LKML Archive on
help / color / mirror / Atom feed
From: Mathieu Desnoyers <>
To: Lai Jiangshan <>
	"Martin J. Bligh" <>,
	Steven Rostedt <>, Ingo Molnar <>,
	Linus Torvalds <>,
	Andrew Morton <>,
	Peter Zijlstra <>,,,
Subject: Re: [ltt-dev] LTTng 0.44 and LTTV 0.11.3
Date: Wed, 29 Oct 2008 13:40:01 -0400	[thread overview]
Message-ID: <20081029174001.GA30796@Krystal> (raw)
In-Reply-To: <>

* Lai Jiangshan ( wrote:
> Mathieu Desnoyers wrote:
> > - I have also vastly simplified locking in the markers and tracepoints
> >   by using _only_ the modules mutex. I actually took this mutex out of
> >   module.c and created its own file so tracepoints and markers can use
> >   it. That should please Lai Jiangshan. Although he may have some work
> >   to do to see how his new probes manager might benefit from it.
> > 
> >   See :
> >;a=commitdiff;h=7aea87ac46df7613d68034f5904bc8d575069076
> >   and
> >;a=commitdiff;h=5f6814237f7a67650e7b6214d916825e3f8fc1b7
> >;a=commitdiff;h=410ba66a1cbe27a611e1c18c0a53e87b4652a2c9
> > 
> Hi, Mathieu,
> I strongly reject for removing tracepoint_mutex and marker_mutex.
> As an independent subsystem, we should use our own locks. Do not use others.
> otherwise coupling will be increased in linux kernel.
> I condemn unnecessary coupling.
> Our tracepoint & marker had tied to modules(for traveling all tracepoints
> or markers). The best thing is that we do not increase the coupling.
> [PATCH 2/2] tracepoint: introduce *_noupdate APIs.
> is helpful for auto-active-tracepoint-mechanism.
> 		Thanx, Lai.

Hi Lai,

The approach you propose looks interesting. Please see below to make
sure we are on the same page.

The problem is that when we want to connect
markers/tracepoints/immediate values together, it results in a real
locking mess between


When we want to take care of a marker at module load, we have to insure
the following calling scenario is correct :

  call markers_update_probes_range() (on the module markers)
    call tracepoint register (to automatically enable a tracepoint
                               when a marker is connected to it)
      call tracepoints_update_probe_range (on kernel core and all modules)
        call imv_update_range (on kernel core and all modules)

The current locking status of tracepoints vs markers does not currently
allow tracepoints_register to be called from the marker update because
it would take the modules_mutex twice.

What you propose is something like this :

  call markers_update_probes_range()
    call tracepoint_register_noupdate (to automatically enable a tracepoint
                                       when a marker is connected to it)
  call tracepoints_update_all() (for core kernel and all modules (*))
    name##__imv = (i)
  call imv_update_all() (for core kernel and all modules (*))

(*) This is required because registering a tracepoint might have impact
    outside of the module in which the marker is located. Same for
    changing an immediate value.

And on marker_register_probe() :
  call markers_update_probes_range()
    call tracepoint_register_noupdate
  call tracepoints_update_all()
    name##__imv = (i)
  call imv_update_all()

Which basically uses the same trick I used for immediate values : it
separates the "backing data" update (name##_imv = (i)) from the actual
update that needs to iterate on the modules.

The only thing we have to be aware of is that it actually couples
markers/tracepoints/immediate values much more thightly to keep separate
locking for each, because, as the example above shows, the markers have
to be aware that they must call tracepoints_update_all and
imv_update_all explicitely. On the plus side, it requires much less
iterations on the module sections, which is a clear win.

So the expected mutex nesting order is (indent implies "nested in"):

On load_module :


On marker register :


On tracepoint register :


On imv_update :


So yes, I think your approach is good, although there are some
implementation quirks in the patch you submitted. I'll comment by
replying to your other post.



> > So hopefully everyone will be happy with this new release. :)
> > 
> > Mathieu
> > 
> --
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to
> More majordomo info at
> Please read the FAQ at

Mathieu Desnoyers
OpenPGP key fingerprint: 8CD5 52C3 8E3C 4140 715F  BA06 3F25 A8FE 3BAE 9A68

  reply	other threads:[~2008-10-29 17:45 UTC|newest]

Thread overview: 5+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2008-10-24  0:45 Mathieu Desnoyers
2008-10-28  3:55 ` [ltt-dev] " Lai Jiangshan
2008-10-29 17:40   ` Mathieu Desnoyers [this message]
2008-10-30  3:32     ` [sadump 04308] " Lai Jiangshan
2008-10-30  6:01       ` [ltt-dev] [sadump 04308] " Mathieu Desnoyers

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20081029174001.GA30796@Krystal \ \ \ \ \ \ \ \ \ \ \ \ \
    --subject='Re: [ltt-dev] LTTng 0.44 and LTTV 0.11.3' \

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).