Re: [PATCH tracing/kprobes 5/7] tracing/kprobes: Add argument name support

[Masami Hiramatsu <mhiramat@redhat.com>]

Masami Hiramatsu wrote:
> Steven Rostedt wrote:
>> On Thu, 2009-09-10 at 19:53 -0400, Masami Hiramatsu wrote:
>>
>>> @@ -870,9 +892,13 @@ print_kprobe_event(struct trace_iterator *iter, int flags)
>>>    {
>>>    	struct kprobe_trace_entry *field;
>>>    	struct trace_seq *s =&iter->seq;
>>> +	struct trace_event *event;
>>> +	struct trace_probe *tp;
>>>    	int i;
>>>
>>>    	field = (struct kprobe_trace_entry *)iter->ent;
>>> +	event = ftrace_find_event(field->ent.type);
>>> +	tp = container_of(event, struct trace_probe, event);
>>
>> Can this function be called the data is in the ring buffer, but the
>> probe has been unregistered? If so, the result of ftrace_find_event be
>> NULL?
> 
> Hmm, it will depend on ftrace implementation. Before releasing
> trace_probe, kprobe tracer tries to unregister event call.
> If it's correctly locking mutex or some rw_lock for both of
> unregistering and printing, it will be safe.
> 
> Unfortunately, it seems not :-(.
> 
> In trace_events.c,
> 1054 static void __trace_remove_event_call(struct ftrace_event_call *call)
> 1055 {
> 1056         ftrace_event_enable_disable(call, 0);
> 1057         if (call->event)
> 1058                 __unregister_ftrace_event(call->event);
> 
> What we need to do is calling unregister_ftrace_event() instead of
> __unregister_ftrace_event.

Aah, NO. the caller of trace_remove_event_call() should have
trace_event_mutex before calling!


-- 
Masami Hiramatsu

Software Engineer
Hitachi Computer Products (America), Inc.
Software Solutions Division

e-mail: mhiramat@redhat.com