Re: [PATCH 03/11] jump label: Base patch for jump label

[Jason Baron <jbaron@redhat.com>]

On Fri, Sep 24, 2010 at 04:54:35PM -0400, Mathieu Desnoyers wrote:
> * Steven Rostedt (rostedt@goodmis.org) wrote:
> > On Thu, 2010-09-23 at 14:40 -0400, Jason Baron wrote:
> > 
> > > I agree that we this synchronization for the module __init section.
> > > 
> > > However, I believe we are ok for module removal case. free_module() is
> > > called *after* blocking_notifier_call_chain() call. The
> > > blocking_notifier_call_chain() is going to call back into the jump label
> > > code, grab the jump_label_mutex and remove the reference to the module that
> > > is about to freed. Thus, the jump label code can no longer reference it.
> > > 
> > > So I think the following patch is all that is required here (lightly
> > > tested).
> > > 
> > > Steve, I'll re-post as a separate patch, if we agree on this fix.
> > 
> > Ug, I'm struggling to get ready for my Tokyo trip. I'll try to look at
> > it on the flight. But I still need to write my presentation :-)
> > 
> > I may have a response back till Monday or Tuesday.
> 
> As a note for when you review this patch, I have a strong preference for
> adding a "pre-init-free notifier callback" rather than taking the rcu
> read lock + adding a synchronize_rcu() in module.c. This will make our
> life much easier when we end up doing modifications down in text_poke.
> 
> Thanks,
> 
> Mathieu
> 

Yes, I think we can make this work with a "pre-init-free notifier
callback". In fact, we can use MODULE_STATE_LIVE state that is already
in module.c. I'm in the process of testing a patch using this
alternative method.

thanks,

-Jason