linux-kernel - Re: [PATCH] KVM: nSVM: fix nested tsc scaling when not enabled but MSR_AMD64_TSC

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <44604447-9686-24b3-4216-71d52eb1a6c2@redhat.com>
Date:   Mon, 21 Feb 2022 13:33:13 +0100
From:   Paolo Bonzini <pbonzini@...hat.com>
To:     Maxim Levitsky <mlevitsk@...hat.com>, kvm@...r.kernel.org
Cc:     linux-kernel@...r.kernel.org, stable@...r.kernel.org,
        Dāvis Mosāns <davispuh@...il.com>
Subject: Re: [PATCH] KVM: nSVM: fix nested tsc scaling when not enabled but
 MSR_AMD64_TSC_RATIO set

On 2/21/22 11:33, Maxim Levitsky wrote:
> For compatibility with userspace the MSR_AMD64_TSC_RATIO can be set
> even when the feature is not exposed to the guest, which breaks assumptions
> that it has the default value in this case.
> 
> Fixes: 5228eb96a487 ("KVM: x86: nSVM: implement nested TSC scaling")
> Cc: stable@...r.kernel.org
> 
> Reported-by: Dāvis Mosāns <davispuh@...il.com>
> Signed-off-by: Maxim Levitsky <mlevitsk@...hat.com>

It's not clear how QEMU ends up writing MSR_AMD64_TSC_RATIO_DEFAULT 
rather than 0, but we clearly have a bug in KVM.  It should not allow 
writing 0 in the first place if tsc-ratio is not available in the VM.

If QEMU really can get itself in this situation, we cannot fix this 
except with KVM_CAP_DISABLE_QUIRKS (a quirk that would accept and ignore 
host-initiated writes if the CPUID bit is not available) or perhaps with 
a pr_warn_ratelimited and a quick deprecation cycle, until some time 
after 6.2.1 is released.

Hmmm... maybe the issue is actually that KVM *returns* 0 from 
KVM_GET_MSRS?  And in this case, fixing KVM would also prevent QEMU from 
sending the bogus KVM_SET_MSRS?

Thanks,

Paolo

> ---
>   arch/x86/kvm/svm/nested.c | 10 ++++------
>   arch/x86/kvm/svm/svm.c    |  5 +++--
>   arch/x86/kvm/svm/svm.h    |  1 +
>   3 files changed, 8 insertions(+), 8 deletions(-)
> 
> diff --git a/arch/x86/kvm/svm/nested.c b/arch/x86/kvm/svm/nested.c
> index 1218b5a342fc..a2e2436057dc 100644
> --- a/arch/x86/kvm/svm/nested.c
> +++ b/arch/x86/kvm/svm/nested.c
> @@ -574,14 +574,12 @@ static void nested_vmcb02_prepare_control(struct vcpu_svm *svm)
>   	vcpu->arch.tsc_offset = kvm_calc_nested_tsc_offset(
>   			vcpu->arch.l1_tsc_offset,
>   			svm->nested.ctl.tsc_offset,
> -			svm->tsc_ratio_msr);
> +			svm_get_l2_tsc_multiplier(vcpu));
>   
>   	svm->vmcb->control.tsc_offset = vcpu->arch.tsc_offset;
>   
> -	if (svm->tsc_ratio_msr != kvm_default_tsc_scaling_ratio) {
> -		WARN_ON(!svm->tsc_scaling_enabled);
> +	if (svm_get_l2_tsc_multiplier(vcpu) != kvm_default_tsc_scaling_ratio)
>   		nested_svm_update_tsc_ratio_msr(vcpu);
> -	}
>   
>   	svm->vmcb->control.int_ctl             =
>   		(svm->nested.ctl.int_ctl & int_ctl_vmcb12_bits) |
> @@ -867,8 +865,7 @@ int nested_svm_vmexit(struct vcpu_svm *svm)
>   		vmcb_mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
>   	}
>   
> -	if (svm->tsc_ratio_msr != kvm_default_tsc_scaling_ratio) {
> -		WARN_ON(!svm->tsc_scaling_enabled);
> +	if (svm_get_l2_tsc_multiplier(vcpu) != kvm_default_tsc_scaling_ratio) {
>   		vcpu->arch.tsc_scaling_ratio = vcpu->arch.l1_tsc_scaling_ratio;
>   		svm_write_tsc_multiplier(vcpu, vcpu->arch.tsc_scaling_ratio);
>   	}
> @@ -1264,6 +1261,7 @@ void nested_svm_update_tsc_ratio_msr(struct kvm_vcpu *vcpu)
>   {
>   	struct vcpu_svm *svm = to_svm(vcpu);
>   
> +	WARN_ON_ONCE(!svm->tsc_scaling_enabled);
>   	vcpu->arch.tsc_scaling_ratio =
>   		kvm_calc_nested_tsc_multiplier(vcpu->arch.l1_tsc_scaling_ratio,
>   					       svm->tsc_ratio_msr);
> diff --git a/arch/x86/kvm/svm/svm.c b/arch/x86/kvm/svm/svm.c
> index 975be872cd1a..5cf6bb5bfd3e 100644
> --- a/arch/x86/kvm/svm/svm.c
> +++ b/arch/x86/kvm/svm/svm.c
> @@ -911,11 +911,12 @@ static u64 svm_get_l2_tsc_offset(struct kvm_vcpu *vcpu)
>   	return svm->nested.ctl.tsc_offset;
>   }
>   
> -static u64 svm_get_l2_tsc_multiplier(struct kvm_vcpu *vcpu)
> +u64 svm_get_l2_tsc_multiplier(struct kvm_vcpu *vcpu)
>   {
>   	struct vcpu_svm *svm = to_svm(vcpu);
>   
> -	return svm->tsc_ratio_msr;
> +	return svm->tsc_scaling_enabled ? svm->tsc_ratio_msr :
> +	       kvm_default_tsc_scaling_ratio;
>   }
>   
>   static void svm_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
> diff --git a/arch/x86/kvm/svm/svm.h b/arch/x86/kvm/svm/svm.h
> index 852b12aee03d..006571dd30df 100644
> --- a/arch/x86/kvm/svm/svm.h
> +++ b/arch/x86/kvm/svm/svm.h
> @@ -542,6 +542,7 @@ int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
>   			       bool has_error_code, u32 error_code);
>   int nested_svm_exit_special(struct vcpu_svm *svm);
>   void nested_svm_update_tsc_ratio_msr(struct kvm_vcpu *vcpu);
> +u64 svm_get_l2_tsc_multiplier(struct kvm_vcpu *vcpu);
>   void svm_write_tsc_multiplier(struct kvm_vcpu *vcpu, u64 multiplier);
>   void nested_copy_vmcb_control_to_cache(struct vcpu_svm *svm,
>   				       struct vmcb_control_area *control);