Re: arm64: fp-stress: BUG: KFENCE: memory corruption in fpsimd_release_task

[Mark Brown]

On Mon, May 22, 2023 at 05:41:17PM +0100, Will Deacon wrote:
> On Tue, May 16, 2023 at 02:44:49PM +0100, Will Deacon wrote:
> > On Tue, May 16, 2023 at 11:58:40AM +0530, Naresh Kamboju wrote:

> > > Following kernel BUG noticed while running selftests arm64 fp-stress
> > > running stable rc kernel versions 6.1.29-rc1 and 6.3.3-rc1.

> > Is there a known-good build so that we could attempt a bisection?

> FWIW, I've been trying (and failing) all day to reproduce this in QEMU.
> I matched the same VL configuration as you have in the fastmodel and
> tried enabling additional memory debugging options too, but I'm yet to
> see a kfence splat (or any other splat fwiw).

> How often do you see this?

As I said in another mail I've also been unable to reproduce this.  FWIW
I *suspect* that it might need to be run in the context of a full
kselftest run to manifest rather than just running fp-stress in
isolation, that's mostly a guess but given that the kfence trap appeared
to be happening on free at a point where the test program shouldn't be
exiting any tasks and should not be changing vector lengths on tasks
that have used either of the vector extensions.

Attachment: signature.asc
Description: PGP signature