On 26.06.2025 17:45, Jan Beulich wrote: > On 26.06.2025 17:31, Andrew Cooper wrote: >> On 16/06/2025 1:59 pm, Jan Beulich wrote: >>> + if ( !raw_cpu_policy.feat.erms ) >> >> This wants to be the host policy, not the raw policy. That's why >> `cpuid=no-erms` isn't working in the way you'd expect. >> >> cpuid=no-$foo means "Xen should behave as if $foo wasn't reported by >> hardware", and that includes not giving it to guests by default. > > Hmm, interesting. That's definitely not the meaning I give this. To me it > means merely Xen shouldn't use the feature (with an impact on guests only > when the feature in hardware is required to surface it to guests). I > don't think we have the precise meaning of this option written down > anywhere?
Then again I notice what you ask for is in line with uses of cpu_has_* (where available) elsewhere (e.g. in your "x86/cpu-policy: Simplify logic in guest_common_default_feature_adjustments()"), with calculate_host_policy() simply copying boot_cpu_data.x86_capability into the host policy. So yes, I guess I'll need to adjust. Ftaod, this still leaves open what exact meaning "cpuid=no-..." ought to have. While kind of an example in the opposite direction, consider e.g. "cpuid=no-lm": This cannot sensibly have any effect on Xen itself. It could plausibly mean to permit only 32-bit guests. Jan
