On Fri, May 07, 2021, Reiji Watanabe wrote:
> > -static void vmx_setup_uret_msr(struct vcpu_vmx *vmx, unsigned int msr)
> > +static void vmx_setup_uret_msr(struct vcpu_vmx *vmx, unsigned int msr,
> > + bool load_into_hardware)
> > {
> > - struct vmx_uret_msr tmp;
> > - int from, to;
> > + struct vmx_uret_msr *uret_msr;
> >
> > - from = __vmx_find_uret_msr(vmx, msr);
> > - if (from < 0)
> > + uret_msr = vmx_find_uret_msr(vmx, msr);
> > + if (!uret_msr)
> > return;
> > - to = vmx->nr_active_uret_msrs++;
> >
> > - tmp = vmx->guest_uret_msrs[to];
> > - vmx->guest_uret_msrs[to] = vmx->guest_uret_msrs[from];
> > - vmx->guest_uret_msrs[from] = tmp;
> > + uret_msr->load_into_hardware = load_into_hardware;
> > }
> >
> > /*
> > @@ -1785,30 +1781,36 @@ static void vmx_setup_uret_msr(struct vcpu_vmx *vmx, unsigned int msr)
> > */
> > static void setup_msrs(struct vcpu_vmx *vmx)
> > {
> > - vmx->guest_uret_msrs_loaded = false;
> > - vmx->nr_active_uret_msrs = 0;
> > #ifdef CONFIG_X86_64
> > + bool load_syscall_msrs;
> > +
> > /*
> > * The SYSCALL MSRs are only needed on long mode guests, and only
> > * when EFER.SCE is set.
> > */
> > - if (is_long_mode(&vmx->vcpu) && (vmx->vcpu.arch.efer & EFER_SCE)) {
> > - vmx_setup_uret_msr(vmx, MSR_STAR);
> > - vmx_setup_uret_msr(vmx, MSR_LSTAR);
> > - vmx_setup_uret_msr(vmx, MSR_SYSCALL_MASK);
> > - }
> > + load_syscall_msrs = is_long_mode(&vmx->vcpu) &&
> > + (vmx->vcpu.arch.efer & EFER_SCE);
> > +
> > + vmx_setup_uret_msr(vmx, MSR_STAR, load_syscall_msrs);
> > + vmx_setup_uret_msr(vmx, MSR_LSTAR, load_syscall_msrs);
> > + vmx_setup_uret_msr(vmx, MSR_SYSCALL_MASK, load_syscall_msrs);
> > #endif
> > - if (update_transition_efer(vmx))
> > - vmx_setup_uret_msr(vmx, MSR_EFER);
> > + vmx_setup_uret_msr(vmx, MSR_EFER, update_transition_efer(vmx));
> >
> > - if (guest_cpuid_has(&vmx->vcpu, X86_FEATURE_RDTSCP) ||
> > - guest_cpuid_has(&vmx->vcpu, X86_FEATURE_RDPID))
> > - vmx_setup_uret_msr(vmx, MSR_TSC_AUX);
> > + vmx_setup_uret_msr(vmx, MSR_TSC_AUX,
> > + guest_cpuid_has(&vmx->vcpu, X86_FEATURE_RDTSCP) ||
> > + guest_cpuid_has(&vmx->vcpu, X86_FEATURE_RDPID));
>
> Shouldn't vmx_setup_uret_msr(,MSR_TSC_AUX,) be called to update
> the new flag load_into_hardware for MSR_TSC_AUX when CPUID
> (X86_FEATURE_RDTSCP/X86_FEATURE_RDPID) of the vCPU is updated ?
Yes. I have a patch in the massive vCPU RESET/INIT series to do exactly that.
I honestly can't remember if there was a dependency that "required" the fix to
be buried in the middle of that series. I suspect not; I'm guessing I just
didn't think it was worth backporting to stable kernels and so didn't prioritize
hoisting the patch out of that mess.
https://lkml.kernel.org/r/20210424004645.3950558-34-seanjc@xxxxxxxxxx
