Re: [PATCH 8/9] x86/apic: match destination id with destination mode

[Ingo Molnar <mingo@elte.hu>]

* Eric W. Biederman <ebiederm@xmission.com> wrote:

> Ingo Molnar <mingo@elte.hu> writes:
> 
> > The question here is whether this should layer on top of 
> > Jeremy's IO-APIC driverization patches. I think it should.
> 
> The patch is a bad hack that is totally misdocumented.  A bit like 
> the Xen apic changes in that respect.
> 
> I haven't seen Jeremy's IO-APIC driverization patches.
> 
> I am stumped why we need any driverization in this area.  x86_64 
> and has had for years a mechanism that is perfectly fine for 
> abstracting this. i386 also has had something similar and last I 
> looked we just about had that code merged.

We have the local apic abstracted out into struct apic on both 
32-bit and 64-bit, but not the IO-APIC methods.

> Xen doesn't have ioapics so it doesn't need us faking writes to 
> ioapics.  Xen either needs to parse the ioapic tables itself or 
> Xen needs a proper interface to be given the table information.
> 
> I this patch can be replaced by a 2 line change to the apic mode 
> logic to force us into physflat mode on moorestown.

Yes, probably. I havent looked deeply into all the merits yet, there 
were so many other structural problems with these patches.

	Ingo