linux-kernel - Re: [PATCH v2] KVM: SVM: Delay restoration of host MSR_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: <71038b23-0902-1074-bd94-d5b30311faf0@redhat.com>
Date:   Thu, 22 Apr 2021 08:45:29 +0200
From:   Paolo Bonzini <pbonzini@...hat.com>
To:     Sean Christopherson <seanjc@...gle.com>
Cc:     Vitaly Kuznetsov <vkuznets@...hat.com>,
        Wanpeng Li <wanpengli@...cent.com>,
        Jim Mattson <jmattson@...gle.com>,
        Joerg Roedel <joro@...tes.org>, kvm@...r.kernel.org,
        linux-kernel@...r.kernel.org
Subject: Re: [PATCH v2] KVM: SVM: Delay restoration of host MSR_TSC_AUX until
 return to userspace

On 22/04/21 02:17, Sean Christopherson wrote:
> Use KVM's "user return MSRs" framework to defer restoring the host's
> MSR_TSC_AUX until the CPU returns to userspace.  Add/improve comments to
> clarify why MSR_TSC_AUX is intercepted on both RDMSR and WRMSR, and why
> it's safe for KVM to keep the guest's value loaded even if KVM is
> scheduled out.
> 
> Signed-off-by: Sean Christopherson <seanjc@...gle.com>
> ---
> 
> v2: Rebased to kvm/queue (ish), commit 0e91d1992235 ("KVM: SVM: Allocate SEV
>      command structures on local stack")
>   
>   arch/x86/kvm/svm/svm.c | 50 ++++++++++++++++++------------------------
>   arch/x86/kvm/svm/svm.h |  7 ------
>   2 files changed, 21 insertions(+), 36 deletions(-)
> 
> diff --git a/arch/x86/kvm/svm/svm.c b/arch/x86/kvm/svm/svm.c
> index cd8c333ed2dc..596361449f25 100644
> --- a/arch/x86/kvm/svm/svm.c
> +++ b/arch/x86/kvm/svm/svm.c
> @@ -213,6 +213,15 @@ struct kvm_ldttss_desc {
>   
>   DEFINE_PER_CPU(struct svm_cpu_data *, svm_data);
>   
> +/*
> + * Only MSR_TSC_AUX is switched via the user return hook.  EFER is switched via
> + * the VMCB, and the SYSCALL/SYSENTER MSRs are handled by VMLOAD/VMSAVE.
> + *
> + * RDTSCP and RDPID are not used in the kernel, specifically to allow KVM to
> + * defer the restoration of TSC_AUX until the CPU returns to userspace.
> + */
> +#define TSC_AUX_URET_SLOT	0
> +
>   static const u32 msrpm_ranges[] = {0, 0xc0000000, 0xc0010000};
>   
>   #define NUM_MSR_MAPS ARRAY_SIZE(msrpm_ranges)
> @@ -958,6 +967,9 @@ static __init int svm_hardware_setup(void)
>   		kvm_tsc_scaling_ratio_frac_bits = 32;
>   	}
>   
> +	if (boot_cpu_has(X86_FEATURE_RDTSCP))
> +		kvm_define_user_return_msr(TSC_AUX_URET_SLOT, MSR_TSC_AUX);
> +
>   	/* Check for pause filtering support */
>   	if (!boot_cpu_has(X86_FEATURE_PAUSEFILTER)) {
>   		pause_filter_count = 0;
> @@ -1423,19 +1435,10 @@ static void svm_prepare_guest_switch(struct kvm_vcpu *vcpu)
>   {
>   	struct vcpu_svm *svm = to_svm(vcpu);
>   	struct svm_cpu_data *sd = per_cpu(svm_data, vcpu->cpu);
> -	unsigned int i;
>   
>   	if (svm->guest_state_loaded)
>   		return;
>   
> -	/*
> -	 * Certain MSRs are restored on VMEXIT (sev-es), or vmload of host save
> -	 * area (non-sev-es). Save ones that aren't so we can restore them
> -	 * individually later.
> -	 */
> -	for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
> -		rdmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
> -
>   	/*
>   	 * Save additional host state that will be restored on VMEXIT (sev-es)
>   	 * or subsequent vmload of host save area.
> @@ -1454,29 +1457,15 @@ static void svm_prepare_guest_switch(struct kvm_vcpu *vcpu)
>   		}
>   	}
>   
> -	/* This assumes that the kernel never uses MSR_TSC_AUX */
>   	if (static_cpu_has(X86_FEATURE_RDTSCP))
> -		wrmsrl(MSR_TSC_AUX, svm->tsc_aux);
> +		kvm_set_user_return_msr(TSC_AUX_URET_SLOT, svm->tsc_aux, -1ull);
>   
>   	svm->guest_state_loaded = true;
>   }
>   
>   static void svm_prepare_host_switch(struct kvm_vcpu *vcpu)
>   {
> -	struct vcpu_svm *svm = to_svm(vcpu);
> -	unsigned int i;
> -
> -	if (!svm->guest_state_loaded)
> -		return;
> -
> -	/*
> -	 * Certain MSRs are restored on VMEXIT (sev-es), or vmload of host save
> -	 * area (non-sev-es). Restore the ones that weren't.
> -	 */
> -	for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
> -		wrmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
> -
> -	svm->guest_state_loaded = false;
> +	to_svm(vcpu)->guest_state_loaded = false;
>   }
>   
>   static void svm_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
> @@ -2893,12 +2882,15 @@ static int svm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
>   			return 1;
>   
>   		/*
> -		 * This is rare, so we update the MSR here instead of using
> -		 * direct_access_msrs.  Doing that would require a rdmsr in
> -		 * svm_vcpu_put.
> +		 * TSC_AUX is usually changed only during boot and never read
> +		 * directly.  Intercept TSC_AUX instead of exposing it to the
> +		 * guest via direct_acess_msrs, and switch it via user return.
>   		 */
>   		svm->tsc_aux = data;
> -		wrmsrl(MSR_TSC_AUX, svm->tsc_aux);
> +
> +		preempt_disable();
> +		kvm_set_user_return_msr(TSC_AUX_URET_SLOT, data, -1ull);
> +		preempt_enable();
>   		break;
>   	case MSR_IA32_DEBUGCTLMSR:
>   		if (!boot_cpu_has(X86_FEATURE_LBRV)) {
> diff --git a/arch/x86/kvm/svm/svm.h b/arch/x86/kvm/svm/svm.h
> index 454da1c1d9b7..9dce6f290041 100644
> --- a/arch/x86/kvm/svm/svm.h
> +++ b/arch/x86/kvm/svm/svm.h
> @@ -23,11 +23,6 @@
>   
>   #define __sme_page_pa(x) __sme_set(page_to_pfn(x) << PAGE_SHIFT)
>   
> -static const u32 host_save_user_msrs[] = {
> -	MSR_TSC_AUX,
> -};
> -#define NR_HOST_SAVE_USER_MSRS ARRAY_SIZE(host_save_user_msrs)
> -
>   #define	IOPM_SIZE PAGE_SIZE * 3
>   #define	MSRPM_SIZE PAGE_SIZE * 2
>   
> @@ -128,8 +123,6 @@ struct vcpu_svm {
>   
>   	u64 next_rip;
>   
> -	u64 host_user_msrs[NR_HOST_SAVE_USER_MSRS];
> -
>   	u64 spec_ctrl;
>   	/*
>   	 * Contains guest-controlled bits of VIRT_SPEC_CTRL, which will be
> 

Queued, thanks.

Paolo