linux-kernel - Re: [PATCH v2 5/5] x86/pti: Do not enable PTI on fixed Intel processors

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: <CA+55aFxAM0NHYXxXdpKjO=Swer2-Mvjs2J6ekZo_GwPAFRC+jA@mail.gmail.com>
Date:   Tue, 23 Jan 2018 12:38:47 -0800
From:   Linus Torvalds <torvalds@...ux-foundation.org>
To:     Alan Cox <gnomes@...rguk.ukuu.org.uk>
Cc:     David Woodhouse <dwmw@...zon.co.uk>,
        Linux Kernel Mailing List <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH v2 5/5] x86/pti: Do not enable PTI on fixed Intel processors

On Tue, Jan 23, 2018 at 10:45 AM, Alan Cox <gnomes@...rguk.ukuu.org.uk> wrote:
>
> static int in_order_cpu(void)
> {
>         /* Processors with CPU id etc */
>         if (x86_match_cpu(cpu_in_order))
>                 return 1;
>         /* Other rules here */
>         return 0;
> }

Note that "in-order" does *not* necessarily mean "no speculation". You
still do branch prediction, you still have a BTB for all but the
simplest in-order things, and you may still end up running
instructions at least part-way through a pipeline. Whether it gets to
a cache replacement or not is unclear.

So that naming is very misleading.

It is quite possible (even likely) that the old in-order Atoms do not
have this issue, because it's definitely true that in-order tends to
limit speculation a lot, and any mis-predict will _probably_ kill
instructions early enough that you'd never see this.

But meltdown, for example, is based entirely on uarch details in the
memory pipeline, not on in-order vs ooo. Again, an in-order CPU is
probably less aggressive in the memory pipeline too, so there is
probably strong correlation, but there is no causal relationship.

              Linus