Subject: Re: [PATCH 2/3] tracing/profile: add ref count for registering
 profile events
From: Steven Rostedt <rostedt@goodmis.org>
Reply-To: rostedt@goodmis.org
To: Peter Zijlstra <peterz@infradead.org>
Cc: linux-kernel@vger.kernel.org, Ingo Molnar <mingo@elte.hu>,
       Andrew Morton <akpm@linux-foundation.org>,
       Frederic Weisbecker <fweisbec@gmail.com>,
       Li Zefan <lizf@cn.fujitsu.com>,
       Mathieu Desnoyers <compudj@krystal.dyndns.org>
In-Reply-To: <1252678183.7126.34.camel@laptop>
References: <20090911135452.866274568@goodmis.org>
	 <20090911135626.421489476@goodmis.org>  <1252677861.7126.32.camel@laptop>
	 <1252678183.7126.34.camel@laptop>
Content-Type: text/plain
Organization: Kihon Technologies Inc.
Date: Fri, 11 Sep 2009 10:33:26 -0400
Message-Id: <1252679606.18996.681.camel@gandalf.stny.rr.com>
Mime-Version: 1.0
Content-Transfer-Encoding: 7bit
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 2031
Lines: 60

On Fri, 2009-09-11 at 16:09 +0200, Peter Zijlstra wrote:
> On Fri, 2009-09-11 at 16:04 +0200, Peter Zijlstra wrote:
> > On Fri, 2009-09-11 at 09:54 -0400, Steven Rostedt wrote:
> > 
> > > +#ifdef MODULE
> > > +# define event_trace_up_ref()					\
> > > +	do {							\
> > > +		if (!try_module_get(THIS_MODULE)) {		\
> > > +			atomic_dec(&event_call->profile_count);	\
> > > +			return -ENOENT;				\
> > > +		}						\
> > > +	} while (0)
> > > +# define event_trace_down_ref() module_put(THIS_MODULE)
> > > +#else
> > > +# define event_trace_up_ref() do { } while (0)
> > > +# define event_trace_down_ref() do { } while (0)
> > > +#endif
> > 
> > That's like truely gruesomely ugly.
> > 
> > At the very least write it like:
> > 
> > int event_trace_up_ref(struct ftrace_event_call *call)
> > {
> > 	if (!try_module_get(THIS_MODULE)) {
> > 		atomic_dev(&call->profile_count);
> > 		return -ENOENT;
> > 	}
> > 	return 0;
> > }
> 
> Or we can go with Li's original patch, that was less ugly.

I can go back to Li's original patch, but the talk on that was
"fragile". If you no longer feel that way, then I'll use his instead.

For now, I'll pull out this patch altogether, and resubmit the pull
request without it. I'd like the other changes to not be held up by
this.

> 
> I still think tracepoints/markers should sort this out, because we now
> have a sematic difference between the two wrt modules.

I originally tried to do it in the tracepoint logic, but that broke a
lot of assumptions about tracepoints that Mathieu pointed out. This is
not a normal use of tracepoints. It is expected that if you register a
probe in a module, you will unregister it before exiting.

I can't remember all the details, but at the end, it seemed that the fix
belonged at the ftrace level.

-- Steve


--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/