Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::3:8 as permitted sender) client-ip=2620:137:e000::3:8;
MIME-Version: 1.0
References: <20230912090051.4014114-17-ardb@google.com> <ZQQiUxh5vmeZnp7s@gmail.com>
 <ZQRAckHVxQZRNEGA@gmail.com> <CAGnOC3Zw49_30FkGY=RRLn-sCHNgFY_T0ugNJZCgg_T3opHm+Q@mail.gmail.com>
 <CAMj1kXEVA5iL79HwXdYOAXAr0ZemS6repMRi9b20VSU=u0Ee=Q@mail.gmail.com> <ZQVxIIh1byq65lK/@gmail.com>
In-Reply-To: <ZQVxIIh1byq65lK/@gmail.com>
From:   Ard Biesheuvel <ardb@kernel.org>
Date:   Sat, 16 Sep 2023 21:14:35 +0200
Message-ID: <CAMj1kXH7b+TRajewcY0hzh_j0qMV+X9Nm2PNFftWTTEBp9xr5g@mail.gmail.com>
Subject: Re: [PATCH v2 00/15] x86/boot: Rework PE header generation
To:     Ingo Molnar <mingo@kernel.org>
Cc:     Ard Biesheuvel <ardb@google.com>, linux-efi@vger.kernel.org,
        linux-kernel@vger.kernel.org, Evgeniy Baskov <baskov@ispras.ru>,
        Borislav Petkov <bp@alien8.de>,
        Dave Hansen <dave.hansen@linux.intel.com>,
        Ingo Molnar <mingo@redhat.com>,
        Thomas Gleixner <tglx@linutronix.de>,
        Peter Jones <pjones@redhat.com>,
        Matthew Garrett <mjg59@srcf.ucam.org>,
        Gerd Hoffmann <kraxel@redhat.com>,
        Kees Cook <keescook@chromium.org>,
        "H. Peter Anvin" <hpa@zytor.com>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Sat, 16 Sept 2023 at 11:11, Ingo Molnar <mingo@kernel.org> wrote:
>
>
> * Ard Biesheuvel <ardb@kernel.org> wrote:
>
> > > > So, the first 8 patches broke the x86-64-defconfig-ish Qemu bzImage bootup,
> > > > due to the 8th patch:
> > > >
> > > >   988b52b207a9fe74c3699bda8c2256714926b94b is the first bad commit
> > > >   commit 988b52b207a9fe74c3699bda8c2256714926b94b
> > > >   Author: Ard Biesheuvel <ardb@kernel.org>
> > > >   Date:   Tue Sep 12 09:01:01 2023 +0000
> > > >
> > > >       x86/boot: Define setup size in linker script
> > > >
> > > > I've removed it for now - but this side effect was not expected.
> > > >
> > >
> > > No, definitely not expected. I tested various combinations of i386 /
> > > x86_64 built with GCC / Clang doing EFI or BIOS boot.
> > >
> > > I'll rebase the remaining stuff onto -tip and see if I can reproduce this.
> >
> > This is actually quite bizarre. x86_64_defconfig has
> > CONFIG_EFI_MIXED=y and i tested that this change produces the exact
> > same bzImage binary in that case.
> >
> > Could you send me the .config and the QEMU command line perhaps?
>
> So the patch below is the delta between v2 and v3 - that is expected
> to fix the bzImage boot crash, right?
>

Yes.

ld.bfd does something unexpected [to me] here, and the resulting value
turns out not to be a multiple of 512 at all.

With this tweak, my claim that this patch does not affect the binary
bzImage at all actually holds for ld.bfd as well (provided that
CONFIG_EFI_MIXED=y and CONFIG_LOCAL_VERSION_AUTO is disabled)

So if this still does not work in your test, could you please disable
CONFIG_LOCAL_VERSION_AUTO and compare the md5sums of the two builds?

Thanks,


> --- tip.orig/arch/x86/boot/setup.ld
> +++ tip/arch/x86/boot/setup.ld
> @@ -41,7 +41,7 @@ SECTIONS
>                 LONG(0x5a5aaa55)
>
>                 /* Reserve some extra space for the reloc and compat sections */
> -               setup_size = ABSOLUTE(ALIGN(. + 64, 512));
> +               setup_size = ALIGN(ABSOLUTE(.) + 64, 512);
>                 setup_sects = ABSOLUTE(setup_size / 512);
>         }
>
>