Re: [PATCH 05/15] KVM: VMX: Disable preemption when probing user return MSRs

[Maxim Levitsky <mlevitsk@redhat.com>]

On Tue, 2021-05-04 at 10:17 -0700, Sean Christopherson wrote:
> Disable preemption when probing a user return MSR via RDSMR/WRMSR.  If
> the MSR holds a different value per logical CPU, the WRMSR could corrupt
> the host's value if KVM is preempted between the RDMSR and WRMSR, and
> then rescheduled on a different CPU.
> 
> Opportunistically land the helper in common x86, SVM will use the helper
> in a future commit.
> 
> Fixes: 4be534102624 ("KVM: VMX: Initialize vmx->guest_msrs[] right after allocation")
> Cc: stable@vger.kernel.org
> Cc: Xiaoyao Li <xiaoyao.li@intel.com>
> Signed-off-by: Sean Christopherson <seanjc@google.com>
> ---
>  arch/x86/include/asm/kvm_host.h |  1 +
>  arch/x86/kvm/vmx/vmx.c          |  5 +----
>  arch/x86/kvm/x86.c              | 16 ++++++++++++++++
>  3 files changed, 18 insertions(+), 4 deletions(-)
> 
> diff --git a/arch/x86/include/asm/kvm_host.h b/arch/x86/include/asm/kvm_host.h
> index 3e5fc80a35c8..a02c9bf3f7f1 100644
> --- a/arch/x86/include/asm/kvm_host.h
> +++ b/arch/x86/include/asm/kvm_host.h
> @@ -1778,6 +1778,7 @@ int kvm_pv_send_ipi(struct kvm *kvm, unsigned long ipi_bitmap_low,
>  		    unsigned long icr, int op_64_bit);
>  
>  void kvm_define_user_return_msr(unsigned index, u32 msr);
> +int kvm_probe_user_return_msr(u32 msr);
>  int kvm_set_user_return_msr(unsigned index, u64 val, u64 mask);
>  
>  u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc);
> diff --git a/arch/x86/kvm/vmx/vmx.c b/arch/x86/kvm/vmx/vmx.c
> index 99591e523b47..990ee339a05f 100644
> --- a/arch/x86/kvm/vmx/vmx.c
> +++ b/arch/x86/kvm/vmx/vmx.c
> @@ -6914,12 +6914,9 @@ static int vmx_create_vcpu(struct kvm_vcpu *vcpu)
>  
>  	for (i = 0; i < ARRAY_SIZE(vmx_uret_msrs_list); ++i) {
>  		u32 index = vmx_uret_msrs_list[i];
> -		u32 data_low, data_high;
>  		int j = vmx->nr_uret_msrs;
>  
> -		if (rdmsr_safe(index, &data_low, &data_high) < 0)
> -			continue;
> -		if (wrmsr_safe(index, data_low, data_high) < 0)
> +		if (kvm_probe_user_return_msr(index))
>  			continue;
>  
>  		vmx->guest_uret_msrs[j].slot = i;
> diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
> index 3bf52ba5f2bb..e304447be42d 100644
> --- a/arch/x86/kvm/x86.c
> +++ b/arch/x86/kvm/x86.c
> @@ -339,6 +339,22 @@ static void kvm_on_user_return(struct user_return_notifier *urn)
>  	}
>  }
>  
> +int kvm_probe_user_return_msr(u32 msr)
> +{
> +	u64 val;
> +	int ret;
> +
> +	preempt_disable();
> +	ret = rdmsrl_safe(msr, &val);
> +	if (ret)
> +		goto out;
> +	ret = wrmsrl_safe(msr, val);
> +out:
> +	preempt_enable();
> +	return ret;
> +}
> +EXPORT_SYMBOL_GPL(kvm_probe_user_return_msr);
> +
>  void kvm_define_user_return_msr(unsigned slot, u32 msr)
>  {
>  	BUG_ON(slot >= KVM_MAX_NR_USER_RETURN_MSRS);

Reviewed-by: Maxim Levitsky <mlevitsk@redhat.com>

One note though: every time we probe for a MSR existance via
checking for a #GP on write, we risk getting nonsense results 
if the L1 has ignore_msrs=1.

Thus if possible to use the CPUID instead, 
that would be preferred.

Best regards,
	Maxim Levitsky