Re: [PATCH] KVM: nVMX: Don't expose TSC scaling to L1 when on Hyper-V

From: Sean Christopherson <seanjc@google.com>
To: Anirudh Rayabharam <anrayabh@linux.microsoft.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>,
	Vitaly Kuznetsov <vkuznets@redhat.com>,
	Wanpeng Li <wanpengli@tencent.com>,
	Jim Mattson <jmattson@google.com>, Joerg Roedel <joro@8bytes.org>,
	Thomas Gleixner <tglx@linutronix.de>,
	Ingo Molnar <mingo@redhat.com>, Borislav Petkov <bp@alien8.de>,
	Dave Hansen <dave.hansen@linux.intel.com>,
	x86@kernel.org, "H. Peter Anvin" <hpa@zytor.com>,
	Ilias Stamatis <ilstam@amazon.com>,
	Maxim Levitsky <mlevitsk@redhat.com>,
	mail@anirudhrb.com, kumarpraveen@linux.microsoft.com,
	wei.liu@kernel.org, robert.bradford@intel.com,
	liuwe@microsoft.com, kvm@vger.kernel.org,
	linux-kernel@vger.kernel.org
Subject: Re: [PATCH] KVM: nVMX: Don't expose TSC scaling to L1 when on Hyper-V
Date: Mon, 13 Jun 2022 16:41:27 +0000	[thread overview]
Message-ID: <Yqdot/66J7RMRw2B@google.com> (raw)
In-Reply-To: <20220613161611.3567556-1-anrayabh@linux.microsoft.com>

On Mon, Jun 13, 2022, Anirudh Rayabharam wrote:
> As per the comments in arch/x86/kvm/vmx/evmcs.h, TSC multiplier field is

It's not just the comments, it's also the code.  It would be helpful to call out
in the changelog that KVM clears unsupported controls via evmcs_sanitize_exec_ctrls()
when using eVMCS.  

> currently not supported in EVMCS. As a result, there is no TSC scaling
> support when KVM is running on Hyper-V i.e. kvm_has_tsc_control is
> false.
> 
> However, in nested_vmx_setup_ctls_msrs(), TSC scaling is exposed to L1.
> When L1 tries to launch an L2 guest, vmcs12 has TSC scaling enabled.
> This propagates to vmcs02. But KVM doesn't set the TSC multiplier value
> because kvm_has_tsc_control is false. Due to this, VM entry for L2 guest
> fails. (VM entry fails if "use TSC scaling" is 1 and TSC multiplier is 0.)
>
> To fix, expose TSC scaling to L1 only if kvm_has_tsc_control.
> 
> Fixes: d041b5ea93352 ("KVM: nVMX: Enable nested TSC scaling")
> Signed-off-by: Anirudh Rayabharam <anrayabh@linux.microsoft.com>
> ---
>  arch/x86/kvm/vmx/nested.c | 6 ++++--
>  1 file changed, 4 insertions(+), 2 deletions(-)
> 
> diff --git a/arch/x86/kvm/vmx/nested.c b/arch/x86/kvm/vmx/nested.c
> index f5cb18e00e78..d773ddc6422b 100644
> --- a/arch/x86/kvm/vmx/nested.c
> +++ b/arch/x86/kvm/vmx/nested.c
> @@ -6656,6 +6656,9 @@ void nested_vmx_setup_ctls_msrs(struct nested_vmx_msrs *msrs, u32 ept_caps)
>  		      msrs->secondary_ctls_low,
>  		      msrs->secondary_ctls_high);
>  
> +	if (!kvm_has_tsc_control)
> +		msrs->secondary_ctls_high &= ~SECONDARY_EXEC_TSC_SCALING;

I would much rather we fix the root of the problem and not play whack-a-mole,
e.g. all of the other controls that aren't supported by eVMCS have the same bug,.

nested_vmx_setup_ctls_msrs() should use vmcs_config to get the base MSR values,
not read the MSRs from hardware.  And it's not just eVMCS, e.g. the manipulation
of VM_{ENTRY,EXIT}_IA32_PERF_GLOBAL_CTRL for a CPU errata isn't handled either.

>  	msrs->secondary_ctls_low = 0;
>  	msrs->secondary_ctls_high &=
>  		SECONDARY_EXEC_DESC |
> @@ -6667,8 +6670,7 @@ void nested_vmx_setup_ctls_msrs(struct nested_vmx_msrs *msrs, u32 ept_caps)
>  		SECONDARY_EXEC_RDRAND_EXITING |
>  		SECONDARY_EXEC_ENABLE_INVPCID |
>  		SECONDARY_EXEC_RDSEED_EXITING |
> -		SECONDARY_EXEC_XSAVES |
> -		SECONDARY_EXEC_TSC_SCALING;
> +		SECONDARY_EXEC_XSAVES;
>  
>  	/*
>  	 * We can emulate "VMCS shadowing," even if the hardware
> -- 
> 2.34.1
> 