Re: Non power-of-two cache sizes and page colouring

[Christoph Egger <Christoph_Egger%gmx.de@localhost>]

On Friday 18 April 2008 17:21:47 Simon Burge wrote:
> Izumi Tsutsui wrote:
> > simonb%NetBSD.org@localhost wrote:
> > > I can't correlate anything in that document about working out cache
> > > info with what x86/x86/cacheinfo.c does.  At this stage I'm entirely
> > > prepared to believe NetBSD doesn't know what it's doing with cache info
> > > :-)
> >
> > i386/i386/identcpu.c may have the structure for it?
>
> Ah, intel_cpuid_cache_info[] is exactly what we want, although
> somewhat out of date.
>
> > Should we move it under arch/x86?
>
> Definitely.  I'll look into this.
>
> > > CPUID(0x80000006) doesn't list 24-way as an option in Table 3-17, and
> > > yet Table 3-7 does (the line you quoted above).  It looks like we
> > > should use CPUID(2) and then look that up in a table built from Table
> > > 3-7.
> >
> > Agreed.
> >
> > Maybe Intel guys didn't expect non-power-of-two way cache
> > because it was a bit unlikely in general hardware design.
> > Then it actually appeares but CPUID(0x80000006) can't
> > describe such value, I guess.
>
> I think the idea for Intel CPUs is to only use CPUID(2) to identify
> cache info?  I wonder if CPUID(0x80000006) is more for AMD CPUs?

On AMD, CPUID(0x80000005) is for L1 caches.

CPUID(0x80000006)  %eax, %ebx and %ecx contain L2 cache information
and %edx contains L3 cache information.

Christoph