Re: [PATCH] memcg: restore ss->id_lock to spinlock, using RCU for next

[Eric Dumazet <eric.dumazet@gmail.com>]

Le jeudi 19 janvier 2012 A  04:28 -0800, Tejun Heo a A(C)crit :
> Hello,
> 
> On Wed, Jan 18, 2012 at 11:33 PM, Eric Dumazet <eric.dumazet@gmail.com> wrote:
> > Interesting, but should be a patch on its own.
> 
> Yeap, agreed.
> 
> > Maybe other idr users can benefit from your idea as well, if patch is
> > labeled  "idr: allow idr_get_next() from rcu_read_lock" or something...
> >
> > I suggest introducing idr_get_next_rcu() helper to make the check about
> > rcu cleaner.
> >
> > idr_get_next_rcu(...)
> > {
> >        WARN_ON_ONCE(!rcu_read_lock_held());
> >        return idr_get_next(...);
> > }
> 
> Hmmm... I don't know. Does having a separate set of interface help
> much?  It's easy to avoid/miss the test by using the other one.  If we
> really worry about it, maybe indicating which locking is to be used
> during init is better? We can remember the lockdep map and trigger
> WARN_ON_ONCE() if neither the lock or RCU read lock is held.


There is a rcu_dereference_raw(ptr) in idr_get_next()

This could be changed to rcu_dereference_check(ptr, condition) to get
lockdep support for free :)

[ condition would be the appropriate
lockdep_is_held(&the_lock_protecting_my_idr) or 'I use the rcu variant'
and I hold rcu_read_lock ]

This would need to add a 'condition' parameter to idr_gen_next(), but we
have very few users in kernel at this moment.



--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>