[PATCH 04/12] asm-generic/mmiowb: Mark accesses to fix KCSAN warnings

[Nicholas Piggin]

On Mon May 8, 2023 at 12:01 PM AEST, Rohan McLure wrote:
> Prior to this patch, data races are detectable by KCSAN of the following
> forms:
>
> [1] Asynchronous calls to mmiowb_set_pending() from an interrupt context
>     or otherwise outside of a critical section
> [2] Interrupted critical sections, where the interrupt will itself
>     acquire a lock
>
> In case [1], calling context does not need an mmiowb() call to be
> issued, otherwise it would do so itself. Such calls to
> mmiowb_set_pending() are either idempotent or no-ops.
>
> In case [2], irrespective of when the interrupt occurs, the interrupt
> will acquire and release its locks prior to its return, nesting_count
> will continue balanced. In the worst case, the interrupted critical
> section during a mmiowb_spin_unlock() call observes an mmiowb to be
> pending and afterward is interrupted, leading to an extraneous call to
> mmiowb(). This data race is clearly innocuous.
>
> Mark all potentially asynchronous memory accesses with READ_ONCE or
> WRITE_ONCE, including increments and decrements to nesting_count. This
> has the effect of removing KCSAN warnings at consumer's callsites.
>
> Signed-off-by: Rohan McLure <rmclure at linux.ibm.com>
> Reported-by: Michael Ellerman <mpe at ellerman.id.au>
> Reported-by: Gautam Menghani <gautammenghani201 at gmail.com>
> ---
>  include/asm-generic/mmiowb.h | 17 +++++++++++------
>  1 file changed, 11 insertions(+), 6 deletions(-)
>
> diff --git a/include/asm-generic/mmiowb.h b/include/asm-generic/mmiowb.h
> index 5698fca3bf56..0b8b794150db 100644
> --- a/include/asm-generic/mmiowb.h
> +++ b/include/asm-generic/mmiowb.h
> @@ -35,27 +35,32 @@ DECLARE_PER_CPU(struct mmiowb_state, __mmiowb_state);
>  static inline void mmiowb_set_pending(void)
>  {
>  	struct mmiowb_state *ms = __mmiowb_state();
> +	u16 nesting_count = READ_ONCE(ms->nesting_count);

The nesting_count is invariant from the point of view of this context,
so READ_ONCE shouldn't be required AFAIKS? It's sort of not even a
data race.

mmiowb_pending is a data race. I think we could get away without using
READ/WRITE_ONCE, but maybe a bit subtle to bother doing that and
explaining why it's okay.

Thanks,
Nick