linux-kernel - Re: [RFC 29/33] KVM: VMX: Save instruction length on EPT violation

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]

Date:   Wed, 8 Nov 2023 09:20:37 -0800
From:   Sean Christopherson <seanjc@...gle.com>
To:     Nicolas Saenz Julienne <nsaenz@...zon.com>
Cc:     kvm@...r.kernel.org, linux-kernel@...r.kernel.org,
        linux-hyperv@...r.kernel.org, pbonzini@...hat.com,
        vkuznets@...hat.com, anelkz@...zon.com, graf@...zon.com,
        dwmw@...zon.co.uk, jgowans@...zon.com, corbert@....net,
        kys@...rosoft.com, haiyangz@...rosoft.com, decui@...rosoft.com,
        x86@...nel.org, linux-doc@...r.kernel.org
Subject: Re: [RFC 29/33] KVM: VMX: Save instruction length on EPT violation

On Wed, Nov 08, 2023, Nicolas Saenz Julienne wrote:
> Save the length of the instruction that triggered an EPT violation in
> struct kvm_vcpu_arch. This will be used to populate Hyper-V VSM memory
> intercept messages.

This is silly and unnecessarily obfuscates *why* (as my response regarding SVM
shows), i.e. that this is "needed" becuase the value is consumed by a *different*
vCPU, not because of performance concerns.

It's also broken, AFAICT nothing prevents the intercepted vCPU from hitting a
different EPT violation before the target vCPU consumes exit_instruction_len.

Holy cow.  All of deliver_gpa_intercept() is wildly unsafe.  Aside from race
conditions, which in and of themselves are a non-starter, nothing guarantees that
the intercepted vCPU actually cached all of the information that is held in its VMCS.

The sane way to do this is to snapshot *all* information on the intercepted vCPU,
and then hand that off as a payload to the target vCPU.  That is, assuming the
cross-vCPU stuff is actually necessary.  At a glance, I don't see anything that
explains *why*.