mac80211: update to 2009-11-13 and add some of my new performance improvement patches. fixes excessive cpu usage of hostapd in ap mode, should also fix the ath9k memleak

git-svn-id: svn://svn.openwrt.org/openwrt/trunk@18415 3c298f89-4303-0410-b956-a3cf2f4a3e73

1 files changed, 460 insertions, 0 deletions

diff --git a/package/mac80211/patches/530-ath9k_cleanup.patch b/package/mac80211/patches/530-ath9k_cleanup.patch
new file mode 100644
index 0000000000..49a1143ab0
--- /dev/null
+++ b/package/mac80211/patches/530-ath9k_cleanup.patch
@@ -0,0 +1,460 @@
+--- a/drivers/net/wireless/ath/ath9k/common.h
++++ b/drivers/net/wireless/ath/ath9k/common.h
+@@ -81,6 +81,7 @@ struct ath_buf {
+ 	u16 bf_flags;
+ 	struct ath_buf_state bf_state;
+ 	dma_addr_t bf_dmacontext;
++	struct ath_wiphy *aphy;
+ };
+ 
+ struct ath_atx_tid {
+--- a/drivers/net/wireless/ath/ath9k/rc.h
++++ b/drivers/net/wireless/ath/ath9k/rc.h
+@@ -167,24 +167,18 @@ struct ath_rate_priv {
+ 	struct ath_rate_softc *asc;
+ };
+ 
++#define ATH_TX_INFO_FRAME_TYPE_INTERNAL	(1 << 0)
++#define ATH_TX_INFO_FRAME_TYPE_PAUSE	(1 << 1)
++#define ATH_TX_INFO_UPDATE_RC		(1 << 2)
++#define ATH_TX_INFO_XRETRY		(1 << 3)
++#define ATH_TX_INFO_UNDERRUN		(1 << 4)
++
+ enum ath9k_internal_frame_type {
+ 	ATH9K_NOT_INTERNAL,
+ 	ATH9K_INT_PAUSE,
+ 	ATH9K_INT_UNPAUSE
+ };
+ 
+-struct ath_tx_info_priv {
+-	struct ath_wiphy *aphy;
+-	struct ath_tx_status tx;
+-	int n_frames;
+-	int n_bad_frames;
+-	bool update_rc;
+-	enum ath9k_internal_frame_type frame_type;
+-};
+-
+-#define ATH_TX_INFO_PRIV(tx_info) \
+-	((struct ath_tx_info_priv *)((tx_info)->rate_driver_data[0]))
+-
+ void ath_rate_attach(struct ath_softc *sc);
+ u8 ath_rate_findrateix(struct ath_softc *sc, u8 dot11_rate);
+ int ath_rate_control_register(void);
+--- a/drivers/net/wireless/ath/ath9k/virtual.c
++++ b/drivers/net/wireless/ath/ath9k/virtual.c
+@@ -338,13 +338,11 @@ void ath9k_wiphy_chan_work(struct work_s
+ void ath9k_tx_status(struct ieee80211_hw *hw, struct sk_buff *skb)
+ {
+ 	struct ath_wiphy *aphy = hw->priv;
+-	struct ieee80211_tx_info *info = IEEE80211_SKB_CB(skb);
+ 	struct ieee80211_tx_info *tx_info = IEEE80211_SKB_CB(skb);
+-	struct ath_tx_info_priv *tx_info_priv = ATH_TX_INFO_PRIV(tx_info);
+ 
+-	if (tx_info_priv && tx_info_priv->frame_type == ATH9K_INT_PAUSE &&
++	if ((tx_info->driver_flags & ATH_TX_INFO_FRAME_TYPE_PAUSE) &&
+ 	    aphy->state == ATH_WIPHY_PAUSING) {
+-		if (!(info->flags & IEEE80211_TX_STAT_ACK)) {
++		if (!(tx_info->flags & IEEE80211_TX_STAT_ACK)) {
+ 			printk(KERN_DEBUG "ath9k: %s: no ACK for pause "
+ 			       "frame\n", wiphy_name(hw->wiphy));
+ 			/*
+@@ -363,9 +361,6 @@ void ath9k_tx_status(struct ieee80211_hw
+ 		}
+ 	}
+ 
+-	kfree(tx_info_priv);
+-	tx_info->rate_driver_data[0] = NULL;
+-
+ 	dev_kfree_skb(skb);
+ }
+ 
+--- a/drivers/net/wireless/ath/ath9k/xmit.c
++++ b/drivers/net/wireless/ath/ath9k/xmit.c
+@@ -251,6 +251,7 @@ static struct ath_buf* ath_clone_txbuf(s
+ 
+ 	ATH_TXBUF_RESET(tbf);
+ 
++	tbf->aphy = bf->aphy;
+ 	tbf->bf_mpdu = bf->bf_mpdu;
+ 	tbf->bf_buf_addr = bf->bf_buf_addr;
+ 	*(tbf->bf_desc) = *(bf->bf_desc);
+@@ -270,7 +271,6 @@ static void ath_tx_complete_aggr(struct 
+ 	struct ieee80211_hw *hw;
+ 	struct ieee80211_hdr *hdr;
+ 	struct ieee80211_tx_info *tx_info;
+-	struct ath_tx_info_priv *tx_info_priv;
+ 	struct ath_atx_tid *tid = NULL;
+ 	struct ath_buf *bf_next, *bf_last = bf->bf_lastbf;
+ 	struct ath_desc *ds = bf_last->bf_desc;
+@@ -284,8 +284,7 @@ static void ath_tx_complete_aggr(struct 
+ 	hdr = (struct ieee80211_hdr *)skb->data;
+ 
+ 	tx_info = IEEE80211_SKB_CB(skb);
+-	tx_info_priv = (struct ath_tx_info_priv *) tx_info->rate_driver_data[0];
+-	hw = tx_info_priv->aphy->hw;
++	hw = bf->aphy->hw;
+ 
+ 	rcu_read_lock();
+ 
+@@ -464,7 +463,6 @@ static u32 ath_lookup_rate(struct ath_so
+ 	struct sk_buff *skb;
+ 	struct ieee80211_tx_info *tx_info;
+ 	struct ieee80211_tx_rate *rates;
+-	struct ath_tx_info_priv *tx_info_priv;
+ 	u32 max_4ms_framelen, frmlen;
+ 	u16 aggr_limit, legacy = 0;
+ 	int i;
+@@ -472,7 +470,6 @@ static u32 ath_lookup_rate(struct ath_so
+ 	skb = bf->bf_mpdu;
+ 	tx_info = IEEE80211_SKB_CB(skb);
+ 	rates = tx_info->control.rates;
+-	tx_info_priv = (struct ath_tx_info_priv *)tx_info->rate_driver_data[0];
+ 
+ 	/*
+ 	 * Find the lowest frame length among the rate series that will have a
+@@ -1560,21 +1557,26 @@ static int ath_tx_setup_buffer(struct ie
+ 	struct ath_softc *sc = aphy->sc;
+ 	struct ieee80211_tx_info *tx_info = IEEE80211_SKB_CB(skb);
+ 	struct ieee80211_hdr *hdr = (struct ieee80211_hdr *)skb->data;
+-	struct ath_tx_info_priv *tx_info_priv;
+ 	int hdrlen;
+ 	__le16 fc;
+ 
+-	tx_info_priv = kzalloc(sizeof(*tx_info_priv), GFP_ATOMIC);
+-	if (unlikely(!tx_info_priv))
+-		return -ENOMEM;
+-	tx_info->rate_driver_data[0] = tx_info_priv;
+-	tx_info_priv->aphy = aphy;
+-	tx_info_priv->frame_type = txctl->frame_type;
++	tx_info->driver_flags = 0;
++	switch (txctl->frame_type) {
++	case ATH9K_NOT_INTERNAL:
++		break;
++	case ATH9K_INT_PAUSE:
++		tx_info->driver_flags |= ATH_TX_INFO_FRAME_TYPE_PAUSE;
++		/* fall through */
++	case ATH9K_INT_UNPAUSE:
++		tx_info->driver_flags |= ATH_TX_INFO_FRAME_TYPE_INTERNAL;
++		break;
++	}
+ 	hdrlen = ieee80211_get_hdrlen_from_skb(skb);
+ 	fc = hdr->frame_control;
+ 
+ 	ATH_TXBUF_RESET(bf);
+ 
++	bf->aphy = aphy;
+ 	bf->bf_frmlen = skb->len + FCS_LEN - (hdrlen & 3);
+ 
+ 	if (conf_is_ht(&hw->conf) && !is_pae(skb))
+@@ -1599,8 +1601,6 @@ static int ath_tx_setup_buffer(struct ie
+ 					   skb->len, DMA_TO_DEVICE);
+ 	if (unlikely(dma_mapping_error(sc->dev, bf->bf_dmacontext))) {
+ 		bf->bf_mpdu = NULL;
+-		kfree(tx_info_priv);
+-		tx_info->rate_driver_data[0] = NULL;
+ 		ath_print(ath9k_hw_common(sc->sc_ah), ATH_DBG_FATAL,
+ 			  "dma_mapping_error() on TX\n");
+ 		return -ENOMEM;
+@@ -1781,27 +1781,17 @@ exit:
+ /*****************/
+ 
+ static void ath_tx_complete(struct ath_softc *sc, struct sk_buff *skb,
+-			    int tx_flags)
++			    struct ath_wiphy *aphy, int tx_flags)
+ {
+ 	struct ieee80211_hw *hw = sc->hw;
+ 	struct ieee80211_tx_info *tx_info = IEEE80211_SKB_CB(skb);
+-	struct ath_tx_info_priv *tx_info_priv = ATH_TX_INFO_PRIV(tx_info);
+ 	struct ath_common *common = ath9k_hw_common(sc->sc_ah);
+ 	int hdrlen, padsize;
+-	int frame_type = ATH9K_NOT_INTERNAL;
+ 
+ 	ath_print(common, ATH_DBG_XMIT, "TX complete: skb: %p\n", skb);
+ 
+-	if (tx_info_priv) {
+-		hw = tx_info_priv->aphy->hw;
+-		frame_type = tx_info_priv->frame_type;
+-	}
+-
+-	if (tx_info->flags & IEEE80211_TX_CTL_NO_ACK ||
+-	    tx_info->flags & IEEE80211_TX_STAT_TX_FILTERED) {
+-		kfree(tx_info_priv);
+-		tx_info->rate_driver_data[0] = NULL;
+-	}
++	if (aphy)
++		hw = aphy->hw;
+ 
+ 	if (tx_flags & ATH_TX_BAR)
+ 		tx_info->flags |= IEEE80211_TX_STAT_AMPDU_NO_BACK;
+@@ -1833,10 +1823,10 @@ static void ath_tx_complete(struct ath_s
+ 					SC_OP_WAIT_FOR_TX_ACK));
+ 	}
+ 
+-	if (frame_type == ATH9K_NOT_INTERNAL)
+-		ieee80211_tx_status(hw, skb);
+-	else
++	if (unlikely(tx_info->driver_flags & ATH_TX_INFO_FRAME_TYPE_INTERNAL))
+ 		ath9k_tx_status(hw, skb);
++	else
++		ieee80211_tx_status(hw, skb);
+ }
+ 
+ static void ath_tx_complete_buf(struct ath_softc *sc, struct ath_buf *bf,
+@@ -1859,7 +1849,7 @@ static void ath_tx_complete_buf(struct a
+ 	}
+ 
+ 	dma_unmap_single(sc->dev, bf->bf_dmacontext, skb->len, DMA_TO_DEVICE);
+-	ath_tx_complete(sc, skb, tx_flags);
++	ath_tx_complete(sc, skb, bf->aphy, tx_flags);
+ 	ath_debug_stat_tx(sc, txq, bf);
+ 
+ 	/*
+@@ -1907,8 +1897,7 @@ static void ath_tx_rc_status(struct ath_
+ 	struct sk_buff *skb = bf->bf_mpdu;
+ 	struct ieee80211_hdr *hdr = (struct ieee80211_hdr *)skb->data;
+ 	struct ieee80211_tx_info *tx_info = IEEE80211_SKB_CB(skb);
+-	struct ath_tx_info_priv *tx_info_priv = ATH_TX_INFO_PRIV(tx_info);
+-	struct ieee80211_hw *hw = tx_info_priv->aphy->hw;
++	struct ieee80211_hw *hw = bf->aphy->hw;
+ 	u8 i, tx_rateindex;
+ 
+ 	if (txok)
+@@ -1917,17 +1906,22 @@ static void ath_tx_rc_status(struct ath_
+ 	tx_rateindex = ds->ds_txstat.ts_rateindex;
+ 	WARN_ON(tx_rateindex >= hw->max_rates);
+ 
+-	tx_info_priv->update_rc = update_rc;
++	if (update_rc)
++		tx_info->driver_flags |= ATH_TX_INFO_UPDATE_RC;
+ 	if (ds->ds_txstat.ts_status & ATH9K_TXERR_FILT)
+ 		tx_info->flags |= IEEE80211_TX_STAT_TX_FILTERED;
+ 
+ 	if ((ds->ds_txstat.ts_status & ATH9K_TXERR_FILT) == 0 &&
+ 	    (bf->bf_flags & ATH9K_TXDESC_NOACK) == 0 && update_rc) {
+ 		if (ieee80211_is_data(hdr->frame_control)) {
+-			memcpy(&tx_info_priv->tx, &ds->ds_txstat,
+-			       sizeof(tx_info_priv->tx));
+-			tx_info_priv->n_frames = bf->bf_nframes;
+-			tx_info_priv->n_bad_frames = nbad;
++			if (ds->ds_txstat.ts_flags &
++			    (ATH9K_TX_DATA_UNDERRUN | ATH9K_TX_DELIM_UNDERRUN))
++				tx_info->driver_flags |= ATH_TX_INFO_UNDERRUN;
++			if ((ds->ds_txstat.ts_status & ATH9K_TXERR_XRETRY) ||
++			    (ds->ds_txstat.ts_status & ATH9K_TXERR_FIFO))
++				tx_info->driver_flags |= ATH_TX_INFO_XRETRY;
++			tx_info->ampdu_len = bf->bf_nframes;
++			tx_info->status.ampdu_ack_len = bf->bf_nframes - nbad;
+ 		}
+ 	}
+ 
+--- a/drivers/net/wireless/ath/ath9k/debug.c
++++ b/drivers/net/wireless/ath/ath9k/debug.c
+@@ -257,14 +257,17 @@ static const struct file_operations fops
+ 
+ void ath_debug_stat_rc(struct ath_softc *sc, struct sk_buff *skb)
+ {
+-	struct ath_tx_info_priv *tx_info_priv = NULL;
+ 	struct ieee80211_tx_info *tx_info = IEEE80211_SKB_CB(skb);
+ 	struct ieee80211_tx_rate *rates = tx_info->status.rates;
+-	int final_ts_idx, idx;
++	int final_ts_idx = 0, idx, i;
+ 	struct ath_rc_stats *stats;
+ 
+-	tx_info_priv = ATH_TX_INFO_PRIV(tx_info);
+-	final_ts_idx = tx_info_priv->tx.ts_rateindex;
++	for (i = 0; i < IEEE80211_TX_MAX_RATES; i++) {
++		if (!rates[i].count)
++			break;
++
++		final_ts_idx = i;
++	}
+ 	idx = rates[final_ts_idx].idx;
+ 	stats = &sc->debug.stats.rcstats[idx];
+ 	stats->success++;
+--- a/drivers/net/wireless/ath/ath9k/rc.c
++++ b/drivers/net/wireless/ath/ath9k/rc.c
+@@ -859,12 +859,12 @@ static void ath_get_rate(void *priv, str
+ static bool ath_rc_update_per(struct ath_softc *sc,
+ 			      const struct ath_rate_table *rate_table,
+ 			      struct ath_rate_priv *ath_rc_priv,
+-			      struct ath_tx_info_priv *tx_info_priv,
++				  struct ieee80211_tx_info *tx_info,
+ 			      int tx_rate, int xretries, int retries,
+ 			      u32 now_msec)
+ {
+ 	bool state_change = false;
+-	int count;
++	int count, n_bad_frames;
+ 	u8 last_per;
+ 	static u32 nretry_to_per_lookup[10] = {
+ 		100 * 0 / 1,
+@@ -880,6 +880,7 @@ static bool ath_rc_update_per(struct ath
+ 	};
+ 
+ 	last_per = ath_rc_priv->per[tx_rate];
++	n_bad_frames = tx_info->ampdu_len - tx_info->status.ampdu_ack_len;
+ 
+ 	if (xretries) {
+ 		if (xretries == 1) {
+@@ -907,7 +908,7 @@ static bool ath_rc_update_per(struct ath
+ 		if (retries >= count)
+ 			retries = count - 1;
+ 
+-		if (tx_info_priv->n_bad_frames) {
++		if (n_bad_frames) {
+ 			/* new_PER = 7/8*old_PER + 1/8*(currentPER)
+ 			 * Assuming that n_frames is not 0.  The current PER
+ 			 * from the retries is 100 * retries / (retries+1),
+@@ -920,14 +921,14 @@ static bool ath_rc_update_per(struct ath
+ 			 * the above PER.  The expression below is a
+ 			 * simplified version of the sum of these two terms.
+ 			 */
+-			if (tx_info_priv->n_frames > 0) {
+-				int n_frames, n_bad_frames;
++			if (tx_info->ampdu_len > 0) {
++				int n_frames, n_bad_tries;
+ 				u8 cur_per, new_per;
+ 
+-				n_bad_frames = retries * tx_info_priv->n_frames +
+-					tx_info_priv->n_bad_frames;
+-				n_frames = tx_info_priv->n_frames * (retries + 1);
+-				cur_per = (100 * n_bad_frames / n_frames) >> 3;
++				n_bad_tries = retries * tx_info->ampdu_len +
++					n_bad_frames;
++				n_frames = tx_info->ampdu_len * (retries + 1);
++				cur_per = (100 * n_bad_tries / n_frames) >> 3;
+ 				new_per = (u8)(last_per - (last_per >> 3) + cur_per);
+ 				ath_rc_priv->per[tx_rate] = new_per;
+ 			}
+@@ -943,8 +944,7 @@ static bool ath_rc_update_per(struct ath
+ 		 * this was a probe.  Otherwise, ignore the probe.
+ 		 */
+ 		if (ath_rc_priv->probe_rate && ath_rc_priv->probe_rate == tx_rate) {
+-			if (retries > 0 || 2 * tx_info_priv->n_bad_frames >
+-				tx_info_priv->n_frames) {
++			if (retries > 0 || 2 * n_bad_frames > tx_info->ampdu_len) {
+ 				/*
+ 				 * Since we probed with just a single attempt,
+ 				 * any retries means the probe failed.  Also,
+@@ -1003,7 +1003,7 @@ static bool ath_rc_update_per(struct ath
+ 
+ static void ath_rc_update_ht(struct ath_softc *sc,
+ 			     struct ath_rate_priv *ath_rc_priv,
+-			     struct ath_tx_info_priv *tx_info_priv,
++			     struct ieee80211_tx_info *tx_info,
+ 			     int tx_rate, int xretries, int retries)
+ {
+ 	u32 now_msec = jiffies_to_msecs(jiffies);
+@@ -1020,7 +1020,7 @@ static void ath_rc_update_ht(struct ath_
+ 
+ 	/* Update PER first */
+ 	state_change = ath_rc_update_per(sc, rate_table, ath_rc_priv,
+-					 tx_info_priv, tx_rate, xretries,
++					 tx_info, tx_rate, xretries,
+ 					 retries, now_msec);
+ 
+ 	/*
+@@ -1098,7 +1098,6 @@ static void ath_rc_tx_status(struct ath_
+ 			     struct ieee80211_tx_info *tx_info,
+ 			     int final_ts_idx, int xretries, int long_retry)
+ {
+-	struct ath_tx_info_priv *tx_info_priv = ATH_TX_INFO_PRIV(tx_info);
+ 	const struct ath_rate_table *rate_table;
+ 	struct ieee80211_tx_rate *rates = tx_info->status.rates;
+ 	u8 flags;
+@@ -1124,9 +1123,8 @@ static void ath_rc_tx_status(struct ath_
+ 					return;
+ 
+ 				rix = ath_rc_get_rateindex(rate_table, &rates[i]);
+-				ath_rc_update_ht(sc, ath_rc_priv,
+-						tx_info_priv, rix,
+-						xretries ? 1 : 2,
++				ath_rc_update_ht(sc, ath_rc_priv, tx_info,
++						rix, xretries ? 1 : 2,
+ 						rates[i].count);
+ 			}
+ 		}
+@@ -1149,8 +1147,7 @@ static void ath_rc_tx_status(struct ath_
+ 		return;
+ 
+ 	rix = ath_rc_get_rateindex(rate_table, &rates[i]);
+-	ath_rc_update_ht(sc, ath_rc_priv, tx_info_priv, rix,
+-			 xretries, long_retry);
++	ath_rc_update_ht(sc, ath_rc_priv, tx_info, rix, xretries, long_retry);
+ }
+ 
+ static const
+@@ -1301,23 +1298,30 @@ static void ath_tx_status(void *priv, st
+ {
+ 	struct ath_softc *sc = priv;
+ 	struct ath_rate_priv *ath_rc_priv = priv_sta;
+-	struct ath_tx_info_priv *tx_info_priv = NULL;
+ 	struct ieee80211_tx_info *tx_info = IEEE80211_SKB_CB(skb);
+ 	struct ieee80211_hdr *hdr;
+-	int final_ts_idx, tx_status = 0, is_underrun = 0;
++	int final_ts_idx = 0, tx_status = 0, is_underrun = 0;
++	int long_retry = 0;
+ 	__le16 fc;
++	int i;
+ 
+ 	hdr = (struct ieee80211_hdr *)skb->data;
+ 	fc = hdr->frame_control;
+-	tx_info_priv = ATH_TX_INFO_PRIV(tx_info);
+-	final_ts_idx = tx_info_priv->tx.ts_rateindex;
++	for (i = 0; i < IEEE80211_TX_MAX_RATES; i++) {
++		struct ieee80211_tx_rate *rate = &tx_info->status.rates[i];
++		if (!rate->count)
++			break;
++
++		final_ts_idx = i;
++		long_retry = rate->count - 1;
++	}
+ 
+ 	if (!priv_sta || !ieee80211_is_data(fc) ||
+-	    !tx_info_priv->update_rc)
+-		goto exit;
++	    !(tx_info->driver_flags & ATH_TX_INFO_UPDATE_RC))
++		return;
+ 
+-	if (tx_info_priv->tx.ts_status & ATH9K_TXERR_FILT)
+-		goto exit;
++	if (tx_info->flags & IEEE80211_TX_STAT_TX_FILTERED)
++		return;
+ 
+ 	/*
+ 	 * If underrun error is seen assume it as an excessive retry only
+@@ -1325,20 +1329,17 @@ static void ath_tx_status(void *priv, st
+ 	 * Adjust the long retry as if the frame was tried hw->max_rate_tries
+ 	 * times. This affects how ratectrl updates PER for the failed rate.
+ 	 */
+-	if (tx_info_priv->tx.ts_flags &
+-	    (ATH9K_TX_DATA_UNDERRUN | ATH9K_TX_DELIM_UNDERRUN) &&
+-	    ((sc->sc_ah->tx_trig_level) >= ath_rc_priv->tx_triglevel_max)) {
++	if ((tx_info->driver_flags & ATH_TX_INFO_UNDERRUN) &&
++	    (sc->sc_ah->tx_trig_level >= ath_rc_priv->tx_triglevel_max)) {
+ 		tx_status = 1;
+ 		is_underrun = 1;
+ 	}
+ 
+-	if ((tx_info_priv->tx.ts_status & ATH9K_TXERR_XRETRY) ||
+-	    (tx_info_priv->tx.ts_status & ATH9K_TXERR_FIFO))
++	if (tx_info->driver_flags & ATH_TX_INFO_XRETRY)
+ 		tx_status = 1;
+ 
+ 	ath_rc_tx_status(sc, ath_rc_priv, tx_info, final_ts_idx, tx_status,
+-			 (is_underrun) ? sc->hw->max_rate_tries :
+-			 tx_info_priv->tx.ts_longretry);
++			 (is_underrun) ? sc->hw->max_rate_tries : long_retry);
+ 
+ 	/* Check if aggregation has to be enabled for this tid */
+ 	if (conf_is_ht(&sc->hw->conf) &&
+@@ -1357,8 +1358,6 @@ static void ath_tx_status(void *priv, st
+ 	}
+ 
+ 	ath_debug_stat_rc(sc, skb);
+-exit:
+-	kfree(tx_info_priv);
+ }
+ 
+ static void ath_rate_init(void *priv, struct ieee80211_supported_band *sband,