On Mon, 2020-09-28 at 21:56 -0700, Sean Christopherson wrote:
> I think you want "extract", not "abstract".
>
>
> On Fri, Jul 31, 2020 at 10:42:19AM +0800, Robert Hoo wrote:
> > Add below functions, whose aggregation equals
> > kvm_update_cpuid_runtime() and
> > kvm_vcpu_after_set_cpuid().
> >
> > void kvm_osxsave_update_cpuid(struct kvm_vcpu *vcpu, bool set)
> > void kvm_pke_update_cpuid(struct kvm_vcpu *vcpu, bool set)
> > void kvm_apic_base_update_cpuid(struct kvm_vcpu *vcpu, bool set)
> > void kvm_mwait_update_cpuid(struct kvm_vcpu *vcpu, bool set)
> > void kvm_xcr0_update_cpuid(struct kvm_vcpu *vcpu)
> > static void kvm_pv_unhalt_update_cpuid(struct kvm_vcpu *vcpu)
> > static void kvm_update_maxphyaddr(struct kvm_vcpu *vcpu)
> > static void kvm_update_lapic_timer_mode(struct kvm_vcpu *vcpu)
> >
> > And, for some among the above, avoid double check set or clear
> > inside function
> > body, but provided by caller.
> >
> > Signed-off-by: Robert Hoo <robert.hu@xxxxxxxxxxxxxxx>
> > ---
> > arch/x86/kvm/cpuid.c | 99
> > ++++++++++++++++++++++++++++++++++++++++++++++++++++
> > arch/x86/kvm/cpuid.h | 6 ++++
> > 2 files changed, 105 insertions(+)
> >
> > diff --git a/arch/x86/kvm/cpuid.c b/arch/x86/kvm/cpuid.c
> > index 7d92854..efa7182 100644
> > --- a/arch/x86/kvm/cpuid.c
> > +++ b/arch/x86/kvm/cpuid.c
> > @@ -73,6 +73,105 @@ static int kvm_check_cpuid(struct kvm_vcpu
> > *vcpu)
> > return 0;
> > }
> >
> > +void kvm_osxsave_update_cpuid(struct kvm_vcpu *vcpu, bool set)
> > +{
> > + struct kvm_cpuid_entry2 *best;
>
> I vote that we opportunistically move away from the "best"
> terminology. Either
> there's a matching entry or there's not. Using "e" would probably
> shave a few
> lines of code.
>
> > +
> > + best = kvm_find_cpuid_entry(vcpu, 1, 0);
> > +
> > + if (best && boot_cpu_has(X86_FEATURE_XSAVE)) {
>
> Braces not needed. We should also check boot_cpu_has() first, it's
> constant
> time and maybe even in the cache, whereas finding CPUID entries is
> linear
> time and outright slow.
>
> Actually, can you add a helper to handle this? With tht
> boot_cpu_has() check
> outside of the helper? That'll allow the helper to be used for more
> features,
> and will force checking boot_cpu_has() first. Hmm, and not all of
> the callers
> will need the boot_cpu_has() check, e.g. toggling PKE from
> kvm_set_cr4()
> doesn't need to be guarded because KVM disallows setting PKE if it's
> not
> supported by the host.
Do you mean because in kvm_set_cr4(), it has kvm_valid_cr4(vcpu, cr4)
check first?
Then how about the other 2 callers of kvm_pke_update_cpuid()?
enter_smm() -- I think it can ommit boot_cpu_has() check as well.
because it unconditionally cleared all CR4 bit before calls
kvm_set_cr4().
__set_sregs() -- looks like it doesn't valid host PKE status before
call kvm_pke_update_cpuid(). Can I ommit boot_cpu_has() as well?
So, I don't think kvm_pke_update_cpuid() can leverage the helper. Am I
right?
>
> static inline void guest_cpuid_change(struct kvm_vcpu *vcpu, u32
> function,
> u32 index, unsigned int feature,
> bool set)
> {
> struct kvm_cpuid_entry2 *e = kvm_find_cpuid_entry(vcpu,
> function, index);
>
> if (e)
> cpuid_entry_change(best, X86_FEATURE_OSXSAVE, set);
> }
Thanks Sean, I'm going to have this helper in v2 and have your signed-
off-by.
>
> > + cpuid_entry_change(best, X86_FEATURE_OSXSAVE, set);
> > + }
> > +}
> > +
> > +void kvm_pke_update_cpuid(struct kvm_vcpu *vcpu, bool set)
> > +{
> > + struct kvm_cpuid_entry2 *best;
> > +
> > + best = kvm_find_cpuid_entry(vcpu, 7, 0);
> > +
> > + if (best && boot_cpu_has(X86_FEATURE_PKU)) {
> > + cpuid_entry_change(best, X86_FEATURE_OSPKE, set);
> > + }
> > +}
> > +
> > +void kvm_xcr0_update_cpuid(struct kvm_vcpu *vcpu)
> > +{
> > + struct kvm_cpuid_entry2 *best;a
> > +
> > + best = kvm_find_cpuid_entry(vcpu, 0xD, 0);
> > + if (!best) {
> > + vcpu->arch.guest_supported_xcr0 = 0;
> > + } else {
> > + vcpu->arch.guest_supported_xcr0 =
> > + (best->eax | ((u64)best->edx << 32)) &
> > supported_xcr0;
> > + best->ebx = xstate_required_size(vcpu->arch.xcr0,
> > false);
> > + }
> > +
> > + best = kvm_find_cpuid_entry(vcpu, 0xD, 1);
> > + if (!best)
> > + return;
> > + if (cpuid_entry_has(best, X86_FEATURE_XSAVES) ||
> > + cpuid_entry_has(best,
> > X86_FEATURE_XSAVEC))
>
> Indentation should be aligned, e.g.
>
> if (cpuid_entry_has(best, X86_FEATURE_XSAVES) ||
> cpuid_entry_has(best, X86_FEATURE_XSAVEC))
>
>
> > + best->ebx = xstate_required_size(vcpu->arch.xcr0,
> > true);
> > +}
> > +
> > +static void kvm_pv_unhalt_update_cpuid(struct kvm_vcpu *vcpu)
> > +{
> > + struct kvm_cpuid_entry2 *best;
> > +
> > + best = kvm_find_cpuid_entry(vcpu, KVM_CPUID_FEATURES, 0);
> > + if (kvm_hlt_in_guest(vcpu->kvm) && best &&
> > + (best->eax & (1 <<
> > KVM_FEATURE_PV_UNHALT)))
> > + best->eax &= ~(1 << KVM_FEATURE_PV_UNHALT);
> > +}
> > +
> > +void kvm_mwait_update_cpuid(struct kvm_vcpu *vcpu, bool set)
> > +{
> > + struct kvm_cpuid_entry2 *best;
> > +
> > + best = kvm_find_cpuid_entry(vcpu, 0x1, 0);
> > + if (best)
> > + cpuid_entry_change(best, X86_FEATURE_MWAIT, set);
> > +}
> > +
> > +static void kvm_update_maxphyaddr(struct kvm_vcpu *vcpu)
> > +{
> > +
> > + /* Note, maxphyaddr must be updated before tdp_level. */
> > + vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
> > + vcpu->arch.tdp_level = kvm_x86_ops.get_tdp_level(vcpu);
> > + kvm_mmu_reset_context(vcpu);
> > +}
> > +
> > +void kvm_apic_base_update_cpuid(struct kvm_vcpu *vcpu, bool set)
> > +{
> > + struct kvm_cpuid_entry2 *best;
> > +
> > + best = kvm_find_cpuid_entry(vcpu, 1, 0);
> > + if (!best)
> > + return;
> > +
> > + cpuid_entry_change(best, X86_FEATURE_APIC, set);
> > +}
> > +
> > +static void kvm_update_lapic_timer_mode(struct kvm_vcpu *vcpu)
> > +{
> > + struct kvm_cpuid_entry2 *best;
> > + struct kvm_lapic *apic = vcpu->arch.apic;
> > +
> > + best = kvm_find_cpuid_entry(vcpu, 1, 0);
> > + if (!best)
> > + return;
> > +
> > + if (apic) {
>
> Check apic before the lookup.
>
> > + if (cpuid_entry_has(best,
> > X86_FEATURE_TSC_DEADLINE_TIMER))
> > + apic->lapic_timer.timer_mode_mask = 3 << 17;
> > + else
> > + apic->lapic_timer.timer_mode_mask = 1 << 17;
> > + }
> > +}
> > +
> > void kvm_update_cpuid_runtime(struct kvm_vcpu *vcpu)
> > {
...
> > --
> > 1.8.3.1
> >
