Re: [PATCH 7/8] cpufreq: Add support for x86 cpuinfo auto loading v3

[Dave Jones <davej@redhat.com>]

On Wed, Dec 21, 2011 at 12:14:57PM -0800, H. Peter Anvin wrote:
 > On 12/21/2011 12:12 PM, Andi Kleen wrote:
 > >>> +static const struct x86_cpu_id powernow_k8_ids[] = {
 > >>> +	{ X86_VENDOR_AMD, 0xf, },
 > >>> +	{ X86_VENDOR_AMD, 0x10, },
 > >>> +	{ X86_VENDOR_AMD, 0x11, },
 > >>> +	{ X86_VENDOR_AMD, 0x12, },
 > >>> +	{ X86_VENDOR_AMD, 0x13, },
 > >>> +	{ X86_VENDOR_AMD, 0x14, },
 > >> { X86_VENDOR_AMD, 0x15, },
 > >>
 > >> Should also have BullDozer family included.
 > > 
 > > I think it really needs a cpuid feature bit for the hwpstate.  This is
 > > not in one of the normal CPUID leaves, so the amd specific CPU init
 > > code would need to set it.
 > > 
 > > I added 0x15 for now, but it would be good if someone with the hardware
 > > fixed it like this.
 > > 
 > > Then could just add a match on "AMD-hw pstate" or so.
 > > 
 > 
 > Agreed.  This is literally screaming for a synthetic CPUID bit.

A synthetic one is pointless, because you'd just be duplicating something
similar to the above in generic code that would need updating for every new stepping.

If anything other than powernow-k8 cared, there might be a more
compelling argument, but I don't see the point in moving something that
only this driver cares about to common code.

	Dave