linux-kernel - Re: AMDGPU and 16B stack alignment

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] [day] [month] [year] [list]
Message-ID: <CAKwvOdkXDyqMA-mOz_PE9x0V0ePtA9uFcJS+1ibpBqde6MXX4g@mail.gmail.com>
Date:   Wed, 16 Oct 2019 16:05:27 -0700
From:   Nick Desaulniers <ndesaulniers@...gle.com>
To:     Arvind Sankar <nivedita@...m.mit.edu>
Cc:     Arnd Bergmann <arnd@...db.de>, "S, Shirish" <sshankar@....com>,
        "Wentland, Harry" <Harry.Wentland@....com>,
        "Deucher, Alexander" <Alexander.Deucher@....com>,
        "yshuiv7@...il.com" <yshuiv7@...il.com>,
        "andrew.cooper3@...rix.com" <andrew.cooper3@...rix.com>,
        clang-built-linux <clang-built-linux@...glegroups.com>,
        Matthias Kaehlcke <mka@...gle.com>,
        "S, Shirish" <Shirish.S@....com>,
        "Zhou, David(ChunMing)" <David1.Zhou@....com>,
        "Koenig, Christian" <Christian.Koenig@....com>,
        amd-gfx list <amd-gfx@...ts.freedesktop.org>,
        LKML <linux-kernel@...r.kernel.org>
Subject: Re: AMDGPU and 16B stack alignment

On Wed, Oct 16, 2019 at 11:55 AM Arvind Sankar <nivedita@...m.mit.edu> wrote:
>
> On Tue, Oct 15, 2019 at 06:51:26PM -0700, Nick Desaulniers wrote:
> > On Tue, Oct 15, 2019 at 1:26 PM Arvind Sankar <nivedita@...m.mit.edu> wrote:
> > >
> > > On Tue, Oct 15, 2019 at 11:05:56AM -0700, Nick Desaulniers wrote:
> > > > Hmmm...I would have liked to remove it outright, as it is an ABI
> > > > mismatch that is likely to result in instability and non-fun-to-debug
> > > > runtime issues in the future.  I suspect my patch does work for GCC
> > > > 7.1+.  The question is: Do we want to either:
> > > > 1. mark AMDGPU broken for GCC < 7.1, or
> > > > 2. continue supporting it via stack alignment mismatch?
> > > >
> > > > 2 is brittle, and may break at any point in the future, but if it's
> > > > working for someone it does make me feel bad to outright disable it.
> > > > What I'd image 2 looks like is (psuedo code in a Makefile):
> > > >
> > > > if CC_IS_GCC && GCC_VERSION < 7.1:
> > > >   set stack alignment to 16B and hope for the best
> > > >
> > > > So my diff would be amended to keep the stack alignment flags, but
> > > > only to support GCC < 7.1.  And that assumes my change compiles with
> > > > GCC 7.1+. (Looks like it does for me locally with GCC 8.3, but I would
> > > > feel even more confident if someone with hardware to test on and GCC
> > > > 7.1+ could boot test).
> > > > --
> > > > Thanks,
> > > > ~Nick Desaulniers
> > >
> > > If we do keep it, would adding -mstackrealign make it more robust?
> > > That's simple and will only add the alignment to functions that require
> > > 16-byte alignment (at least on gcc).
> >
> > I think there's also `-mincoming-stack-boundary=`.
> > https://github.com/ClangBuiltLinux/linux/issues/735#issuecomment-540038017
>
> Yes, but -mstackrealign looks like it's supported by clang as well.

Good to know, but I want less duct tape, not more.

> >
> > >
> > > Alternative is to use
> > > __attribute__((force_align_arg_pointer)) on functions that might be
> > > called from 8-byte-aligned code.
> >
> > Which is hard to automate and easy to forget.  Likely a large diff to fix today.
>
> Right, this is a no-go, esp to just fix old compilers.
> >
> > >
> > > It looks like -mstackrealign should work from gcc 5.3 onwards.
> >
> > The kernel would generally like to support GCC 4.9+.
> >
> > There's plenty of different ways to keep layering on duct tape and
> > bailing wire to support differing ABIs, but that's just adding
> > technical debt that will have to be repaid one day.  That's why the
> > cleanest solution IMO is mark the driver broken for old toolchains,
> > and use a code-base-consistent stack alignment.  Bending over
> > backwards to support old toolchains means accepting stack alignment
> > mismatches, which is in the "unspecified behavior" ring of the
> > "undefined behavior" Venn diagram.  I have the same opinion on relying
> > on explicitly undefined behavior.
> >
> > I'll send patches for fixing up Clang, but please consider my strong
> > advice to generally avoid stack alignment mismatches, regardless of
> > compiler.
> > --
> > Thanks,
> > ~Nick Desaulniers
>
> What I suggested was in reference to your proposal for dropping the
> -mpreferred-stack-boundary=4 for modern compilers, but keeping it for
> <7.1. -mstackrealign would at least let 5.3 onwards be less likely to
> break (and it doesn't error before then, I think it just doesn't
> actually do anything, so no worse than now at least).
>
> Simply dropping support for <7.1 would be cleanest, yes, but it sounds
> like people don't want to go that far.

That's fair.  I've included your suggestions in the commit message of
02/03 of a series I just sent but forgot to in reply to this thread:
https://lkml.org/lkml/2019/10/16/1700

Also, I do appreciate the suggestions and understand the value of brainstorming.
-- 
Thanks,
~Nick Desaulniers