[PATCH] powerpc: avoid broken GCC __attribute__((optimize))

[Michael Ellerman]

Ard Biesheuvel <ardb at kernel.org> writes:
> On Wed, 28 Oct 2020 at 09:04, Ard Biesheuvel <ardb at kernel.org> wrote:
>>
>> Commit 7053f80d9696 ("powerpc/64: Prevent stack protection in early boot")
>> introduced a couple of uses of __attribute__((optimize)) with function
>> scope, to disable the stack protector in some early boot code.
>>
>> Unfortunately, and this is documented in the GCC man pages [0], overriding
>> function attributes for optimization is broken, and is only supported for
>> debug scenarios, not for production: the problem appears to be that
>> setting GCC -f flags using this method will cause it to forget about some
>> or all other optimization settings that have been applied.
>>
>> So the only safe way to disable the stack protector is to disable it for
>> the entire source file.
>>
>> [0] https://gcc.gnu.org/onlinedocs/gcc/Common-Function-Attributes.html
>>
>> Cc: Michael Ellerman <mpe at ellerman.id.au>
>> Cc: Benjamin Herrenschmidt <benh at kernel.crashing.org>
>> Cc: Paul Mackerras <paulus at samba.org>
>> Cc: Nick Desaulniers <ndesaulniers at google.com>
>> Cc: Arvind Sankar <nivedita at alum.mit.edu>
>> Cc: Randy Dunlap <rdunlap at infradead.org>
>> Cc: Josh Poimboeuf <jpoimboe at redhat.com>
>> Cc: Thomas Gleixner <tglx at linutronix.de>
>> Cc: Alexei Starovoitov <ast at kernel.org>
>> Cc: Daniel Borkmann <daniel at iogearbox.net>
>> Cc: Peter Zijlstra (Intel) <peterz at infradead.org>
>> Cc: Geert Uytterhoeven <geert at linux-m68k.org>
>> Cc: Kees Cook <keescook at chromium.org>
>> Fixes: 7053f80d9696 ("powerpc/64: Prevent stack protection in early boot")
>> Signed-off-by: Ard Biesheuvel <ardb at kernel.org>
>> ---
>> Related discussion here:
>> https://lore.kernel.org/lkml/CAMuHMdUg0WJHEcq6to0-eODpXPOywLot6UD2=GFHpzoj_hCoBQ@mail.gmail.com/
>>
>> TL;DR using __attribute__((optimize("-fno-gcse"))) in the BPF interpreter
>> causes the compiler to forget about -fno-asynchronous-unwind-tables passed
>> on the command line, resulting in unexpected .eh_frame sections in vmlinux.
>>
>>  arch/powerpc/kernel/Makefile   | 3 +++
>>  arch/powerpc/kernel/paca.c     | 2 +-
>>  arch/powerpc/kernel/setup.h    | 6 ------
>>  arch/powerpc/kernel/setup_64.c | 2 +-
>>  4 files changed, 5 insertions(+), 8 deletions(-)

Thanks for the patch.

> FYI i was notified by one of the robots that I missed one occurrence
> of __nostackprotector in arch/powerpc/kernel/paca.c
>
> Let me know if I need to resend.

That's fine I'll fix it up when applying.

With the existing code, with STACKPROTECTOR_STRONG=y, I see two
functions in setup_64.c that are triggering stack protection. One is
__init, and the other takes no parameters and is not easily reachable
from userspace, so I don't think losing the stack canary on either of
those is a concern.

I don't see anything in paca.c triggering stack protection.

I don't think there's any evidence this is causing a bug for us, so I'll
plan to put this in next for v5.11.

cheers