20650:b543acc1aaad you made clear_IO_APIC() call both
> clear_IO_APIC_pin() and clear_IO_APIC_raw(), and I don't really
> follow why the former would be necessary here - the only code
> paths we care about are boot and shutdown/crash, and hence
> all we want is clearing the actual IO-APIC RTEs (without regard
> to interrupt re-mapping).
> 
> Furthermore, even clear_IO_APIC_pin() is only called during early
> boot, so it would seem to me that we should only need the "raw"
> variant.
> 
> The context of the question is the need to add some extra code
> here to clear the remoteIRR bit if still set (e.g. during a crash), as
> that bit remaining set can otherwise lead to later confusion (during
> crash handling in the secondary kernel, and when booted through
> kexec in Xen itself).

I think Alex is @ Red Hat now, so unlikely to be of help here. You can feel
free to propose whatever fixups and cleanups you like on that function.

 -- Keir

> Jan
>