Re: Building arm64 EFI stub with -fpie breaks build of 4.9.x (undefined reference to `__efistub__GLOBAL_OFFSET_TABLE_')

From: Ard Biesheuvel <ard.biesheuvel@linaro.org>
To: Nick Desaulniers <ndesaulniers@google.com>
Cc: Greg KH <gregkh@linuxfoundation.org>,
	Rolf Eike Beer <eb@emlix.com>,
	Linus Torvalds <torvalds@linux-foundation.org>,
	Matt Fleming <matt@codeblueprint.co.uk>,
	Peter Zijlstra <peterz@infradead.org>,
	Thomas Gleixner <tglx@linutronix.de>,
	linux-efi <linux-efi@vger.kernel.org>,
	Linux Kernel Developers List <linux-kernel@vger.kernel.org>,
	stable <stable@vger.kernel.org>,
	clang-built-linux <clang-built-linux@googlegroups.com>
Subject: Re: Building arm64 EFI stub with -fpie breaks build of 4.9.x (undefined reference to `__efistub__GLOBAL_OFFSET_TABLE_')
Date: Thu, 6 Jun 2019 08:55:29 +0200	[thread overview]
Message-ID: <CAKv+Gu9Leaq_s2kVNzHx+zkdKFXgQVkouN3M56u5nou5WX=cKg@mail.gmail.com> (raw)
In-Reply-To: <CAKwvOdnPcjESFrQRR_=cCVag3ZSnC0nBqF7+LFHrcDArT_segA@mail.gmail.com>

On Wed, 5 Jun 2019 at 22:48, Nick Desaulniers <ndesaulniers@google.com> wrote:
>
> On Wed, Jun 5, 2019 at 11:42 AM Ard Biesheuvel
> <ard.biesheuvel@linaro.org> wrote:
> > For the record, this is an example of why I think backporting those
> > clang enablement patches is a bad idea.
>
> There's always a risk involved with backports of any kind; more CI
> coverage can help us mitigate some of these risks in an automated
> fashion before we get user reports like this.  I meet with the
> KernelCI folks weekly, so I'll double check on the coverage of the
> stable tree's branches.  The 0day folks are also very responsive and
> I've spoken with them a few times, so I'll try to get to the bottom of
> why this wasn't reported by either of those.
>
> Also, these patches help keep Android, CrOS, and Google internal
> production kernels closer to their upstream sources.
>
> > We can't actually build those
> > kernels with clang, can we? So what is the point? </grumpy>
>
> Here's last night's build:
> https://travis-ci.com/ClangBuiltLinux/continuous-integration/builds/114388434
>

If you are saying that plain upstream 4.9-stable defconfig can be
built with Clang, then I am pleasantly surprised.

> Also, Android and CrOS have shipped X million devices w/ 4.9 kernels
> built with Clang.  I think this number will grow at least one order of
> magnitude imminently.
>

I know that (since you keep reminding me :-)), but obviously, Google
does not care about changes that regress GCC support.

> > Alternatively, we can just revert this patch from 4.9
>
> That would break at least the above devices next time Android and CrOS
> pulled from stable.
>
> > It would be helpful to get a relocation dump (objdump -r) of
> > arm64-stub.o to figure out which symbol needs a 'hidden' annotation to
> > prevent GCC from emitting it as a PIC reference requiring a GOT.
>
> Sounds like the best way forward, as well as having more info on which
> config/toolchain reliably reproduces the issue.

Let me know once you can reproduce it, I will have a look as well.