Re: [PATCH] compiler.h: Raise minimum version of GCC to 5.1 for arm64

From: Ard Biesheuvel <ardb@kernel.org>
To: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Russell King - ARM Linux admin <linux@armlinux.org.uk>,
	Florian Weimer <fweimer@redhat.com>,
	Arnd Bergmann <arnd@kernel.org>,
	"Cc: Android Kernel" <kernel-team@android.com>,
	"Theodore Ts'o" <tytso@mit.edu>,
	Peter Zijlstra <peterz@infradead.org>,
	Catalin Marinas <catalin.marinas@arm.com>,
	Masahiro Yamada <masahiroy@kernel.org>,
	Nick Desaulniers <ndesaulniers@google.com>,
	Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
	Will Deacon <will@kernel.org>,
	linux-arm-kernel <linux-arm-kernel@lists.infradead.org>
Subject: Re: [PATCH] compiler.h: Raise minimum version of GCC to 5.1 for arm64
Date: Thu, 14 Jan 2021 09:18:07 +0100	[thread overview]
Message-ID: <CAMj1kXFaDNRbDvr43VLEHFRHHS0sGGcF=iTfES4sxPg-rZ34NA@mail.gmail.com> (raw)
In-Reply-To: <CAHk-=wjqGRXUp6KOdx-eHYEotGvY=a5tSY1mF-BkAcX2YAuBYw@mail.gmail.com>

On Wed, 13 Jan 2021 at 23:09, Linus Torvalds
<torvalds@linux-foundation.org> wrote:
>
> On Wed, Jan 13, 2021 at 1:44 PM Russell King - ARM Linux admin
> <linux@armlinux.org.uk> wrote:
> >
> > So, maybe the Sparc issue was just a similar but different bug in gcc
> > 4.9.x.
>
> Good catch. And I know this bug has happened independently on
> different architectures several times (I remember this on x86-64 as
> well), so I started looking around.
>
> And in fact, 4.9 was buggy on x86-64 too:
>
>     https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61904
>
> And yeah, _that_ gcc bug wasn't actually x86-64 specific, but
> apparently a generic instruction scheduling bug.
>
> So it's an independent bug, but I do have to admit that the arguments
> against 4.9 are piling up (even if that particular fix apparently got
> fixed in the gcc branches and apparently backported to distro
> compilers too).
>

So if the arguments are piling up, what is holding us back, other than
inertia? RHEL 7 used to be a factor, but it ships with 4.8 not 4.9, so
its users already need to upgrade. Is anyone aware of a good reason to
keep 4.9 supported? Are any other long term supported distros using
4.9.x?

I know that distros probably backported fixes for all of these issues,
but without a way to interrogate the compiler about this, that doesn't
really make a difference IMHO.

Note that banning 4.9 for arm64 and banning it in general should be
two different changes in any case, as the former will need to be
backported to -stable kernels as well.

 --
Ard.