Re: Aarch64 EXT4FS inode checksum failures - seems to be weak memory ordering issues - Russell King

From: Russell King - ARM Linux admin <linux@armlinux.org.uk>
To: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Will Deacon <will@kernel.org>,
	Peter Zijlstra <peterz@infradead.org>,
	Arnd Bergmann <arnd@kernel.org>,
	linux-toolchains@vger.kernel.org,
	Mark Rutland <mark.rutland@arm.com>,
	Theodore Ts'o <tytso@mit.edu>,
	"linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>,
	Andreas Dilger <adilger.kernel@dilger.ca>,
	Ext4 Developers List <linux-ext4@vger.kernel.org>,
	Linux ARM <linux-arm-kernel@lists.infradead.org>
Subject: Re: Aarch64 EXT4FS inode checksum failures - seems to be weak memory ordering issues
Date: Fri, 8 Jan 2021 20:29:24 +0000	[thread overview]
Message-ID: <20210108202924.GY1551@shell.armlinux.org.uk> (raw)
In-Reply-To: <CAHk-=whnKkj5CSbj-uG_MVVUsPZ6ppd_MFhZf_kpXDkh2MAVRA@mail.gmail.com>

On Fri, Jan 08, 2021 at 12:02:53PM -0800, Linus Torvalds wrote:
> Well, honestly, I'm always in favor of having people not use ancient
> compilers, but both of the issues at hand do seem to be specific to
> arm64.
> 
> The "gcc before 5.1 generates incorrect stack pointer writes on arm64"
> sounds pretty much deadly, and I think means that yes, for arm64 we
> simply need to require 5.1 or newer.
> 
> I also suspect there is much less reason to use old gcc's on arm64. I
> can't imagine that people really run very old setups, Is some old RHEL
> version even relevant for arm64?

For me, six years old for a compiler is really not "very old" - and,
when I first encountered this problem, it was over 12 months ago.
Apart from the kernel, I am not in the habbit of upgrading stuff for
the sake of upgrading - I tend to stick with what I have and what
works. Not everyone on this planet has a desire to have the latest
and greatest all the time.

Since then, I've _not_ wanted to change the compiler in case the
problem vanishes without explanation - it had the feeling of being
way more serious than a compiler bug, potentially a memory ordering
bug.

It took about a year just to start being able to work out what was
going on - it would take up to about three months to show for me,
and when it did, it spat out an ext4 inode checksum error and made
the rootfs read-only.

To "hide" that by upgrading the compiler, and then to be in the
situation where you do not trust any aarch64 machine with your data
is no real solution. That's exactly where I was until this had been
found. The aarch64 architecture had completely lost my trust as a
viable computing platform - and I was at the point of considering
disposing of all my aarch64 hardware and replacing it with x86.

-- 
RMK's Patch system: https://www.armlinux.org.uk/developer/patches/
FTTP is here! 40Mbps down 10Mbps up. Decent connectivity at last!