From mboxrd@z Thu Jan  1 00:00:00 1970
From: mingo@elte.hu (Ingo Molnar)
Date: Wed, 25 Mar 2009 09:40:54 +0100
Subject: [ltt-dev] [patch 7/9] LTTng instrumentation - kernel
In-Reply-To: <200903251143.12472.rusty@rustcorp.com.au>
References: <20090324155625.420966314@polymtl.ca>
	<20090324160148.872305963@polymtl.ca>
	<20090324183313.GH31117@elte.hu>
	<200903251143.12472.rusty@rustcorp.com.au>
Message-ID: <20090325084054.GB11217@elte.hu>


* Rusty Russell <rusty at rustcorp.com.au> wrote:

> On Wednesday 25 March 2009 05:03:13 Ingo Molnar wrote:
> > 
> > (Rusty Cc:-ed - for the module.c tracepoints below)
> 
> Thanks, tho they look fine and non-intrusive to me.

Thanks - i'll take this as an Acked-by :-)

( Mathieu, mind re-sending a variant that does them via
  TRACE_EVENT(), against the tracing tree? That makes it useful not 
  just to LTTng but the default mainline kernel as well. Thanks! )

> > I believe that to have a complete picture of module usage, module 
> > refcount get/put events should be included as well, beyond the basic 
> > load/free events.
> > 
> > These both have performance impact (a module get/put in a fastpath 
> > hurts scalability), and are informative in terms of establishing the 
> > module dependency graph.
> 
> A module_get()/put() should not hurt scalability at all!  I went 
> to great and horrible lengths to ensure that was the case since 
> the rewrite in 2.4.

i know, it uses percpu refcounts :) Still it can be somewhat 
non-trivial in a fastpath:

static inline int try_module_get(struct module *module)
{
        int ret = 1;

        if (module) {
                unsigned int cpu = get_cpu();
                if (likely(module_is_live(module)))
                        local_inc(__module_ref_addr(module, cpu));
                else
                        ret = 0;
                put_cpu();
        }
        return ret;
}

So we want to reduce excessive uses of it.

[ And please forget i mentioned scalability - that bit is fine and 
  you were offended rightfully :) ]

This reminds me. Couldnt we now:

 #ifdef CONFIG_SMP
         char *refptr;
 #else
         local_t ref;
 #endif

... unify these bits to just standardize on a per-cpu refptr all the 
time, with Tejun's ueber-cool percpu changes in place?

> But a module dependency graph et. al. would be kind of cool.

Yeah. Can be in a separate patch as well - load/unload events are 
meaningful in isolation as well.

	Ingo