On 23.02.2026 13:36, Andrew Cooper wrote:
> On 16/02/2026 11:56 am, Jan Beulich wrote:
>> The records reported used to contain Linux enumerators. We first broke
>> that connection when purging unused ones, and then again when switching to
>> the bit mask forms.
>>
>> Fixes: 408413051144 ("x86/cpu: Drop unused X86_VENDOR_* values")
>> Fixes: 0cd074144cbb ("x86/cpu: Renumber X86_VENDOR_* to form a bitmap")
>> Signed-off-by: Jan Beulich <[email protected]>
>
> I see you've committed this, but it's not a viable approach.
Well, depends on what the goal is, I suppose: It's viable to restore previously
regressed behavior. We have been in agreement, I think, that vMCE itself is
doomed.
>> --- a/xen/arch/x86/cpu/mcheck/mce.c
>> +++ b/xen/arch/x86/cpu/mcheck/mce.c
>> @@ -858,6 +858,22 @@ void *x86_mcinfo_reserve(struct mc_info
>> return mic_index;
>> }
>>
>> +/* Linux values need using when reporting CPU vendors to the outside. */
>> +static unsigned int xen2linux_vendor(unsigned int vendor)
>> +{
>> + switch ( vendor )
>> + {
>> + case X86_VENDOR_INTEL: return 0;
>> + case X86_VENDOR_AMD: return 2;
>> + case X86_VENDOR_CENTAUR: return 5;
>> + case X86_VENDOR_HYGON: return 9;
>> + case X86_VENDOR_SHANGHAI: return 10; /* X86_VENDOR_ZHAOXIN */
>> + default: break;
>> + }
>> +
>> + return 0xff; /* X86_VENDOR_UNKNOWN */
>> +}
>
> The Linux x86 maintainers say that the numbers are internal and not an
> ABI, even they haven't changed since 2008.
Yet the numbers are propagated to the outside world there as well, aiui.
>> @@ -866,7 +882,7 @@ static void x86_mcinfo_apei_save(
>> memset(&m, 0, sizeof(struct mce));
>>
>> m.cpu = mc_global->mc_coreid;
>> - m.cpuvendor = boot_cpu_data.x86_vendor;
>> + m.cpuvendor = xen2linux_vendor(boot_cpu_data.x86_vendor);
>
> This is an internal Xen structure. This hunk needs reverting.
It's not, unless I'm missing something. The struct is passed to
apei_write_mce(), to then be passed to erst_write(). ERST isn't Xen-
internal anymore, is it?
Jan
