Re: [RFC PATCH v4 10/36] i386/kvm: Move architectural CPUID leaf generation to separate helper

[Xiaoyao Li]

On 5/13/2022 1:48 AM, Isaku Yamahata wrote:
> On Thu, May 12, 2022 at 11:17:37AM +0800,
> Xiaoyao Li <xiaoyao.li@intel.com> wrote:
>
> > diff --git a/target/i386/kvm/kvm_i386.h b/target/i386/kvm/kvm_i386.h
> > index b434feaa6b1d..5c7972f617e8 100644
> > --- a/target/i386/kvm/kvm_i386.h
> > +++ b/target/i386/kvm/kvm_i386.h
> > @@ -24,6 +24,10 @@
> >   #define kvm_ioapic_in_kernel() \
> >       (kvm_irqchip_in_kernel() && !kvm_irqchip_is_split())
> >   
> > +#define KVM_MAX_CPUID_ENTRIES  100
>
> In Linux side, the value was bumped to 256.  Opportunistically let's make it
> same.
>
> 3f4e3eb417b1 KVM: x86: bump KVM_MAX_CPUID_ENTRIES

I don't think so.

In KVM, KVM_MAX_CPUID_ENTRIES is used to guard IOCTL 
KVM_SET_CPUID/KVM_SET_CPUID2/KVM_GET_SUPPORTED_CPUID/KVM_GET_EMULATED_CPUID, 
that KVM handles at most
the number of KVM_MAX_CPUID_ENTRIES entries.

However, in QEMU, KVM_MAX_CPUID_ENTRIES is used as the maximum total 
number of CPUID entries that generated by QEMU. It's used to guard the 
number in kvm_x86_arch_cpuid().

I think we can increase the number when we actually hit the check in 
kvm_x86_arch_cupid().

> > +uint32_t kvm_x86_arch_cpuid(CPUX86State *env, struct kvm_cpuid_entry2 *entries,
> > +                            uint32_t cpuid_i);
> > +
> >   #else
> >   
> >   #define kvm_pit_in_kernel()      0
> > --
> > 2.27.0