Re: [PATCH] x86/topo: Map vendor CPU types to generic Linux such types

On Thu, Aug 27, 2026 at 01:44:25PM -0700, Borislav Petkov wrote:
> Here's the cleanup ontop, lemme know pls whether that makes sense.
> 
> Some noteworthy things:
> 
> * this basically switches to TOPO_CPU_TYPE and only the detection code knows
> about the vendor-specific ones. The generic types are union set of both.
> 
> * We do some CPUID calls where needed:
> 
> -			native_id = c->topo.intel_native_model_id;
> +			native_id = cpuid_eax(0x1a) & GENMASK(23, 0);
> 
> and in get_topology_cpu_type() but those will go away too with Ahmed's rework
> 
> The rest is just manual conversion labour. Oh, and TOPO_CPU_TYPE_ANY is gone
> too, so that Pawan can sleep at night :-P

And put my OCD to rest :-)
(BTW, some of that is acquired occupational hazard.)

> Thoughts?

Overall, looks like a good cleanup to me.

> I think it is better this way instead of carrying intel_type and amd_type in
> cpuinfo - now the topology code is preparing things for us.
> 
> I'll split it into proper patches if people are ok with it.
> 
> Thx.
...
> --- a/arch/x86/kernel/cpu/topology_common.c
> +++ b/arch/x86/kernel/cpu/topology_common.c
> @@ -35,13 +35,17 @@ void topology_set_dom(struct topo_scan *tscan, enum x86_topology_domains dom,
>  enum x86_topology_cpu_type get_topology_cpu_type(struct cpuinfo_x86 *c)
>  {
>  	if (c->x86_vendor == X86_VENDOR_INTEL) {
> -		switch (c->topo.intel_type) {
> +		unsigned int type = (cpuid_eax(0x1a) >> 24) & 0xff;

You already mentioned about the cpuid taken care elsewhere, just noting
that this needs a (c->cpuid_level >= 0x1a) check.

> +		switch (type) {
>  		case INTEL_CPU_TYPE_ATOM: return TOPO_CPU_TYPE_EFFICIENCY;
>  		case INTEL_CPU_TYPE_CORE: return TOPO_CPU_TYPE_PERFORMANCE;
>  		}
>  	}

Since x86_match_cpu() matches with boot_cpu_data, one rough edge remaining
is, a secondary cpu_type in a hybrid system can never be matched. For
this reason X86_MATCH_VFM_CPU_TYPE() is of little use now.