linux-kernel - Re: [PATCH 4/8] kvm: vmx: Set IBPB when running a different VCPU

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: <1515835795.22302.517.camel@amazon.co.uk>
Date:   Sat, 13 Jan 2018 09:29:55 +0000
From:   "Woodhouse, David" <dwmw@...zon.co.uk>
To:     Jim Mattson <jmattson@...gle.com>, Wanpeng Li <kernellwp@...il.com>
CC:     Paolo Bonzini <pbonzini@...hat.com>,
        LKML <linux-kernel@...r.kernel.org>, kvm <kvm@...r.kernel.org>,
        Radim Krcmar <rkrcmar@...hat.com>,
        Liran Alon <liran.alon@...cle.com>,
        Anthony Liguori <aliguori@...zon.com>,
        Tom Lendacky <thomas.lendacky@....com>,
        Borislav Petkov <bp@...en8.de>,
        "the arch/x86 maintainers" <x86@...nel.org>,
        Tim Chen <tim.c.chen@...ux.intel.com>
Subject: Re: [PATCH 4/8] kvm: vmx: Set IBPB when running a different VCPU

On Fri, 2018-01-12 at 09:03 -0800, Jim Mattson wrote:
> The point behind the IPBP in vmx_vcpu_load is to prevent one VCPU from
> steering the speculative execution of the next. If the VMCS address is
> recycled, vmx_vcpu_load doesn't realize that the VCPUs are different,
> and so it won't issue the IPBP.

I don't understand the sequence of events that could lead to this.

If the VMCS is freed, surely per_cpu(current_vmcs, cpu) has to be
cleared? If the VMCS is freed while it's still *active* on a CPU,
that's a bug, surely? And if that CPU is later offlined and clears the
VMCS, it's going to scribble on freed (and potentially re-used) memory.

So vmx_cpu_load() *will* realise that it's different, won't it?

>> +       if (have_spec_ctrl)
>> +               wrmsrl(MSR_IA32_PRED_CMD, PRED_CMD_IBPB);

Also, I think the same condition applies to the conditional branches
over the IBPB-frobbing, as it does to setting IBRS. You can eschew the
'else lfence' only if you put in a comment showing that you've proved
it's safe. Many of the other bits like this are being done with
alternatives, which avoids that concern completely.

But really, I don't like this series much. Don't say "let's do this
until upstream supports...". Just fix it up properly, and add the
generic X86_FEATURE_IBPB bit and use it. We have *too* many separate
tiny patch sets, and we need to be getting our act together and putting
it all in one.

Download attachment "smime.p7s" of type "application/x-pkcs7-signature" (5210 bytes)