Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
MIME-Version: 1.0
References: <20200303221205.4048668-5-nivedita@alum.mit.edu>
 <20200511170149.1363260-1-mike@fireburn.co.uk> <20200511183645.GA596850@rani.riverdale.lan>
In-Reply-To: <20200511183645.GA596850@rani.riverdale.lan>
From:   Ard Biesheuvel <ardb@kernel.org>
Date:   Mon, 11 May 2020 23:13:00 +0200
Message-ID: <CAMj1kXEmJRpVyWDVkBn9eL0y0J4iVrUkYZd_pk_oKOeQPH661g@mail.gmail.com>
Subject: Re: [PATCH v2 4/5] efi/x86: Remove extra headroom for setup block
To:     Arvind Sankar <nivedita@alum.mit.edu>
Cc:     Mike Lothian <mike@fireburn.co.uk>,
        linux-efi <linux-efi@vger.kernel.org>,
        Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
        X86 ML <x86@kernel.org>
Content-Type: text/plain; charset="UTF-8"
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Mon, 11 May 2020 at 20:36, Arvind Sankar <nivedita@alum.mit.edu> wrote:
>
> On Mon, May 11, 2020 at 06:01:49PM +0100, Mike Lothian wrote:
> > Hi
> >
> > This patch has been causing issues for me since switching to GCC 10.1:
> >
> >   CALL    scripts/checksyscalls.sh
> >   CALL    scripts/atomic/check-atomics.sh
> >   DESCEND  objtool
> >   CHK     include/generated/compile.h
> >   HOSTCC  arch/x86/boot/tools/build
> > /usr/lib/gcc/x86_64-pc-linux-gnu/10.1.0/../../../../x86_64-pc-linux-gnu/bin/ld: error: linker defined: multiple definition of '_end'
> > /usr/lib/gcc/x86_64-pc-linux-gnu/10.1.0/../../../../x86_64-pc-linux-gnu/bin/ld: /tmp/ccEkW0jM.o: previous definition here
> > collect2: error: ld returned 1 exit status
> > make[1]: *** [scripts/Makefile.host:103: arch/x86/boot/tools/build] Error 1
> > make: *** [arch/x86/Makefile:303: bzImage] Error 2
> >
> > Cheers
> >
> > Mike
>
> I'm not getting an error even with gcc 10 for some reason, but I can see
> that it is busted. It's using the linker-defined _end symbol which is
> just pass the end of the .bss.
>
> Does adding "static" to the declaration of _end fix your error?

This is in a host tool, so it depends on the builtin linker script the
toolchain decides to use. This is risky, though, as it may be using
PROVIDE() for _end, which means that in cases where it doesn't break,
other references to _end that may exist will be linked to the wrong
symbol. I don't think 'build' should be expected to do anything
interesting with its own representation in memory, but better fix it
nonetheless.

Arvind: mind sending a fix for this, please?