diff mbox

[03/10] mac80211: Use rcu_barrier() on unload.

Message ID 1245837965.6695.45.camel@localhost.localdomain
State Not Applicable, archived
Delegated to: David Miller
Headers show

Commit Message

Jesper Dangaard Brouer June 24, 2009, 10:06 a.m. UTC 
On Tue, 2009-06-23 at 17:15 +0200, Johannes Berg wrote:
> On Tue, 2009-06-23 at 17:04 +0200, Jesper Dangaard Brouer wrote:
> > The mac80211 module uses rcu_call() thus it should use rcu_barrier()
> > on module unload.
> > 
> > I'm having a hardtime verifying that no more call_rcu() callbacks can
> > be schedules in the module unload path.  Could a maintainer please
> > look into this?
> > 
> > Signed-off-by: Jesper Dangaard Brouer <hawk@comx.dk>
> > ---
> > 
> >  net/mac80211/main.c |    2 ++
> >  1 files changed, 2 insertions(+), 0 deletions(-)
> > 
> > diff --git a/net/mac80211/main.c b/net/mac80211/main.c
> > index 092a017..e9f70ce 100644
> > --- a/net/mac80211/main.c
> > +++ b/net/mac80211/main.c
> > @@ -1100,6 +1100,8 @@ static void __exit ieee80211_exit(void)
> >  		ieee80211s_stop();
> >  
> >  	ieee80211_debugfs_netdev_exit();
> > +
> > +	rcu_barrier(); /* Wait for completion of call_rcu()'s */
> >  }
> 
> I don't think this is correct at all -- note that call_rcu() is done in
> some of the mesh code, so I would think you need to do this in
> ieee80211_stop() since the call_rcu() code requires the interface to
> still be around. And when it's stopped everything should be idle.

Should it then not be in mesh.c ieee80211_stop_mesh().  We can replace
the synchronize_rcu() in this function with a rcu_barrier().

> I can fix it later, but I'm deep in some other stuff right now.

Yes, I noticed you seem quite active :-)
I can also do a repost... what about the patch below?

Comments

Johannes Berg June 24, 2009, 10:21 a.m. UTC | #1 
On Wed, 2009-06-24 at 12:06 +0200, Jesper Dangaard Brouer wrote:
> On Tue, 2009-06-23 at 17:15 +0200, Johannes Berg wrote:
> > On Tue, 2009-06-23 at 17:04 +0200, Jesper Dangaard Brouer wrote:
> > > The mac80211 module uses rcu_call() thus it should use rcu_barrier()
> > > on module unload.
> > > 
> > > I'm having a hardtime verifying that no more call_rcu() callbacks can
> > > be schedules in the module unload path.  Could a maintainer please
> > > look into this?
> > > 
> > > Signed-off-by: Jesper Dangaard Brouer <hawk@comx.dk>
> > > ---
> > > 
> > >  net/mac80211/main.c |    2 ++
> > >  1 files changed, 2 insertions(+), 0 deletions(-)
> > > 
> > > diff --git a/net/mac80211/main.c b/net/mac80211/main.c
> > > index 092a017..e9f70ce 100644
> > > --- a/net/mac80211/main.c
> > > +++ b/net/mac80211/main.c
> > > @@ -1100,6 +1100,8 @@ static void __exit ieee80211_exit(void)
> > >  		ieee80211s_stop();
> > >  
> > >  	ieee80211_debugfs_netdev_exit();
> > > +
> > > +	rcu_barrier(); /* Wait for completion of call_rcu()'s */
> > >  }
> > 
> > I don't think this is correct at all -- note that call_rcu() is done in
> > some of the mesh code, so I would think you need to do this in
> > ieee80211_stop() since the call_rcu() code requires the interface to
> > still be around. And when it's stopped everything should be idle.
> 
> Should it then not be in mesh.c ieee80211_stop_mesh().  We can replace
> the synchronize_rcu() in this function with a rcu_barrier().

Yes, this seems correct.

johannes
Jesper Dangaard Brouer June 24, 2009, 11:32 a.m. UTC | #2 
On Wed, 2009-06-24 at 12:21 +0200, Johannes Berg wrote:
> On Wed, 2009-06-24 at 12:06 +0200, Jesper Dangaard Brouer wrote:
> > On Tue, 2009-06-23 at 17:15 +0200, Johannes Berg wrote:
> > > On Tue, 2009-06-23 at 17:04 +0200, Jesper Dangaard Brouer wrote:
> > > > The mac80211 module uses rcu_call() thus it should use rcu_barrier()
> > > > on module unload.
> > > > 
> > > > I'm having a hardtime verifying that no more call_rcu() callbacks can
> > > > be schedules in the module unload path.  Could a maintainer please
> > > > look into this?
> > > > 
> > > > Signed-off-by: Jesper Dangaard Brouer <hawk@comx.dk>
> > > > ---
> > > > 
> > > >  net/mac80211/main.c |    2 ++
> > > >  1 files changed, 2 insertions(+), 0 deletions(-)
> > > > 
> > > > diff --git a/net/mac80211/main.c b/net/mac80211/main.c
> > > > index 092a017..e9f70ce 100644
> > > > --- a/net/mac80211/main.c
> > > > +++ b/net/mac80211/main.c
> > > > @@ -1100,6 +1100,8 @@ static void __exit ieee80211_exit(void)
> > > >  		ieee80211s_stop();
> > > >  
> > > >  	ieee80211_debugfs_netdev_exit();
> > > > +
> > > > +	rcu_barrier(); /* Wait for completion of call_rcu()'s */
> > > >  }
> > > 
> > > I don't think this is correct at all -- note that call_rcu() is done in
> > > some of the mesh code, so I would think you need to do this in
> > > ieee80211_stop() since the call_rcu() code requires the interface to
> > > still be around. And when it's stopped everything should be idle.
> > 
> > Should it then not be in mesh.c ieee80211_stop_mesh().  We can replace
> > the synchronize_rcu() in this function with a rcu_barrier().
> 
> Yes, this seems correct.
> 
> johannes

Can I consider this a:

Acked-by: Johannes Berg <johannes@sipsolutions.net>

???

DaveM seems to like this as patchwork.ozlabs.org picks up this
automatically...
Johannes Berg June 24, 2009, 11:39 a.m. UTC | #3 
On Wed, 2009-06-24 at 13:32 +0200, Jesper Dangaard Brouer wrote:

> > > Should it then not be in mesh.c ieee80211_stop_mesh().  We can replace
> > > the synchronize_rcu() in this function with a rcu_barrier().
> > 
> > Yes, this seems correct.
> > 
> > johannes
> 
> Can I consider this a:
> 
> Acked-by: Johannes Berg <johannes@sipsolutions.net>

Yeah, 
Acked-by: Johannes Berg <johannes@sipsolutions.net>

johannes
diff mbox

Patch

diff --git a/net/mac80211/mesh.c b/net/mac80211/mesh.c
index fc712e6..11cf45b 100644
--- a/net/mac80211/mesh.c
+++ b/net/mac80211/mesh.c
@@ -494,7 +494,7 @@  void ieee80211_stop_mesh(struct ieee80211_sub_if_data *sdata)
 	 * should it be using the interface and enqueuing
 	 * frames at this very time on another CPU.
 	 */
-	synchronize_rcu();
+	rcu_barrier(); /* Wait for RX path and call_rcu()'s */
 	skb_queue_purge(&sdata->u.mesh.skb_queue);
 }