[PATCH] mac80211: use synchronize_rcu() with rcu_barrier()

[PATCH] mac80211: use synchronize_rcu() with rcu_barrier()

[Bob Copeland]

The RCU docs used to state that rcu_barrier() included a wait
for an RCU grace period; however the comments for rcu_barrier()
as of commit f0a0e6f... "rcu: Clarify memory-ordering properties
of grace-period primitives" contradict this.

So add back synchronize_{rcu,net}() to where they once were,
but keep the rcu_barrier()s for the call_rcu() callbacks.

Cc: stable <stable@vger.kernel.org>
Signed-off-by: Bob Copeland <bob@cozybit.com>
---
 net/mac80211/iface.c |    5 +++--
 net/mac80211/pm.c    |    1 +
 2 files changed, 4 insertions(+), 2 deletions(-)

diff --git a/net/mac80211/iface.c b/net/mac80211/iface.c
index b6abaaa..c634aff 100644
--- a/net/mac80211/iface.c
+++ b/net/mac80211/iface.c
@@ -836,11 +836,12 @@ static void ieee80211_do_stop(struct ieee80211_sub_if_data *sdata,
 		 *
 		 * sta_info_flush_cleanup() requires rcu_barrier()
 		 * first to wait for the station call_rcu() calls
-		 * to complete, here we need at least sychronize_rcu()
-		 * it to wait for the RX path in case it is using the
+		 * to complete, and we also need synchronize_rcu()
+		 * to wait for the RX path in case it is using the
 		 * interface and enqueuing frames at this very time on
 		 * another CPU.
 		 */
+		synchronize_rcu();
 		rcu_barrier();
 		sta_info_flush_cleanup(sdata);
 
diff --git a/net/mac80211/pm.c b/net/mac80211/pm.c
index 4431f0f..7fc5d0d 100644
--- a/net/mac80211/pm.c
+++ b/net/mac80211/pm.c
@@ -38,6 +38,7 @@ int __ieee80211_suspend(struct ieee80211_hw *hw, struct cfg80211_wowlan *wowlan)
 					IEEE80211_QUEUE_STOP_REASON_SUSPEND);
 
 	/* flush out all packets and station cleanup call_rcu()s */
+	synchronize_net();
 	rcu_barrier();
 
 	ieee80211_flush_queues(local, NULL);
-- 
1.7.10.4

-- 
Bob Copeland %% www.bobcopeland.com

Re: [PATCH] mac80211: use synchronize_rcu() with rcu_barrier()

[Paul E. McKenney]

On Thu, Apr 18, 2013 at 06:04:43PM -0400, Bob Copeland wrote:
> The RCU docs used to state that rcu_barrier() included a wait
> for an RCU grace period; however the comments for rcu_barrier()
> as of commit f0a0e6f... "rcu: Clarify memory-ordering properties
> of grace-period primitives" contradict this.
> 
> So add back synchronize_{rcu,net}() to where they once were,
> but keep the rcu_barrier()s for the call_rcu() callbacks.
> 
> Cc: stable <stable@vger.kernel.org>
> Signed-off-by: Bob Copeland <bob@cozybit.com>

Reviewed-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>

> ---
>  net/mac80211/iface.c |    5 +++--
>  net/mac80211/pm.c    |    1 +
>  2 files changed, 4 insertions(+), 2 deletions(-)
> 
> diff --git a/net/mac80211/iface.c b/net/mac80211/iface.c
> index b6abaaa..c634aff 100644
> --- a/net/mac80211/iface.c
> +++ b/net/mac80211/iface.c
> @@ -836,11 +836,12 @@ static void ieee80211_do_stop(struct ieee80211_sub_if_data *sdata,
>  		 *
>  		 * sta_info_flush_cleanup() requires rcu_barrier()
>  		 * first to wait for the station call_rcu() calls
> -		 * to complete, here we need at least sychronize_rcu()
> -		 * it to wait for the RX path in case it is using the
> +		 * to complete, and we also need synchronize_rcu()
> +		 * to wait for the RX path in case it is using the
>  		 * interface and enqueuing frames at this very time on
>  		 * another CPU.
>  		 */
> +		synchronize_rcu();
>  		rcu_barrier();
>  		sta_info_flush_cleanup(sdata);
> 
> diff --git a/net/mac80211/pm.c b/net/mac80211/pm.c
> index 4431f0f..7fc5d0d 100644
> --- a/net/mac80211/pm.c
> +++ b/net/mac80211/pm.c
> @@ -38,6 +38,7 @@ int __ieee80211_suspend(struct ieee80211_hw *hw, struct cfg80211_wowlan *wowlan)
>  					IEEE80211_QUEUE_STOP_REASON_SUSPEND);
> 
>  	/* flush out all packets and station cleanup call_rcu()s */
> +	synchronize_net();
>  	rcu_barrier();
> 
>  	ieee80211_flush_queues(local, NULL);
> -- 
> 1.7.10.4
> 
> -- 
> Bob Copeland %% www.bobcopeland.com
> 

[PATCH v2] mac80211: use synchronize_rcu() with rcu_barrier()

[Bob Copeland]

The RCU docs used to state that rcu_barrier() included a wait
for an RCU grace period; however the comments for rcu_barrier()
as of commit f0a0e6f... "rcu: Clarify memory-ordering properties
of grace-period primitives" contradict this.

So add back synchronize_{rcu,net}() to where they once were,
but keep the rcu_barrier()s for the call_rcu() callbacks.

Cc: stable <stable@vger.kernel.org>
Signed-off-by: Bob Copeland <bob@cozybit.com>

---
v2: also in cfg.c

 net/mac80211/cfg.c   |    1 +
 net/mac80211/iface.c |    5 +++--
 net/mac80211/pm.c    |    1 +
 3 files changed, 5 insertions(+), 2 deletions(-)

diff --git a/net/mac80211/cfg.c b/net/mac80211/cfg.c
index 72ab1c0..dcffca4 100644
--- a/net/mac80211/cfg.c
+++ b/net/mac80211/cfg.c
@@ -1043,6 +1043,7 @@ static int ieee80211_stop_ap(struct wiphy *wiphy, struct net_device *dev)
 	list_for_each_entry(vlan, &sdata->u.ap.vlans, u.vlan.list)
 		sta_info_flush_defer(vlan);
 	sta_info_flush_defer(sdata);
+	synchronize_net();
 	rcu_barrier();
 	list_for_each_entry(vlan, &sdata->u.ap.vlans, u.vlan.list) {
 		sta_info_flush_cleanup(vlan);
diff --git a/net/mac80211/iface.c b/net/mac80211/iface.c
index b6abaaa..c634aff 100644
--- a/net/mac80211/iface.c
+++ b/net/mac80211/iface.c
@@ -836,11 +836,12 @@ static void ieee80211_do_stop(struct ieee80211_sub_if_data *sdata,
 		 *
 		 * sta_info_flush_cleanup() requires rcu_barrier()
 		 * first to wait for the station call_rcu() calls
-		 * to complete, here we need at least sychronize_rcu()
-		 * it to wait for the RX path in case it is using the
+		 * to complete, and we also need synchronize_rcu()
+		 * to wait for the RX path in case it is using the
 		 * interface and enqueuing frames at this very time on
 		 * another CPU.
 		 */
+		synchronize_rcu();
 		rcu_barrier();
 		sta_info_flush_cleanup(sdata);
 
diff --git a/net/mac80211/pm.c b/net/mac80211/pm.c
index 4431f0f..7fc5d0d 100644
--- a/net/mac80211/pm.c
+++ b/net/mac80211/pm.c
@@ -38,6 +38,7 @@ int __ieee80211_suspend(struct ieee80211_hw *hw, struct cfg80211_wowlan *wowlan)
 					IEEE80211_QUEUE_STOP_REASON_SUSPEND);
 
 	/* flush out all packets and station cleanup call_rcu()s */
+	synchronize_net();
 	rcu_barrier();
 
 	ieee80211_flush_queues(local, NULL);
-- 
1.7.10.4

Re: [PATCH v2] mac80211: use synchronize_rcu() with rcu_barrier()

[Johannes Berg]

On Thu, 2013-04-18 at 18:26 -0400, Bob Copeland wrote:
> The RCU docs used to state that rcu_barrier() included a wait
> for an RCU grace period; however the comments for rcu_barrier()
> as of commit f0a0e6f... "rcu: Clarify memory-ordering properties
> of grace-period primitives" contradict this.
> 
> So add back synchronize_{rcu,net}() to where they once were,
> but keep the rcu_barrier()s for the call_rcu() callbacks.

Applied, thanks for spotting this!

johannes