On Fri, Feb 21, 2020 at 04:16:31PM +0100, Vitaly Kuznetsov wrote:
> Sean Christopherson <sean.j.christopherson@xxxxxxxxx> writes:
>
> > diff --git a/arch/x86/kvm/vmx/vmx.h b/arch/x86/kvm/vmx/vmx.h
> > index a4f7f737c5d4..70eafa88876a 100644
> > --- a/arch/x86/kvm/vmx/vmx.h
> > +++ b/arch/x86/kvm/vmx/vmx.h
> > @@ -449,7 +449,7 @@ static inline void vmx_segment_cache_clear(struct vcpu_vmx *vmx)
> > static inline u32 vmx_vmentry_ctrl(void)
> > {
> > u32 vmentry_ctrl = vmcs_config.vmentry_ctrl;
> > - if (pt_mode == PT_MODE_SYSTEM)
> > + if (vmx_pt_mode_is_system())
>
> Just wondering, would it rather be better to say
> if (!vmx_pt_supported())
> here?
>
> > vmentry_ctrl &= ~(VM_ENTRY_PT_CONCEAL_PIP |
> > VM_ENTRY_LOAD_IA32_RTIT_CTL);
> > /* Loading of EFER and PERF_GLOBAL_CTRL are toggled dynamically */
> > @@ -460,7 +460,7 @@ static inline u32 vmx_vmentry_ctrl(void)
> > static inline u32 vmx_vmexit_ctrl(void)
> > {
> > u32 vmexit_ctrl = vmcs_config.vmexit_ctrl;
> > - if (pt_mode == PT_MODE_SYSTEM)
> > + if (vmx_pt_mode_is_system())
>
> ... and here? I.e. to cover the currently unsupported 'host-only' mode.
Hmm, good question. I don't think so? On VM-Enter, RTIT_CTL would need to
be loaded to disable PT. Clearing RTIT_CTL on VM-Exit would be redundant
at that point[1]. And AIUI, the PIP for VM-Enter/VM-Exit isn't needed
because there is no context switch from the decoder's perspective.
Note, the original upstreaming series also used "pt_mode == PT_MODE_SYSTEM"
logic for this check when "host-only mode" was supported[2].
[1] Arguably, KVM should use the VM-Exit MSR load list to atomically
reenable tracing, but that's feedback for a non-existence patch :-).
[2] https://patchwork.kernel.org/patch/10104533/
>
> > vmexit_ctrl &= ~(VM_EXIT_PT_CONCEAL_PIP |
> > VM_EXIT_CLEAR_IA32_RTIT_CTL);
> > /* Loading of EFER and PERF_GLOBAL_CTRL are toggled dynamically */
