[PATCH v4] arm64: Enable KCSAN

[Mark Rutland]

On Tue, Dec 14, 2021 at 08:11:24PM +0100, Marco Elver wrote:
> On Tue, 14 Dec 2021 at 19:24, Mark Rutland <mark.rutland at arm.com> wrote:
> >
> >   Reviewed-by: Mark Rutland <mark.rutland at arm.com>
> >   Tested-by: Mark Rutland <mark.rutland at arm.com>
> 
> Thanks for taking a look and testing! Thought I'd update you re status
> of some of the reports below. :-)

Thanks for the pointers below! I'll fold those into a test branch, so that I
can soak this under Syzkaller over the holiday break.

I've given precise refernces below in case I've misunderstood or missed
something.

> > * BUG: KCSAN: data-race in mutex_spin_on_owner+0xcc/0x150
> > * BUG: KCSAN: data-race in rwsem_spin_on_owner+0xa8/0x13c
> 
> These are going away, fixes are already in -tip/-next.

Cool! I assume that means commits:

* c0bed69daf4b6780 ("locking: Make owner_on_cpu() into <linux/sched.h>")
  from https://lore.kernel.org/r/20211203075935.136808-2-wangkefeng.wang@huawei.com

* 4cf75fd4a2545ca4 ("locking: Mark racy reads of owner->on_cpu")
  from https://lore.kernel.org/r/20211203075935.136808-3-wangkefeng.wang@huawei.com

... in the tip locking/core branch (the head commit of which is currently
5fb6e8cf53b005d2 ("locking/atomic: atomic64: Remove unusable atomic ops")).

> > * UBSAN: object-size-mismatch in net/unix/af_unix.c:977:14
> 
> The UBSAN object-size-mismatch warnings are going away, as
> fsanitize=object-size is broken/incomplete as it turns out --
> UBSAN_OBJECT_SIZE will be removed from 5.17:
> https://bugzilla.kernel.org/show_bug.cgi?id=214861#c4

I see the relevant patch is:

* "[PATCH] ubsan: Remove CONFIG_UBSAN_OBJECT_SIZE"
  from https://lore.kernel.org/all/20211203235346.110809-1-keescook@chromium.org/

... and from looking at next-20211214 that's been picked up by Andrew Morton.

Thanks,
Mark.