Re: [PATCH] mac80211: hoist sta->lock from reorder release timer

[Christian Lamparter <chunkeey@googlemail.com>]

On Wednesday 06 October 2010 12:10:26 Johannes Berg wrote:
> On Wed, 2010-10-06 at 12:00 +0200, Christian Lamparter wrote:
> 
> > The timer itself is part of the station's private struct.
> > The clean-up routine will deactivate the timer as soon as
> > the station is removed. Therefore the extra sta->lock
> > protection should not be necessary.
> 
> >  	rcu_read_lock();
> > -	spin_lock(&sta->lock);
> >  	ieee80211_release_reorder_timeout(sta, *ptid);
> > -	spin_unlock(&sta->lock);
> >  	rcu_read_unlock();
> 
> There's a comment on ieee80211_release_reorder_timeout() saying that the
> lock must be held -- which is probably not true? We don't generally hold
> that lock on the RX path...?

That comment is more or less a 1:1 copy from the comment about
struct tid_ampdu_rx (in sta_info.h).

> * This structure is protected by RCU and the per-station
> * spinlock. Assignments to the array holding it must hold
> * the spinlock, only the RX path can access it under RCU
> * lock-free.

thing is: we now have the reorder_lock which protects the
reorder buffer against "destructive access". So, is it "ok"
to trim the comments a bit?