On 7/11/25 08:40, Xiaoyao Li wrote:
On 7/11/2025 2:35 PM, Paolo Bonzini wrote:
Il ven 11 lug 2025, 04:18 Xiaoyao Li <xiaoyao...@intel.com> ha scritto:
Besides, do we need to call host_cpu_max_instance_init() for the case of
xcc->max_features, like what has been done for kvm and hvf?
I am intentionally skipping that because it would not have any effect and
there is no equivalent to KVM_GET_SUPPORTED_CPUID2 implemented for those
accelerators.
I meant host_cpu_max_instance_init(), not the upper function like
kvm_cpu_max_instance_init() or hvf_cpu_max_instance_init().
host_cpu_max_instance_init() is for the case "-cpu max/host", which not
only sets "vendor" to the host value, but also the "host-phys-bits",
"family" "model" "stepping" and "model-id"
Ah, thanks - it also does not have any effect so I didn't think about
it. But the separation between host_cpu_instance_init() and
host_cpu_max_instance_init() is confusing. I'll send a patch to merge
them into one function, which should resolve your doubt.
Paolo
