Message-ID: <BDF4AF6B-DB1F-4BAF-999E-72CDDF408F37@nutanix.com>
Date:   Fri, 7 May 2021 18:26:18 +0000
From:   Jon Kohler <jon@...anix.com>
To:     Sean Christopherson <seanjc@...gle.com>
CC:     Venkatesh Srinivas <venkateshs@...omium.org>,
        Jon Kohler <jon@...anix.com>,
        Paolo Bonzini <pbonzini@...hat.com>,
        Vitaly Kuznetsov <vkuznets@...hat.com>,
        Wanpeng Li <wanpengli@...cent.com>,
        Jim Mattson <jmattson@...gle.com>,
        Joerg Roedel <joro@...tes.org>,
        Thomas Gleixner <tglx@...utronix.de>,
        Ingo Molnar <mingo@...hat.com>, Borislav Petkov <bp@...en8.de>,
        "x86@...nel.org" <x86@...nel.org>,
        "H. Peter Anvin" <hpa@...or.com>,
        "kvm@...r.kernel.org" <kvm@...r.kernel.org>,
        "linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH] KVM: x86: use X86_FEATURE_RSB_CTXSW for RSB stuffing in
 vmexit



> On May 7, 2021, at 1:58 PM, Sean Christopherson <seanjc@...gle.com> wrote:
> 
> On Fri, May 07, 2021, Venkatesh Srinivas wrote:
>> On Fri, May 7, 2021 at 8:08 AM Jon Kohler <jon@...anix.com> wrote:
>>> 
>>> cpufeatures.h defines X86_FEATURE_RSB_CTXSW as "Fill RSB on context
>>> switches" which seems more accurate than using X86_FEATURE_RETPOLINE
>>> in the vmxexit path for RSB stuffing.
>>> 
>>> X86_FEATURE_RSB_CTXSW is used for FILL_RETURN_BUFFER in
>>> arch/x86/entry/entry_{32|64}.S. This change makes KVM vmx and svm
>>> follow that same pattern. This pairs up nicely with the language in
>>> bugs.c, where this cpu_cap is enabled, which indicates that RSB
>>> stuffing should be unconditional with spectrev2 enabled.
>>>        /*
>>>         * If spectre v2 protection has been enabled, unconditionally fill
>>>         * RSB during a context switch; this protects against two independent
>>>         * issues:
>>>         *
>>>         *      - RSB underflow (and switch to BTB) on Skylake+
>>>         *      - SpectreRSB variant of spectre v2 on X86_BUG_SPECTRE_V2 CPUs
>>>         */
>>>        setup_force_cpu_cap(X86_FEATURE_RSB_CTXSW);
>>> 
>>> Furthermore, on X86_FEATURE_IBRS_ENHANCED CPUs && SPECTRE_V2_CMD_AUTO,
>>> we're bypassing setting X86_FEATURE_RETPOLINE, where as far as I could
>>> find, we should still be doing RSB stuffing no matter what when
>>> CONFIG_RETPOLINE is enabled and spectrev2 is set to auto.
>> 
>> If I'm reading https://urldefense.proofpoint.com/v2/url?u=https-3A__software.intel.com_security-2Dsoftware-2Dguidance_deep-2Ddives_deep-2Ddive-2Dindirect-2Dbranch-2Drestricted-2Dspeculation&d=DwIBAg&c=s883GpUCOChKOHiocYtGcg&r=NGPRGGo37mQiSXgHKm5rCQ&m=d1CkIBCdVFwUtKYx3SRW9dZD0kA_IX9VKEPG2-x4kBo&s=KlK_T41o6UVpLhMKDcK9iZfsJnop72K3CveJVIak5K8&e= 
>> correctly, I don't think an RSB fill sequence is required on VMExit on
>> processors w/ Enhanced IBRS. Specifically:
>> """
>> On processors with enhanced IBRS, an RSB overwrite sequence may not
>> suffice to prevent the predicted target of a near return from using an
>> RSB entry created in a less privileged predictor mode.  Software can
>> prevent this by enabling SMEP (for transitions from user mode to
>> supervisor mode) and by having IA32_SPEC_CTRL.IBRS set during VM exits
>> """
>> On Enhanced IBRS processors, it looks like SPEC_CTRL.IBRS is set
>> across all #VMExits via x86_virt_spec_ctrl in kvm.
>> 
>> So is this patch needed?
> 
> Venkatesh belatedly pointed out (off list) that KVM VMX stops intercepting
> MSR_IA32_SPEC_CTRL after the first (successful) write by the guest.  But, I 
> believe that's a non-issue for ENHANCED_IBRS because of this blurb in Intel's
> documentation[*]:
> 
>  Processors with enhanced IBRS still support the usage model where IBRS is set
>  only in the OS/VMM for OSes that enable SMEP. To do this, such processors will
>  ensure that guest behavior cannot control the RSB after a VM exit once IBRS is
>  set, even if IBRS was not set at the time of the VM exit.
> 
> The code and changelog for commit 706d51681d63 ("x86/speculation: Support
> Enhanced IBRS on future CPUs") is more than a little confusing:
> 
>  spectre_v2_select_mitigation():
> 	if (boot_cpu_has(X86_FEATURE_IBRS_ENHANCED)) {
> 		mode = SPECTRE_V2_IBRS_ENHANCED;
> 		/* Force it so VMEXIT will restore correctly */
> 		x86_spec_ctrl_base |= SPEC_CTRL_IBRS;
> 		wrmsrl(MSR_IA32_SPEC_CTRL, x86_spec_ctrl_base);
> 		goto specv2_set_mode;
> 	}

Thanks Sean, that makes sense. The handling of the MSR part of things
is separate from what I was looking at, which is where it seems like
Intel is still recommending doing an RSB overwrite/stuff even with
eIBRS; however, in KVM we use X86_FEATURE_RETPOLINE to figure
that out, but in bugs.c eIBRS systems skip having set, as
goto specv2_set_mode skips cover retpoline_auto now.

> 
>  changelog:
> 	Kernel also has to make sure that IBRS bit remains set after
> 	VMEXIT because the guest might have cleared the bit. This is already
> 	covered by the existing x86_spec_ctrl_set_guest() and
> 	x86_spec_ctrl_restore_host() speculation control functions.
> 
> but I _think_ that is simply saying that MSR_IA32_SPEC_CTRL.IBRS needs to be
> restored in order to keep the mitigations active in the host.   I don't think it
> contradicts the documentation that says VM-Exit is automagically mitigated if
> IBRS has _ever_ been set.
> 
> [*] https://urldefense.proofpoint.com/v2/url?u=https-3A__software.intel.com_security-2Dsoftware-2Dguidance_deep-2Ddives_deep-2Ddive-2Dindirect-2Dbranch-2Drestricted-2Dspeculation&d=DwIBAg&c=s883GpUCOChKOHiocYtGcg&r=NGPRGGo37mQiSXgHKm5rCQ&m=d1CkIBCdVFwUtKYx3SRW9dZD0kA_IX9VKEPG2-x4kBo&s=KlK_T41o6UVpLhMKDcK9iZfsJnop72K3CveJVIak5K8&e=

Powered by blists - more mailing lists