Improvements to software scanning
Subsystems (5)
cover Improvements to software scanning
This patch series fixes a number of problems observed during software scanning, as described in [1].
The first four patches implement improved queue handling during off-channel operation and add some needed flushing of the hardware queues, as suggested by Johannes in [2]. This includes adding a new queue stop reason of off-channel operation, a transmit path for frames which do need to be transmitted when off-channel, and some improvements in error handling.
The last three fix a problem specific to brcmsmac (and likely b43, though I don't have hardware for testing b43). Broadcom hardware actively clears the PM bit in frame control whenever the hardware is not configured correctly, and since wireless drivers have no knowledge of off-channel powersave such configuration cannot be done. The patches expand the driver powersave configuration with an off-channel state and update brcmsmac to make use of it.
Johannes: I have a couple of comments/questions for you related to these patches.
First, in the patches I've added an offchan_tx_ok argument to the tx operations, but this seems a little awkward to me since it has to be propogated down through a fairly deep call stack. The alternative idea that occurred to me is to use a tx control flag, but that seems to be pretty crowded. Any thoughts?
Second, I attempted to test these patches with iwlwifi (Centrino Advanced-N 6235) to verify that I didn't break anything for drivers with hw scanning. My standard test for this involves running iperf while triggering nearly continuous scans, but I'm seeing lots of problems running a tcp iperf test even with unpatched 3.8-rc4. iperf with udp does fine in either direction. I haven't had time to do any kind of debugging yet, but I thought you'd want to know.
Thanks,
Seth
[1] http://marc.info/?l=linux-wireless&m=135766865110986&w=2
[2] http://marc.info/?l=linux-wireless&m=135838252227053&w=2
Seth Forshee (7):
mac80211: Return a status for tx operations
mac80211: Fix tx queue handling during scans
mac80211: Improve error handling for off-channel operation
mac80211: Add flushes before going off-channel
mac80211: Expand powersave configuration flag to be two bits
mac80211: Add off-channel powersave state
brcmsmac: Add support for off-channel powersave
drivers/net/wireless/ath/ath9k/htc_drv_main.c | 2 +- drivers/net/wireless/ath/ath9k/main.c | 2 +- drivers/net/wireless/ath/carl9170/main.c | 2 +- drivers/net/wireless/ath/carl9170/rx.c | 2 +- .../net/wireless/brcm80211/brcmsmac/mac80211_if.c | 15 ++- drivers/net/wireless/brcm80211/brcmsmac/main.c | 9 ++ drivers/net/wireless/brcm80211/brcmsmac/pub.h | 1 + drivers/net/wireless/iwlwifi/dvm/power.c | 2 +- drivers/net/wireless/mac80211_hwsim.c | 2 +- drivers/net/wireless/p54/fwio.c | 2 +- drivers/net/wireless/p54/txrx.c | 2 +- drivers/net/wireless/rt2x00/rt2400pci.c | 3 +- drivers/net/wireless/rt2x00/rt2500pci.c | 3 +- drivers/net/wireless/rt2x00/rt2500usb.c | 3 +- drivers/net/wireless/rt2x00/rt2800lib.c | 3 +- drivers/net/wireless/rt2x00/rt2x00config.c | 4 +- drivers/net/wireless/rt2x00/rt2x00dev.c | 2 +- drivers/net/wireless/rt2x00/rt61pci.c | 3 +- drivers/net/wireless/rt2x00/rt73usb.c | 3 +- drivers/net/wireless/rtlwifi/core.c | 2 +- drivers/net/wireless/rtlwifi/ps.c | 4 +- drivers/net/wireless/ti/wl1251/main.c | 7 +- drivers/net/wireless/ti/wlcore/main.c | 6 +- include/net/mac80211.h | 125 +++++++++++++++---- net/mac80211/agg-tx.c | 4 +- net/mac80211/cfg.c | 2 +- net/mac80211/ht.c | 2 +- net/mac80211/ieee80211_i.h | 72 ++++++++--- net/mac80211/mlme.c | 32 ++--- net/mac80211/offchannel.c | 114 +++++++++-------- net/mac80211/scan.c | 44 +++++-- net/mac80211/sta_info.c | 2 +- net/mac80211/tx.c | 131 +++++++++++--------- net/mac80211/util.c | 13 +- 34 files changed, 400 insertions(+), 225 deletions(-)
PATCH 1/7 mac80211: Return a status for tx operations
For off-channel operation we would like to know whether or not frames submitted for tx are dropped, but most tx interfaces don't return any status. The low-level interfaces do return bool but make no distinction between dropping a packet and sucessfully handing it off to the driver.
Add an enum named ieee80211_tx_status which defines status codes to indicate whether a tx request results in successfully handing the frame off to the driver, queueing the frame, or dropping it. Convert all tx interfaces to return one of the values.
--- net/mac80211/ieee80211_i.h | 53 ++++++++++++++++------- net/mac80211/tx.c | 103 +++++++++++++++++++++++--------------------- 2 files changed, 92 insertions(+), 64 deletions(-)
net/mac80211/ieee80211_i.h +38 -15
@@ -1495,46 +1495,69 @@ void mac80211_ev_michael_mic_failure(struct ieee80211_sub_if_data *sdata, int ke gfp_t gfp);void ieee80211_set_wmm_default (struct ieee80211_sub_if_data *sdata,bool bss_notify); -void ieee80211_xmit (struct ieee80211_sub_if_data *sdata,struct sk_buff *skb, -enum ieee80211_band band); -void __ieee80211_tx_skb_tid_band (struct ieee80211_sub_if_data *sdata, -struct sk_buff *skb,int tid, -enum ieee80211_band band); +/** +* enum ieee80211_tx_status - Tx request status +* @IEEE80211_TX_OK: Tx request handed off to driver +* @IEEE80211_TX_DROPPED: Tx request dropped +* @IEEE80211_TX_QUEUED: Tx requeust queued +*/ +enum ieee80211_tx_status { +IEEE80211_TX_OK , +IEEE80211_TX_DROPPED , +IEEE80211_TX_QUEUED , +}; -static inline void +enum ieee80211_tx_status +ieee80211_xmit (struct ieee80211_sub_if_data *sdata,struct sk_buff *skb, +enum ieee80211_band band); + +enum ieee80211_tx_status +__ieee80211_tx_skb_tid_band (struct ieee80211_sub_if_data *sdata, +struct sk_buff *skb,int tid, +enum ieee80211_band band); + +static inline enum ieee80211_tx_status ieee80211_tx_skb_tid_band (struct ieee80211_sub_if_data *sdata,struct sk_buff *skb,int tid,enum ieee80211_band band) { +enum ieee80211_tx_status ret; +rcu_read_lock (); -__ieee80211_tx_skb_tid_band (sdata, skb, tid, band); + ret =__ieee80211_tx_skb_tid_band (sdata, skb, tid, band);rcu_read_unlock (); + +return ret; } -static inline void ieee80211_tx_skb_tid (struct ieee80211_sub_if_data *sdata, -struct sk_buff *skb,int tid) +static inline enum ieee80211_tx_status +ieee80211_tx_skb_tid (struct ieee80211_sub_if_data *sdata,struct sk_buff *skb, +int tid) {struct ieee80211_chanctx_conf *chanctx_conf; +enum ieee80211_tx_status ret;rcu_read_lock (); chanctx_conf =rcu_dereference (sdata->vif.chanctx_conf);if (WARN_ON (!chanctx_conf)) {rcu_read_unlock ();kfree_skb (skb); -return ; +return IEEE80211_TX_DROPPED ; } -__ieee80211_tx_skb_tid_band (sdata, skb, tid, - chanctx_conf->def.chan->band); + ret =__ieee80211_tx_skb_tid_band (sdata, skb, tid, + chanctx_conf->def.chan->band);rcu_read_unlock (); + +return ret; } -static inline void ieee80211_tx_skb (struct ieee80211_sub_if_data *sdata, -struct sk_buff *skb) +static inline enum ieee80211_tx_status +ieee80211_tx_skb (struct ieee80211_sub_if_data *sdata,struct sk_buff *skb) {/* Send all internal mgmt frames on VO. Accordingly set TID to 7. */ -ieee80211_tx_skb_tid (sdata, skb,7 ); +return ieee80211_tx_skb_tid (sdata, skb,7 ); } void ieee802_11_parse_elems(u8 *start, size_t len,
net/mac80211/tx.c +54 -49
235 lines not on this page · raw message
Signature · 1 line
1.7.9.5
PATCH 2/7 mac80211: Fix tx queue handling during scans
In reply to Seth ForsheeScans currently work by stopping the netdev tx queues but leaving the mac80211 queues active. This stops the flow of incoming packets while still allowing mac80211 to transmit nullfunc and probe request frames to facilitate scanning. However, the driver may try to wake the mac80211 queues while in this state, which will also wake the netdev queues.
To prevent this, add a new queue stop reason, IEEE80211_QUEUE_STOP_REASON_OFFCHANNEL, to be used when stopping the tx queues for off-channel operation. This prevents the netdev queues from waking when a driver wakes the mac80211 queues.
This also stops all frames from being transmitted, even those required for scanning. To get around this, add a new offchan_tx_ok argument to most of the tx interfaces. This flag can be set for frames which need to be transmitted during off-channel operation, allowing off-channel frames to always be passed down to the driver as long as the driver hasn't stopped the queues. Add ieee80211_tx_skb_offchannel() for transmitting off-channel frames with this flag set.
--- net/mac80211/agg-tx.c | 4 ++-- net/mac80211/cfg.c | 2 +- net/mac80211/ht.c | 2 +- net/mac80211/ieee80211_i.h | 25 ++++++++++++++++++------- net/mac80211/mlme.c | 2 +- net/mac80211/offchannel.c | 32 +++++++++++--------------------- net/mac80211/sta_info.c | 2 +- net/mac80211/tx.c | 44 +++++++++++++++++++++++++++++--------------- net/mac80211/util.c | 3 ++- 9 files changed, 66 insertions(+), 50 deletions(-)
net/mac80211/agg-tx.c +2 -2
@@ -107,7 +107,7 @@ static void ieee80211_send_addba_request(struct ieee80211_sub_if_data *sdata, mgmt->u.action.u.addba_req.start_seq_num =cpu_to_le16 (start_seq_num <<4 ); -ieee80211_tx_skb_tid (sdata, skb, tid); +ieee80211_tx_skb_tid (sdata, skb, tid, false); }void ieee80211_send_bar (struct ieee80211_vif *vif,u8 *ra,u16 tid,u16 ssn)
@@ -137,7 +137,7 @@ void ieee80211_send_bar(struct ieee80211_vif *vif, u8 *ra, u16 tid, u16 ssn)IEEE80211_SKB_CB (skb)->flags |=IEEE80211_TX_INTFL_DONT_ENCRYPT |IEEE80211_TX_CTL_REQ_TX_STATUS ; -ieee80211_tx_skb_tid (sdata, skb, tid); +ieee80211_tx_skb_tid (sdata, skb, tid, false); }EXPORT_SYMBOL (ieee80211_send_bar);
net/mac80211/cfg.c +1 -1
@@ -3211,7 +3211,7 @@ static int ieee80211_probe_client(struct wiphy *wiphy, struct net_device *dev, nullfunc->qos_ctrl =cpu_to_le16 (7 );local_bh_disable (); -ieee80211_xmit (sdata, skb, band); +ieee80211_xmit (sdata, skb, band, false);local_bh_enable ();rcu_read_unlock ();
net/mac80211/ht.c +1 -1
@@ -311,7 +311,7 @@ void ieee80211_send_delba(struct ieee80211_sub_if_data *sdata, mgmt->u.action.u.delba.params =cpu_to_le16 (params); mgmt->u.action.u.delba.reason_code =cpu_to_le16 (reason_code); -ieee80211_tx_skb_tid (sdata, skb, tid); +ieee80211_tx_skb_tid (sdata, skb, tid, false); } void ieee80211_process_delba(struct ieee80211_sub_if_data *sdata,
net/mac80211/ieee80211_i.h +18 -7
@@ -842,6 +842,7 @@ enum queue_stop_reason {IEEE80211_QUEUE_STOP_REASON_AGGREGATION ,IEEE80211_QUEUE_STOP_REASON_SUSPEND ,IEEE80211_QUEUE_STOP_REASON_SKB_ADD , + IEEE80211_QUEUE_STOP_REASON_OFFCHANNEL, };#ifdef CONFIG_MAC80211_LEDS
@@ -1510,22 +1511,23 @@ enum ieee80211_tx_status {enum ieee80211_tx_status ieee80211_xmit (struct ieee80211_sub_if_data *sdata,struct sk_buff *skb, -enum ieee80211_band band); +enum ieee80211_band band,bool offchan_tx_ok);enum ieee80211_tx_status __ieee80211_tx_skb_tid_band (struct ieee80211_sub_if_data *sdata,struct sk_buff *skb,int tid, -enum ieee80211_band band); +enum ieee80211_band band,bool offchan_tx_ok);static inline enum ieee80211_tx_status ieee80211_tx_skb_tid_band (struct ieee80211_sub_if_data *sdata,struct sk_buff *skb,int tid, -enum ieee80211_band band) +enum ieee80211_band band,bool offchan_tx_ok) {enum ieee80211_tx_status ret;rcu_read_lock (); - ret =__ieee80211_tx_skb_tid_band (sdata, skb, tid, band); + ret =__ieee80211_tx_skb_tid_band (sdata, skb, tid, band, + offchan_tx_ok);rcu_read_unlock ();return ret;
@@ -1533,7 +1535,7 @@ ieee80211_tx_skb_tid_band(struct ieee80211_sub_if_data *sdata,static inline enum ieee80211_tx_status ieee80211_tx_skb_tid (struct ieee80211_sub_if_data *sdata,struct sk_buff *skb, -int tid) +int tid,bool offchan_tx_ok) {struct ieee80211_chanctx_conf *chanctx_conf;enum ieee80211_tx_status ret;
@@ -1547,7 +1549,8 @@ ieee80211_tx_skb_tid(struct ieee80211_sub_if_data *sdata, struct sk_buff *skb, } ret =__ieee80211_tx_skb_tid_band (sdata, skb, tid, - chanctx_conf->def.chan->band); + chanctx_conf->def.chan->band, + offchan_tx_ok);rcu_read_unlock ();return ret;
@@ -1557,7 +1560,15 @@ static inline enum ieee80211_tx_status ieee80211_tx_skb(struct ieee80211_sub_if_data *sdata, struct sk_buff *skb) {/* Send all internal mgmt frames on VO. Accordingly set TID to 7. */ -return ieee80211_tx_skb_tid (sdata, skb,7 ); +return ieee80211_tx_skb_tid (sdata, skb,7 , false); +} + +static inline enum ieee80211_tx_status +ieee80211_tx_skb_offchannel (struct ieee80211_sub_if_data *sdata, +struct sk_buff *skb) +{ +/* Send all internal mgmt frames on VO. Accordingly set TID to 7. */ +return ieee80211_tx_skb_tid (sdata, skb,7 , true); } void ieee802_11_parse_elems(u8 *start, size_t len,
net/mac80211/mlme.c +1 -1
@@ -685,7 +685,7 @@ void ieee80211_send_nullfunc(struct ieee80211_local *local, IEEE80211_STA_CONNECTION_POLL)) IEEE80211_SKB_CB(skb)->flags |= IEEE80211_TX_CTL_USE_MINRATE; - ieee80211_tx_skb(sdata, skb); + ieee80211_tx_skb_offchannel(sdata, skb); } static void ieee80211_send_4addr_nullfunc(struct ieee80211_local *local,
net/mac80211/offchannel.c +11 -21
@@ -113,6 +113,10 @@ void ieee80211_offchannel_stop_vifs(struct ieee80211_local *local)* notify the AP about us leaving the channel and stop all * STA interfaces. */ + +ieee80211_stop_queues_by_reason (&local->hw, +IEEE80211_QUEUE_STOP_REASON_OFFCHANNEL ); +mutex_lock (&local->iflist_mtx);list_for_each_entry (sdata, &local->interfaces, list) {if (!ieee80211_sdata_running (sdata))
@@ -133,12 +137,9 @@ void ieee80211_offchannel_stop_vifs(struct ieee80211_local *local) sdata, BSS_CHANGED_BEACON_ENABLED); } -if (sdata->vif.type !=NL80211_IFTYPE_MONITOR ) { -netif_tx_stop_all_queues (sdata->dev); -if (sdata->vif.type ==NL80211_IFTYPE_STATION && - sdata->u.mgd.associated) -ieee80211_offchannel_ps_enable (sdata); - } +if (sdata->vif.type ==NL80211_IFTYPE_STATION && + sdata->u.mgd.associated) +ieee80211_offchannel_ps_enable (sdata); }mutex_unlock (&local->iflist_mtx); }
@@ -166,20 +167,6 @@ void ieee80211_offchannel_return(struct ieee80211_local *local) sdata->u.mgd.associated)ieee80211_offchannel_ps_disable (sdata); -if (sdata->vif.type !=NL80211_IFTYPE_MONITOR ) { -/* -* This may wake up queues even though the driver -* currently has them stopped. This is not very -* likely, since the driver won't have gotten any -* (or hardly any) new packets while we weren't -* on the right channel, and even if it happens -* it will at most lead to queueing up one more -* packet per queue in mac80211 rather than on -* the interface qdisc. -*/ -netif_tx_wake_all_queues (sdata->dev); - } -if (test_and_clear_bit (SDATA_STATE_OFFCHANNEL_BEACON_STOPPED , &sdata->state)) { sdata->vif.bss_conf.enable_beacon = true;
@@ -188,6 +175,9 @@ void ieee80211_offchannel_return(struct ieee80211_local *local) } }mutex_unlock (&local->iflist_mtx); + +ieee80211_wake_queues_by_reason (&local->hw, +IEEE80211_QUEUE_STOP_REASON_OFFCHANNEL ); }void ieee80211_handle_roc_started (struct ieee80211_roc_work *roc)
@@ -198,7 +188,7 @@ void ieee80211_handle_roc_started(struct ieee80211_roc_work *roc)if (roc->mgmt_tx_cookie) {if (!WARN_ON (!roc->frame)) {ieee80211_tx_skb_tid_band (roc->sdata, roc->frame,7 , - roc->chan->band); + roc->chan->band, true); roc->frame =NULL ; } }else {
net/mac80211/sta_info.c +1 -1
@@ -1122,7 +1122,7 @@ static void ieee80211_send_null_response(struct ieee80211_sub_if_data *sdata,return ; } -ieee80211_xmit (sdata, skb, chanctx_conf->def.chan->band); +ieee80211_xmit (sdata, skb, chanctx_conf->def.chan->band, false);rcu_read_unlock (); }
net/mac80211/tx.c +29 -15
128 lines not on this page · raw message
net/mac80211/util.c +2 -1
@@ -1290,7 +1290,8 @@ void ieee80211_send_probe_req(struct ieee80211_sub_if_data *sdata, u8 *dst,IEEE80211_SKB_CB (skb)->flags |=IEEE80211_TX_CTL_NO_CCK_RATE ;if (scan) -ieee80211_tx_skb_tid_band (sdata, skb,7 , channel->band); +ieee80211_tx_skb_tid_band (sdata, skb,7 , + channel->band, true);else ieee80211_tx_skb (sdata, skb); }
Signature · 1 line
1.7.9.5
Johannes Berg
maintainerRe: [PATCH 2/7] mac80211: Fix tx queue handling during scans
net/mac80211/tx.c ieee80211_tx_frags(struct ieee80211_local *local, struct ieee80211_vif *vif, · line 1225
Quoted 1 line · Seth Forshee
net/mac80211/tx.c +1
-0+if (offchan_tx_ok) {
unlikely(), this is one of the hottest code paths in mac80211 after all :-)
net/mac80211/tx.c ieee80211_tx_frags(struct ieee80211_local *local, struct ieee80211_vif *vif, · line 1231
Quoted 6 lines
net/mac80211/tx.c +6
-0+/* +* Always directly transmit off-channel frames +* unless the driver has stopped the queues. +*/ +if (test_bit (IEEE80211_QUEUE_STOP_REASON_DRIVER , + &local->queue_stop_reasons[q])) {
might there be other reasons to TX, i.e. ignore only STOP_REASON_OFFCHANNEL?
That might be more efficient too:
offchflag = offchan_tx_ok << log2(STOP_REASON_OFFCHANNEL);
net/mac80211/tx.c ieee80211_tx_frags(struct ieee80211_local *local, struct ieee80211_vif *vif, · line 1237
Quoted 1 line
net/mac80211/tx.c +1
-0+ } else if (local->queue_stop_reasons[q] ||
if (local->queue_stop_reasons[q] & ~offchflag || ...
OTOH, I guess you want a different return value too. But that's in a relatively unlikely code path again, so might still be better to then again differentiate within that if.
johannes
Re: [PATCH 2/7] mac80211: Fix tx queue handling during scans
Quoted 6 lines · Johannes Berg
net/mac80211/tx.c ieee80211_tx_frags(struct ieee80211_local *local, struct ieee80211_vif *vif, · line 1225
Quoted 1 line · Seth Forshee
net/mac80211/tx.c +1
-0+if (offchan_tx_ok) {
unlikely(), this is one of the hottest code paths in mac80211 after all :-)
Ack, I'll change this for v2.
Quoted 9 lines
net/mac80211/tx.c ieee80211_tx_frags(struct ieee80211_local *local, struct ieee80211_vif *vif, · line 1231
Quoted 6 lines
net/mac80211/tx.c +6
-0+/* +* Always directly transmit off-channel frames +* unless the driver has stopped the queues. +*/ +if (test_bit (IEEE80211_QUEUE_STOP_REASON_DRIVER , + &local->queue_stop_reasons[q])) {might there be other reasons to TX, i.e. ignore only
STOP_REASON_OFFCHANNEL?
I originally wrote it that way, but I'm not sure that all of the stop reasons need to block tx here. I'll take another look.
Quoted 11 lines
That might be more efficient too:
offchflag =
offchan_tx_ok<<log2(STOP_REASON_OFFCHANNEL);net/mac80211/tx.c ieee80211_tx_frags(struct ieee80211_local *local, struct ieee80211_vif *vif, · line 1237
Quoted 1 line
net/mac80211/tx.c +1
-0+ } else if (local->queue_stop_reasons[q] ||if (local->queue_stop_reasons[q] & ~offchflag || ...
OTOH, I guess you want a different return value too. But that's in a relatively unlikely code path again, so might still be better to then again differentiate within that if.
Good suggestions, I'll take a look at applying them in some form.
Seth
PATCH 3/7 mac80211: Improve error handling for off-channel operation
In reply to Seth ForsheeErrors in sending nullfunc or probe request frames during off-channel operation may have undesirable consequences, e.g. failure to set powersave at the AP. Add error handling for failures to transmit these frames. In the case of a nullfunc failure, fail to go off-channel and return an error to userspace. In the case of a failed probe request, abort the scan.
--- net/mac80211/ieee80211_i.h | 6 +++--- net/mac80211/mlme.c | 6 +++--- net/mac80211/offchannel.c | 24 ++++++++++++++++++------ net/mac80211/scan.c | 41 ++++++++++++++++++++++++++++------------- net/mac80211/util.c | 12 ++++++++---- 5 files changed, 60 insertions(+), 29 deletions(-)
net/mac80211/ieee80211_i.h +3 -3
@@ -1369,7 +1369,7 @@ int ieee80211_request_sched_scan_stop(struct ieee80211_sub_if_data *sdata);void ieee80211_sched_scan_stopped_work (struct work_struct *work);/* off-channel helpers */ -void ieee80211_offchannel_stop_vifs (struct ieee80211_local *local); +bool ieee80211_offchannel_stop_vifs (struct ieee80211_local *local);void ieee80211_offchannel_return (struct ieee80211_local *local);void ieee80211_roc_setup (struct ieee80211_local *local);void ieee80211_start_next_roc (struct ieee80211_local *local);
@@ -1582,7 +1582,7 @@ u32 ieee80211_mandatory_rates(struct ieee80211_local *local,void ieee80211_dynamic_ps_enable_work (struct work_struct *work);void ieee80211_dynamic_ps_disable_work (struct work_struct *work);void ieee80211_dynamic_ps_timer (unsigned long data); -void ieee80211_send_nullfunc (struct ieee80211_local *local, +bool 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,
@@ -1627,7 +1627,7 @@ struct sk_buff *ieee80211_build_probe_req(struct ieee80211_sub_if_data *sdata, const u8 *ssid, size_t ssid_len, const u8 *ie, size_t ie_len, bool directed); -void ieee80211_send_probe_req(struct ieee80211_sub_if_data *sdata, u8 *dst, +bool ieee80211_send_probe_req(struct ieee80211_sub_if_data *sdata, u8 *dst, const u8 *ssid, size_t ssid_len, const u8 *ie, size_t ie_len, u32 ratemask, bool directed, bool no_cck,
net/mac80211/mlme.c +3 -3
@@ -664,7 +664,7 @@ void ieee80211_send_pspoll(struct ieee80211_local *local,ieee80211_tx_skb (sdata, skb); } -void ieee80211_send_nullfunc (struct ieee80211_local *local, +bool ieee80211_send_nullfunc (struct ieee80211_local *local,struct ieee80211_sub_if_data *sdata,int powersave) {
@@ -674,7 +674,7 @@ void ieee80211_send_nullfunc(struct ieee80211_local *local, skb =ieee80211_nullfunc_get (&local->hw, &sdata->vif);if (!skb) -return ; +return false; nullfunc = (struct ieee80211_hdr_3addr *) skb->data;if (powersave)
@@ -685,7 +685,7 @@ void ieee80211_send_nullfunc(struct ieee80211_local *local, IEEE80211_STA_CONNECTION_POLL)) IEEE80211_SKB_CB(skb)->flags |= IEEE80211_TX_CTL_USE_MINRATE; - ieee80211_tx_skb_offchannel(sdata, skb); + return ieee80211_tx_skb_offchannel(sdata, skb) == IEEE80211_TX_OK; } static void ieee80211_send_4addr_nullfunc(struct ieee80211_local *local,
net/mac80211/offchannel.c +18 -6
@@ -24,10 +24,11 @@* because we *may* be doing work on-operating channel, and want our * hardware unconditionally awake, but still let the AP send us normal frames. */ -static void ieee80211_offchannel_ps_enable (struct ieee80211_sub_if_data *sdata) +static bool ieee80211_offchannel_ps_enable (struct ieee80211_sub_if_data *sdata) {struct ieee80211_local *local = sdata->local;struct ieee80211_if_managed *ifmgd = &sdata->u.mgd; +bool ret = true; local->offchannel_ps_enabled = false;
@@ -57,7 +58,9 @@ static void ieee80211_offchannel_ps_enable(struct ieee80211_sub_if_data *sdata)* to send a new nullfunc frame to inform the AP that we * are again sleeping. */ -ieee80211_send_nullfunc (local, sdata,1 ); + ret =ieee80211_send_nullfunc (local, sdata,1 ); + +return ret; }/* inform AP that we are awake again, unless power save is enabled */
@@ -102,12 +105,13 @@ static void ieee80211_offchannel_ps_disable(struct ieee80211_sub_if_data *sdata)ieee80211_sta_reset_conn_monitor (sdata); } -void ieee80211_offchannel_stop_vifs (struct ieee80211_local *local) +bool ieee80211_offchannel_stop_vifs (struct ieee80211_local *local) {struct ieee80211_sub_if_data *sdata; +bool ret = true;if (WARN_ON (local->use_chanctx)) -return ; +return false;/* * notify the AP about us leaving the channel and stop all
@@ -138,10 +142,18 @@ void ieee80211_offchannel_stop_vifs(struct ieee80211_local *local) }if (sdata->vif.type ==NL80211_IFTYPE_STATION && - sdata->u.mgd.associated) -ieee80211_offchannel_ps_enable (sdata); + sdata->u.mgd.associated) { + ret =ieee80211_offchannel_ps_enable (sdata); +if (!ret) +break ; + } }mutex_unlock (&local->iflist_mtx); + +/* Attempt to recover if failed */ +if (!ret) +ieee80211_offchannel_return (local); +return ret; }void ieee80211_offchannel_return (struct ieee80211_local *local)
net/mac80211/scan.c +28 -13
74 lines not on this page · raw message
net/mac80211/util.c +8 -4
@@ -1274,13 +1274,14 @@ struct sk_buff *ieee80211_build_probe_req(struct ieee80211_sub_if_data *sdata,return skb; } -void ieee80211_send_probe_req (struct ieee80211_sub_if_data *sdata,u8 *dst, +bool ieee80211_send_probe_req (struct ieee80211_sub_if_data *sdata,u8 *dst,const u8 *ssid,size_t ssid_len,const u8 *ie,size_t ie_len,u32 ratemask,bool directed,bool no_cck,struct ieee80211_channel *channel,bool scan) {struct sk_buff *skb; +enum ieee80211_tx_status tx_stat =IEEE80211_TX_DROPPED ; skb = ieee80211_build_probe_req(sdata, dst, ratemask, channel, ssid, ssid_len,
@@ -1290,11 +1291,14 @@ void ieee80211_send_probe_req(struct ieee80211_sub_if_data *sdata, u8 *dst,IEEE80211_SKB_CB (skb)->flags |=IEEE80211_TX_CTL_NO_CCK_RATE ;if (scan) -ieee80211_tx_skb_tid_band (sdata, skb,7 , - channel->band, true); + tx_stat =ieee80211_tx_skb_tid_band (sdata, skb,7 , + channel->band, + true);else -ieee80211_tx_skb (sdata, skb); + tx_stat =ieee80211_tx_skb (sdata, skb); } + +return tx_stat !=IEEE80211_TX_DROPPED ; } u32 ieee80211_sta_get_rates(struct ieee80211_local *local,
Signature · 1 line
1.7.9.5
Johannes Berg
maintainerRe: [PATCH 3/7] mac80211: Improve error handling for off-channel operation
Quoted 6 lines · Seth Forshee
Errors in sending nullfunc or probe request frames during off-channel operation may have undesirable consequences, e.g. failure to set powersave at the AP. Add error handling for failures to transmit these frames. In the case of a nullfunc failure, fail to go off-channel and return an error to userspace. In the case of a failed probe request, abort the scan.
That latter part seems excessive? Maybe increase the time to use a passive scan? But if there are multiple scan requests ...
Is all of this really worth the effort? The driver queues should be empty after the flush, after all, and the driver doesn't return any TX status. So what can really happen?
johannes
Re: [PATCH 3/7] mac80211: Improve error handling for off-channel operation
Quoted 8 lines · Johannes Berg
Quoted 6 lines · Seth Forshee
Errors in sending nullfunc or probe request frames during off-channel operation may have undesirable consequences, e.g. failure to set powersave at the AP. Add error handling for failures to transmit these frames. In the case of a nullfunc failure, fail to go off-channel and return an error to userspace. In the case of a failed probe request, abort the scan.
Quoted 6 lines
That latter part seems excessive? Maybe increase the time to use a passive scan? But if there are multiple scan requests ...
Is all of this really worth the effort? The driver queues should be empty after the flush, after all, and the driver doesn't return any TX status. So what can really happen?
Hmm, yeah, maybe it is a bit excessive. The idea of falling back to a passive scan is interesting though. I'll rip out the scan abort stuff for v2 and think about doing the passive scan fallback as a future enhancement.
Seth
PATCH 4/7 mac80211: Add flushes before going off-channel
In reply to Seth ForsheeWe've got a couple of races when enabling powersave with an AP for off-channel operation. The first is fairly simple. If we go off-channel before the nullfunc frame to enable PS is transmitted then it may not be received by the AP. Add a flush after enabling off-channel PS to prevent this from happening.
The second race is a bit more subtle. If the driver supports QoS and has frames queued when the nullfunc frame is queued, those frames may get transmitted after the nullfunc frame. If PM is not set then the AP is being told that we've exited PS before we go off-channel and may try to deliver frames. To prevent this, add a flush after stopping the queues but before passing the nullfunc frame to the driver.
--- net/mac80211/offchannel.c | 5 +++++ net/mac80211/scan.c | 3 +++ 2 files changed, 8 insertions(+)
net/mac80211/offchannel.c +5 -0
@@ -118,8 +118,13 @@ bool ieee80211_offchannel_stop_vifs(struct ieee80211_local *local)* STA interfaces. */ +/* +* Stop queues and transmit all frames queued by the driver +* before sending nullfunc to enable powersave at the AP. +*/ ieee80211_stop_queues_by_reason (&local->hw,IEEE80211_QUEUE_STOP_REASON_OFFCHANNEL ); +drv_flush (local, false);mutex_lock (&local->iflist_mtx);list_for_each_entry (sdata, &local->interfaces, list) {
net/mac80211/scan.c +3 -0
@@ -343,6 +343,9 @@ static int ieee80211_start_sw_scan(struct ieee80211_local *local)if (!ieee80211_offchannel_stop_vifs (local)) goto error; +/* ensure nullfunc is transmitted before leaving operating channel */ +drv_flush (local, false); +ieee80211_configure_filter (local);/* We need to set power level at maximum rate for scanning. */
Signature · 1 line
1.7.9.5
PATCH 5/7 mac80211: Expand powersave configuration flag to be two bits
In reply to Seth ForsheeThe powersave configuration flag is a single bit and can thus only indicate that powersave is either on or off. This is inadequate for off-channel powersave, where the hardware must remain powered on but some configuration of the hardware may be required (e.g. to ensure that PM is set in frame control in frames sent to the associated AP, as with brcmsmac).
In preparation for supporting an off-channel powersave state, expand the powersave configuration flag to two bits. Also introduce some interfaces which functionally replace the current open-coded setting and checking of the power state and update mac80211 and the drivers to use these interfaces. Update the documentation to refer to the new interfaces.
--- drivers/net/wireless/ath/ath9k/htc_drv_main.c | 2 +- drivers/net/wireless/ath/ath9k/main.c | 2 +- drivers/net/wireless/ath/carl9170/main.c | 2 +- drivers/net/wireless/ath/carl9170/rx.c | 2 +- .../net/wireless/brcm80211/brcmsmac/mac80211_if.c | 2 +- drivers/net/wireless/iwlwifi/dvm/power.c | 2 +- drivers/net/wireless/mac80211_hwsim.c | 2 +- drivers/net/wireless/p54/fwio.c | 2 +- drivers/net/wireless/p54/txrx.c | 2 +- drivers/net/wireless/rt2x00/rt2400pci.c | 3 +- drivers/net/wireless/rt2x00/rt2500pci.c | 3 +- drivers/net/wireless/rt2x00/rt2500usb.c | 3 +- drivers/net/wireless/rt2x00/rt2800lib.c | 3 +- drivers/net/wireless/rt2x00/rt2x00config.c | 4 +- drivers/net/wireless/rt2x00/rt2x00dev.c | 2 +- drivers/net/wireless/rt2x00/rt61pci.c | 3 +- drivers/net/wireless/rt2x00/rt73usb.c | 3 +- drivers/net/wireless/rtlwifi/core.c | 2 +- drivers/net/wireless/rtlwifi/ps.c | 4 +- drivers/net/wireless/ti/wl1251/main.c | 7 +- drivers/net/wireless/ti/wlcore/main.c | 6 +- include/net/mac80211.h | 96 +++++++++++++++----- net/mac80211/mlme.c | 26 +++--- net/mac80211/offchannel.c | 10 +- net/mac80211/tx.c | 2 +- net/mac80211/util.c | 2 +- 26 files changed, 123 insertions(+), 74 deletions(-)
drivers/net/wireless/ath/ath9k/htc_drv_main.c +1 -1
@@ -1212,7 +1212,7 @@ static int ath9k_htc_config(struct ieee80211_hw *hw, u32 changed) }if (changed &IEEE80211_CONF_CHANGE_PS ) { -if (conf->flags &IEEE80211_CONF_PS ) { +if (ieee80211_is_ps_enabled (conf)) {ath9k_htc_setpower (priv,ATH9K_PM_NETWORK_SLEEP ); priv->ps_enabled = true; }else {
drivers/net/wireless/ath/ath9k/main.c +1 -1
@@ -1180,7 +1180,7 @@ static int ath9k_config(struct ieee80211_hw *hw, u32 changed)if (changed &IEEE80211_CONF_CHANGE_PS ) {unsigned long flags;spin_lock_irqsave (&sc->sc_pm_lock, flags); -if (conf->flags &IEEE80211_CONF_PS ) +if (ieee80211_is_ps_enabled (conf))ath9k_enable_ps (sc);else ath9k_disable_ps (sc);
drivers/net/wireless/ath/carl9170/main.c +1 -1
@@ -836,7 +836,7 @@ static int carl9170_ps_update(struct ar9170 *ar)int err =0 ;if (!ar->ps.off_override) - ps = (ar->hw->conf.flags &IEEE80211_CONF_PS ); + ps =ieee80211_is_ps_enabled (&ar->hw->conf);if (ps != ar->ps.state) { err =carl9170_powersave (ar, ps);
drivers/net/wireless/ath/carl9170/rx.c +1 -1
@@ -524,7 +524,7 @@ static void carl9170_ps_beacon(struct ar9170 *ar, void *data, unsigned int len)u8 tim_len;bool cam; -if (likely (!(ar->hw->conf.flags &IEEE80211_CONF_PS ))) +if (likely (ieee80211_is_ps_disabled (&ar->hw->conf)))return ;/* check if this really is a beacon */
drivers/net/wireless/brcm80211/brcmsmac/mac80211_if.c +1 -1
@@ -395,7 +395,7 @@ static int brcms_ops_config(struct ieee80211_hw *hw, u32 changed) "true" : "false");if (changed &IEEE80211_CONF_CHANGE_PS )brcms_err (core,"%s: change power-save mode: %s (implement)\n" , - __func__, conf->flags &IEEE80211_CONF_PS ? + __func__,ieee80211_is_ps_enabled (conf) ?"true" :"false" );if (changed &IEEE80211_CONF_CHANGE_POWER ) {
drivers/net/wireless/iwlwifi/dvm/power.c +1 -1
@@ -286,7 +286,7 @@ static int iwl_set_power(struct iwl_priv *priv, struct iwl_powertable_cmd *cmd)static void iwl_power_build_cmd (struct iwl_priv *priv,struct iwl_powertable_cmd *cmd) { -bool enabled = priv->hw->conf.flags &IEEE80211_CONF_PS ; +bool enabled =ieee80211_is_ps_enabled (&priv->hw->conf);int dtimper; dtimper = priv->hw->conf.ps_dtim_period ?:1 ;
drivers/net/wireless/mac80211_hwsim.c +1 -1
@@ -1080,7 +1080,7 @@ static int mac80211_hwsim_config(struct ieee80211_hw *hw, u32 changed) conf->channel ? conf->channel->center_freq :0 , hwsim_chantypes[conf->channel_type], !!(conf->flags &IEEE80211_CONF_IDLE ), - !!(conf->flags &IEEE80211_CONF_PS ), +ieee80211_is_ps_enabled (conf), smps_modes[conf->smps_mode]); data->idle = !!(conf->flags &IEEE80211_CONF_IDLE );
drivers/net/wireless/p54/fwio.c +1 -1
@@ -601,7 +601,7 @@ int p54_set_ps(struct p54_common *priv)unsigned int i;u16 mode; -if (priv->hw->conf.flags &IEEE80211_CONF_PS && +if (ieee80211_is_ps_enabled (&priv->hw->conf) && !priv->powersave_override) mode =P54_PSM |P54_PSM_BEACON_TIMEOUT |P54_PSM_DTIM |P54_PSM_CHECKSUM |P54_PSM_MCBC ;
drivers/net/wireless/p54/txrx.c +1 -1
@@ -380,7 +380,7 @@ static int p54_rx_data(struct p54_common *priv, struct sk_buff *skb)skb_pull (skb, header_len);skb_trim (skb,le16_to_cpu (hdr->len)); -if (unlikely (priv->hw->conf.flags &IEEE80211_CONF_PS )) +if (unlikely (ieee80211_is_ps_enabled (&priv->hw->conf)))p54_pspoll_workaround (priv, skb);ieee80211_rx_irqsafe (priv->hw, skb);
drivers/net/wireless/rt2x00/rt2400pci.c +1 -2
@@ -521,8 +521,7 @@ static void rt2400pci_config_ps(struct rt2x00_dev *rt2x00dev, struct rt2x00lib_conf *libconf) {enum dev_state state = - (libconf->conf->flags &IEEE80211_CONF_PS ) ? -STATE_SLEEP :STATE_AWAKE ; +ieee80211_is_ps_enabled (libconf->conf) ?STATE_SLEEP :STATE_AWAKE ;u32 reg;if (state ==STATE_SLEEP ) {
drivers/net/wireless/rt2x00/rt2500pci.c +1 -2
@@ -570,8 +570,7 @@ static void rt2500pci_config_ps(struct rt2x00_dev *rt2x00dev, struct rt2x00lib_conf *libconf) {enum dev_state state = - (libconf->conf->flags &IEEE80211_CONF_PS ) ? -STATE_SLEEP :STATE_AWAKE ; +ieee80211_is_ps_enabled (libconf->conf) ?STATE_SLEEP :STATE_AWAKE ;u32 reg;if (state ==STATE_SLEEP ) {
drivers/net/wireless/rt2x00/rt2500usb.c +1 -2
@@ -649,8 +649,7 @@ static void rt2500usb_config_ps(struct rt2x00_dev *rt2x00dev, struct rt2x00lib_conf *libconf) {enum dev_state state = - (libconf->conf->flags &IEEE80211_CONF_PS ) ? -STATE_SLEEP :STATE_AWAKE ; +ieee80211_is_ps_enabled (libconf->conf) ?STATE_SLEEP :STATE_AWAKE ;u16 reg;if (state ==STATE_SLEEP ) {
drivers/net/wireless/rt2x00/rt2800lib.c +1 -2
@@ -2862,8 +2862,7 @@ static void rt2800_config_ps(struct rt2x00_dev *rt2x00dev, struct rt2x00lib_conf *libconf) {enum dev_state state = - (libconf->conf->flags &IEEE80211_CONF_PS ) ? -STATE_SLEEP :STATE_AWAKE ; +ieee80211_is_ps_enabled (libconf->conf) ?STATE_SLEEP :STATE_AWAKE ;u32 reg;if (state ==STATE_SLEEP ) {
drivers/net/wireless/rt2x00/rt2x00config.c +2 -2
@@ -261,7 +261,7 @@ void rt2x00lib_config(struct rt2x00_dev *rt2x00dev,if (test_bit (DEVICE_STATE_PRESENT , &rt2x00dev->flags) &&test_bit (REQUIRE_PS_AUTOWAKE , &rt2x00dev->cap_flags) && (ieee80211_flags &IEEE80211_CONF_CHANGE_PS ) && - (conf->flags &IEEE80211_CONF_PS )) { +ieee80211_is_ps_enabled (conf)) { beacon_diff = (long )jiffies - (long )rt2x00dev->last_beacon; beacon_int =msecs_to_jiffies (rt2x00dev->beacon_int);
@@ -274,7 +274,7 @@ void rt2x00lib_config(struct rt2x00_dev *rt2x00dev, autowake_timeout - 15); } -if (conf->flags &IEEE80211_CONF_PS ) +if (ieee80211_is_ps_enabled (conf))set_bit (CONFIG_POWERSAVING , &rt2x00dev->flags);else clear_bit (CONFIG_POWERSAVING , &rt2x00dev->flags);
drivers/net/wireless/rt2x00/rt2x00dev.c +1 -1
@@ -596,7 +596,7 @@ static void rt2x00lib_rxdone_check_ps(struct rt2x00_dev *rt2x00dev,* configured, or if the device is already in powersaving mode * we can exit now. */ if (likely (!ieee80211_is_beacon (hdr->frame_control) || - !(rt2x00dev->hw->conf.flags &IEEE80211_CONF_PS ))) +ieee80211_is_ps_disabled (&rt2x00dev->hw->conf)))return ;/* min. beacon length + FCS_LEN */
drivers/net/wireless/rt2x00/rt61pci.c +1 -2
@@ -942,8 +942,7 @@ static void rt61pci_config_ps(struct rt2x00_dev *rt2x00dev, struct rt2x00lib_conf *libconf) {enum dev_state state = - (libconf->conf->flags &IEEE80211_CONF_PS ) ? -STATE_SLEEP :STATE_AWAKE ; +ieee80211_is_ps_enabled (libconf->conf) ?STATE_SLEEP :STATE_AWAKE ;u32 reg;if (state ==STATE_SLEEP ) {
drivers/net/wireless/rt2x00/rt73usb.c +1 -2
@@ -827,8 +827,7 @@ static void rt73usb_config_ps(struct rt2x00_dev *rt2x00dev, struct rt2x00lib_conf *libconf) {enum dev_state state = - (libconf->conf->flags &IEEE80211_CONF_PS ) ? -STATE_SLEEP :STATE_AWAKE ; +ieee80211_is_ps_enabled (libconf->conf) ?STATE_SLEEP :STATE_AWAKE ;u32 reg;if (state ==STATE_SLEEP ) {
drivers/net/wireless/rtlwifi/core.c +1 -1
@@ -289,7 +289,7 @@ static int rtl_op_config(struct ieee80211_hw *hw, u32 changed)if (changed &IEEE80211_CONF_CHANGE_PS ) {cancel_delayed_work (&rtlpriv->works.ps_work);cancel_delayed_work (&rtlpriv->works.ps_rfon_wq); -if (conf->flags &IEEE80211_CONF_PS ) { +if (ieee80211_is_ps_enabled (conf)) { rtlpriv->psc.sw_ps_enabled = true;/* sleep here is must, or we may recv the beacon and * cause mac80211 into wrong ps state, this will cause
drivers/net/wireless/rtlwifi/ps.c +2 -2
@@ -468,7 +468,7 @@ void rtl_swlps_beacon(struct ieee80211_hw *hw, void *data, unsigned int len)if (rtlpriv->psc.fwctrl_lps)return ; -if (likely (!(hw->conf.flags &IEEE80211_CONF_PS ))) +if (likely (ieee80211_is_ps_disabled (&hw->conf)))return ;/* check if this really is a beacon */
@@ -624,7 +624,7 @@ void rtl_swlps_wq_callback(void *data)struct rtl_priv *rtlpriv =rtl_priv (hw);bool ps = false; - ps = (hw->conf.flags &IEEE80211_CONF_PS ); + ps =ieee80211_is_ps_enabled (&hw->conf);/* we can sleep after ps null send ok */ if (rtlpriv->psc.state_inap) {
drivers/net/wireless/ti/wl1251/main.c +3 -4
@@ -576,7 +576,7 @@ static int wl1251_op_config(struct ieee80211_hw *hw, u32 changed)wl1251_debug (DEBUG_MAC80211 ,"mac80211 config ch %d psm %s power %d" , channel, - conf->flags &IEEE80211_CONF_PS ?"on" :"off" , +ieee80211_is_ps_enabled (conf) ?"on" :"off" , conf->power_level);mutex_lock (&wl->mutex);
@@ -594,7 +594,7 @@ static int wl1251_op_config(struct ieee80211_hw *hw, u32 changed) goto out_sleep; } -if (conf->flags &IEEE80211_CONF_PS && !wl->psm_requested) { +if (ieee80211_is_ps_enabled (conf) && !wl->psm_requested) {wl1251_debug (DEBUG_PSM ,"psm enabled" ); wl->psm_requested = true;
@@ -610,8 +610,7 @@ static int wl1251_op_config(struct ieee80211_hw *hw, u32 changed) ret =wl1251_ps_set_mode (wl,STATION_POWER_SAVE_MODE );if (ret <0 ) goto out_sleep; - }else if (!(conf->flags &IEEE80211_CONF_PS ) && - wl->psm_requested) { + }else if (ieee80211_is_ps_disabled (conf) && wl->psm_requested) {wl1251_debug (DEBUG_PSM ,"psm disabled" ); wl->psm_requested = false;
drivers/net/wireless/ti/wlcore/main.c +3 -3
@@ -2670,7 +2670,7 @@ static int wl12xx_config_vif(struct wl1271 *wl, struct wl12xx_vif *wlvif,if ((changed &IEEE80211_CONF_CHANGE_PS ) && !is_ap) { -if ((conf->flags &IEEE80211_CONF_PS ) && +if (ieee80211_is_ps_enabled (conf) &&test_bit (WLVIF_FLAG_STA_ASSOCIATED , &wlvif->flags) && !test_bit (WLVIF_FLAG_IN_PS , &wlvif->flags)) {
@@ -2693,7 +2693,7 @@ static int wl12xx_config_vif(struct wl1271 *wl, struct wl12xx_vif *wlvif,wl1271_warning ("enter %s ps failed %d" , ps_mode_str, ret); - }else if (!(conf->flags &IEEE80211_CONF_PS ) && + }else if (ieee80211_is_ps_disabled (conf) &&test_bit (WLVIF_FLAG_IN_PS , &wlvif->flags)) {wl1271_debug (DEBUG_PSM ,"auto ps disabled" );
@@ -2728,7 +2728,7 @@ static int wl1271_op_config(struct ieee80211_hw *hw, u32 changed)wl1271_debug (DEBUG_MAC80211 ,"mac80211 config ch %d psm %s power %d %s" " changed 0x%x" , channel, - conf->flags &IEEE80211_CONF_PS ?"on" :"off" , +ieee80211_is_ps_enabled (conf) ?"on" :"off" , conf->power_level, conf->flags &IEEE80211_CONF_IDLE ?"idle" :"in use" , changed);
include/net/mac80211.h +73 -23
151 lines not on this page · raw message
net/mac80211/mlme.c +15 -11
@@ -1004,7 +1004,7 @@ static void ieee80211_enable_ps(struct ieee80211_local *local, (local->hw.flags & IEEE80211_HW_REPORTS_TX_ACK_STATUS)) return; - conf->flags |=IEEE80211_CONF_PS ; +ieee80211_set_ps_state (conf,IEEE80211_CONF_PS_ENABLED );ieee80211_hw_config (local,IEEE80211_CONF_CHANGE_PS ); } }
@@ -1015,8 +1015,8 @@ static void ieee80211_change_ps(struct ieee80211_local *local)if (local->ps_sdata) {ieee80211_enable_ps (local, local->ps_sdata); - }else if (conf->flags &IEEE80211_CONF_PS ) { - conf->flags &= ~IEEE80211_CONF_PS ; + }else if (ieee80211_is_ps_enabled (conf)) { +ieee80211_set_ps_state (conf,IEEE80211_CONF_PS_DISABLED );ieee80211_hw_config (local,IEEE80211_CONF_CHANGE_PS );del_timer_sync (&local->dynamic_ps_timer);cancel_work_sync (&local->dynamic_ps_enable_work);
@@ -1150,8 +1150,9 @@ void ieee80211_dynamic_ps_disable_work(struct work_struct *work) container_of(work, struct ieee80211_local, dynamic_ps_disable_work); -if (local->hw.conf.flags &IEEE80211_CONF_PS ) { - local->hw.conf.flags &= ~IEEE80211_CONF_PS ; +if (ieee80211_is_ps_enabled (&local->hw.conf)) { +ieee80211_set_ps_state (&local->hw.conf, +IEEE80211_CONF_PS_DISABLED );ieee80211_hw_config (local,IEEE80211_CONF_CHANGE_PS ); }
@@ -1175,7 +1176,7 @@ void ieee80211_dynamic_ps_enable_work(struct work_struct *work) ifmgd = &sdata->u.mgd; -if (local->hw.conf.flags &IEEE80211_CONF_PS ) +if (ieee80211_is_ps_enabled (&local->hw.conf))return ; if (!local->disable_dynamic_ps &&
@@ -1226,7 +1227,8 @@ void ieee80211_dynamic_ps_enable_work(struct work_struct *work) (local->hw.flags & IEEE80211_HW_PS_NULLFUNC_STACK)) || (ifmgd->flags & IEEE80211_STA_NULLFUNC_ACKED)) { ifmgd->flags &= ~IEEE80211_STA_NULLFUNC_ACKED ; - local->hw.conf.flags |=IEEE80211_CONF_PS ; +ieee80211_set_ps_state (&local->hw.conf, +IEEE80211_CONF_PS_ENABLED );ieee80211_hw_config (local,IEEE80211_CONF_CHANGE_PS ); }
@@ -1534,8 +1536,9 @@ static void ieee80211_set_disassoc(struct ieee80211_sub_if_data *sdata,* to do it before sending disassoc, as otherwise the null-packet * won't be valid. */ -if (local->hw.conf.flags &IEEE80211_CONF_PS ) { - local->hw.conf.flags &= ~IEEE80211_CONF_PS ; +if (!ieee80211_is_ps_disabled (&local->hw.conf)) { +ieee80211_set_ps_state (&local->hw.conf, +IEEE80211_CONF_PS_DISABLED );ieee80211_hw_config (local,IEEE80211_CONF_CHANGE_PS ); } local->ps_sdata =NULL ;
@@ -2671,8 +2674,9 @@ static void ieee80211_rx_mgmt_beacon(struct ieee80211_sub_if_data *sdata, ifmgd->aid);if (directed_tim) {if (local->hw.conf.dynamic_ps_timeout >0 ) { -if (local->hw.conf.flags &IEEE80211_CONF_PS ) { - local->hw.conf.flags &= ~IEEE80211_CONF_PS ; +if (ieee80211_is_ps_enabled (&local->hw.conf)) { +ieee80211_set_ps_state (&local->hw.conf, +IEEE80211_CONF_PS_DISABLED );ieee80211_hw_config (local,IEEE80211_CONF_CHANGE_PS ); }
net/mac80211/offchannel.c +6 -4
@@ -40,9 +40,10 @@ static bool ieee80211_offchannel_ps_enable(struct ieee80211_sub_if_data *sdata)cancel_work_sync (&local->dynamic_ps_enable_work); -if (local->hw.conf.flags &IEEE80211_CONF_PS ) { +if (ieee80211_is_ps_enabled (&local->hw.conf)) { local->offchannel_ps_enabled = true; - local->hw.conf.flags &= ~IEEE80211_CONF_PS ; +ieee80211_set_ps_state (&local->hw.conf, +IEEE80211_CONF_PS_DISABLED );ieee80211_hw_config (local,IEEE80211_CONF_CHANGE_PS ); }
@@ -87,11 +88,12 @@ static void ieee80211_offchannel_ps_disable(struct ieee80211_sub_if_data *sdata)/* TODO: Only set hardware if CONF_PS changed? * TODO: Should we set offchannel_ps_enabled to false? */ - local->hw.conf.flags |=IEEE80211_CONF_PS ; +ieee80211_set_ps_state (&local->hw.conf, +IEEE80211_CONF_PS_ENABLED );ieee80211_hw_config (local,IEEE80211_CONF_CHANGE_PS ); }else if (local->hw.conf.dynamic_ps_timeout >0 ) {/* -* If IEEE80211_CONF_PS was not set and the dynamic_ps_timer +* If power save was not enabled and the dynamic_ps_timer * had been running before leaving the operating channel, * restart the timer now and send a nullfunc frame to inform * the AP that we are awake.
net/mac80211/tx.c +1 -1
@@ -231,7 +231,7 @@ ieee80211_tx_h_dynamic_ps(struct ieee80211_tx_data *tx) skb_get_queue_mapping(tx->skb) == IEEE80211_AC_VO)return TX_CONTINUE ; -if (local->hw.conf.flags &IEEE80211_CONF_PS ) { +if (ieee80211_is_ps_enabled (&local->hw.conf)) {ieee80211_stop_queues_by_reason (&local->hw,IEEE80211_QUEUE_STOP_REASON_PS ); ifmgd->flags &= ~IEEE80211_STA_NULLFUNC_ACKED ;
net/mac80211/util.c +1 -1
@@ -1594,7 +1594,7 @@ int ieee80211_reconfig(struct ieee80211_local *local)* explicitly send a null packet in order to make sure * it'll sync against the ap (and get out of psm). */ -if (!(local->hw.conf.flags &IEEE80211_CONF_PS )) { +if (ieee80211_is_ps_disabled (&local->hw.conf)) {list_for_each_entry (sdata, &local->interfaces, list) {if (sdata->vif.type !=NL80211_IFTYPE_STATION )continue ;
Signature · 1 line
1.7.9.5
Johannes Berg
maintainerPATCH 5/7 mac80211: Expand powersave configuration flag to be two bits
include/net/mac80211.h struct ieee80211_conf { · line 1009
Quoted 3 lines · Seth Forshee
include/net/mac80211.h +2
-0+static inline bool ieee80211_is_ps_disabled(struct ieee80211_conf *conf) +static inline bool ieee80211_is_ps_enabled(struct ieee80211_conf *conf)
Huh, is that worth the confusion? It seems !enabled should be the same as disabled, but it's not quite the same, which might be confusing.
Quoted 6 lines
include/net/mac80211.h +6
-0+/** +* ieee80211_set_ps_state - set device powersave state +* +* Sets the powersave state in the supplied device configuration to the +* specified state. +*
include/net/mac80211.h struct ieee80211_conf { · line 1030
Quoted 10 lines
include/net/mac80211.h +10
-0+* @conf: device configuration +* @state: new powersave state. Must be one of the IEEE80211_CONF_PS_* +* flags from enum ieee80211_conf_flags. +*/ +static inline void ieee80211_set_ps_state (struct ieee80211_conf *conf, +u32 state) +{ + conf->flags = (conf->flags & ~IEEE80211_CONF_PS_MASK ) | + (state &IEEE80211_CONF_PS_MASK ); +}
I don't think the driver should do this, so the inline shouldn't be here?
johannes
PATCH 5/7 mac80211: Expand powersave configuration flag to be two bits
Quoted 8 lines · Johannes Berg
include/net/mac80211.h struct ieee80211_conf { · line 1009
Quoted 3 lines · Seth Forshee
include/net/mac80211.h +2
-0+static inline bool ieee80211_is_ps_disabled(struct ieee80211_conf *conf) +static inline bool ieee80211_is_ps_enabled(struct ieee80211_conf *conf)Huh, is that worth the confusion? It seems !enabled should be the same as disabled, but it's not quite the same, which might be confusing.
In this patch there's no distinction, but after adding the off-channel powersave state there is -- disabled == !enabled && !offchannel. Actually one of the last bugs I fixed before sending these was a place where I had used disabled instead of !enabled, and the frames ended up with PM set when it shouldn't have been.
I agree though that the distinction is confusing. Maybe some better state names are needed. Perhaps awake, offchannel, and doze?
Quoted 11 lines
Quoted 11 lines
include/net/mac80211.h +11
-0+/** +* ieee80211_set_ps_state - set device powersave state +* +* Sets the powersave state in the supplied device configuration to the +* specified state. +* +* @conf: device configuration +* @state: new powersave state. Must be one of the IEEE80211_CONF_PS_* +* flags from enum ieee80211_conf_flags. +*/ +static inline void ieee80211_set_ps_state(struct ieee80211_conf *conf,
Quoted 8 lines
include/net/mac80211.h struct ieee80211_conf { · line 1030
Quoted 5 lines
include/net/mac80211.h +5
-0+ u32 state) +{ + conf->flags = (conf->flags & ~IEEE80211_CONF_PS_MASK ) | + (state &IEEE80211_CONF_PS_MASK ); +}I don't think the driver should do this, so the inline shouldn't be here?
That's true. Would moving it to ieee80211_i.h be appropriate, or is there somewhere better?
Seth
Johannes Berg
maintainerPATCH 5/7 mac80211: Expand powersave configuration flag to be two bits
Quoted 5 lines · Seth Forshee
Quoted 4 lines · Johannes Berg
Quoted 1 line · Seth Forshee
+static inline bool
ieee80211_is_ps_disabled(struct ieee80211_conf *conf)
Quoted 7 lines
Quoted 4 lines
Quoted 1 line
+static inline bool
ieee80211_is_ps_enabled(struct ieee80211_conf *conf)Huh, is that worth the confusion? It seems !enabled should be the same as disabled, but it's not quite the same, which might be confusing.
In this patch there's no distinction, but after adding the off-channel powersave state there is -- disabled == !enabled && !offchannel.
I thought it was something like that, yeah.
Quoted 6 lines
Actually one of the last bugs I fixed before sending these was a place where I had used disabled instead of !enabled, and the frames ended up with PM set when it shouldn't have been.
I agree though that the distinction is confusing. Maybe some better state names are needed. Perhaps awake, offchannel, and doze?
I think what you really want is to distinguish between "HW can go to powersave" and "PM bit should be set"? That's pretty much what your CONF_PS_ENABLED and CONF_PS_OFFCHANNEL means, respectively, but maybe putting it in different terms would make it less confusing?
Quoted 20 lines
Quoted 11 lines
Quoted 11 lines
include/net/mac80211.h +11
-0+/** +* ieee80211_set_ps_state - set device powersave state +* +* Sets the powersave state in the supplied device configuration to the +* specified state. +* +* @conf: device configuration +* @state: new powersave state. Must be one of the IEEE80211_CONF_PS_* +* flags from enum ieee80211_conf_flags. +*/ +static inline void ieee80211_set_ps_state(struct ieee80211_conf *conf,Quoted 8 lines
include/net/mac80211.h struct ieee80211_conf { · line 1030
Quoted 5 lines
include/net/mac80211.h +5
-0+ u32 state) +{ + conf->flags = (conf->flags & ~IEEE80211_CONF_PS_MASK ) | + (state &IEEE80211_CONF_PS_MASK ); +}I don't think the driver should do this, so the inline shouldn't be here?
Quoted 2 lines
That's true. Would moving it to
ieee80211_i.hbe appropriate, or is there somewhere better?
ieee80211_i.h is good
johannes
PATCH 5/7 mac80211: Expand powersave configuration flag to be two bits
Quoted 20 lines · Johannes Berg
Quoted 5 lines · Seth Forshee
Quoted 4 lines · Johannes Berg
Quoted 1 line · Seth Forshee
+static inline bool
ieee80211_is_ps_disabled(struct ieee80211_conf *conf)Quoted 7 lines
Quoted 4 lines
Quoted 1 line
+static inline bool
ieee80211_is_ps_enabled(struct ieee80211_conf *conf)Huh, is that worth the confusion? It seems !enabled should be the same as disabled, but it's not quite the same, which might be confusing.
In this patch there's no distinction, but after adding the off-channel powersave state there is -- disabled == !enabled && !offchannel.
I thought it was something like that, yeah.
Quoted 4 lines
Actually one of the last bugs I fixed before sending these was a place where I had used disabled instead of !enabled, and the frames ended up with PM set when it shouldn't have been.
Quoted 6 lines
Quoted 2 lines
I agree though that the distinction is confusing. Maybe some better state names are needed. Perhaps awake, offchannel, and doze?
I think what you really want is to distinguish between "HW can go to powersave" and "PM bit should be set"? That's pretty much what your
CONF_PS_ENABLEDandCONF_PS_OFFCHANNELmeans, respectively, but maybe
Correct, with the understanding that "HW can go to powersave" also implies "PM bit should be set."
Another approach would be to keep the CONF_PS flag the same and add a CONF_PM flag or similar. I didn't go with this approach because CONF_PS && !CONF_PM really doesn't make any sense, which really doesn't help with reducing confusion. The advantage is that it separates setting PM from PS for those driver that don't support PS but need to configure the hardware to set PM for off-channel.
Quoted 1 line
putting it in different terms would make it less confusing?
Yes. I think the reason enabled/disabled is confusing because it would usually imply a binary state.
Seth
Johannes Berg
maintainerPATCH 5/7 mac80211: Expand powersave configuration flag to be two bits
Quoted 14 lines · Seth Forshee
Quoted 10 lines
Quoted 6 lines
Actually one of the last bugs I fixed before sending these was a place where I had used disabled instead of !enabled, and the frames ended up with PM set when it shouldn't have been.
I agree though that the distinction is confusing. Maybe some better state names are needed. Perhaps awake, offchannel, and doze?
I think what you really want is to distinguish between "HW can go to powersave" and "PM bit should be set"? That's pretty much what your
CONF_PS_ENABLEDandCONF_PS_OFFCHANNELmeans, respectively, but maybeCorrect, with the understanding that "HW can go to powersave" also implies "PM bit should be set."
Quoted 6 lines
Another approach would be to keep the
CONF_PSflag the same and add aCONF_PMflag or similar. I didn't go with this approach becauseCONF_PS&& !CONF_PM really doesn't make any sense, which really doesn't help with reducing confusion. The advantage is that it separates setting PM from PS for those driver that don't support PS but need to configure the hardware to set PM for off-channel.
Good point, that'd work too. PS && !PM would never be used, I guess? It'd also have the advantage of not having to touch all the drivers?
It doesn't really matter all that much to me though, I just think what you have right now is (too) confusing.
johannes
PATCH 5/7 mac80211: Expand powersave configuration flag to be two bits
Quoted 23 lines · Johannes Berg
Quoted 14 lines · Seth Forshee
Quoted 10 lines
Quoted 6 lines
Actually one of the last bugs I fixed before sending these was a place where I had used disabled instead of !enabled, and the frames ended up with PM set when it shouldn't have been.
I agree though that the distinction is confusing. Maybe some better state names are needed. Perhaps awake, offchannel, and doze?
I think what you really want is to distinguish between "HW can go to powersave" and "PM bit should be set"? That's pretty much what your
CONF_PS_ENABLEDandCONF_PS_OFFCHANNELmeans, respectively, but maybeCorrect, with the understanding that "HW can go to powersave" also implies "PM bit should be set."
Quoted 6 lines
Another approach would be to keep the
CONF_PSflag the same and add aCONF_PMflag or similar. I didn't go with this approach becauseCONF_PS&& !CONF_PM really doesn't make any sense, which really doesn't help with reducing confusion. The advantage is that it separates setting PM from PS for those driver that don't support PS but need to configure the hardware to set PM for off-channel.
Quoted 5 lines
Good point, that'd work too. PS && !PM would never be used, I guess? It'd also have the advantage of not having to touch all the drivers?
It doesn't really matter all that much to me though, I just think what you have right now is (too) confusing.
Hi Johannes,
I've been thinking about and playing around with these ideas. I've implemented the CONF_PM idea, and it does end up with fewer changes, but I just don't think separating powersave from setting PM makes much sense. In the end it just seems like a kludge to fix a problem with Broadcom chips, and if I want a kludge I can do it entirely within the driver.
So what I'm planning to do know is implement the awake/doze/offchannel powersave modes like I had mentioned, but taking things a bit further. I'd change IEEE80211_HW_SUPPORTS_PS to SUPPORTS_PS_DOZE, indicating support for a low-power powersave state rather than support for powersave in general. All hardware will be reconfigured for awake<->offchannel transitions (though most drivers can simply ignore these transitions), but hardware will only be put in the doze state if it indicates PS_DOZE support.
This will make it compulsory for all drivers to indicate whether or not they require IEEE802111_HW_PS_NULLFUNC_STACK. I'll simply set this flag for any drivers not currently supporting powersave.
In practice the changes shouldn't end up much different than what I have in these patches, but I think it's conceptually cleaner (and less confusing!). Does this sound reasonable to you?
Thanks,
Seth
Johannes Berg
maintainerPATCH 5/7 mac80211: Expand powersave configuration flag to be two bits
Hi Seth,
Quoted 7 lines
I've been thinking about and playing around with these ideas. I've implemented the
CONF_PMidea, and it does end up with fewer changes, but I just don't think separating powersave from setting PM makes much sense. In the end it just seems like a kludge to fix a problem with Broadcom chips, and if I want a kludge I can do it entirely within the driver.
Quoted 5 lines
So what I'm planning to do know is implement the awake/doze/offchannel powersave modes like I had mentioned, but taking things a bit further. I'd change
IEEE80211_HW_SUPPORTS_PStoSUPPORTS_PS_DOZE, indicating support for a low-power powersave state rather than support for powersave in general.
Hm, I'm not sure I fully understand this. What does PS_DOZE mean then? I think in the spec doze is a specific term for mesh?
Does it just mean "I support actually turning off the radio"? But then what's the difference to supporting powersave? Can we maybe just disregard wl1251, which has the stupidest powersave implementation on the planet, and solve the "normal" problems first? :)
Quoted 8 lines
All hardware will be reconfigured for
awake<->offchannel transitions (though most drivers can simply ignore these transitions), but hardware will only be put in the doze state if it indicatesPS_DOZEsupport.This will make it compulsory for all drivers to indicate whether or not they require
IEEE802111_HW_PS_NULLFUNC_STACK. I'll simply set this flag for any drivers not currently supporting powersave.
That seems odd, why would they advertise they want null-data packets for powersave, when they don't have powersave? Just for the interaction with scan/offchannel?
Quoted 3 lines
In practice the changes shouldn't end up much different than what I have in these patches, but I think it's conceptually cleaner (and less confusing!). Does this sound reasonable to you?
Not really sure I understand it enough to comment :)
johannes
PATCH 5/7 mac80211: Expand powersave configuration flag to be two bits
Quoted 15 lines · Johannes Berg
Hi Seth,
Quoted 7 lines
I've been thinking about and playing around with these ideas. I've implemented the
CONF_PMidea, and it does end up with fewer changes, but I just don't think separating powersave from setting PM makes much sense. In the end it just seems like a kludge to fix a problem with Broadcom chips, and if I want a kludge I can do it entirely within the driver.Quoted 5 lines
So what I'm planning to do know is implement the awake/doze/offchannel powersave modes like I had mentioned, but taking things a bit further. I'd change
IEEE80211_HW_SUPPORTS_PStoSUPPORTS_PS_DOZE, indicating support for a low-power powersave state rather than support for powersave in general.
Quoted 7 lines
Hm, I'm not sure I fully understand this. What does
PS_DOZEmean then? I think in the spec doze is a specific term for mesh?Does it just mean "I support actually turning off the radio"? But then what's the difference to supporting powersave? Can we maybe just disregard wl1251, which has the stupidest powersave implementation on the planet, and solve the "normal" problems first? :)
PS_DOZE means it actually supports putting the hardware into a low-power state for powersave. I did take the term from the spec (802.11-2012). It is usually used with regard to mesh, but it also appears wrt infrastructure BSS (see especially 10.2.1.2 which defines both awake and doze in the context of infrastructure networks).
I'm open to other terms, doze just seems to be consistent with the spec.
I haven't considered wl1251 specifically, only enough to update it so that it continues to build.
Quoted 5 lines
Quoted 5 lines
All hardware will be reconfigured for
awake<->offchannel transitions (though most drivers can simply ignore these transitions), but hardware will only be put in the doze state if it indicatesPS_DOZEsupport.
Quoted 7 lines
Quoted 3 lines
This will make it compulsory for all drivers to indicate whether or not they require
IEEE802111_HW_PS_NULLFUNC_STACK. I'll simply set this flag for any drivers not currently supporting powersave.That seems odd, why would they advertise they want null-data packets for powersave, when they don't have powersave? Just for the interaction with scan/offchannel?
Yes.
Maybe what's confusing here is that I'm making a differentiation between "powersave" as a mac-level feature and "powersave" as a low-power hardware state. Which is why I'm trying to change the mac80211 terminology around so that "doze" now means the low-power state and "powersave" refers only to the state in which the AP is bufferring frames for us.
So using these definition powersave is already a mandatory feature for any hardware which uses software scanning. All I'm really doing is making this explicit, and drivers would now opt in to being placed into the doze state rather than opting into powersave in general. And of course drivers are now told about transitions to the non-doze powersave state (what I'm calling offchannel), so that drivers which need to do hardware configuration for this state can do so.
Quoted 5 lines
Quoted 3 lines
In practice the changes shouldn't end up much different than what I have in these patches, but I think it's conceptually cleaner (and less confusing!). Does this sound reasonable to you?
Not really sure I understand it enough to comment :)
I've got working patches now, so maybe those will make everything clear. I need to do a little more testing and give them a quick review, then I'll send them out.
Seth
Johannes Berg
maintainerPATCH 5/7 mac80211: Expand powersave configuration flag to be two bits
Quoted 5 lines · Seth Forshee
Quoted 4 lines
Does it just mean "I support actually turning off the radio"? But then what's the difference to supporting powersave? Can we maybe just disregard wl1251, which has the stupidest powersave implementation on the planet, and solve the "normal" problems first? :)
Quoted 7 lines
PS_DOZEmeans it actually supports putting the hardware into a low-power state for powersave. I did take the term from the spec (802.11-2012). It is usually used with regard to mesh, but it also appears wrt infrastructure BSS (see especially 10.2.1.2 which defines both awake and doze in the context of infrastructure networks).I'm open to other terms, doze just seems to be consistent with the spec.
Thanks for the reference. I had actually thought the term seemed familiar but then only found mesh references ... need to brush up my spec knowledge I guess :)
Quoted 2 lines
I haven't considered wl1251 specifically, only enough to update it so that it continues to build.
Ah, wl1251 is the weird thing that wants to be told when to wake up/go to sleep, and then sends nulldata packets itself ... so when we send nulldata packets *again* for off-channel, because it also uses software scanning. Like I said before -- "stupidest powersave implementation on the planet".
Quoted 6 lines
Maybe what's confusing here is that I'm making a differentiation between "powersave" as a mac-level feature and "powersave" as a low-power hardware state. Which is why I'm trying to change the mac80211 terminology around so that "doze" now means the low-power state and "powersave" refers only to the state in which the AP is bufferring frames for us.
Right, ok. Maybe another term would be worthwhile? I mean, *all* hardware has to support "MAC-level powersave" since you always want background scanning. The question is how it's implemented ... I'm off for dinner in a minute but I'll think about it a bit.
Quoted 2 lines
So using these definition powersave is already a mandatory feature for any hardware which uses software scanning.
Even offloaded scanning, it's just not visible to mac80211 then.
Quoted 6 lines
All I'm really doing is
making this explicit, and drivers would now opt in to being placed into the doze state rather than opting into powersave in general. And of course drivers are now told about transitions to the non-doze powersave state (what I'm calling offchannel), so that drivers which need to do hardware configuration for this state can do so.
Ok, so maybe just calling it offchannel makes more sense, although that really interacts only this way with powersave in managed mode, in P2P GO mode .... oh well.
Quoted 9 lines
Quoted 5 lines
Quoted 3 lines
In practice the changes shouldn't end up much different than what I have in these patches, but I think it's conceptually cleaner (and less confusing!). Does this sound reasonable to you?
Not really sure I understand it enough to comment :)
I've got working patches now, so maybe those will make everything clear. I need to do a little more testing and give them a quick review, then I'll send them out.
Ok.
johannes
PATCH 5/7 mac80211: Expand powersave configuration flag to be two bits
Quoted 8 lines · Johannes Berg
Quoted 2 lines
I haven't considered wl1251 specifically, only enough to update it so that it continues to build.
Ah, wl1251 is the weird thing that wants to be told when to wake up/go to sleep, and then sends nulldata packets itself ... so when we send nulldata packets *again* for off-channel, because it also uses software scanning. Like I said before -- "stupidest powersave implementation on the planet".
Ah, I had wondered what hardware that code was there for :-)
So I think the approach I'm advocating could possibly make things better for wl1251. If it's told about going to off-channel powersave rather than being told it's leaving powersave, maybe it can do the right thing with the nullfunc frames without mac80211 generating the extra nullfunc. I don't know that for sure though because I don't understand how the hardware works.
Quoted 4 lines
Quoted 2 lines
So using these definition powersave is already a mandatory feature for any hardware which uses software scanning.
Even offloaded scanning, it's just not visible to mac80211 then.
Okay, sure. I'm just thinking of things from the mac80211 perspective.
Quoted 10 lines
Quoted 6 lines
All I'm really doing is
making this explicit, and drivers would now opt in to being placed into the doze state rather than opting into powersave in general. And of course drivers are now told about transitions to the non-doze powersave state (what I'm calling offchannel), so that drivers which need to do hardware configuration for this state can do so.Ok, so maybe just calling it offchannel makes more sense, although that really interacts only this way with powersave in managed mode, in P2P GO mode .... oh well.
I admit I've only been considering managed mode and know next to nothing about P2P GO mode. If I'm creating problems for any other modes I'd definitely like to know about it so I can take it into consideration.
Seth
Johannes Berg
maintainerPATCH 5/7 mac80211: Expand powersave configuration flag to be two bits
Quoted 7 lines · Seth Forshee
Quoted 5 lines
Ah, wl1251 is the weird thing that wants to be told when to wake up/go to sleep, and then sends nulldata packets itself ... so when we send nulldata packets *again* for off-channel, because it also uses software scanning. Like I said before -- "stupidest powersave implementation on the planet".
Ah, I had wondered what hardware that code was there for :-)
I was actually wrong -- wl1251 has HW scan so it doesn't matter. However, iwlegacy does this...
Quoted 6 lines
So I think the approach I'm advocating could possibly make things better for wl1251. If it's told about going to off-channel powersave rather than being told it's leaving powersave, maybe it can do the right thing with the nullfunc frames without mac80211 generating the extra nullfunc. I don't know that for sure though because I don't understand how the hardware works.
Probably not. For iwlegacy anyway it won't help.
Quoted 6 lines
Quoted 4 lines
Quoted 2 lines
So using these definition powersave is already a mandatory feature for any hardware which uses software scanning.
Even offloaded scanning, it's just not visible to mac80211 then.
Okay, sure. I'm just thinking of things from the mac80211 perspective.
Right, sure.
Quoted 7 lines
Quoted 7 lines
Quoted 6 lines
All I'm really doing is
making this explicit, and drivers would now opt in to being placed into the doze state rather than opting into powersave in general. And of course drivers are now told about transitions to the non-doze powersave state (what I'm calling offchannel), so that drivers which need to do hardware configuration for this state can do so.
Quoted 7 lines
Quoted 3 lines
Ok, so maybe just calling it offchannel makes more sense, although that really interacts only this way with powersave in managed mode, in P2P GO mode .... oh well.
I admit I've only been considering managed mode and know next to nothing about P2P GO mode. If I'm creating problems for any other modes I'd definitely like to know about it so I can take it into consideration.
No no, I'm just saying there "powersave" could have even more meanings ...
johannes
PATCH 6/7 mac80211: Add off-channel powersave state
In reply to Seth ForsheeSome driviers need to perform hardware configuration or other activities in order to properly support off-channel powersave. Add an off-channel powersave state to inform drivers that MAC-level support for powersave is required while keeping the hardware powered on for off-channel operation, and update the documentation with informtion about this state.
Also modify the conditions for sending nullfunc frames from mac80211 for off-channel operation. Hardware which supports sending the frames will now have accurate information about the powersave state and should no longer require that mac80211 generate the frames. Continue sending them for all drivers which do not support powersave, however.
--- include/net/mac80211.h | 29 ++++++++++++++++++++++++ net/mac80211/offchannel.c | 55 +++++++++++++++++++++------------------------ 2 files changed, 55 insertions(+), 29 deletions(-)
include/net/mac80211.h +29 -0
@@ -871,6 +871,11 @@ struct ieee80211_rx_status {* up for beacons, is able to transmit frames and receive the possible * acknowledgement frames. Not to be confused with hardware specific * wakeup/sleep states; the driver is responsible for that. +* @IEEE80211_CONF_PS_OFFCHANNEL: Power save is enabled at the AP, but the +* hardware remains powered on for transmitting and receiving frames. This +* is used to request that the AP buffer frames while the STA leaves the +* operating channel. The hardware must be configured to set PM in +* frame control, if applicable. * @IEEE80211_CONF_IDLE: The device is running, but idle; if the flag is set * the driver should be prepared to handle configuration requests but * may turn the device off as much as possible. Typically, this flag will
@@ -884,6 +889,7 @@ enum ieee80211_conf_flags {IEEE80211_CONF_PS_MASK = (3 <<1 ),IEEE80211_CONF_PS_DISABLED = (0 <<1 ),IEEE80211_CONF_PS_ENABLED = (1 <<1 ), +IEEE80211_CONF_PS_OFFCHANNEL = (2 <<1 ),IEEE80211_CONF_IDLE = (1 <<3 ), IEEE80211_CONF_OFFCHANNEL = (1<<4), };
@@ -1013,6 +1019,20 @@ static inline bool ieee80211_is_ps_enabled(struct ieee80211_conf *conf) }/** +* ieee80211_is_ps_offchannel - check if off-channel powersave is enabled +* +* Returns true if off-channel powersave is enabled in the supplied +* configuration. +* +* @conf: device configuration +*/ +static inline bool ieee80211_is_ps_offchannel (struct ieee80211_conf *conf) +{ +return (conf->flags &IEEE80211_CONF_PS_MASK ) == +IEEE80211_CONF_PS_OFFCHANNEL ; +} + +/** * ieee80211_set_ps_state - set device powersave state * * Sets the powersave state in the supplied device configuration to the
@@ -1712,6 +1732,15 @@ void ieee80211_free_txskb(struct ieee80211_hw *hw, struct sk_buff *skb);* waking up for multicast traffic; if it cannot the driver must handle that * as best as it can, mac80211 is too slow to do that. * +* If ieee80211_is_ps_offchannel() returns true, off-channel powersave is +* enabled. During off-channel powersave the AP is instructed to buffer frames +* as if the hardware is in powersave, but the hardware must remain fully +* operational to support off-channel operation. If necessary the hardware +* must be configured to set PM in frame control for frames destined for the +* AP. Drivers will be told about entering and leaving off-channel powersave +* even if %IEEE80211_HW_SUPPORTS_PS is cleared; drivers which do not need to +* do any configuration can safely ignore it. +* * Dynamic powersave is an extension to normal powersave in which the * hardware stays awake for a user-specified period of time after sending a * frame so that reply frames need not be buffered and therefore delayed to
net/mac80211/offchannel.c +26 -29
82 lines not on this page · raw message
Signature · 1 line
1.7.9.5
PATCH 7/7 brcmsmac: Add support for off-channel powersave
In reply to Seth ForsheeBroadcom hardware sets the value of the PM flag in frame control based on the value of MCTL_HPS, which seems to include actively clearing PM if MCTL_HPS is not set. brcmsmac needs to suppor the off-channel powersave state in order to enable powersave at the AP at all.
Add limited support for powersave to brcmsmac to prevent frame loss during background scans. Full powersave supoprt remains unimplemented, but switching between the off-channel and disabled states is possible.
--- .../net/wireless/brcm80211/brcmsmac/mac80211_if.c | 15 +++++++++++---- drivers/net/wireless/brcm80211/brcmsmac/main.c | 9 +++++++++ drivers/net/wireless/brcm80211/brcmsmac/pub.h | 1 + 3 files changed, 21 insertions(+), 4 deletions(-)
drivers/net/wireless/brcm80211/brcmsmac/mac80211_if.c +11 -4
@@ -393,10 +393,17 @@ static int brcms_ops_config(struct ieee80211_hw *hw, u32 changed)brcms_dbg_info (core,"%s: change monitor mode: %s\n" , __func__, conf->flags &IEEE80211_CONF_MONITOR ?"true" :"false" ); -if (changed &IEEE80211_CONF_CHANGE_PS ) -brcms_err (core,"%s: change power-save mode: %s (implement)\n" , - __func__,ieee80211_is_ps_enabled (conf) ? -"true" :"false" ); +if (changed &IEEE80211_CONF_CHANGE_PS ) { +/* +* brcmsmac doesn't support powersave, but it does support +* setting the PM bit in frame control for off-channel PS +*/ +if (ieee80211_is_ps_enabled (conf)) +brcms_err (core,"%s: cannot enable power-save mode (implement)\n" , + __func__); +else +brcms_c_set_ps (wl->wlc); + }if (changed &IEEE80211_CONF_CHANGE_POWER ) { err =brcms_c_set_tx_power (wl->wlc, conf->power_level);
drivers/net/wireless/brcm80211/brcmsmac/main.c +9 -0
@@ -3054,6 +3054,10 @@ static bool brcms_c_ps_allowed(struct brcms_c_info *wlc) {struct brcms_bss_cfg *cfg = wlc->bsscfg; +/* allow PS when off-channel PS is enabled */ +if (ieee80211_is_ps_offchannel (&wlc->pub->ieee_hw->conf)) +return true; +/* disallow PS when one of the following global conditions meets */ if (!wlc->pub->associated)return false;
@@ -7545,6 +7549,11 @@ void brcms_c_set_beacon_listen_interval(struct brcms_c_info *wlc, u8 interval)brcms_c_bcn_li_upd (wlc); } +void brcms_c_set_ps (struct brcms_c_info *wlc) +{ +brcms_c_set_ps_ctrl (wlc); +} +int brcms_c_set_tx_power (struct brcms_c_info *wlc,int txpwr) {uint qdbm;
drivers/net/wireless/brcm80211/brcmsmac/pub.h +1 -0
@@ -328,6 +328,7 @@ extern void brcms_c_set_shortslot_override(struct brcms_c_info *wlc, s8 sslot_override);extern void brcms_c_set_beacon_listen_interval (struct brcms_c_info *wlc,u8 interval); +extern void brcms_c_set_ps (struct brcms_c_info *wlc);extern int brcms_c_set_tx_power (struct brcms_c_info *wlc,int txpwr);extern int brcms_c_get_tx_power (struct brcms_c_info *wlc);extern bool brcms_c_check_radio_disabled (struct brcms_c_info *wlc);
Signature · 1 line
1.7.9.5
Re: [PATCH 7/7] brcmsmac: Add support for off-channel powersave
Hi Seth,
Quoted 21 lines · Seth Forshee
Broadcom hardware sets the value of the PM flag in frame control based on the value of
MCTL_HPS, which seems to include actively clearing PM ifMCTL_HPSis not set. brcmsmac needs to suppor the off-channel powersave state in order to enable powersave at the AP at all.Add limited support for powersave to brcmsmac to prevent frame loss during background scans. Full powersave supoprt remains unimplemented, but switching between the off-channel and disabled states is possible.
Signed-off-by Seth Forshee <[email protected]>--- .../net/wireless/brcm80211/brcmsmac/mac80211_if.c | 15 +++++++++++---- drivers/net/wireless/brcm80211/brcmsmac/main.c | 9 +++++++++ drivers/net/wireless/brcm80211/brcmsmac/pub.h | 1 + 3 files changed, 21 insertions(+), 4 deletions(-)drivers/net/wireless/brcm80211/brcmsmac/main.c +0
-0
drivers/net/wireless/brcm80211/brcmsmac/main.c void brcms_c_set_beacon_listen_interval(struct brcms_c_info *wlc, u8 interval) · line 7556
Quoted 8 lines
drivers/net/wireless/brcm80211/brcmsmac/main.c +5
-0@@ -7545,6 +7549,11 @@ void brcms_c_set_beacon_listen_interval(struct brcms_c_info *wlc, u8 interval)brcms_c_bcn_li_upd (wlc); } +void brcms_c_set_ps (struct brcms_c_info *wlc) +{ +brcms_c_set_ps_ctrl (wlc); +} +
Why not just use brcms_c_set_ps_ctrl() directly?
Thanks,
Signature · 5 lines
Julian Calaby Email: [email protected] Profile: http://www.google.com/profiles/julian.calaby/ .Plan: http://sites.google.com/site/juliancalaby/
Re: [PATCH 7/7] brcmsmac: Add support for off-channel powersave
Quoted 28 lines · Julian Calaby
Hi Seth,
Quoted 24 lines · Seth Forshee
Broadcom hardware sets the value of the PM flag in frame control based on the value of
MCTL_HPS, which seems to include actively clearing PM ifMCTL_HPSis not set. brcmsmac needs to suppor the off-channel powersave state in order to enable powersave at the AP at all.Add limited support for powersave to brcmsmac to prevent frame loss during background scans. Full powersave supoprt remains unimplemented, but switching between the off-channel and disabled states is possible.
Signed-off-by Seth Forshee <[email protected]>--- .../net/wireless/brcm80211/brcmsmac/mac80211_if.c | 15 +++++++++++---- drivers/net/wireless/brcm80211/brcmsmac/main.c | 9 +++++++++ drivers/net/wireless/brcm80211/brcmsmac/pub.h | 1 + 3 files changed, 21 insertions(+), 4 deletions(-)drivers/net/wireless/brcm80211/brcmsmac/main.c +0
-0@@ -7545,6 +7549,11 @@ void brcms_c_set_beacon_listen_interval(struct brcms_c_info *wlc, u8 interval)brcms_c_bcn_li_upd (wlc); }
Quoted 7 lines
drivers/net/wireless/brcm80211/brcmsmac/main.c void brcms_c_set_beacon_listen_interval(struct brcms_c_info *wlc, u8 interval) · line 7556
Quoted 5 lines
drivers/net/wireless/brcm80211/brcmsmac/main.c +5
-0+void brcms_c_set_ps (struct brcms_c_info *wlc) +{ +brcms_c_set_ps_ctrl (wlc); +} +Why not just use
brcms_c_set_ps_ctrl()directly?
It's a remnant from an earlier version of the patches. Looks like I just didn't take things to the logical conclusion when I refactored the code.
Seth
Quoted 42 lines · Seth Forshee
This patch series fixes a number of problems observed during software scanning, as described in [1].
The first four patches implement improved queue handling during off-channel operation and add some needed flushing of the hardware queues, as suggested by Johannes in [2]. This includes adding a new queue stop reason of off-channel operation, a transmit path for frames which do need to be transmitted when off-channel, and some improvements in error handling.
The last three fix a problem specific to brcmsmac (and likely b43, though I don't have hardware for testing b43). Broadcom hardware actively clears the PM bit in frame control whenever the hardware is not configured correctly, and since wireless drivers have no knowledge of off-channel powersave such configuration cannot be done. The patches expand the driver powersave configuration with an off-channel state and update brcmsmac to make use of it.
Johannes: I have a couple of comments/questions for you related to these patches.
First, in the patches I've added an
offchan_tx_okargument to the tx operations, but this seems a little awkward to me since it has to be propogated down through a fairly deep call stack. The alternative idea that occurred to me is to use a tx control flag, but that seems to be pretty crowded. Any thoughts?Second, I attempted to test these patches with iwlwifi (Centrino Advanced-N 6235) to verify that I didn't break anything for drivers with hw scanning. My standard test for this involves running iperf while triggering nearly continuous scans, but I'm seeing lots of problems running a tcp iperf test even with unpatched 3.8-rc4. iperf with udp does fine in either direction. I haven't had time to do any kind of debugging yet, but I thought you'd want to know.
Thanks,
Seth[1] http://marc.info/?l=linux-wireless&m=135766865110986&w=2
[2] http://marc.info/?l=linux-wireless&m=135838252227053&w=2
Quoted 8 lines
Seth Forshee (7):
mac80211: Return a status for tx operations
mac80211: Fix tx queue handling during scans
mac80211: Improve error handling for off-channel operation
mac80211: Add flushes before going off-channel
mac80211: Expand powersave configuration flag to be two bits
mac80211: Add off-channel powersave state
brcmsmac: Add support for off-channel powersave
Johannes,
Please take the brcsmac patch through your tree if/when you take the others.
Thanks!
John
Signature · 2 lines
John W. Linville Someday the world will need a hero, and you [email protected] might be all we have. Be ready.
Quoted 2 lines · Seth Forshee
This patch series fixes a number of problems observed during software scanning, as described in [1].
Hi Seth,
Not sure why but it seems I only received patch 5, 7, and the cover letter. Could you resend them to me so I can give it a spin over here?
Gr. AvS
Quoted 8 lines · Arend van Spriel
Quoted 2 lines · Seth Forshee
This patch series fixes a number of problems observed during software scanning, as described in [1].
Hi Seth,
Not sure why but it seems I only received patch 5, 7, and the cover letter. Could you resend them to me so I can give it a spin over here?
I tried to automate who was on Cc by using get_maintainer.pl, so probably those are the only ones it said you needed to receive ;-)
All the patches should be on the linux-wireless list, but I'll send you a private mail with all the patches as well.
Seth
Johannes Berg
maintainerQuoted 8 lines · Seth Forshee
Johannes: I have a couple of comments/questions for you related to these patches.
First, in the patches I've added an
offchan_tx_okargument to the tx operations, but this seems a little awkward to me since it has to be propogated down through a fairly deep call stack. The alternative idea that occurred to me is to use a tx control flag, but that seems to be pretty crowded. Any thoughts?
Maybe you can bypass by using a flag in struct ieee80211_tx_data, so only the first few functions in the call chain need the argument? Otherwise, I guess adding a flag should be OK. I know it's crowded, but if we really run out I guess we could move all the internal flags etc. wholesale ...
Quoted 7 lines
Second, I attempted to test these patches with iwlwifi (Centrino Advanced-N 6235) to verify that I didn't break anything for drivers with hw scanning. My standard test for this involves running iperf while triggering nearly continuous scans, but I'm seeing lots of problems running a tcp iperf test even with unpatched 3.8-rc4. iperf with udp does fine in either direction. I haven't had time to do any kind of debugging yet, but I thought you'd want to know.
Hm, ok, thanks.
johannes
Johannes Berg
maintainerQuoted 11 lines · Johannes Berg
Quoted 8 lines · Seth Forshee
Johannes: I have a couple of comments/questions for you related to these patches.
First, in the patches I've added an
offchan_tx_okargument to the tx operations, but this seems a little awkward to me since it has to be propogated down through a fairly deep call stack. The alternative idea that occurred to me is to use a tx control flag, but that seems to be pretty crowded. Any thoughts?
Quoted 5 lines
Maybe you can bypass by using a flag in struct
ieee80211_tx_data, so only the first few functions in the call chain need the argument? Otherwise, I guess adding a flag should be OK. I know it's crowded, but if we really run out I guess we could move all the internal flags etc. wholesale ...
Ok no that was wrong ... we can't do that because many flags need to survive queueing.
johannes
Quoted 12 lines · Johannes Berg
Quoted 11 lines · Johannes Berg
Quoted 8 lines · Seth Forshee
Johannes: I have a couple of comments/questions for you related to these patches.
First, in the patches I've added an
offchan_tx_okargument to the tx operations, but this seems a little awkward to me since it has to be propogated down through a fairly deep call stack. The alternative idea that occurred to me is to use a tx control flag, but that seems to be pretty crowded. Any thoughts?
Quoted 8 lines
Quoted 5 lines
Maybe you can bypass by using a flag in struct
ieee80211_tx_data, so only the first few functions in the call chain need the argument? Otherwise, I guess adding a flag should be OK. I know it's crowded, but if we really run out I guess we could move all the internal flags etc. wholesale ...Ok no that was wrong ... we can't do that because many flags need to survive queueing.
An ieee80211_tx_data flag would work for this case, though it doesn't quite have the effect I was hoping for since it really only gets rid of the argument internal to tx.c. I avoided most of the pain by leaving ieee80211_tx_skb() unmodified, but it all still seems a bit ugly.
If you're okay with the way I've got things now I'll just stick with it. I hate to gobble up valuable real estate in the tx control flags just to satisfy my sense of aesthetics ;-)
Seth
Quoted 9 lines · Johannes Berg
Quoted 7 lines
Second, I attempted to test these patches with iwlwifi (Centrino Advanced-N 6235) to verify that I didn't break anything for drivers with hw scanning. My standard test for this involves running iperf while triggering nearly continuous scans, but I'm seeing lots of problems running a tcp iperf test even with unpatched 3.8-rc4. iperf with udp does fine in either direction. I haven't had time to do any kind of debugging yet, but I thought you'd want to know.
Hm, ok, thanks.
I just tried moving this card to a different slot in the same machine, and it's working fine. Hardware problem I guess. Sorry for the false alarm.
Seth