Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Subject: Re: [PATCH 2/2] x86/speculation: Support "Enhanced IBRS" on future
 CPUs
To:     David Woodhouse <dwmw@amazon.co.uk>, tglx@linutronix.de,
        x86@kernel.org, kvm@vger.kernel.org, torvalds@linux-foundation.org,
        linux-kernel@vger.kernel.org, arjan.van.de.ven@intel.com,
        dave.hansen@intel.com
References: <1518449255-2182-1-git-send-email-dwmw@amazon.co.uk>
 <1518449255-2182-2-git-send-email-dwmw@amazon.co.uk>
From:   Paolo Bonzini <pbonzini@redhat.com>
Message-ID: <7e2e5ad1-49b6-1fdb-4a62-8ad6aefc30a0@redhat.com>
Date:   Tue, 13 Feb 2018 09:02:25 +0100
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:52.0) Gecko/20100101
 Thunderbird/52.5.2
MIME-Version: 1.0
In-Reply-To: <1518449255-2182-2-git-send-email-dwmw@amazon.co.uk>
Content-Type: text/plain; charset=utf-8
Content-Language: en-US
Content-Transfer-Encoding: 7bit
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On 12/02/2018 16:27, David Woodhouse wrote:
> The original IBRS hack in microcode is horribly slow. For the next
> generation of CPUs, as a stopgap until we get a proper fix, Intel
> promise an "Enhanced IBRS" which will be fast.
> 
> The assumption is that predictions in the BTB/RSB will be tagged with
> the VMX mode and ring that they were learned in, and thus the CPU will
> avoid consuming unsafe predictions without a performance penalty.
> 
> Intel's documentation says that it is still required to set the IBRS bit
> in the SPEC_CTRL MSR and ensure that it remains set.
> 
> Cope with this by trapping and emulating *all* access to SPEC_CTRL from
> KVM guests when the IBRS_ALL feature is present, so it can never be
> turned off. Guests who see IBRS_ALL should never do anything except
> turn it on at boot anyway. And if they didn't know about IBRS_ALL and
> they keep frobbing IBRS on every kernel entry/exit... well the vmexit
> for a no-op is probably going to be faster than they were expecting
> anyway, so they'll live.
> 
> Signed-off-by: David Woodhouse <dwmw@amazon.co.uk>
> Acked-by: Arjan van de Ven <arjan.van.de.ven@intel.com>
> ---
>  arch/x86/include/asm/nospec-branch.h |  9 ++++++++-
>  arch/x86/kernel/cpu/bugs.c           | 16 ++++++++++++++--
>  arch/x86/kvm/vmx.c                   | 17 ++++++++++-------
>  3 files changed, 32 insertions(+), 10 deletions(-)
> 
> diff --git a/arch/x86/include/asm/nospec-branch.h b/arch/x86/include/asm/nospec-branch.h
> index 788c4da..524bb86 100644
> --- a/arch/x86/include/asm/nospec-branch.h
> +++ b/arch/x86/include/asm/nospec-branch.h
> @@ -140,9 +140,16 @@ enum spectre_v2_mitigation {
>  	SPECTRE_V2_RETPOLINE_MINIMAL_AMD,
>  	SPECTRE_V2_RETPOLINE_GENERIC,
>  	SPECTRE_V2_RETPOLINE_AMD,
> -	SPECTRE_V2_IBRS,
> +	SPECTRE_V2_IBRS_ALL,
>  };
>  
> +extern enum spectre_v2_mitigation spectre_v2_enabled;
> +
> +static inline bool spectre_v2_ibrs_all(void)
> +{
> +	return spectre_v2_enabled == SPECTRE_V2_IBRS_ALL;
> +}
> +
>  extern char __indirect_thunk_start[];
>  extern char __indirect_thunk_end[];
>  
> diff --git a/arch/x86/kernel/cpu/bugs.c b/arch/x86/kernel/cpu/bugs.c
> index debcdda..047538a 100644
> --- a/arch/x86/kernel/cpu/bugs.c
> +++ b/arch/x86/kernel/cpu/bugs.c
> @@ -88,12 +88,13 @@ static const char *spectre_v2_strings[] = {
>  	[SPECTRE_V2_RETPOLINE_MINIMAL_AMD]	= "Vulnerable: Minimal AMD ASM retpoline",
>  	[SPECTRE_V2_RETPOLINE_GENERIC]		= "Mitigation: Full generic retpoline",
>  	[SPECTRE_V2_RETPOLINE_AMD]		= "Mitigation: Full AMD retpoline",
> +	[SPECTRE_V2_IBRS_ALL]			= "Mitigation: Enhanced IBRS",
>  };
>  
>  #undef pr_fmt
>  #define pr_fmt(fmt)     "Spectre V2 : " fmt
>  
> -static enum spectre_v2_mitigation spectre_v2_enabled = SPECTRE_V2_NONE;
> +enum spectre_v2_mitigation spectre_v2_enabled = SPECTRE_V2_NONE;
>  
>  #ifdef RETPOLINE
>  static bool spectre_v2_bad_module;
> @@ -237,6 +238,16 @@ static void __init spectre_v2_select_mitigation(void)
>  
>  	case SPECTRE_V2_CMD_FORCE:
>  	case SPECTRE_V2_CMD_AUTO:
> +		if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES)) {
> +			u64 ia32_cap = 0;
> +
> +			rdmsrl(MSR_IA32_ARCH_CAPABILITIES, ia32_cap);
> +			if (ia32_cap & ARCH_CAP_IBRS_ALL) {
> +				mode = SPECTRE_V2_IBRS_ALL;
> +				wrmsrl(MSR_IA32_SPEC_CTRL, SPEC_CTRL_IBRS);
> +				goto ibrs_all;
> +			}
> +		}
>  		if (IS_ENABLED(CONFIG_RETPOLINE))
>  			goto retpoline_auto;
>  		break;
> @@ -274,6 +285,7 @@ static void __init spectre_v2_select_mitigation(void)
>  		setup_force_cpu_cap(X86_FEATURE_RETPOLINE);
>  	}
>  
> + ibrs_all:
>  	spectre_v2_enabled = mode;
>  	pr_info("%s\n", spectre_v2_strings[mode]);
>  
> @@ -306,7 +318,7 @@ static void __init spectre_v2_select_mitigation(void)
>  	 * branches. But we don't know whether the firmware is safe, so
>  	 * use IBRS to protect against that:
>  	 */
> -	if (boot_cpu_has(X86_FEATURE_IBRS)) {
> +	if (mode != SPECTRE_V2_IBRS_ALL && boot_cpu_has(X86_FEATURE_IBRS)) {
>  		setup_force_cpu_cap(X86_FEATURE_USE_IBRS_FW);
>  		pr_info("Spectre mitigation: Restricting branch speculation (enabling IBRS) for firmware calls\n");
>  	}
> diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c
> index 91e3539..d99ba9b 100644
> --- a/arch/x86/kvm/vmx.c
> +++ b/arch/x86/kvm/vmx.c
> @@ -3419,13 +3419,14 @@ static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
>  
>  		vmx->spec_ctrl = data;
>  
> -		if (!data)
> +		if (!data && !spectre_v2_ibrs_all())
>  			break;

This should check the value of IBRS_ALL in the VM, not in the host.

>  		/*
>  		 * For non-nested:
>  		 * When it's written (to non-zero) for the first time, pass
> -		 * it through.
> +		 * it through unless we have IBRS_ALL and it should just be
> +		 * set for ever.
>  		 *
>  		 * For nested:
>  		 * The handling of the MSR bitmap for L2 guests is done in
> @@ -9441,7 +9442,7 @@ static void __noclone vmx_vcpu_run(struct kvm_vcpu *vcpu)
>  	 * is no need to worry about the conditional branch over the wrmsr
>  	 * being speculatively taken.
>  	 */
> -	if (vmx->spec_ctrl)
> +	if (vmx->spec_ctrl && !spectre_v2_ibrs_all())
>  		wrmsrl(MSR_IA32_SPEC_CTRL, vmx->spec_ctrl);

Same here, and it should also be

	if (vmx->spec_ctrl != host_spec_ctrl())

where

static inline int host_spec_ctrl()
{
	return spectre_v2_enabled != SPECTRE_V2_NONE;
}

Likewise below.

Paolo


>  	vmx->__launched = vmx->loaded_vmcs->launched;
> @@ -9577,11 +9578,13 @@ static void __noclone vmx_vcpu_run(struct kvm_vcpu *vcpu)
>  	 * If the L02 MSR bitmap does not intercept the MSR, then we need to
>  	 * save it.
>  	 */
> -	if (!msr_write_intercepted(vcpu, MSR_IA32_SPEC_CTRL))
> -		rdmsrl(MSR_IA32_SPEC_CTRL, vmx->spec_ctrl);
> +	if (!spectre_v2_ibrs_all) {
> +		if (!msr_write_intercepted(vcpu, MSR_IA32_SPEC_CTRL))
> +			rdmsrl(MSR_IA32_SPEC_CTRL, vmx->spec_ctrl);
>  
> -	if (vmx->spec_ctrl)
> -		wrmsrl(MSR_IA32_SPEC_CTRL, 0);
> +		if (vmx->spec_ctrl)
> +			wrmsrl(MSR_IA32_SPEC_CTRL, 0);
> +	}
>  
>  	/* Eliminate branch target predictions from guest mode */
>  	vmexit_fill_RSB();
>