linux-kernel - Re: [PATCH v2] x86: bring back rep movsq for user access on CPUs without ERMS

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: <CAHk-=wgADyL9i8r1=YkRTehKG8T89TzqAFMXDJV1Ag+_4_25Cw@mail.gmail.com>
Date:   Wed, 30 Aug 2023 13:00:44 -0700
From:   Linus Torvalds <torvalds@...ux-foundation.org>
To:     Mateusz Guzik <mjguzik@...il.com>
Cc:     linux-kernel@...r.kernel.org, linux-arch@...r.kernel.org,
        bp@...en8.de
Subject: Re: [PATCH v2] x86: bring back rep movsq for user access on CPUs
 without ERMS

Side note, if you want to play around with the user copy routines (or
maybe Borislav wants to), I have a patch that handles a couple of
common cases statically.

It requires that we inline copy_to/from_user() in order to get
constant size information, but almost all other architectures do that
anyway, and it's not as horrid as it used to be with the current
access_ok() that doesn't need to do that nasty dynamic task size
check.

In particular, it should help with copying structures - notably the
'stat' structure in cp_new_stat().

The attached patch is entirely untested, except for me checking code
generation for some superficial sanity in a couple of places.

I'm not convinced that

    len >= 64 && !(len & 7)

is necessarily the "correct" option, but I resurrected an older patch
for this, and decided to use that as the "this is what
rep_movs_alternative would do anyway" test.

And obviously I expect that FSRM also does ok with "rep movsq", even
if technically "movsb" is the simpler case (because it doesn't have
the alignment issues that "rep movsq" has).

                 Linus

View attachment "patch.diff" of type "text/x-patch" (1889 bytes)