Re: [RFC PATCH 4/12 v2] ptrace: introduce the empty "struct ptrace_context"

[Oleg Nesterov <oleg@redhat.com>]

On 05/28, Roland McGrath wrote:
>
> > +	task_lock(tsk);
> > +	if (likely(!tsk->ptrace_ctx))
> > +		tsk->ptrace_ctx = ptrace_ctx;
> > +	else
> > +		kfree(ptrace_ctx);
> > +	task_unlock(tsk);
>
> I don't think it's kosher call kfree with task_lock() held.

Yes, it always makes sense to move as much as possible outside of
spin_lock/etc, but

> The kfree case is a very unlikely path,

Exactly, this kfree() "never" runs under task_lock(), but the code
looks a bit simpler and shorter.

> so it can be:
>
> 	task_lock(tsk);
> 	if (likely(!tsk->ptrace_ctx)) {
> 		tsk->ptrace_ctx = ptrace_ctx;
> 		task_unlock(tsk);
> 		return 0;
> 	}
>
> 	task_unlock(tsk);
> 	kfree(ptrace_ctx);

OK, will change.

Oleg.