Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
MIME-Version: 1.0
References: <20221122161017.2426828-1-ardb@kernel.org> <5750d157-43dd-6f3d-1407-f41af3cff207@amd.com>
 <CAMj1kXHUQFAcRKzRkuGG3Rsyrexdi7_NUS1-aXrS36LP4Q=rxw@mail.gmail.com>
 <26c34f9e-3b09-7b10-09a2-993a50790447@amd.com> <Y336yE4P3NjCGavN@zn.tnic>
 <CAMj1kXGnGz+V3tmonitY+3TiWuWJFWj-mQLUbRo+xp8UwZ_SpQ@mail.gmail.com> <Y33/VIsB5HGREY4i@zn.tnic>
In-Reply-To: <Y33/VIsB5HGREY4i@zn.tnic>
From:   Ard Biesheuvel <ardb@kernel.org>
Date:   Wed, 23 Nov 2022 12:22:55 +0100
Message-ID: <CAMj1kXFMc_1ZVeiBJmUwPjPh-_LCU8Jqr_iSOEhVHT=PJFstcQ@mail.gmail.com>
Subject: Re: [PATCH v3 00/17] x86: head_64.S spring cleaning
To:     Borislav Petkov <bp@alien8.de>
Cc:     Tom Lendacky <thomas.lendacky@amd.com>, linux-efi@vger.kernel.org,
        linux-kernel@vger.kernel.org, Thomas Gleixner <tglx@linutronix.de>,
        Ingo Molnar <mingo@redhat.com>,
        Dave Hansen <dave.hansen@linux.intel.com>,
        Michael Roth <michael.roth@amd.com>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Wed, 23 Nov 2022 at 12:09, Borislav Petkov <bp@alien8.de> wrote:
>
> On Wed, Nov 23, 2022 at 11:52:32AM +0100, Ard Biesheuvel wrote:
> > The patch moves it from .data to .bss inadvertently, and I am not
> > convinced Tom's analysis is entirely accurate: we may simply have
> > garbage in image_offset if we access it before .bss gets cleared.
>
> That should not be too hard to find out: add an endless loop in asm in
> the guest right after the first image_offset access:
>
> 1:
>         jmp 1b
>
> and then dump its value.
>
> Or Tom might have an even better solution.
>
> But looking at the code, BSS clearing happens later, at .Lrelocated and
> the EFI stub comes before it. AFAICT.
>

Indeed. And moving it back into .data makes the most sense in any case
- the point of the patch is to drop the duplicate definitions from asm
code, not to move it into a different section.

The reason I hadn't spotted this is because my boot chain always sets
the value of image_offset during the boot, and does not rely on the
statically initialized value at all.