Re: [PATCH v1 07/12] tools/x86/kcpuid: Add rudimentary CPU vendor detection

[Andrew Cooper <andrew.cooper3@citrix.com>]

On 06/03/2025 8:49 pm, Ahmed S. Darwish wrote:
> @@ -145,6 +166,40 @@ static inline bool has_subleafs(u32 f)
>  	return false;
>  }
>  
> +/*
> + * Leaf 0x0 EDX output, CPU vendor ID string bytes 4 - 7.
> + */
> +enum {
> +	EDX_INTEL	= fourcc('i', 'n', 'e', 'I'),	/* Genu_ineI_ntel */
> +	EDX_AMD		= fourcc('e', 'n', 't', 'i'),	/* Auth_enti_cAMD */
> +	EDX_HYGON	= fourcc('n', 'G', 'e', 'n'),	/* Hygo_nGen_uine */
> +	EDX_TRANSMETA	= fourcc('i', 'n', 'e', 'T'),	/* Genu_ineT_Mx86 */
> +	EDX_CENTAUR	= fourcc('a', 'u', 'r', 'H'),	/* Cent_aurH_auls */
> +	EDX_ZHAOXIN	= fourcc('a', 'n', 'g', 'h'),	/*   Sh_angh_ai	  */
> +};
> +
> +static enum cpu_vendor identify_cpu_vendor(void)
> +{
> +	u32 eax = 0, ebx, ecx = 0, edx;
> +
> +	cpuid(&eax, &ebx, &ecx, &edx);
> +
> +	switch (edx) {
> +	case EDX_INTEL:
> +		return VENDOR_INTEL;
> +	case EDX_AMD:
> +	case EDX_HYGON:
> +		return VENDOR_AMD;
> +	case EDX_TRANSMETA:
> +		return VENDOR_TRANSMETA;
> +	case EDX_CENTAUR:
> +	case EDX_ZHAOXIN:
> +		return VENDOR_CENTAUR;
> +	default:
> +		return VENDOR_UNKNOWN;
> +	}

Many CPUs have ways of overriding the vendor string, thanks to
GenuineIntel being hardcoded in too many pieces of software.

I suggest you check all registers, lest you find yourself on a CPU
claiming EBX=0x68747541, ECX=0x6C65746E, EDX=0x6E65476E

~Andrew