2005-05-06 18:00:21

by Protasevich, Natalie

Subject: RE: [patch 1/1] Do not enforce unique IO_APIC_ID for Xeonprocessors in EM64T mode (x86_64)

>
> A run-time solution would be preferable to adding a config
> option that only changes the default behaviour.
>
> In general, the more config options, the more kernels we
> force distros to build and support. We really want to going
> the other way and simplifying, when possible.


Len and Zwane, I sure agree with you. I will take the config option out.
Would the APIC version be a good criteria to make a run-time decision
with Xeons? I know that everything Intel that can run EM64T has front
side bus (APIC version >= 20?). And I guess the boot parameter can still
be useful?

Thanks,
--Natalie

2005-05-06 18:08:30

by Maciej W. Rozycki

Subject: RE: [patch 1/1] Do not enforce unique IO_APIC_ID for Xeonprocessors in EM64T mode (x86_64)

On Fri, 6 May 2005, Protasevich, Natalie wrote:

> Would the APIC version be a good criteria to make a run-time decision
> with Xeons? I know that everything Intel that can run EM64T has front
> side bus (APIC version >= 20?). And I guess the boot parameter can still
> be useful?

Isn't there a bit in one of the I/O APIC registers which denotes that FSB
delivery is used? Hmm, that would be "IO_APIC_reg_00.bits.delivery_type",
actually...

Maciej