Re: [PATCH 3/6] uprobes/tracing: Make uprobe_{trace,perf}_print() uretprobe-friendly

[Srikar Dronamraju <srikar@linux.vnet.ibm.com>]

* Oleg Nesterov <oleg@redhat.com> [2013-04-09 15:33:33]:

> On 04/07, Srikar Dronamraju wrote:
> >
> > * Oleg Nesterov <oleg@redhat.com> [2013-04-01 18:08:51]:
> >
> > > diff --git a/kernel/trace/trace_uprobe.c b/kernel/trace/trace_uprobe.c
> > > index e91a354..db2718a 100644
> > > --- a/kernel/trace/trace_uprobe.c
> > > +++ b/kernel/trace/trace_uprobe.c
> > > @@ -515,15 +515,26 @@ static void uprobe_trace_print(struct trace_uprobe *tu,
> > >  	int size, i;
> > >  	struct ftrace_event_call *call = &tu->call;
> > >
> > > -	size = SIZEOF_TRACE_ENTRY(1) + tu->size;
> > > +	if (is_ret_probe(tu))
> >
> > One nit:
> > Here and couple of places below .. we could check for func instead of
> > is_ret_probe() right?
> 
> Yes we could. And note that we do not really need both uprobe_trace_func()
> and uretprobe_perf_func(), we could use a single function and check "func".
> 
> But:
> 
> > Or is there an advantage of checking is_ret_probe() over func?
> 
> I believe yes. Firstly, we can't use 0ul as "invalid func address" to detect
> the !is_ret_probe() case, we need, say, -1ul which probably needs a symbolic
> name. In fact, I'd prefer to add another "is_return" argument if we want to
> avoid is_ret_probe() and unify 2 functions.
> 
> But more importantly, I think that is_ret_probe() is much more grep-friendly
> and thus more understandable and consistent with other checks which can not
> rely on "func".

Okay, Agree.