Re: [PATCH] rust: DmaFence: Don't drop on signal

[Philipp Stanner <phasta@mailbox.org>]

On Thu, 2026-10-22 at 18:27 +0200, Philipp Stanner wrote:
> +
> +            // SAFETY: `self.data` is valid because `self` is valid.
> +            let fctx = unsafe { self.data.as_ref().fctx };
> +            let _ = fctx.nr_of_unsignaled_fences.fetch_sub(1, Relaxed);
> +

When I think about it we probably want to address this in a series
(preparing for JQ upstreaming) where we also have my proposed
DriverFence::drop(synchronize_rcu). Otherwise this patch would need
both synchronize_rcu() and rcu_barrier() for formal correctness when
the FenceContext drops, which is just overkill.

If only we could fully lock the fence, all this would just go away. But
currently that's what it is.

So forget about this patch. I send sth new soonish.

P.