Aarch64 EXT4FS inode checksum failures - seems to be weak memory ordering issues

[Arnd Bergmann]

On Wed, Jan 6, 2021 at 10:04 PM Arnd Bergmann <arnd at kernel.org> wrote:
> On Wed, Jan 6, 2021 at 6:22 PM Will Deacon <will at kernel.org> wrote:
> > On Wed, Jan 06, 2021 at 01:52:53PM +0000, Russell King - ARM Linux admin wrote:
>
> I tried the Image-5.9.0 on a virtual machine with seven CPUs (two clusters)
> running in an M1 mac mini and ran these commands inside.
>
> > With that, I see the following after ten seconds or so:
> >
> >   EXT4-fs error (device sda2): ext4_lookup:1707: inode #674497: comm md5sum: iget: checksum invalid
> >
> > Russell, Mark -- does this recipe explode reliably for you too?
>
> Negative unfortunately -- no checksum mismatch so far, with 10 minutes
> elapsed. I'll keep it running a bit longer.

I managed to trigger the checksum mismatch once now, after around 40
minutes, with a second run going for 20 minutes without mismatch.

So it's not easily reproducible for me, but it does help to rule out
at least some of the hardware specific theories -- it's not just the
Cortex-A72, nor the CCI doing something weird, as neither of
them are in use here.

This is the output I got:

EXT4-fs error (device vda2): ext4_lookup:1707: inode #1185501: comm
md5sum: iget: checksum invalid
Aborting journal on device vda2-8.
EXT4-fs error (device vda2): ext4_journal_check_start:83: Detected
aborted journal
EXT4-fs (vda2): Remounting filesystem read-only
EXT4-fs (vda2): Remounting filesystem read-only
EXT4-fs error (device vda2): ext4_journal_check_start:83: Detected
aborted journal

      Arnd