Re: [PATCH 04/10] perf/uprobe: RCU-ify find_uprobe()

[Oleg Nesterov]

On 07/09, Peter Zijlstra wrote:

> On Tue, Jul 09, 2024 at 04:32:55PM +0200, Oleg Nesterov wrote:
>
> > Once guard(srcu)(&uprobes_srcu) in handle_swbp() drops the uprobes_srcu lock,
> > utask->active_uprobe can be freed.
>
> Yeah, I've fixed all those already. It's a bit of churn, adding
> inc_not_zero all over the place

I am wondering if we can move delayed_uprobe_remove() from put_uprobe()
to delete_uprobe()... probably not, I forgot everything.

But if we can, then we can probably do

	put_uprobe(uprobe)
	{
		if (refcount_dec_and_test(&uprobe->ref))
			kfree(uprobe);
	}

	uprobe_put_rcu(struct rcu_head *rcu)
	{
		uprobe = container_of(...);
		put_uprobe(uprobe);
	}

	delete_uprobe(uprobe)
	{
		rb_erase(...);
		delayed_uprobe_remove(...);
		...

		call_srcu(&uprobes_srcu, &uprobe->rcu, uprobe_put_rcu);
	}

and avoid inc_not_zero.

Not sure, I am already exhausted ;)

Oleg.