Re: [PATCH 1/2] module: use rcu to protect module list read

[Rusty Russell <rusty@rustcorp.com.au>]

On Sat, 10 Mar 2012 07:25:57 -0800, Eric Dumazet <eric.dumazet@gmail.com> wrote:
> Le samedi 10 mars 2012 à 22:20 +0800, Cong Wang a écrit :
> > Now the read of module list is protected by preempt disable + *_rcu
> > list operations, this is odd, as RCU read lock should be able to
> > protect it directly. This patch makes the read of module list
> > protected by RCU read lock and the write still protected by
> > module_mutex.
> > 
> 
> Problem is that your patch does more than that.
> 
> In set_all_modules_text_rw() and set_all_modules_text_ro() you removed
> the mutex in favor of rcu_read_lock()
> 
> Also, module code uses synchronize_sched(), not synchronize_rcu()

Yes, but only for paranoia.  Really, it's vs. stop_machine().

> Take a look at Documentation/RCU/whatisRCU.txt and see that
> preempt_disable() / preempt_enable() are documented as a right protect
> code, in line 333.
> 
> You added races in /proc/modules as well.

I'm surprised that patch didn't warn... CONFIG_DEBUG_ATOMIC_SLEEP=y
might help here....

Cheers,
Rusty.
-- 
  How could I marry someone with more hair than me?  http://baldalex.org