2023-05-03 04:26:05

by Mingwei Zhang

Subject: [PATCH] KVM: VMX: add MSR_IA32_TSX_CTRL into msrs_to_save

Add MSR_IA32_TSX_CTRL into msrs_to_save[] to explicitly tell userspace to
save/restore the register value during migration. Missing this may cause
userspace that relies on KVM ioctl(KVM_GET_MSR_INDEX_LIST) fail to port the
value to the target VM.

Fixes: b07a5c53d42a ("KVM: vmx: use MSR_IA32_TSX_CTRL to hard-disable TSX on guest that lack it")
Reported-by: Jim Mattson <[email protected]>
Signed-off-by: Mingwei Zhang <[email protected]>
---
arch/x86/kvm/x86.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index 237c483b1230..2236cfee4b7a 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -1431,7 +1431,7 @@ static const u32 msrs_to_save_base[] = {
#endif
MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
MSR_IA32_FEAT_CTL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
- MSR_IA32_SPEC_CTRL,
+ MSR_IA32_SPEC_CTRL, MSR_IA32_TSX_CTRL,
MSR_IA32_RTIT_CTL, MSR_IA32_RTIT_STATUS, MSR_IA32_RTIT_CR3_MATCH,
MSR_IA32_RTIT_OUTPUT_BASE, MSR_IA32_RTIT_OUTPUT_MASK,
MSR_IA32_RTIT_ADDR0_A, MSR_IA32_RTIT_ADDR0_B,
--
2.40.1.495.gc816e09b53d-goog

2023-05-03 16:28:19

by Sean Christopherson

Subject: Re: [PATCH] KVM: VMX: add MSR_IA32_TSX_CTRL into msrs_to_save

On Wed, May 03, 2023, Mingwei Zhang wrote:
> Add MSR_IA32_TSX_CTRL into msrs_to_save[] to explicitly tell userspace to
> save/restore the register value during migration. Missing this may cause
> userspace that relies on KVM ioctl(KVM_GET_MSR_INDEX_LIST) fail to port the
> value to the target VM.
>
> Fixes: b07a5c53d42a ("KVM: vmx: use MSR_IA32_TSX_CTRL to hard-disable TSX on guest that lack it")
> Reported-by: Jim Mattson <[email protected]>
> Signed-off-by: Mingwei Zhang <[email protected]>
> ---
> arch/x86/kvm/x86.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
> index 237c483b1230..2236cfee4b7a 100644
> --- a/arch/x86/kvm/x86.c
> +++ b/arch/x86/kvm/x86.c
> @@ -1431,7 +1431,7 @@ static const u32 msrs_to_save_base[] = {
> #endif
> MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
> MSR_IA32_FEAT_CTL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
> - MSR_IA32_SPEC_CTRL,
> + MSR_IA32_SPEC_CTRL, MSR_IA32_TSX_CTRL,
> MSR_IA32_RTIT_CTL, MSR_IA32_RTIT_STATUS, MSR_IA32_RTIT_CR3_MATCH,
> MSR_IA32_RTIT_OUTPUT_BASE, MSR_IA32_RTIT_OUTPUT_MASK,
> MSR_IA32_RTIT_ADDR0_A, MSR_IA32_RTIT_ADDR0_B,
> --

Hmm, KVM shouldn't report the MSR if it can't be written by the guest. Over-
reporting won't cause functional issues, and the odds of the MSR existing but not
being reported in ARCH_CAPILIBITES are basically zilch, but IMO it's worth adding
the check if only to document when the MSRs is fully supported.

diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index e7f78fe79b32..d8608c6753ff 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -7152,6 +7152,10 @@ static void kvm_probe_msr_to_save(u32 msr_index)
if (!kvm_cpu_cap_has(X86_FEATURE_XFD))
return;
break;
+ case MSR_IA32_TSX_CTRL:
+ if (!(kvm_get_arch_capabilities() & ARCH_CAP_TSX_CTRL_MSR))
+ return;
+ break;
default:
break;
}

2023-05-03 17:04:38

by Jim Mattson

Subject: Re: [PATCH] KVM: VMX: add MSR_IA32_TSX_CTRL into msrs_to_save

On Wed, May 3, 2023 at 9:25 AM Sean Christopherson <[email protected]> wrote:
>
> On Wed, May 03, 2023, Mingwei Zhang wrote:
> > Add MSR_IA32_TSX_CTRL into msrs_to_save[] to explicitly tell userspace to
> > save/restore the register value during migration. Missing this may cause
> > userspace that relies on KVM ioctl(KVM_GET_MSR_INDEX_LIST) fail to port the
> > value to the target VM.
> >
> > Fixes: b07a5c53d42a ("KVM: vmx: use MSR_IA32_TSX_CTRL to hard-disable TSX on guest that lack it")
> > Reported-by: Jim Mattson <[email protected]>
> > Signed-off-by: Mingwei Zhang <[email protected]>
> > ---
> > arch/x86/kvm/x86.c | 2 +-
> > 1 file changed, 1 insertion(+), 1 deletion(-)
> >
> > diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
> > index 237c483b1230..2236cfee4b7a 100644
> > --- a/arch/x86/kvm/x86.c
> > +++ b/arch/x86/kvm/x86.c
> > @@ -1431,7 +1431,7 @@ static const u32 msrs_to_save_base[] = {
> > #endif
> > MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
> > MSR_IA32_FEAT_CTL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
> > - MSR_IA32_SPEC_CTRL,
> > + MSR_IA32_SPEC_CTRL, MSR_IA32_TSX_CTRL,
> > MSR_IA32_RTIT_CTL, MSR_IA32_RTIT_STATUS, MSR_IA32_RTIT_CR3_MATCH,
> > MSR_IA32_RTIT_OUTPUT_BASE, MSR_IA32_RTIT_OUTPUT_MASK,
> > MSR_IA32_RTIT_ADDR0_A, MSR_IA32_RTIT_ADDR0_B,
> > --
>
> Hmm, KVM shouldn't report the MSR if it can't be written by the guest. Over-

I think you mean to say that KVM shouldn't report the MSR if it can't
be written by *any* guest. KVM_GET_MSR_INDEX_LIST is a device ioctl,
so it isn't capable of filtering out MSRs that can't be written by
*the* guest, for some occurrence of "the."

> reporting won't cause functional issues, and the odds of the MSR existing but not
> being reported in ARCH_CAPILIBITES are basically zilch, but IMO it's worth adding
> the check if only to document when the MSRs is fully supported.
>
> diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
> index e7f78fe79b32..d8608c6753ff 100644
> --- a/arch/x86/kvm/x86.c
> +++ b/arch/x86/kvm/x86.c
> @@ -7152,6 +7152,10 @@ static void kvm_probe_msr_to_save(u32 msr_index)
> if (!kvm_cpu_cap_has(X86_FEATURE_XFD))
> return;
> break;
> + case MSR_IA32_TSX_CTRL:
> + if (!(kvm_get_arch_capabilities() & ARCH_CAP_TSX_CTRL_MSR))
> + return;
> + break;
> default:
> break;
> }

2023-05-03 17:49:50

by Sean Christopherson

Subject: Re: [PATCH] KVM: VMX: add MSR_IA32_TSX_CTRL into msrs_to_save

On Wed, May 03, 2023, Jim Mattson wrote:
> On Wed, May 3, 2023 at 9:25 AM Sean Christopherson <[email protected]> wrote:
> >
> > On Wed, May 03, 2023, Mingwei Zhang wrote:
> > > Add MSR_IA32_TSX_CTRL into msrs_to_save[] to explicitly tell userspace to
> > > save/restore the register value during migration. Missing this may cause
> > > userspace that relies on KVM ioctl(KVM_GET_MSR_INDEX_LIST) fail to port the
> > > value to the target VM.
> > >
> > > Fixes: b07a5c53d42a ("KVM: vmx: use MSR_IA32_TSX_CTRL to hard-disable TSX on guest that lack it")
> > > Reported-by: Jim Mattson <[email protected]>
> > > Signed-off-by: Mingwei Zhang <[email protected]>
> > > ---
> > > arch/x86/kvm/x86.c | 2 +-
> > > 1 file changed, 1 insertion(+), 1 deletion(-)
> > >
> > > diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
> > > index 237c483b1230..2236cfee4b7a 100644
> > > --- a/arch/x86/kvm/x86.c
> > > +++ b/arch/x86/kvm/x86.c
> > > @@ -1431,7 +1431,7 @@ static const u32 msrs_to_save_base[] = {
> > > #endif
> > > MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
> > > MSR_IA32_FEAT_CTL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
> > > - MSR_IA32_SPEC_CTRL,
> > > + MSR_IA32_SPEC_CTRL, MSR_IA32_TSX_CTRL,
> > > MSR_IA32_RTIT_CTL, MSR_IA32_RTIT_STATUS, MSR_IA32_RTIT_CR3_MATCH,
> > > MSR_IA32_RTIT_OUTPUT_BASE, MSR_IA32_RTIT_OUTPUT_MASK,
> > > MSR_IA32_RTIT_ADDR0_A, MSR_IA32_RTIT_ADDR0_B,
> > > --
> >
> > Hmm, KVM shouldn't report the MSR if it can't be written by the guest. Over-
>
> I think you mean to say that KVM shouldn't report the MSR if it can't
> be written by *any* guest. KVM_GET_MSR_INDEX_LIST is a device ioctl,
> so it isn't capable of filtering out MSRs that can't be written by
> *the* guest, for some occurrence of "the."

Doh, yes, "the guest" was a handwavy reference to any/all guests.