linux/net/mac80211/agg-tx.c

814 lines
23 KiB
C
Raw Normal View History

/*
* HT handling
*
* Copyright 2003, Jouni Malinen <jkmaline@cc.hut.fi>
* Copyright 2002-2005, Instant802 Networks, Inc.
* Copyright 2005-2006, Devicescape Software, Inc.
* Copyright 2006-2007 Jiri Benc <jbenc@suse.cz>
* Copyright 2007, Michael Wu <flamingice@sourmilk.net>
* Copyright 2007-2010, Intel Corporation
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License version 2 as
* published by the Free Software Foundation.
*/
#include <linux/ieee80211.h>
include cleanup: Update gfp.h and slab.h includes to prepare for breaking implicit slab.h inclusion from percpu.h percpu.h is included by sched.h and module.h and thus ends up being included when building most .c files. percpu.h includes slab.h which in turn includes gfp.h making everything defined by the two files universally available and complicating inclusion dependencies. percpu.h -> slab.h dependency is about to be removed. Prepare for this change by updating users of gfp and slab facilities include those headers directly instead of assuming availability. As this conversion needs to touch large number of source files, the following script is used as the basis of conversion. http://userweb.kernel.org/~tj/misc/slabh-sweep.py The script does the followings. * Scan files for gfp and slab usages and update includes such that only the necessary includes are there. ie. if only gfp is used, gfp.h, if slab is used, slab.h. * When the script inserts a new include, it looks at the include blocks and try to put the new include such that its order conforms to its surrounding. It's put in the include block which contains core kernel includes, in the same order that the rest are ordered - alphabetical, Christmas tree, rev-Xmas-tree or at the end if there doesn't seem to be any matching order. * If the script can't find a place to put a new include (mostly because the file doesn't have fitting include block), it prints out an error message indicating which .h file needs to be added to the file. The conversion was done in the following steps. 1. The initial automatic conversion of all .c files updated slightly over 4000 files, deleting around 700 includes and adding ~480 gfp.h and ~3000 slab.h inclusions. The script emitted errors for ~400 files. 2. Each error was manually checked. Some didn't need the inclusion, some needed manual addition while adding it to implementation .h or embedding .c file was more appropriate for others. This step added inclusions to around 150 files. 3. The script was run again and the output was compared to the edits from #2 to make sure no file was left behind. 4. Several build tests were done and a couple of problems were fixed. e.g. lib/decompress_*.c used malloc/free() wrappers around slab APIs requiring slab.h to be added manually. 5. The script was run on all .h files but without automatically editing them as sprinkling gfp.h and slab.h inclusions around .h files could easily lead to inclusion dependency hell. Most gfp.h inclusion directives were ignored as stuff from gfp.h was usually wildly available and often used in preprocessor macros. Each slab.h inclusion directive was examined and added manually as necessary. 6. percpu.h was updated not to include slab.h. 7. Build test were done on the following configurations and failures were fixed. CONFIG_GCOV_KERNEL was turned off for all tests (as my distributed build env didn't work with gcov compiles) and a few more options had to be turned off depending on archs to make things build (like ipr on powerpc/64 which failed due to missing writeq). * x86 and x86_64 UP and SMP allmodconfig and a custom test config. * powerpc and powerpc64 SMP allmodconfig * sparc and sparc64 SMP allmodconfig * ia64 SMP allmodconfig * s390 SMP allmodconfig * alpha SMP allmodconfig * um on x86_64 SMP allmodconfig 8. percpu.h modifications were reverted so that it could be applied as a separate patch and serve as bisection point. Given the fact that I had only a couple of failures from tests on step 6, I'm fairly confident about the coverage of this conversion patch. If there is a breakage, it's likely to be something in one of the arch headers which should be easily discoverable easily on most builds of the specific arch. Signed-off-by: Tejun Heo <tj@kernel.org> Guess-its-ok-by: Christoph Lameter <cl@linux-foundation.org> Cc: Ingo Molnar <mingo@redhat.com> Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
2010-03-24 11:04:11 +03:00
#include <linux/slab.h>
#include <net/mac80211.h>
#include "ieee80211_i.h"
#include "driver-ops.h"
#include "wme.h"
/**
* DOC: TX A-MPDU aggregation
*
* Aggregation on the TX side requires setting the hardware flag
* %IEEE80211_HW_AMPDU_AGGREGATION. The driver will then be handed
* packets with a flag indicating A-MPDU aggregation. The driver
* or device is responsible for actually aggregating the frames,
* as well as deciding how many and which to aggregate.
*
* When TX aggregation is started by some subsystem (usually the rate
* control algorithm would be appropriate) by calling the
* ieee80211_start_tx_ba_session() function, the driver will be
* notified via its @ampdu_action function, with the
* %IEEE80211_AMPDU_TX_START action.
*
* In response to that, the driver is later required to call the
* ieee80211_start_tx_ba_cb_irqsafe() function, which will really
* start the aggregation session after the peer has also responded.
* If the peer responds negatively, the session will be stopped
* again right away. Note that it is possible for the aggregation
* session to be stopped before the driver has indicated that it
* is done setting it up, in which case it must not indicate the
* setup completion.
*
* Also note that, since we also need to wait for a response from
* the peer, the driver is notified of the completion of the
* handshake by the %IEEE80211_AMPDU_TX_OPERATIONAL action to the
* @ampdu_action callback.
*
* Similarly, when the aggregation session is stopped by the peer
* or something calling ieee80211_stop_tx_ba_session(), the driver's
* @ampdu_action function will be called with the action
* %IEEE80211_AMPDU_TX_STOP. In this case, the call must not fail,
* and the driver must later call ieee80211_stop_tx_ba_cb_irqsafe().
*/
static void ieee80211_send_addba_request(struct ieee80211_sub_if_data *sdata,
const u8 *da, u16 tid,
u8 dialog_token, u16 start_seq_num,
u16 agg_size, u16 timeout)
{
struct ieee80211_local *local = sdata->local;
struct sk_buff *skb;
struct ieee80211_mgmt *mgmt;
u16 capab;
skb = dev_alloc_skb(sizeof(*mgmt) + local->hw.extra_tx_headroom);
if (!skb)
return;
skb_reserve(skb, local->hw.extra_tx_headroom);
mgmt = (struct ieee80211_mgmt *) skb_put(skb, 24);
memset(mgmt, 0, 24);
memcpy(mgmt->da, da, ETH_ALEN);
memcpy(mgmt->sa, sdata->vif.addr, ETH_ALEN);
if (sdata->vif.type == NL80211_IFTYPE_AP ||
sdata->vif.type == NL80211_IFTYPE_AP_VLAN)
memcpy(mgmt->bssid, sdata->vif.addr, ETH_ALEN);
else if (sdata->vif.type == NL80211_IFTYPE_STATION)
memcpy(mgmt->bssid, sdata->u.mgd.bssid, ETH_ALEN);
mgmt->frame_control = cpu_to_le16(IEEE80211_FTYPE_MGMT |
IEEE80211_STYPE_ACTION);
skb_put(skb, 1 + sizeof(mgmt->u.action.u.addba_req));
mgmt->u.action.category = WLAN_CATEGORY_BACK;
mgmt->u.action.u.addba_req.action_code = WLAN_ACTION_ADDBA_REQ;
mgmt->u.action.u.addba_req.dialog_token = dialog_token;
capab = (u16)(1 << 1); /* bit 1 aggregation policy */
capab |= (u16)(tid << 2); /* bit 5:2 TID number */
capab |= (u16)(agg_size << 6); /* bit 15:6 max size of aggergation */
mgmt->u.action.u.addba_req.capab = cpu_to_le16(capab);
mgmt->u.action.u.addba_req.timeout = cpu_to_le16(timeout);
mgmt->u.action.u.addba_req.start_seq_num =
cpu_to_le16(start_seq_num << 4);
ieee80211_tx_skb(sdata, skb);
}
void ieee80211_send_bar(struct ieee80211_vif *vif, u8 *ra, u16 tid, u16 ssn)
{
struct ieee80211_sub_if_data *sdata = vif_to_sdata(vif);
struct ieee80211_local *local = sdata->local;
struct sk_buff *skb;
struct ieee80211_bar *bar;
u16 bar_control = 0;
skb = dev_alloc_skb(sizeof(*bar) + local->hw.extra_tx_headroom);
if (!skb)
return;
skb_reserve(skb, local->hw.extra_tx_headroom);
bar = (struct ieee80211_bar *)skb_put(skb, sizeof(*bar));
memset(bar, 0, sizeof(*bar));
bar->frame_control = cpu_to_le16(IEEE80211_FTYPE_CTL |
IEEE80211_STYPE_BACK_REQ);
memcpy(bar->ra, ra, ETH_ALEN);
memcpy(bar->ta, sdata->vif.addr, ETH_ALEN);
bar_control |= (u16)IEEE80211_BAR_CTRL_ACK_POLICY_NORMAL;
bar_control |= (u16)IEEE80211_BAR_CTRL_CBMTID_COMPRESSED_BA;
bar_control |= (u16)(tid << IEEE80211_BAR_CTRL_TID_INFO_SHIFT);
bar->control = cpu_to_le16(bar_control);
bar->start_seq_num = cpu_to_le16(ssn);
IEEE80211_SKB_CB(skb)->flags |= IEEE80211_TX_INTFL_DONT_ENCRYPT;
ieee80211_tx_skb(sdata, skb);
}
EXPORT_SYMBOL(ieee80211_send_bar);
void ieee80211_assign_tid_tx(struct sta_info *sta, int tid,
struct tid_ampdu_tx *tid_tx)
{
lockdep_assert_held(&sta->ampdu_mlme.mtx);
lockdep_assert_held(&sta->lock);
rcu_assign_pointer(sta->ampdu_mlme.tid_tx[tid], tid_tx);
}
int ___ieee80211_stop_tx_ba_session(struct sta_info *sta, u16 tid,
enum ieee80211_back_parties initiator,
bool tx)
{
struct ieee80211_local *local = sta->local;
struct tid_ampdu_tx *tid_tx;
int ret;
lockdep_assert_held(&sta->ampdu_mlme.mtx);
spin_lock_bh(&sta->lock);
tid_tx = rcu_dereference_protected_tid_tx(sta, tid);
if (!tid_tx) {
spin_unlock_bh(&sta->lock);
return -ENOENT;
}
/* if we're already stopping ignore any new requests to stop */
if (test_bit(HT_AGG_STATE_STOPPING, &tid_tx->state)) {
spin_unlock_bh(&sta->lock);
return -EALREADY;
}
if (test_bit(HT_AGG_STATE_WANT_START, &tid_tx->state)) {
/* not even started yet! */
ieee80211_assign_tid_tx(sta, tid, NULL);
spin_unlock_bh(&sta->lock);
kfree_rcu(tid_tx, rcu_head);
return 0;
}
set_bit(HT_AGG_STATE_STOPPING, &tid_tx->state);
spin_unlock_bh(&sta->lock);
mac80211: fix spurious delBA handling Lennert Buytenhek noticed that delBA handling in mac80211 was broken and has remotely triggerable problems, some of which are due to some code shuffling I did that ended up changing the order in which things were done -- this was commit d75636ef9c1af224f1097941879d5a8db7cd04e5 Author: Johannes Berg <johannes@sipsolutions.net> Date: Tue Feb 10 21:25:53 2009 +0100 mac80211: RX aggregation: clean up stop session and other parts were already present in the original commit d92684e66091c0f0101819619b315b4bb8b5bcc5 Author: Ron Rindjunsky <ron.rindjunsky@intel.com> Date: Mon Jan 28 14:07:22 2008 +0200 mac80211: A-MPDU Tx add delBA from recipient support The first problem is that I moved a BUG_ON before various checks -- thereby making it possible to hit. As the comment indicates, the BUG_ON can be removed since the ampdu_action callback must already exist when the state is != IDLE. The second problem isn't easily exploitable but there's a race condition due to unconditionally setting the state to OPERATIONAL when a delBA frame is received, even when no aggregation session was ever initiated. All the drivers accept stopping the session even then, but that opens a race window where crashes could happen before the driver accepts it. Right now, a WARN_ON may happen with non-HT drivers, while the race opens only for HT drivers. For this case, there are two things necessary to fix it: 1) don't process spurious delBA frames, and be more careful about the session state; don't drop the lock 2) HT drivers need to be prepared to handle a session stop even before the session was really started -- this is true for all drivers (that support aggregation) but iwlwifi which can be fixed easily. The other HT drivers (ath9k and ar9170) are behaving properly already. Reported-by: Lennert Buytenhek <buytenh@marvell.com> Cc: stable@kernel.org Signed-off-by: Johannes Berg <johannes@sipsolutions.net> Signed-off-by: John W. Linville <linville@tuxdriver.com>
2009-11-22 14:28:41 +03:00
#ifdef CONFIG_MAC80211_HT_DEBUG
printk(KERN_DEBUG "Tx BA session stop requested for %pM tid %u\n",
sta->sta.addr, tid);
#endif /* CONFIG_MAC80211_HT_DEBUG */
del_timer_sync(&tid_tx->addba_resp_timer);
/*
* After this packets are no longer handed right through
* to the driver but are put onto tid_tx->pending instead,
* with locking to ensure proper access.
*/
clear_bit(HT_AGG_STATE_OPERATIONAL, &tid_tx->state);
tid_tx->stop_initiator = initiator;
tid_tx->tx_stop = tx;
ret = drv_ampdu_action(local, sta->sdata,
IEEE80211_AMPDU_TX_STOP,
&sta->sta, tid, NULL, 0);
/* HW shall not deny going back to legacy */
if (WARN_ON(ret)) {
/*
* We may have pending packets get stuck in this case...
* Not bothering with a workaround for now.
*/
}
return ret;
}
/*
* After sending add Block Ack request we activated a timer until
* add Block Ack response will arrive from the recipient.
* If this timer expires sta_addba_resp_timer_expired will be executed.
*/
static void sta_addba_resp_timer_expired(unsigned long data)
{
/* not an elegant detour, but there is no choice as the timer passes
* only one argument, and both sta_info and TID are needed, so init
* flow in sta_info_create gives the TID as data, while the timer_to_id
* array gives the sta through container_of */
u16 tid = *(u8 *)data;
struct sta_info *sta = container_of((void *)data,
struct sta_info, timer_to_tid[tid]);
struct tid_ampdu_tx *tid_tx;
/* check if the TID waits for addBA response */
rcu_read_lock();
tid_tx = rcu_dereference(sta->ampdu_mlme.tid_tx[tid]);
if (!tid_tx ||
test_bit(HT_AGG_STATE_RESPONSE_RECEIVED, &tid_tx->state)) {
rcu_read_unlock();
#ifdef CONFIG_MAC80211_HT_DEBUG
printk(KERN_DEBUG "timer expired on tid %d but we are not "
"(or no longer) expecting addBA response there\n",
tid);
#endif
return;
}
#ifdef CONFIG_MAC80211_HT_DEBUG
printk(KERN_DEBUG "addBA response timer expired on tid %d\n", tid);
#endif
ieee80211_stop_tx_ba_session(&sta->sta, tid);
rcu_read_unlock();
}
mac80211: fix aggregation for hardware with ampdu queues Hardware with AMPDU queues currently has broken aggregation. This patch fixes it by making all A-MPDUs go over the regular AC queues, but keeping track of the hardware queues in mac80211. As a first rough version, it actually stops the AC queue for extended periods of time, which can be removed by adding buffering internal to mac80211, but is currently not a huge problem because people rarely use multiple TIDs that are in the same AC (and iwlwifi currently doesn't operate as AP). This is a short-term fix, my current medium-term plan, which I hope to execute soon as well, but am not sure can finish before .30, looks like this: 1) rework the internal queuing layer in mac80211 that we use for fragments if the driver stopped queue in the middle of a fragmented frame to be able to queue more frames at once (rather than just a single frame with its fragments) 2) instead of stopping the entire AC queue, queue up the frames in a per-station/per-TID queue during aggregation session initiation, when the session has come up take all those frames and put them onto the queue from 1) 3) push the ampdu queue layer abstraction this patch introduces in mac80211 into the driver, and remove the virtual queue stuff from mac80211 again This plan will probably also affect ath9k in that mac80211 queues the frames instead of passing them down, even when there are no ampdu queues. Signed-off-by: Johannes Berg <johannes@sipsolutions.net> Signed-off-by: John W. Linville <linville@tuxdriver.com>
2009-02-12 02:51:53 +03:00
static inline int ieee80211_ac_from_tid(int tid)
{
return ieee802_1d_to_ac[tid & 7];
}
/*
* When multiple aggregation sessions on multiple stations
* are being created/destroyed simultaneously, we need to
* refcount the global queue stop caused by that in order
* to not get into a situation where one of the aggregation
* setup or teardown re-enables queues before the other is
* ready to handle that.
*
* These two functions take care of this issue by keeping
* a global "agg_queue_stop" refcount.
*/
static void __acquires(agg_queue)
ieee80211_stop_queue_agg(struct ieee80211_local *local, int tid)
{
int queue = ieee80211_ac_from_tid(tid);
if (atomic_inc_return(&local->agg_queue_stop[queue]) == 1)
ieee80211_stop_queue_by_reason(
&local->hw, queue,
IEEE80211_QUEUE_STOP_REASON_AGGREGATION);
__acquire(agg_queue);
}
static void __releases(agg_queue)
ieee80211_wake_queue_agg(struct ieee80211_local *local, int tid)
{
int queue = ieee80211_ac_from_tid(tid);
if (atomic_dec_return(&local->agg_queue_stop[queue]) == 0)
ieee80211_wake_queue_by_reason(
&local->hw, queue,
IEEE80211_QUEUE_STOP_REASON_AGGREGATION);
__release(agg_queue);
}
void ieee80211_tx_ba_session_handle_start(struct sta_info *sta, int tid)
{
struct tid_ampdu_tx *tid_tx;
struct ieee80211_local *local = sta->local;
struct ieee80211_sub_if_data *sdata = sta->sdata;
u16 start_seq_num;
int ret;
tid_tx = rcu_dereference_protected_tid_tx(sta, tid);
/*
* While we're asking the driver about the aggregation,
* stop the AC queue so that we don't have to worry
* about frames that came in while we were doing that,
* which would require us to put them to the AC pending
* afterwards which just makes the code more complex.
*/
ieee80211_stop_queue_agg(local, tid);
clear_bit(HT_AGG_STATE_WANT_START, &tid_tx->state);
/*
* make sure no packets are being processed to get
* valid starting sequence number
*/
synchronize_net();
start_seq_num = sta->tid_seq[tid] >> 4;
ret = drv_ampdu_action(local, sdata, IEEE80211_AMPDU_TX_START,
&sta->sta, tid, &start_seq_num, 0);
if (ret) {
#ifdef CONFIG_MAC80211_HT_DEBUG
printk(KERN_DEBUG "BA request denied - HW unavailable for"
" tid %d\n", tid);
#endif
spin_lock_bh(&sta->lock);
ieee80211_assign_tid_tx(sta, tid, NULL);
spin_unlock_bh(&sta->lock);
ieee80211_wake_queue_agg(local, tid);
kfree_rcu(tid_tx, rcu_head);
return;
}
/* we can take packets again now */
ieee80211_wake_queue_agg(local, tid);
/* activate the timer for the recipient's addBA response */
mod_timer(&tid_tx->addba_resp_timer, jiffies + ADDBA_RESP_INTERVAL);
#ifdef CONFIG_MAC80211_HT_DEBUG
printk(KERN_DEBUG "activated addBA response timer on tid %d\n", tid);
#endif
spin_lock_bh(&sta->lock);
sta->ampdu_mlme.addba_req_num[tid]++;
spin_unlock_bh(&sta->lock);
/* send AddBA request */
ieee80211_send_addba_request(sdata, sta->sta.addr, tid,
tid_tx->dialog_token, start_seq_num,
local->hw.max_tx_aggregation_subframes,
tid_tx->timeout);
}
int ieee80211_start_tx_ba_session(struct ieee80211_sta *pubsta, u16 tid,
u16 timeout)
{
struct sta_info *sta = container_of(pubsta, struct sta_info, sta);
struct ieee80211_sub_if_data *sdata = sta->sdata;
struct ieee80211_local *local = sdata->local;
struct tid_ampdu_tx *tid_tx;
int ret = 0;
trace_api_start_tx_ba_session(pubsta, tid);
if (WARN_ON(!local->ops->ampdu_action))
return -EINVAL;
if ((tid >= STA_TID_NUM) ||
!(local->hw.flags & IEEE80211_HW_AMPDU_AGGREGATION) ||
(local->hw.flags & IEEE80211_HW_TX_AMPDU_SETUP_IN_HW))
return -EINVAL;
#ifdef CONFIG_MAC80211_HT_DEBUG
printk(KERN_DEBUG "Open BA session requested for %pM tid %u\n",
pubsta->addr, tid);
#endif /* CONFIG_MAC80211_HT_DEBUG */
/*
* The aggregation code is not prepared to handle
* anything but STA/AP due to the BSSID handling.
* IBSS could work in the code but isn't supported
* by drivers or the standard.
*/
if (sdata->vif.type != NL80211_IFTYPE_STATION &&
sdata->vif.type != NL80211_IFTYPE_AP_VLAN &&
sdata->vif.type != NL80211_IFTYPE_AP)
return -EINVAL;
if (test_sta_flag(sta, WLAN_STA_BLOCK_BA)) {
#ifdef CONFIG_MAC80211_HT_DEBUG
printk(KERN_DEBUG "BA sessions blocked. "
"Denying BA session request\n");
#endif
return -EINVAL;
}
spin_lock_bh(&sta->lock);
/* we have tried too many times, receiver does not want A-MPDU */
if (sta->ampdu_mlme.addba_req_num[tid] > HT_AGG_MAX_RETRIES) {
ret = -EBUSY;
goto err_unlock_sta;
}
tid_tx = rcu_dereference_protected_tid_tx(sta, tid);
/* check if the TID is not in aggregation flow already */
if (tid_tx || sta->ampdu_mlme.tid_start_tx[tid]) {
#ifdef CONFIG_MAC80211_HT_DEBUG
printk(KERN_DEBUG "BA request denied - session is not "
"idle on tid %u\n", tid);
#endif /* CONFIG_MAC80211_HT_DEBUG */
ret = -EAGAIN;
goto err_unlock_sta;
}
/* prepare A-MPDU MLME for Tx aggregation */
tid_tx = kzalloc(sizeof(struct tid_ampdu_tx), GFP_ATOMIC);
if (!tid_tx) {
ret = -ENOMEM;
goto err_unlock_sta;
}
mac80211: fix aggregation for hardware with ampdu queues Hardware with AMPDU queues currently has broken aggregation. This patch fixes it by making all A-MPDUs go over the regular AC queues, but keeping track of the hardware queues in mac80211. As a first rough version, it actually stops the AC queue for extended periods of time, which can be removed by adding buffering internal to mac80211, but is currently not a huge problem because people rarely use multiple TIDs that are in the same AC (and iwlwifi currently doesn't operate as AP). This is a short-term fix, my current medium-term plan, which I hope to execute soon as well, but am not sure can finish before .30, looks like this: 1) rework the internal queuing layer in mac80211 that we use for fragments if the driver stopped queue in the middle of a fragmented frame to be able to queue more frames at once (rather than just a single frame with its fragments) 2) instead of stopping the entire AC queue, queue up the frames in a per-station/per-TID queue during aggregation session initiation, when the session has come up take all those frames and put them onto the queue from 1) 3) push the ampdu queue layer abstraction this patch introduces in mac80211 into the driver, and remove the virtual queue stuff from mac80211 again This plan will probably also affect ath9k in that mac80211 queues the frames instead of passing them down, even when there are no ampdu queues. Signed-off-by: Johannes Berg <johannes@sipsolutions.net> Signed-off-by: John W. Linville <linville@tuxdriver.com>
2009-02-12 02:51:53 +03:00
skb_queue_head_init(&tid_tx->pending);
__set_bit(HT_AGG_STATE_WANT_START, &tid_tx->state);
tid_tx->timeout = timeout;
/* Tx timer */
tid_tx->addba_resp_timer.function = sta_addba_resp_timer_expired;
tid_tx->addba_resp_timer.data = (unsigned long)&sta->timer_to_tid[tid];
init_timer(&tid_tx->addba_resp_timer);
/* assign a dialog token */
sta->ampdu_mlme.dialog_token_allocator++;
tid_tx->dialog_token = sta->ampdu_mlme.dialog_token_allocator;
/*
* Finally, assign it to the start array; the work item will
* collect it and move it to the normal array.
*/
sta->ampdu_mlme.tid_start_tx[tid] = tid_tx;
ieee80211_queue_work(&local->hw, &sta->ampdu_mlme.work);
/* this flow continues off the work */
mac80211: fix aggregation for hardware with ampdu queues Hardware with AMPDU queues currently has broken aggregation. This patch fixes it by making all A-MPDUs go over the regular AC queues, but keeping track of the hardware queues in mac80211. As a first rough version, it actually stops the AC queue for extended periods of time, which can be removed by adding buffering internal to mac80211, but is currently not a huge problem because people rarely use multiple TIDs that are in the same AC (and iwlwifi currently doesn't operate as AP). This is a short-term fix, my current medium-term plan, which I hope to execute soon as well, but am not sure can finish before .30, looks like this: 1) rework the internal queuing layer in mac80211 that we use for fragments if the driver stopped queue in the middle of a fragmented frame to be able to queue more frames at once (rather than just a single frame with its fragments) 2) instead of stopping the entire AC queue, queue up the frames in a per-station/per-TID queue during aggregation session initiation, when the session has come up take all those frames and put them onto the queue from 1) 3) push the ampdu queue layer abstraction this patch introduces in mac80211 into the driver, and remove the virtual queue stuff from mac80211 again This plan will probably also affect ath9k in that mac80211 queues the frames instead of passing them down, even when there are no ampdu queues. Signed-off-by: Johannes Berg <johannes@sipsolutions.net> Signed-off-by: John W. Linville <linville@tuxdriver.com>
2009-02-12 02:51:53 +03:00
err_unlock_sta:
spin_unlock_bh(&sta->lock);
return ret;
}
EXPORT_SYMBOL(ieee80211_start_tx_ba_session);
/*
* splice packets from the STA's pending to the local pending,
* requires a call to ieee80211_agg_splice_finish later
*/
static void __acquires(agg_queue)
ieee80211_agg_splice_packets(struct ieee80211_local *local,
struct tid_ampdu_tx *tid_tx, u16 tid)
{
int queue = ieee80211_ac_from_tid(tid);
unsigned long flags;
ieee80211_stop_queue_agg(local, tid);
if (WARN(!tid_tx, "TID %d gone but expected when splicing aggregates"
" from the pending queue\n", tid))
mac80211: fix panic when splicing unprepared TIDs We splice skbs from the pending queue for a TID onto the local pending queue when tearing down a block ack request. This is not necessary unless we actually have received a request to start a block ack request (rate control, for example). If we never received that request we should not be splicing the tid pending queue as it would be null, causing a panic. Not sure yet how exactly we allowed through a call when the tid state does not have at least HT_ADDBA_REQUESTED_MSK set, that will require some further review as it is not quite obvious. For more information see the bug report: http://bugzilla.kernel.org/show_bug.cgi?id=13922 This fixes this oops: BUG: unable to handle kernel NULL pointer dereference at 00000030 IP: [<f8806c70>] ieee80211_agg_splice_packets+0x40/0xc0 [mac80211] *pdpt = 0000000002d1e001 *pde = 0000000000000000 Thread overran stack, or stack corrupted Oops: 0000 [#1] SMP last sysfs file: /sys/module/aes_generic/initstate Modules linked in: <bleh> Pid: 0, comm: swapper Not tainted (2.6.31-rc5-wl #2) Dell DV051 EIP: 0060:[<f8806c70>] EFLAGS: 00010292 CPU: 0 EIP is at ieee80211_agg_splice_packets+0x40/0xc0 [mac80211] EAX: 00000030 EBX: 0000004c ECX: 00000003 EDX: 00000000 ESI: c1c98000 EDI: f745a1c0 EBP: c076be58 ESP: c076be38 DS: 007b ES: 007b FS: 00d8 GS: 0000 SS: 0068 Process swapper (pid: 0, ti=c076a000 task=c0709160 task.ti=c076a000) Stack: <bleh2> Call Trace: [<f8806edb>] ? ieee80211_stop_tx_ba_cb+0xab/0x150 [mac80211] [<f8802f1e>] ? ieee80211_tasklet_handler+0xce/0x110 [mac80211] [<c04862ff>] ? net_rx_action+0xef/0x1d0 [<c0149378>] ? tasklet_action+0x58/0xc0 [<c014a0f2>] ? __do_softirq+0xc2/0x190 [<c018eb48>] ? handle_IRQ_event+0x58/0x140 [<c01205fe>] ? ack_apic_level+0x7e/0x270 [<c014a1fd>] ? do_softirq+0x3d/0x40 [<c014a345>] ? irq_exit+0x65/0x90 [<c010a6af>] ? do_IRQ+0x4f/0xc0 [<c014a35d>] ? irq_exit+0x7d/0x90 [<c011d547>] ? smp_apic_timer_interrupt+0x57/0x90 [<c01094a9>] ? common_interrupt+0x29/0x30 [<c010fd9e>] ? mwait_idle+0xbe/0x100 [<c0107e42>] ? cpu_idle+0x52/0x90 [<c054b1a5>] ? rest_init+0x55/0x60 [<c077492d>] ? start_kernel+0x315/0x37d [<c07743ce>] ? unknown_bootoption+0x0/0x1f9 [<c0774099>] ? i386_start_kernel+0x79/0x81 Code: <bleh3> EIP: [<f8806c70>] ieee80211_agg_splice_packets+0x40/0xc0 [mac80211] SS:ESP 0068:c076be38 CR2: 0000000000000030 Cc: stable@kernel.org Testedy-by: Jack Lau <jackelectronics@hotmail.com> Signed-off-by: Luis R. Rodriguez <lrodriguez@atheros.com> Signed-off-by: John W. Linville <linville@tuxdriver.com>
2009-08-12 00:10:33 +04:00
return;
if (!skb_queue_empty(&tid_tx->pending)) {
spin_lock_irqsave(&local->queue_stop_reason_lock, flags);
/* copy over remaining packets */
skb_queue_splice_tail_init(&tid_tx->pending,
&local->pending[queue]);
spin_unlock_irqrestore(&local->queue_stop_reason_lock, flags);
}
}
static void __releases(agg_queue)
ieee80211_agg_splice_finish(struct ieee80211_local *local, u16 tid)
{
ieee80211_wake_queue_agg(local, tid);
}
static void ieee80211_agg_tx_operational(struct ieee80211_local *local,
struct sta_info *sta, u16 tid)
{
struct tid_ampdu_tx *tid_tx;
lockdep_assert_held(&sta->ampdu_mlme.mtx);
tid_tx = rcu_dereference_protected_tid_tx(sta, tid);
#ifdef CONFIG_MAC80211_HT_DEBUG
printk(KERN_DEBUG "Aggregation is on for tid %d\n", tid);
#endif
drv_ampdu_action(local, sta->sdata,
IEEE80211_AMPDU_TX_OPERATIONAL,
&sta->sta, tid, NULL, tid_tx->buf_size);
/*
* synchronize with TX path, while splicing the TX path
* should block so it won't put more packets onto pending.
*/
spin_lock_bh(&sta->lock);
ieee80211_agg_splice_packets(local, tid_tx, tid);
/*
* Now mark as operational. This will be visible
* in the TX path, and lets it go lock-free in
* the common case.
*/
set_bit(HT_AGG_STATE_OPERATIONAL, &tid_tx->state);
ieee80211_agg_splice_finish(local, tid);
spin_unlock_bh(&sta->lock);
}
void ieee80211_start_tx_ba_cb(struct ieee80211_vif *vif, u8 *ra, u16 tid)
{
struct ieee80211_sub_if_data *sdata = vif_to_sdata(vif);
struct ieee80211_local *local = sdata->local;
struct sta_info *sta;
struct tid_ampdu_tx *tid_tx;
trace_api_start_tx_ba_cb(sdata, ra, tid);
if (tid >= STA_TID_NUM) {
#ifdef CONFIG_MAC80211_HT_DEBUG
printk(KERN_DEBUG "Bad TID value: tid = %d (>= %d)\n",
tid, STA_TID_NUM);
#endif
return;
}
mutex_lock(&local->sta_mtx);
sta = sta_info_get(sdata, ra);
if (!sta) {
mutex_unlock(&local->sta_mtx);
#ifdef CONFIG_MAC80211_HT_DEBUG
printk(KERN_DEBUG "Could not find station: %pM\n", ra);
#endif
return;
}
mutex_lock(&sta->ampdu_mlme.mtx);
tid_tx = rcu_dereference_protected_tid_tx(sta, tid);
if (WARN_ON(!tid_tx)) {
#ifdef CONFIG_MAC80211_HT_DEBUG
printk(KERN_DEBUG "addBA was not requested!\n");
#endif
goto unlock;
}
if (WARN_ON(test_and_set_bit(HT_AGG_STATE_DRV_READY, &tid_tx->state)))
goto unlock;
if (test_bit(HT_AGG_STATE_RESPONSE_RECEIVED, &tid_tx->state))
ieee80211_agg_tx_operational(local, sta, tid);
mac80211: fix aggregation for hardware with ampdu queues Hardware with AMPDU queues currently has broken aggregation. This patch fixes it by making all A-MPDUs go over the regular AC queues, but keeping track of the hardware queues in mac80211. As a first rough version, it actually stops the AC queue for extended periods of time, which can be removed by adding buffering internal to mac80211, but is currently not a huge problem because people rarely use multiple TIDs that are in the same AC (and iwlwifi currently doesn't operate as AP). This is a short-term fix, my current medium-term plan, which I hope to execute soon as well, but am not sure can finish before .30, looks like this: 1) rework the internal queuing layer in mac80211 that we use for fragments if the driver stopped queue in the middle of a fragmented frame to be able to queue more frames at once (rather than just a single frame with its fragments) 2) instead of stopping the entire AC queue, queue up the frames in a per-station/per-TID queue during aggregation session initiation, when the session has come up take all those frames and put them onto the queue from 1) 3) push the ampdu queue layer abstraction this patch introduces in mac80211 into the driver, and remove the virtual queue stuff from mac80211 again This plan will probably also affect ath9k in that mac80211 queues the frames instead of passing them down, even when there are no ampdu queues. Signed-off-by: Johannes Berg <johannes@sipsolutions.net> Signed-off-by: John W. Linville <linville@tuxdriver.com>
2009-02-12 02:51:53 +03:00
unlock:
mutex_unlock(&sta->ampdu_mlme.mtx);
mutex_unlock(&local->sta_mtx);
}
void ieee80211_start_tx_ba_cb_irqsafe(struct ieee80211_vif *vif,
const u8 *ra, u16 tid)
{
struct ieee80211_sub_if_data *sdata = vif_to_sdata(vif);
struct ieee80211_local *local = sdata->local;
struct ieee80211_ra_tid *ra_tid;
struct sk_buff *skb = dev_alloc_skb(0);
if (unlikely(!skb))
return;
ra_tid = (struct ieee80211_ra_tid *) &skb->cb;
memcpy(&ra_tid->ra, ra, ETH_ALEN);
ra_tid->tid = tid;
skb->pkt_type = IEEE80211_SDATA_QUEUE_AGG_START;
skb_queue_tail(&sdata->skb_queue, skb);
ieee80211_queue_work(&local->hw, &sdata->work);
}
EXPORT_SYMBOL(ieee80211_start_tx_ba_cb_irqsafe);
int __ieee80211_stop_tx_ba_session(struct sta_info *sta, u16 tid,
enum ieee80211_back_parties initiator,
bool tx)
{
int ret;
mutex_lock(&sta->ampdu_mlme.mtx);
ret = ___ieee80211_stop_tx_ba_session(sta, tid, initiator, tx);
mutex_unlock(&sta->ampdu_mlme.mtx);
return ret;
}
int ieee80211_stop_tx_ba_session(struct ieee80211_sta *pubsta, u16 tid)
{
struct sta_info *sta = container_of(pubsta, struct sta_info, sta);
struct ieee80211_sub_if_data *sdata = sta->sdata;
struct ieee80211_local *local = sdata->local;
struct tid_ampdu_tx *tid_tx;
int ret = 0;
trace_api_stop_tx_ba_session(pubsta, tid);
if (!local->ops->ampdu_action)
return -EINVAL;
if (tid >= STA_TID_NUM)
return -EINVAL;
spin_lock_bh(&sta->lock);
tid_tx = rcu_dereference_protected_tid_tx(sta, tid);
if (!tid_tx) {
ret = -ENOENT;
goto unlock;
}
if (test_bit(HT_AGG_STATE_STOPPING, &tid_tx->state)) {
/* already in progress stopping it */
ret = 0;
goto unlock;
}
set_bit(HT_AGG_STATE_WANT_STOP, &tid_tx->state);
ieee80211_queue_work(&local->hw, &sta->ampdu_mlme.work);
unlock:
spin_unlock_bh(&sta->lock);
return ret;
}
EXPORT_SYMBOL(ieee80211_stop_tx_ba_session);
void ieee80211_stop_tx_ba_cb(struct ieee80211_vif *vif, u8 *ra, u8 tid)
{
struct ieee80211_sub_if_data *sdata = vif_to_sdata(vif);
struct ieee80211_local *local = sdata->local;
struct sta_info *sta;
struct tid_ampdu_tx *tid_tx;
trace_api_stop_tx_ba_cb(sdata, ra, tid);
if (tid >= STA_TID_NUM) {
#ifdef CONFIG_MAC80211_HT_DEBUG
printk(KERN_DEBUG "Bad TID value: tid = %d (>= %d)\n",
tid, STA_TID_NUM);
#endif
return;
}
#ifdef CONFIG_MAC80211_HT_DEBUG
printk(KERN_DEBUG "Stopping Tx BA session for %pM tid %d\n",
ra, tid);
#endif /* CONFIG_MAC80211_HT_DEBUG */
mutex_lock(&local->sta_mtx);
sta = sta_info_get(sdata, ra);
if (!sta) {
#ifdef CONFIG_MAC80211_HT_DEBUG
printk(KERN_DEBUG "Could not find station: %pM\n", ra);
#endif
goto unlock;
}
mutex_lock(&sta->ampdu_mlme.mtx);
spin_lock_bh(&sta->lock);
tid_tx = rcu_dereference_protected_tid_tx(sta, tid);
if (!tid_tx || !test_bit(HT_AGG_STATE_STOPPING, &tid_tx->state)) {
#ifdef CONFIG_MAC80211_HT_DEBUG
printk(KERN_DEBUG "unexpected callback to A-MPDU stop\n");
#endif
goto unlock_sta;
}
if (tid_tx->stop_initiator == WLAN_BACK_INITIATOR && tid_tx->tx_stop)
ieee80211_send_delba(sta->sdata, ra, tid,
WLAN_BACK_INITIATOR, WLAN_REASON_QSTA_NOT_USE);
/*
* When we get here, the TX path will not be lockless any more wrt.
* aggregation, since the OPERATIONAL bit has long been cleared.
* Thus it will block on getting the lock, if it occurs. So if we
* stop the queue now, we will not get any more packets, and any
* that might be being processed will wait for us here, thereby
* guaranteeing that no packets go to the tid_tx pending queue any
* more.
*/
ieee80211_agg_splice_packets(local, tid_tx, tid);
mac80211: fix aggregation for hardware with ampdu queues Hardware with AMPDU queues currently has broken aggregation. This patch fixes it by making all A-MPDUs go over the regular AC queues, but keeping track of the hardware queues in mac80211. As a first rough version, it actually stops the AC queue for extended periods of time, which can be removed by adding buffering internal to mac80211, but is currently not a huge problem because people rarely use multiple TIDs that are in the same AC (and iwlwifi currently doesn't operate as AP). This is a short-term fix, my current medium-term plan, which I hope to execute soon as well, but am not sure can finish before .30, looks like this: 1) rework the internal queuing layer in mac80211 that we use for fragments if the driver stopped queue in the middle of a fragmented frame to be able to queue more frames at once (rather than just a single frame with its fragments) 2) instead of stopping the entire AC queue, queue up the frames in a per-station/per-TID queue during aggregation session initiation, when the session has come up take all those frames and put them onto the queue from 1) 3) push the ampdu queue layer abstraction this patch introduces in mac80211 into the driver, and remove the virtual queue stuff from mac80211 again This plan will probably also affect ath9k in that mac80211 queues the frames instead of passing them down, even when there are no ampdu queues. Signed-off-by: Johannes Berg <johannes@sipsolutions.net> Signed-off-by: John W. Linville <linville@tuxdriver.com>
2009-02-12 02:51:53 +03:00
/* future packets must not find the tid_tx struct any more */
ieee80211_assign_tid_tx(sta, tid, NULL);
ieee80211_agg_splice_finish(local, tid);
kfree_rcu(tid_tx, rcu_head);
unlock_sta:
spin_unlock_bh(&sta->lock);
mutex_unlock(&sta->ampdu_mlme.mtx);
unlock:
mutex_unlock(&local->sta_mtx);
}
void ieee80211_stop_tx_ba_cb_irqsafe(struct ieee80211_vif *vif,
const u8 *ra, u16 tid)
{
struct ieee80211_sub_if_data *sdata = vif_to_sdata(vif);
struct ieee80211_local *local = sdata->local;
struct ieee80211_ra_tid *ra_tid;
struct sk_buff *skb = dev_alloc_skb(0);
if (unlikely(!skb))
return;
ra_tid = (struct ieee80211_ra_tid *) &skb->cb;
memcpy(&ra_tid->ra, ra, ETH_ALEN);
ra_tid->tid = tid;
skb->pkt_type = IEEE80211_SDATA_QUEUE_AGG_STOP;
skb_queue_tail(&sdata->skb_queue, skb);
ieee80211_queue_work(&local->hw, &sdata->work);
}
EXPORT_SYMBOL(ieee80211_stop_tx_ba_cb_irqsafe);
void ieee80211_process_addba_resp(struct ieee80211_local *local,
struct sta_info *sta,
struct ieee80211_mgmt *mgmt,
size_t len)
{
struct tid_ampdu_tx *tid_tx;
u16 capab, tid;
u8 buf_size;
capab = le16_to_cpu(mgmt->u.action.u.addba_resp.capab);
tid = (capab & IEEE80211_ADDBA_PARAM_TID_MASK) >> 2;
buf_size = (capab & IEEE80211_ADDBA_PARAM_BUF_SIZE_MASK) >> 6;
mutex_lock(&sta->ampdu_mlme.mtx);
tid_tx = rcu_dereference_protected_tid_tx(sta, tid);
if (!tid_tx)
goto out;
if (mgmt->u.action.u.addba_resp.dialog_token != tid_tx->dialog_token) {
#ifdef CONFIG_MAC80211_HT_DEBUG
printk(KERN_DEBUG "wrong addBA response token, tid %d\n", tid);
#endif
goto out;
}
del_timer_sync(&tid_tx->addba_resp_timer);
#ifdef CONFIG_MAC80211_HT_DEBUG
printk(KERN_DEBUG "switched off addBA timer for tid %d\n", tid);
#endif
/*
* addba_resp_timer may have fired before we got here, and
* caused WANT_STOP to be set. If the stop then was already
* processed further, STOPPING might be set.
*/
if (test_bit(HT_AGG_STATE_WANT_STOP, &tid_tx->state) ||
test_bit(HT_AGG_STATE_STOPPING, &tid_tx->state)) {
#ifdef CONFIG_MAC80211_HT_DEBUG
printk(KERN_DEBUG
"got addBA resp for tid %d but we already gave up\n",
tid);
#endif
goto out;
}
/*
* IEEE 802.11-2007 7.3.1.14:
* In an ADDBA Response frame, when the Status Code field
* is set to 0, the Buffer Size subfield is set to a value
* of at least 1.
*/
if (le16_to_cpu(mgmt->u.action.u.addba_resp.status)
== WLAN_STATUS_SUCCESS && buf_size) {
if (test_and_set_bit(HT_AGG_STATE_RESPONSE_RECEIVED,
&tid_tx->state)) {
/* ignore duplicate response */
goto out;
}
tid_tx->buf_size = buf_size;
if (test_bit(HT_AGG_STATE_DRV_READY, &tid_tx->state))
ieee80211_agg_tx_operational(local, sta, tid);
sta->ampdu_mlme.addba_req_num[tid] = 0;
} else {
___ieee80211_stop_tx_ba_session(sta, tid, WLAN_BACK_INITIATOR,
true);
}
out:
mutex_unlock(&sta->ampdu_mlme.mtx);
}