Re: discussion about registering task_finder in user space probing

[Wenji Huang <wenji dot huang at oracle dot com>]

Hi Mark,

Thanks for your reply.

Mark Wielaard wrote:
> Hi Wenji,
>
> On Tue, 2009-03-24 at 17:06 +0800, Wenji Huang wrote:
> > I found our task_finder embedded in each stap_uprob_spec even it is 
> > same. That will cause many times registering similar task_finder.
> > [...]
> >    for (i=0; i<220; i++) {
> >      struct stap_uprobe_spec *sups = & stap_uprobe_specs[i];
> >      probe_point = sups->pp;
> >      if (sups->finder.pathname) sups->finder.callback = & 
> > stap_uprobe_process_found;
> >      else if (sups->pathname) sups->finder.vm_callback = & 
> > stap_uprobe_vmchange_found;
> >      rc = stap_register_task_finder_target (& sups->finder);
> >      if (rc) break;
> >    }
> > [snip]
> >
> > So we will have 220 times registering similar task_finder. I think
> > the probe point is irrelevant to task_finder. Can we just keep only one 
> > and invoke callbacks to register all the stap_uprobes[i] once interested 
> > process found?
>
> I was just looking at this since I am doing something similar on the
> pr6866 branch (not just for uprobes, but also for utrace and itrace
> probes). My idea was to have stap_register_task_finder_target ()
> recognize when the callback is already registered for that particular
> pathname/pid. stap_register_task_finder_target () already avoids
> duplication when it sees an existing entry for pathname/pid. So where
Yes, it will search the existing entry before adding entry to list.

> there is now a
>
>     // Add this target to the callback list for this task.
>     list_add_tail(&new_tgt->callback_list, &tgt->callback_list_head);
>
> We could do a lookup of whether the callback is already in the list and
> only add it when it isn't already there. What do you think?

Sounds good. Current code is harmless, but it's better if we could make 
it more graceful.

Regards,
Wenji