Re: [PATCH v2 4/4] tracing/kprobes: Turn trace_probe->files into list_head

[Oleg Nesterov <oleg@redhat.com>]

On 07/01, Steven Rostedt wrote:
>
> On Thu, 2013-06-20 at 19:38 +0200, Oleg Nesterov wrote:
>
> > -static int
> > -trace_probe_file_index(struct trace_probe *tp, struct ftrace_event_file *file)
> > +static struct event_file_link *
> > +find_event_file_link(struct trace_probe *tp, struct ftrace_event_file *file)
> >  {
> > -	struct ftrace_event_file **files;
> > -	int i;
> > +	struct event_file_link *link;
> >
> > -	/*
> > -	 * Since all tp->files updater is protected by probe_enable_lock,
> > -	 * we don't need to lock an rcu_read_lock.
> > -	 */
> > -	files = rcu_dereference_raw(tp->files);
> > -	if (files) {
> > -		for (i = 0; files[i]; i++)
> > -			if (files[i] == file)
> > -				return i;
> > -	}
> > +	list_for_each_entry(link, &tp->files, list)
> > +		if (link->file == file)
> > +			return link;
>
> Shouldn't that be list_for_each_entry_rcu()?

No.

This is the writer which modifies the list. enable/disable_trace_probe
should be serialized wrt each other / itself anyway, otherwise they are
buggy in any case.

Oleg.