Skip to contentSkip to conversation

mac80211: beacon filtering

RFC PATCH In review Kalle Valo 5 participants 28 messages Mar 14, 2009 – Mar 20, 2009 mac80211

lore
Share & tools
lore

Kalle Valo

author
#0 lore raw

v2 cover mac80211: beacon filtering

Here's my proposal how to implement beacon filtering in mac80211. The basic idea is simple, the driver enables filtering whenever mac80211 enables power save and the driver calls ieee80211_beacon_loss() whenever hardware informs about beacon loss.

I have tested this with stlc45xx and based on simple tests this seems to work. For reference I earlier sent stlc45xx patch to this list.

v2:
o rebase to 2.6.29-rc7-wl
o rename ieee80211_rx_trigger() to ieee80211_sta_rx_notify()
o move check for sta mode in ieee80211_sta_rx_notify() to rx.c
o move unicast check from rx.c to ieee80211_sta_rx_notify()
o add comment for setting last_beacon in ieee80211_rx_mgmt_assoc_resp()
o rename IEEE80211_HW_BEACON_FILTERING to IEEE80211_HW_BEACON_FILTER
o send probe request from ieee80211_beacon_loss_work()
o API documentation
o enable beacon filtering only when power save is enabled
o disable power save (and hence beacon filtering) while scanning to make it possible to have reliable scan results even when beacon filtering is enabled

TODO:

o test with different hardware (without beacon filtering) to avoid regressions

---

Kalle Valo (4):
mac80211: add beacon filtering support
mac80211: disable power save when scanning
mac80211: track beacons separately from the rx path activity
mac80211: decrease execution of the associated timer

 include/net/mac80211.h     |   30 +++++++++--
 net/mac80211/ieee80211_i.h |    5 ++
 net/mac80211/iface.c       |    3 +
 net/mac80211/mlme.c        |  124 +++++++++++++++++++++++++++++++++-----------
 net/mac80211/rx.c          |    9 +++
 net/mac80211/scan.c        |   54 ++++++++++++++++++-
 6 files changed, 185 insertions(+), 40 deletions(-)
author
#1 lore raw

RFC PATCH v2 1/4 mac80211: decrease execution of the associated timer

Currently the timer is triggering every two seconds (IEEE80211_MONITORING_INTERVAL). Decrease the timer to only trigger when nothing is received to avoid waking up CPU.

Now there's a functional change that probe requests are sent only when the data path is idle, earlier they were sent also while there was activity on the data path.

This is also preparation for beacon filtering support. Thanks to Johannes Berg for the idea.

---

 net/mac80211/ieee80211_i.h |    2 ++
 net/mac80211/mlme.c        |    8 ++++++++
 net/mac80211/rx.c          |    3 +++
 3 files changed, 13 insertions(+), 0 deletions(-)
net/mac80211/ieee80211_i.h +2 -0
index ecbc8e0..74c4e13 100644
@@ -1080,6 +1080,8 @@ void ieee80211_dynamic_ps_timer(unsigned long data);
 void ieee80211_send_nullfunc(struct ieee80211_local *local,
 			     struct ieee80211_sub_if_data *sdata,
 			     int powersave);
+void ieee80211_sta_rx_notify(struct ieee80211_sub_if_data *sdata,
+			     struct ieee80211_hdr *hdr);
 
 void ieee80211_wake_queues_by_reason(struct ieee80211_hw *hw,
 				     enum queue_stop_reason reason);
net/mac80211/mlme.c +8 -0
index 391445c..397a2e7 100644
@@ -911,6 +911,14 @@ static void ieee80211_associate(struct ieee80211_sub_if_data *sdata)
 	mod_timer(&ifmgd->timer, jiffies + IEEE80211_ASSOC_TIMEOUT);
 }
 
+void ieee80211_sta_rx_notify(struct ieee80211_sub_if_data *sdata,
+			     struct ieee80211_hdr *hdr)
+{
+	/* timer triggers only when there is no unicast traffic */
+	if (!is_multicast_ether_addr(hdr->addr1))
+		mod_timer(&sdata->u.mgd.timer,
+			  jiffies + IEEE80211_MONITORING_INTERVAL);
+}
 
 static void ieee80211_associated(struct ieee80211_sub_if_data *sdata)
 {
net/mac80211/rx.c +3 -0
index 66f7ecf..25982c0 100644
@@ -855,6 +855,9 @@ ieee80211_rx_h_sta_process(struct ieee80211_rx_data *rx)
 	if (!(rx->flags & IEEE80211_RX_RA_MATCH))
 		return RX_CONTINUE;
 
+	if (rx->sdata->vif.type == NL80211_IFTYPE_STATION)
+		ieee80211_sta_rx_notify(rx->sdata, hdr);
+
 	sta->rx_fragments++;
 	sta->rx_bytes += rx->skb->len;
 	sta->last_signal = rx->status->signal;
maintainer
#2 lore raw

Re: [RFC PATCH v2 1/4] mac80211: decrease execution of the associated timer

net/mac80211/mlme.c static void ieee80211_associate(struct ieee80211_sub_if_data *sdata) · line 921

Quoted 8 lines · Kalle Valo
net/mac80211/mlme.c +8 -0
+void ieee80211_sta_rx_notify(struct ieee80211_sub_if_data *sdata,
+			     struct ieee80211_hdr *hdr)
+{
+	/* timer triggers only when there is no unicast traffic */
+	if (!is_multicast_ether_addr(hdr->addr1))
+		mod_timer(&sdata->u.mgd.timer,
+			  jiffies + IEEE80211_MONITORING_INTERVAL);
+}

Do we really need the multicast check? The frame will be coming from the AP in managed mode in both cases, so why regard the data path as idle when we're receiving multicast traffic?

johannes

author
#3 lore raw

Re: [RFC PATCH v2 1/4] mac80211: decrease execution of the associated timer

Quoted 14 lines · Johannes Berg

net/mac80211/mlme.c static void ieee80211_associate(struct ieee80211_sub_if_data *sdata) · line 921

Quoted 8 lines · Kalle Valo
net/mac80211/mlme.c +8 -0
+void ieee80211_sta_rx_notify(struct ieee80211_sub_if_data *sdata,
+			     struct ieee80211_hdr *hdr)
+{
+	/* timer triggers only when there is no unicast traffic */
+	if (!is_multicast_ether_addr(hdr->addr1))
+		mod_timer(&sdata->u.mgd.timer,
+			  jiffies + IEEE80211_MONITORING_INTERVAL);
+}

Do we really need the multicast check? The frame will be coming from the AP in managed mode in both cases, so why regard the data path as idle when we're receiving multicast traffic?

It seems that I had a good vacation because I can't remember anymore why I added the check :D Too bad that even the comment I wrote was next to useless.

I'll remove the multicast check because I don't see the point for it. And if in case I finally recall the reason I'll definitely add a better comment.

Thanks for all the review!

Kalle

#4 lore raw

Re: [RFC PATCH v2 1/4] mac80211: decrease execution of the associated timer

On Sat, Mar 14, 2009 at 11:46 AM, Kalle Valo <[email protected]> wro= te:

Quoted 6 lines
Quoted 5 lines · Johannes Berg
Quoted 2 lines · Kalle Valo

+void ieee80211_sta_rx_notify(struct ieee80211_sub_if_data *sdata,
+ =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=

=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0struct ieee80211_hdr *hdr)

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

+{
+ =C2=A0 =C2=A0 =C2=A0 /* timer triggers only when there is no unic=

ast traffic */

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

+ =C2=A0 =C2=A0 =C2=A0 if (!is_multicast_ether_addr(hdr->addr1))
+ =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 mod_timer(&sdata=

->u.mgd.timer,

Quoted 1 line
Quoted 1 line
Quoted 1 line

+ =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=

=A0 =C2=A0 =C2=A0 jiffies + IEEE80211_MONITORING_INTERVAL);

Quoted 3 lines
Quoted 3 lines
Quoted 1 line

+}

Do we really need the multicast check? The frame will be coming from=

the

Quoted 1 line
Quoted 1 line

AP in managed mode in both cases, so why regard the data path as idl=

e

Quoted 3 lines
Quoted 1 line

when we're receiving multicast traffic?

It seems that I had a good vacation because I can't remember anymore =

why I

Quoted 4 lines

added the check :D Too bad that even the comment I wrote was next to useless.

I'll remove the multicast check because I don't see the point for it.=

And if

Quoted 1 line

in case I finally recall the reason I'll definitely add a better comm=

ent.

Quoted 2 lines

Thanks for all the review!

If that's the case better just remove a whole routine just for that.

Luis
--
To unsubscribe from this list: send the line "unsubscribe linux-wireles= s" in
the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html

#5 lore raw

Re: [RFC PATCH v2 1/4] mac80211: decrease execution of the associated timer

In reply to Kalle Valo
Quoted 16 lines · Kalle Valo
Quoted 14 lines · Johannes Berg

net/mac80211/mlme.c static void ieee80211_associate(struct ieee80211_sub_if_data *sdata) · line 921

Quoted 8 lines · Kalle Valo
net/mac80211/mlme.c +8 -0
+void ieee80211_sta_rx_notify(struct ieee80211_sub_if_data *sdata,
+			     struct ieee80211_hdr *hdr)
+{
+	/* timer triggers only when there is no unicast traffic */
+	if (!is_multicast_ether_addr(hdr->addr1))
+		mod_timer(&sdata->u.mgd.timer,
+			  jiffies + IEEE80211_MONITORING_INTERVAL);
+}

Do we really need the multicast check? The frame will be coming from the AP in managed mode in both cases, so why regard the data path as idle when we're receiving multicast traffic?

Quoted 7 lines

It seems that I had a good vacation because I can't remember anymore why I added the check :D Too bad that even the comment I wrote was next to useless.

I'll remove the multicast check because I don't see the point for it. And if in case I finally recall the reason I'll definitely add a better comment.

Now I remember what this test was about. In ieee80211_associated() there are two tests: probe request check after an data idle period (currently 60 secs) and beacon loss check for drivers not supporting filtering (2 secs).

The former test has been originally (even before my beacon filter patches) implemented so that the probe request is sent even though we are receiving beacons. mod_timer() is called only for unicast traffic so that the probe request test will trigger when we are still receiving beacons.

I'll add a proper comment explaining all this.

Also currently ieee80211_associated() is called every two seconds, we need to fix this in future to avoid waking up cpu needlessly. I don't want to make too intrusive changes right now.

Signature · 1 line
Kalle Valo
maintainer
#6 lore raw

Re: [RFC PATCH v2 1/4] mac80211: decrease execution of the associated timer

Quoted 26 lines · Kalle Valo
Quoted 13 lines
Quoted 4 lines
Quoted 4 lines
net/mac80211/mlme.c +4 -0
+void ieee80211_sta_rx_notify(struct ieee80211_sub_if_data *sdata,
+			     struct ieee80211_hdr *hdr)
+{
+	/* timer triggers only when there is no unicast traffic */
Quoted 8 lines

net/mac80211/mlme.c static void ieee80211_associate(struct ieee80211_sub_if_data *sdata) · line 921

Quoted 4 lines
net/mac80211/mlme.c +4 -0
+	if (!is_multicast_ether_addr(hdr->addr1))
+		mod_timer(&sdata->u.mgd.timer,
+			  jiffies + IEEE80211_MONITORING_INTERVAL);
+}

Do we really need the multicast check? The frame will be coming from the AP in managed mode in both cases, so why regard the data path as idle when we're receiving multicast traffic?

Quoted 7 lines

It seems that I had a good vacation because I can't remember anymore why I added the check :D Too bad that even the comment I wrote was next to useless.

I'll remove the multicast check because I don't see the point for it. And if in case I finally recall the reason I'll definitely add a better comment.

Now I remember what this test was about. In ieee80211_associated() there are two tests: probe request check after an data idle period (currently 60 secs) and beacon loss check for drivers not supporting filtering (2 secs).

Quoted 5 lines

The former test has been originally (even before my beacon filter patches) implemented so that the probe request is sent even though we are receiving beacons. mod_timer() is called only for unicast traffic so that the probe request test will trigger when we are still receiving beacons.

Why would we not want to change that? I don't see a reason to send a probe request when we're still receiving beacons. It was meant to be used when beacons got lost _during_ a data idle period to verify that the AP is still there, afaict.

Quoted 5 lines

I'll add a proper comment explaining all this.

Also currently ieee80211_associated() is called every two seconds, we need to fix this in future to avoid waking up cpu needlessly. I don't want to make too intrusive changes right now.

I thought you already did that too :)

johannes

#7 lore raw

Re: [RFC PATCH v2 1/4] mac80211: decrease execution of the associated timer

Quoted 13 lines · Johannes Berg
Quoted 10 lines · Kalle Valo

Now I remember what this test was about. In ieee80211_associated() there are two tests: probe request check after an data idle period (currently 60 secs) and beacon loss check for drivers not supporting filtering (2 secs).

The former test has been originally (even before my beacon filter patches) implemented so that the probe request is sent even though we are receiving beacons. mod_timer() is called only for unicast traffic so that the probe request test will trigger when we are still receiving beacons.

Quoted 4 lines

Why would we not want to change that? I don't see a reason to send a probe request when we're still receiving beacons. It was meant to be used when beacons got lost _during_ a data idle period to verify that the AP is still there, afaict.

The original (or current) implementation sends probe requests even when beacons are received or there is unicast traffic going on. My patches change this so that the probe request is only sent when there is data idle period (ie. no unicast traffic).

Like we discussed earlier, I think the most optimal solution would be to send nullfunc frames during data idle periods just to check that the connection is working bothways.

What should AP do if it receives a nullfunc frame from a station which is not associated? Send a deauth frame back to the station? It would be nice to check the association state also during the data idle period. Sometimes I have noticed that AP might deauthenticate the station without station noticing it. It would be just perfect if nullfunc frame would do the assocition also during data idle periods.

Quoted 7 lines
Quoted 5 lines

I'll add a proper comment explaining all this.

Also currently ieee80211_associated() is called every two seconds, we need to fix this in future to avoid waking up cpu needlessly. I don't want to make too intrusive changes right now.

I thought you already did that too :)

Sorry, I did that only for the period when there is data transmission. To get rid of unncessary wakeups altogether I need to do something more advanced, like using two timers or something. I haven't thought that yet.

I'll mention this in the commit logs.

Signature · 1 line
Kalle Valo
maintainer
#8 lore raw

Re: [RFC PATCH v2 1/4] mac80211: decrease execution of the associated timer

Quoted 4 lines · Kalle Valo

The original (or current) implementation sends probe requests even when beacons are received or there is unicast traffic going on. My patches change this so that the probe request is only sent when there is data idle period (ie. no unicast traffic).

Ok.

Quoted 10 lines

Like we discussed earlier, I think the most optimal solution would be to send nullfunc frames during data idle periods just to check that the connection is working bothways.

What should AP do if it receives a nullfunc frame from a station which is not associated? Send a deauth frame back to the station? It would be nice to check the association state also during the data idle period. Sometimes I have noticed that AP might deauthenticate the station without station noticing it. It would be just perfect if nullfunc frame would do the assocition also during data idle periods.

Yes, I think nullfunc, being data, counts as class 3, and as such the AP will deauth the station.

Quoted 4 lines

Sorry, I did that only for the period when there is data transmission. To get rid of unncessary wakeups altogether I need to do something more advanced, like using two timers or something. I haven't thought that yet.

Ok, guess I got lost in all the timers.

johannes

author
#9 lore raw

RFC PATCH v2 2/4 mac80211: track beacons separately from the rx path activity

In reply to Kalle Valo

Separate beacon and rx path tracking in preparation for the beacon filtering support. At the same time change ieee80211_associated() to look a bit simpler.

Probe requests are now sent only after IEEE80211_PROBE_IDLE_TIME, which is now set to 60 seconds.

---

 net/mac80211/ieee80211_i.h |    1 +
 net/mac80211/mlme.c        |   79 ++++++++++++++++++++++++++------------------
 net/mac80211/rx.c          |    6 +++
 3 files changed, 53 insertions(+), 33 deletions(-)
net/mac80211/ieee80211_i.h +1 -0
index 74c4e13..5d3d0bf 100644
@@ -305,6 +305,7 @@ struct ieee80211_if_managed {
 	unsigned long request;
 
 	unsigned long last_probe;
+	unsigned long last_beacon;
 
 	unsigned int flags;
 
net/mac80211/mlme.c +47 -32

130 lines not on this page · raw message

net/mac80211/rx.c +5 -1
index 25982c0..4990e76 100644
@@ -849,7 +849,11 @@ ieee80211_rx_h_sta_process(struct ieee80211_rx_data *rx)
 		 * Mesh beacons will update last_rx when if they are found to
 		 * match the current local configuration when processed.
 		 */
-		sta->last_rx = jiffies;
+		if (rx->sdata->vif.type == NL80211_IFTYPE_STATION &&
+		    ieee80211_is_beacon(hdr->frame_control)) {
+			rx->sdata->u.mgd.last_beacon = jiffies;
+		} else
+			sta->last_rx = jiffies;
 	}
 
 	if (!(rx->flags & IEEE80211_RX_RA_MATCH))
maintainer
#10 lore raw

Re: [RFC PATCH v2 2/4] mac80211: track beacons separately from the rx path activity

Quoted 5 lines · Kalle Valo

Separate beacon and rx path tracking in preparation for the beacon filtering support. At the same time change ieee80211_associated() to look a bit simpler.

Probe requests are now sent only after IEEE80211_PROBE_IDLE_TIME, which is now set to 60 seconds.

Looks good.

net/mac80211/mlme.c static void ieee80211_associated(struct ieee80211_sub_if_data *sdata) · line 983

Quoted 4 lines
net/mac80211/mlme.c +1 -1
 	else
 		mod_timer(&ifmgd->timer, jiffies +
-				      IEEE80211_MONITORING_INTERVAL);
+			  IEEE80211_MONITORING_INTERVAL);

That seems to be indented a little oddly though :)

johannes

#11 lore raw

Re: [RFC PATCH v2 2/4] mac80211: track beacons separately from the rx path activity

In reply to Kalle Valo
Quoted 7 lines · Kalle Valo

Separate beacon and rx path tracking in preparation for the beacon filtering support. At the same time change ieee80211_associated() to look a bit simpler.

Probe requests are now sent only after IEEE80211_PROBE_IDLE_TIME, which is now set to 60 seconds.

Can this be split into 3 separate patches, to make the one with the actual functional changes easier to review. One is the making of ieee80211_associated() look easier, the other the rename, and lastly the actual functional changes. When reviewing the commit log it would then be easier to see what was done.

Luis

maintainer
#12 lore raw

Re: [RFC PATCH v2 2/4] mac80211: track beacons separately from the rx path activity

Quoted 7 lines · Luis R. Rodriguez
Quoted 6 lines · Kalle Valo

Separate beacon and rx path tracking in preparation for the beacon filtering support. At the same time change ieee80211_associated() to look a bit simpler.

Probe requests are now sent only after IEEE80211_PROBE_IDLE_TIME, which is now set to 60 seconds.

Quoted 7 lines
Quoted 1 line

Can this be split into 3 separate patches, to make the one with the actual functional changes easier to review. One is the making of ieee80211_associated() look easier, the other the rename, and lastly the actual functional changes. When reviewing the commit log it would then be easier to see what was done.

Disagree, the rename is part of the functional change, it's not just a plain rename.

johannes

#13 lore raw

Re: [RFC PATCH v2 2/4] mac80211: track beacons separately from the rx path activity

Quoted 10 lines · Johannes Berg
Quoted 9 lines · Luis R. Rodriguez
Quoted 7 lines · Kalle Valo

Separate beacon and rx path tracking in preparation for the beacon filtering support. At the same time change ieee80211_associated() to look a bit simpler.

Probe requests are now sent only after IEEE80211_PROBE_IDLE_TIME, which is now set to 60 seconds.

Quoted 8 lines
Quoted 5 lines

Can this be split into 3 separate patches, to make the one with the actual functional changes easier to review. One is the making of ieee80211_associated() look easier, the other the rename, and lastly the actual functional changes. When reviewing the commit log it would then be easier to see what was done.

Disagree, the rename is part of the functional change, it's not just a plain rename.

Alright, then 2.

Luis

author
#14 lore raw

Re: [RFC PATCH v2 2/4] mac80211: track beacons separately from the rx path activity

In reply to Luis R. Rodriguez
Quoted 7 lines · Luis R. Rodriguez
Quoted 6 lines · Kalle Valo

Separate beacon and rx path tracking in preparation for the beacon filtering support. At the same time change ieee80211_associated() to look a bit simpler.

Probe requests are now sent only after IEEE80211_PROBE_IDLE_TIME, which is now set to 60 seconds.

Quoted 7 lines
Quoted 1 line

Can this be split into 3 separate patches, to make the one with the actual functional changes easier to review. One is the making of ieee80211_associated() look easier, the other the rename, and lastly the actual functional changes. When reviewing the commit log it would then be easier to see what was done.

Is it really worth the trouble? The patch is relatively small so spotting the changes shouldn't be that difficult.

Signature · 1 line
Kalle Valo
author
#15 lore raw

RFC PATCH v2 3/4 mac80211: disable power save when scanning

In reply to Kalle Valo

When software scanning we need to disable power save so that all possible probe responses and beacons are received. For hardware scanning assume that hardware will take care of that and document that assumption.

---

 include/net/mac80211.h |   12 ++++++-----
 net/mac80211/scan.c    |   54 ++++++++++++++++++++++++++++++++++++++++++++++--
 2 files changed, 59 insertions(+), 7 deletions(-)
include/net/mac80211.h +7 -5
index 12a52ef..9ef131d 100644
@@ -1318,11 +1318,13 @@ enum ieee80211_ampdu_mlme_action {
  *
  * @hw_scan: Ask the hardware to service the scan request, no need to start
  *	the scan state machine in stack. The scan must honour the channel
- *	configuration done by the regulatory agent in the wiphy's registered
- *	bands. When the scan finishes, ieee80211_scan_completed() must be
- *	called; note that it also must be called when the scan cannot finish
- *	because the hardware is turned off! Anything else is a bug!
- *	Returns a negative error code which will be seen in userspace.
+ *	configuration done by the regulatory agent in the wiphy's
+ *	registered bands. The hardware (or the driver) needs to make sure
+ *	that power save is disabled. When the scan finishes,
+ *	ieee80211_scan_completed() must be called; note that it also must
+ *	be called when the scan cannot finish because the hardware is
+ *	turned off! Anything else is a bug! Returns a negative error code
+ *	which will be seen in userspace.
  *
  * @sw_scan_start: Notifier function that is called just before a software scan
  *	is started. Can be NULL, if the driver doesn't need this notification.
net/mac80211/scan.c +52 -2

72 lines not on this page · raw message

maintainer
#16 lore raw

Re: [RFC PATCH v2 3/4] mac80211: disable power save when scanning

net/mac80211/scan.c ieee80211_scan_rx(struct ieee80211_sub_if_data *sdata, struct sk_buff *skb, · line 232

Quoted 8 lines · Kalle Valo
net/mac80211/scan.c +8 -0
+	if (!ps || (local->hw.flags & IEEE80211_HW_PS_NULLFUNC_STACK))
+		/*
+		 * with IEEE80211_HW_PS_NULLFUNC_STACK and power save
+		 * enabled the firmware sent a null frame when power save
+		 * was disabled, so we need to send a new null frame
+		 */
+		ieee80211_send_nullfunc(local, sdata, 1);
+}

Now I'm confused a little. I thought with PS_NULLFUNC_STACK the firmware _didn't_ send such frames.

johannes

author
#17 lore raw

Re: [RFC PATCH v2 3/4] mac80211: disable power save when scanning

Quoted 13 lines · ext Johannes Berg

net/mac80211/scan.c ieee80211_scan_rx(struct ieee80211_sub_if_data *sdata, struct sk_buff *skb, · line 232

Quoted 8 lines · Kalle Valo
net/mac80211/scan.c +8 -0
+	if (!ps || (local->hw.flags & IEEE80211_HW_PS_NULLFUNC_STACK))
+		/*
+		 * with IEEE80211_HW_PS_NULLFUNC_STACK and power save
+		 * enabled the firmware sent a null frame when power save
+		 * was disabled, so we need to send a new null frame
+		 */
+		ieee80211_send_nullfunc(local, sdata, 1);
+}

Now I'm confused a little. I thought with PS_NULLFUNC_STACK the firmware _didn't_ send such frames.

That's a bug, good catch! It should be:

!(local->hw.flags & IEEE80211_HW_PS_NULLFUNC_STACK)

I'll fix that in v3.

Kalle

author
#18 lore raw

RFC PATCH v2 4/4 mac80211: add beacon filtering support

In reply to Kalle Valo

Add IEEE80211_HW_BEACON_FILTERING flag so that driver inform that it supports beacon filtering. Drivers need to call the new function ieee80211_beacon_loss() to notify about beacon loss.

---

 include/net/mac80211.h     |   18 ++++++++++++++++++
 net/mac80211/ieee80211_i.h |    2 ++
 net/mac80211/iface.c       |    3 +++
 net/mac80211/mlme.c        |   39 ++++++++++++++++++++++++++++++++++++++-
 4 files changed, 61 insertions(+), 1 deletions(-)
include/net/mac80211.h +18 -0
index 9ef131d..1262b83 100644
@@ -888,6 +888,13 @@ enum ieee80211_tkip_key_type {
  *
  * @IEEE80211_HW_MFP_CAPABLE:
  *	Hardware supports management frame protection (MFP, IEEE 802.11w).
+ *
+ * @IEEE80211_HW_BEACON_FILTER: Hardware can drop similar beacon frames to
+ *	avoid waking up cpu. Enabling this flag disables the beacon check
+ *	in mac80211 when in power save mode and the hardware must have an
+ *	event to notify that beacons are lost. Use ieee80211_beacon_loss()
+ *	to notify the event to the stack. The filtering must be enabled
+ *	only when IEEE80211_CONF_PS is set.
  */
 enum ieee80211_hw_flags {
 	IEEE80211_HW_RX_INCLUDES_FCS			= 1<<1,
@@ -903,6 +910,7 @@ enum ieee80211_hw_flags {
 	IEEE80211_HW_PS_NULLFUNC_STACK			= 1<<11,
 	IEEE80211_HW_SUPPORTS_DYNAMIC_PS		= 1<<12,
 	IEEE80211_HW_MFP_CAPABLE			= 1<<13,
+	IEEE80211_HW_BEACON_FILTER			= 1<<14,
 };
 
 /**
@@ -1981,6 +1989,16 @@ void ieee80211_stop_tx_ba_cb_irqsafe(struct ieee80211_hw *hw, const u8 *ra,
 struct ieee80211_sta *ieee80211_find_sta(struct ieee80211_hw *hw,
 					 const u8 *addr);
 
+/**
+ * ieee80211_beacon_loss - inform that hardware does not receive beacons
+ *
+ * @hw: pointer as obtained from ieee80211_alloc_hw()
+ *
+ * When beacon filtering is enabled with IEEE80211_HW_BEACON_FILTERING and
+ * IEEE80211_CONF_PS is set, the driver needs to inform whenever the
+ * hardware is not receiving beacons with this function.
+ */
+void ieee80211_beacon_loss(struct ieee80211_hw *hw);
 
 /* Rate control API */
 
net/mac80211/ieee80211_i.h +2 -0
index 5d3d0bf..488c112 100644
@@ -272,6 +272,7 @@ struct ieee80211_if_managed {
 	struct timer_list chswitch_timer;
 	struct work_struct work;
 	struct work_struct chswitch_work;
+	struct work_struct beacon_loss_work;
 
 	u8 bssid[ETH_ALEN], prev_bssid[ETH_ALEN];
 
@@ -1083,6 +1084,7 @@ void ieee80211_send_nullfunc(struct ieee80211_local *local,
 			     int powersave);
 void ieee80211_sta_rx_notify(struct ieee80211_sub_if_data *sdata,
 			     struct ieee80211_hdr *hdr);
+void ieee80211_beacon_loss_work(struct work_struct *work);
 
 void ieee80211_wake_queues_by_reason(struct ieee80211_hw *hw,
 				     enum queue_stop_reason reason);
net/mac80211/iface.c +3 -0
index 2acc416..e0fa20f 100644
@@ -470,6 +470,9 @@ static int ieee80211_stop(struct net_device *dev)
 		 */
 		cancel_work_sync(&sdata->u.mgd.work);
 		cancel_work_sync(&sdata->u.mgd.chswitch_work);
+
+		cancel_work_sync(&sdata->u.mgd.beacon_loss_work);
+
 		/*
 		 * When we get here, the interface is marked down.
 		 * Call synchronize_rcu() to wait for the RX path
net/mac80211/mlme.c +38 -1
index c235d39..053d75b 100644
@@ -920,6 +920,41 @@ void ieee80211_sta_rx_notify(struct ieee80211_sub_if_data *sdata,
 			  jiffies + IEEE80211_MONITORING_INTERVAL);
 }
 
+void ieee80211_beacon_loss_work(struct work_struct *work)
+{
+	struct ieee80211_sub_if_data *sdata =
+		container_of(work, struct ieee80211_sub_if_data,
+			     u.mgd.beacon_loss_work);
+	struct ieee80211_if_managed *ifmgd = &sdata->u.mgd;
+
+	printk(KERN_DEBUG "%s: driver reports beacon loss from AP %pM "
+	       "- sending probe request\n", sdata->dev->name,
+	       sdata->u.mgd.bssid);
+
+	ifmgd->flags |= IEEE80211_STA_PROBEREQ_POLL;
+	ieee80211_send_probe_req(sdata, ifmgd->bssid, ifmgd->ssid,
+				 ifmgd->ssid_len, NULL, 0);
+
+	mod_timer(&ifmgd->timer, jiffies + IEEE80211_MONITORING_INTERVAL);
+}
+
+void ieee80211_beacon_loss(struct ieee80211_hw *hw)
+{
+	struct ieee80211_local *local = hw_to_local(hw);
+	struct ieee80211_sub_if_data *sdata;
+
+	rcu_read_lock();
+	list_for_each_entry(sdata, &local->interfaces, list) {
+		if (sdata->vif.type != NL80211_IFTYPE_STATION)
+			continue;
+
+		queue_work(local->hw.workqueue,
+			   &sdata->u.mgd.beacon_loss_work);
+	}
+	rcu_read_unlock();
+}
+EXPORT_SYMBOL(ieee80211_beacon_loss);
+
 static void ieee80211_associated(struct ieee80211_sub_if_data *sdata)
 {
 	struct ieee80211_if_managed *ifmgd = &sdata->u.mgd;
@@ -954,7 +989,8 @@ static void ieee80211_associated(struct ieee80211_sub_if_data *sdata)
 		goto unlock;
 	}
 
-	if (time_after(jiffies,
+	if (!(local->hw.flags & IEEE80211_HW_BEACON_FILTER) &&
+	    time_after(jiffies,
 		       ifmgd->last_beacon + IEEE80211_MONITORING_INTERVAL)) {
 		printk(KERN_DEBUG "%s: beacon loss from AP %pM "
 		       "- sending probe request\n",
@@ -1839,6 +1875,7 @@ void ieee80211_sta_setup_sdata(struct ieee80211_sub_if_data *sdata)
 	ifmgd = &sdata->u.mgd;
 	INIT_WORK(&ifmgd->work, ieee80211_sta_work);
 	INIT_WORK(&ifmgd->chswitch_work, ieee80211_chswitch_work);
+	INIT_WORK(&ifmgd->beacon_loss_work, ieee80211_beacon_loss_work);
 	setup_timer(&ifmgd->timer, ieee80211_sta_timer,
 		    (unsigned long) sdata);
 	setup_timer(&ifmgd->chswitch_timer, ieee80211_chswitch_timer,
maintainer
#19 lore raw

Re: [RFC PATCH v2 4/4] mac80211: add beacon filtering support

Quoted 18 lines · Kalle Valo
(quoted diff) +16 -0
+/**
+ * ieee80211_beacon_loss - inform that hardware does not receive beacons
+ *
+ * @hw: pointer as obtained from ieee80211_alloc_hw()
+ *
+ * When beacon filtering is enabled with IEEE80211_HW_BEACON_FILTERING and
+ * IEEE80211_CONF_PS is set, the driver needs to inform whenever the
+ * hardware is not receiving beacons with this function.
+ */
+void ieee80211_beacon_loss(struct ieee80211_hw *hw);
 
 
+void ieee80211_beacon_loss(struct ieee80211_hw *hw)
+{
+	struct ieee80211_local *local = hw_to_local(hw);
+	struct ieee80211_sub_if_data *sdata;
+
+	rcu_read_lock();

Review

Quoted 10 lines
(quoted diff) +10 -0
+	list_for_each_entry(sdata, &local->interfaces, list) {
+		if (sdata->vif.type != NL80211_IFTYPE_STATION)
+			continue;
+
+		queue_work(local->hw.workqueue,
+			   &sdata->u.mgd.beacon_loss_work);
+	}
+	rcu_read_unlock();
+}
+EXPORT_SYMBOL(ieee80211_beacon_loss);

Shouldn't the driver just pass the relevant vif pointer?

johannes

author
#20 lore raw

Re: [RFC PATCH v2 4/4] mac80211: add beacon filtering support

Quoted 12 lines · Johannes Berg
Quoted 10 lines · Kalle Valo
net/mac80211/mlme.c +10 -0
+void ieee80211_beacon_loss(struct ieee80211_hw *hw)
+{
+	struct ieee80211_local *local = hw_to_local(hw);
+	struct ieee80211_sub_if_data *sdata;
+
+	rcu_read_lock();
+	list_for_each_entry(sdata, &local->interfaces, list) {
+		if (sdata->vif.type != NL80211_IFTYPE_STATION)
+			continue;
+
Quoted 8 lines

net/mac80211/mlme.c void ieee80211_sta_rx_notify(struct ieee80211_sub_if_data *sdata, · line 956

Quoted 6 lines
net/mac80211/mlme.c +6 -0
+		queue_work(local->hw.workqueue,
+			   &sdata->u.mgd.beacon_loss_work);
+	}
+	rcu_read_unlock();
+}
+EXPORT_SYMBOL(ieee80211_beacon_loss);

Shouldn't the driver just pass the relevant vif pointer?

Nice, that way the implementation would be a lot simpler and consistent with the rest of the driver interface. Yes, I'll change it to use vif pointer.

Kalle

#21 lore raw

Re: [RFC PATCH v2 4/4] mac80211: add beacon filtering support

In reply to Kalle Valo

On Sat, Mar 14, 2009 at 10:14 AM, Kalle Valo <[email protected]> wro= te:

Quoted 1 line

Add IEEE80211_HW_BEACON_FILTERING flag so that driver inform that it =

supports

Quoted 7 lines

beacon filtering. Drivers need to call the new function ieee80211_beacon_loss() to notify about beacon loss.

---

=C2=A0include/net/mac80211.h =C2=A0 =C2=A0 | =C2=A0 18 ++++++++++++++=

++++

Quoted 3 lines

=C2=A0net/mac80211/ieee80211_i.h | =C2=A0 =C2=A02 ++ =C2=A0net/mac80211/iface.c =C2=A0 =C2=A0 =C2=A0 | =C2=A0 =C2=A03 +++ =C2=A0net/mac80211/mlme.c =C2=A0 =C2=A0 =C2=A0 =C2=A0| =C2=A0 39 ++++=

++++++++++++++++++++++++++++++++++-

Quoted 10 lines

=C2=A04 files changed, 61 insertions(+), 1 deletions(-)

include/net/mac80211.h +0 -0
index 9ef131d..1262b83 100644
@@ -888,6 +888,13 @@ enum ieee80211_tkip_key_type {

=C2=A0*
=C2=A0* @IEEE80211_HW_MFP_CAPABLE:
=C2=A0* =C2=A0 =C2=A0 Hardware supports management frame protection (=

MFP, IEEE 802.11w).

Quoted 2 lines

+ *
+ * @IEEE80211_HW_BEACON_FILTER: Hardware can drop similar beacon fra=

mes to

Quoted 1 line

+ * =C2=A0 =C2=A0 avoid waking up cpu. Enabling this flag disables th=

e beacon check

Quoted 1 line

+ * =C2=A0 =C2=A0 in mac80211 when in power save mode and the hardwar=

e must have an

Quoted 1 line

+ * =C2=A0 =C2=A0 event to notify that beacons are lost. Use ieee8021=

1_beacon_loss()

Quoted 1 line

+ * =C2=A0 =C2=A0 to notify the event to the stack. The filtering mus=

t be enabled

Quoted 2 lines

+ * =C2=A0 =C2=A0 only when IEEE80211_CONF_PS is set. =C2=A0*/

Lets add to the documentation a little detail explaining how beacon fil= tering
is expected to work, and elaborate on what we expect from the driver, i= f
anything all.

I'd like to review this just once more.

Last we reviewed this it seemed we were set on only allowing this for hardware which supports beacon filtering with support for checksumming of the beacons. Reason checksumming is important is for considerations of DFS with the channel switch announcements, HT with with channel widt= h
changes. In fact I'm curious why checksumming would be required for 2.4=
GHz
single band devices. Kalle, do you happen to know?

Considering that power saving features are important I think its worth to revisit
this position in detail.

An alternative to this above position is that if the devices do not support checksumming
of the beacons to let the driver figure when this should be enabled dynamically. For example
if devices do not support checksumming but are single 2.4 GHz band non-HT capable
devices we should let drivers enable this feature.

Other devices which do not support checksumming should only enable this=
only
when associating to an AP on non-DFS channels and when not using HT.

In practice this would mean allowing both ath5k and ath9k to take advan= tage
of this feature.

Technically we could move the conditional logic check when this should be enabled
to mac80211 as well. When we reviewed this it was seen as unnecessary complexity,
I actually don't see it as too complex and think the the advantage is worth to consider.

Thoughts?

Luis
--
To unsubscribe from this list: send the line "unsubscribe linux-wireles= s" in
the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html

author
#22 lore raw

Re: [RFC PATCH v2 4/4] mac80211: add beacon filtering support

Quoted 6 lines · Luis R. Rodriguez

Last we reviewed this it seemed we were set on only allowing this for hardware which supports beacon filtering with support for checksumming of the beacons. Reason checksumming is important is for considerations of DFS with the channel switch announcements, HT with with channel width changes. In fact I'm curious why checksumming would be required for 2.4 GHz single band devices. Kalle, do you happen to know?

In my opinion also 2.4 GHz band needs checksum support. At least ERP protection changes come to my mind, but maybe there are also others.

Quoted 11 lines

Considering that power saving features are important I think its worth to revisit this position in detail.

An alternative to this above position is that if the devices do not support checksumming of the beacons to let the driver figure when this should be enabled dynamically. For example if devices do not support checksumming but are single 2.4 GHz band non-HT capable devices we should let drivers enable this feature.

Other devices which do not support checksumming should only enable this only when associating to an AP on non-DFS channels and when not using HT.

I'm not fully convinced about this.

Quoted 2 lines

In practice this would mean allowing both ath5k and ath9k to take advantage of this feature.

Do you mean that they have beacon filter support but do not do any checksumming in hardware?

Other option is that the hardware which does not support checksumming would just pass beacons periodically to the stack, for example every 5 seconds. Even though beacon filter is enabled in mac80211 it does not prevent hardware from sending beacons. This is not the most optimal solution, but there's not much choices if the hardware doesn't support checksumming.

Quoted 6 lines

Technically we could move the conditional logic check when this should be enabled to mac80211 as well. When we reviewed this it was seen as unnecessary complexity, I actually don't see it as too complex and think the the advantage is worth to consider.

Thoughts?

I think this is future stuff, I would like to get the basic (and the simplest) implementation to the tree first and then we can start improving and extending it.

Signature · 1 line
Kalle Valo
maintainer
#23 lore raw

Re: [RFC PATCH v2 4/4] mac80211: add beacon filtering support

Quoted 11 lines · Kalle Valo
Quoted 6 lines · Luis R. Rodriguez

Last we reviewed this it seemed we were set on only allowing this for hardware which supports beacon filtering with support for checksumming of the beacons. Reason checksumming is important is for considerations of DFS with the channel switch announcements, HT with with channel width changes. In fact I'm curious why checksumming would be required for 2.4 GHz single band devices. Kalle, do you happen to know?

In my opinion also 2.4 GHz band needs checksum support. At least ERP protection changes come to my mind, but maybe there are also others.

Also, the virtual hardware support Jouni is working on might use quiet elements, and these would have to be detected by the stations. We really want beacon change detection somewhere.

Quoted 3 lines

I think this is future stuff, I would like to get the basic (and the simplest) implementation to the tree first and then we can start improving and extending it.

Agreed, anything that is more complex than this basic stuff which requires proper beacon change watching by the firmware is too much for the first cut here -- we're having enough trouble as-is, for example with the BSS struct stuff.

johannes

#24 lore raw

Re: [RFC PATCH v2 4/4] mac80211: add beacon filtering support

Quoted 17 lines · Johannes Berg
Quoted 11 lines · Kalle Valo
Quoted 6 lines · Luis R. Rodriguez

Last we reviewed this it seemed we were set on only allowing this for hardware which supports beacon filtering with support for checksumming of the beacons. Reason checksumming is important is for considerations of DFS with the channel switch announcements, HT with with channel width changes. In fact I'm curious why checksumming would be required for 2.4 GHz single band devices. Kalle, do you happen to know?

In my opinion also 2.4 GHz band needs checksum support. At least ERP protection changes come to my mind, but maybe there are also others.

Also, the virtual hardware support Jouni is working on might use quiet elements, and these would have to be detected by the stations. We really want beacon change detection somewhere.

Quoted 8 lines
Quoted 3 lines

I think this is future stuff, I would like to get the basic (and the simplest) implementation to the tree first and then we can start improving and extending it.

Agreed, anything that is more complex than this basic stuff which requires proper beacon change watching by the firmware is too much for the first cut here -- we're having enough trouble as-is, for example with the BSS struct stuff.

Understood, can we just add some text to the doc then elaborating what is expected of the hardware for now.

Luis

author
#25 lore raw

Re: [RFC PATCH v2 4/4] mac80211: add beacon filtering support

Quoted 7 lines · Luis R. Rodriguez
Quoted 5 lines · Johannes Berg
Quoted 3 lines · Kalle Valo

I think this is future stuff, I would like to get the basic (and the simplest) implementation to the tree first and then we can start improving and extending it.

Quoted 7 lines
Quoted 4 lines

Agreed, anything that is more complex than this basic stuff which requires proper beacon change watching by the firmware is too much for the first cut here -- we're having enough trouble as-is, for example with the BSS struct stuff.

Understood, can we just add some text to the doc then elaborating what is expected of the hardware for now.

Sure thing. I already have a todo item to write a better description.

Kalle

#26 lore raw

Re: [RFC PATCH v2 4/4] mac80211: add beacon filtering support

In reply to Kalle Valo
Quoted 2 lines · Kalle Valo

In my opinion also 2.4 GHz band needs checksum support. At least ERP protection changes come to my mind, but maybe there are also others.

Johannes mentioned Quiet IE as one such example. WMM can also change its parameters dynamically. Both of these need to be detected from Beacon frames. In addition, Channel Switch Announcement can be used to move the BSS to another channel, even on 2.4 GHz band.

Quoted 6 lines

Other option is that the hardware which does not support checksumming would just pass beacons periodically to the stack, for example every 5 seconds. Even though beacon filter is enabled in mac80211 it does not prevent hardware from sending beacons. This is not the most optimal solution, but there's not much choices if the hardware doesn't support checksumming.

While it would be possible to do that, it does not really solve many of the use cases where Beacon contents can change and the STA is required to take some action. For example, Channel Switch Announcement IEs would likely be missed if a Beacon frames is passed through that infrequently. I would assume the driver could enable this even if the hardware does not use checksumming to detect Beacon changes, but this may severely limit functionality in number of cases, so I'm not sure I would recommend this to be done in general.

Signature · 1 line
Jouni Malinen                                            PGP id EFC895FA
author
#27 lore raw

Re: [RFC PATCH v2 4/4] mac80211: add beacon filtering support

Quoted 7 lines · Jouni Malinen
Quoted 6 lines

Other option is that the hardware which does not support checksumming would just pass beacons periodically to the stack, for example every 5 seconds. Even though beacon filter is enabled in mac80211 it does not prevent hardware from sending beacons. This is not the most optimal solution, but there's not much choices if the hardware doesn't support checksumming.

Quoted 8 lines

While it would be possible to do that, it does not really solve many of the use cases where Beacon contents can change and the STA is required to take some action. For example, Channel Switch Announcement IEs would likely be missed if a Beacon frames is passed through that infrequently. I would assume the driver could enable this even if the hardware does not use checksumming to detect Beacon changes, but this may severely limit functionality in number of cases, so I'm not sure I would recommend this to be done in general.

Yes, definitely not recommended. I would classify it to "the ugly hack" category.

Signature · 1 line
Kalle Valo