1 files changed, 872 insertions, 257 deletions

diff --git a/drivers/net/wireless/rt2x00/rt2x00queue.c b/drivers/net/wireless/rt2x00/rt2x00queue.c
index 898cdd7f57d..5642ccceca7 100644
--- a/drivers/net/wireless/rt2x00/rt2x00queue.c
+++ b/drivers/net/wireless/rt2x00/rt2x00queue.c
@@ -1,5 +1,7 @@
 /*
-	Copyright (C) 2004 - 2008 rt2x00 SourceForge Project
+	Copyright (C) 2010 Willow Garage <http://www.willowgarage.com>
+	Copyright (C) 2004 - 2010 Ivo van Doorn <IvDoorn@gmail.com>
+	Copyright (C) 2004 - 2009 Gertjan van Wingerde <gwingerde@gmail.com>
 	<http://rt2x00.serialmonkey.com>
 
 	This program is free software; you can redistribute it and/or modify
@@ -13,9 +15,7 @@
 	GNU General Public License for more details.
 
 	You should have received a copy of the GNU General Public License
-	along with this program; if not, write to the
-	Free Software Foundation, Inc.,
-	59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
+	along with this program; if not, see <http://www.gnu.org/licenses/>.
  */
 
 /*
@@ -23,6 +23,7 @@
 	Abstract: rt2x00 queue specific routines.
  */
 
+#include <linux/slab.h>
 #include <linux/kernel.h>
 #include <linux/module.h>
 #include <linux/dma-mapping.h>
@@ -30,35 +31,51 @@
 #include "rt2x00.h"
 #include "rt2x00lib.h"
 
-struct sk_buff *rt2x00queue_alloc_rxskb(struct rt2x00_dev *rt2x00dev,
-					struct queue_entry *entry)
+struct sk_buff *rt2x00queue_alloc_rxskb(struct queue_entry *entry, gfp_t gfp)
 {
-	unsigned int frame_size;
-	unsigned int reserved_size;
+	struct data_queue *queue = entry->queue;
+	struct rt2x00_dev *rt2x00dev = queue->rt2x00dev;
 	struct sk_buff *skb;
 	struct skb_frame_desc *skbdesc;
+	unsigned int frame_size;
+	unsigned int head_size = 0;
+	unsigned int tail_size = 0;
 
 	/*
 	 * The frame size includes descriptor size, because the
 	 * hardware directly receive the frame into the skbuffer.
 	 */
-	frame_size = entry->queue->data_size + entry->queue->desc_size;
+	frame_size = queue->data_size + queue->desc_size + queue->winfo_size;
 
 	/*
 	 * The payload should be aligned to a 4-byte boundary,
 	 * this means we need at least 3 bytes for moving the frame
 	 * into the correct offset.
 	 */
-	reserved_size = 4;
+	head_size = 4;
+
+	/*
+	 * For IV/EIV/ICV assembly we must make sure there is
+	 * at least 8 bytes bytes available in headroom for IV/EIV
+	 * and 8 bytes for ICV data as tailroon.
+	 */
+	if (rt2x00_has_cap_hw_crypto(rt2x00dev)) {
+		head_size += 8;
+		tail_size += 8;
+	}
 
 	/*
 	 * Allocate skbuffer.
 	 */
-	skb = dev_alloc_skb(frame_size + reserved_size);
+	skb = __dev_alloc_skb(frame_size + head_size + tail_size, gfp);
 	if (!skb)
 		return NULL;
 
-	skb_reserve(skb, reserved_size);
+	/*
+	 * Make sure we not have a frame with the requested bytes
+	 * available in the head and tail.
+	 */
+	skb_reserve(skb, head_size);
 	skb_put(skb, frame_size);
 
 	/*
@@ -68,80 +85,354 @@ struct sk_buff *rt2x00queue_alloc_rxskb(struct rt2x00_dev *rt2x00dev,
 	memset(skbdesc, 0, sizeof(*skbdesc));
 	skbdesc->entry = entry;
 
-	if (test_bit(DRIVER_REQUIRE_DMA, &rt2x00dev->flags)) {
-		skbdesc->skb_dma = dma_map_single(rt2x00dev->dev,
-						  skb->data,
-						  skb->len,
-						  DMA_FROM_DEVICE);
+	if (test_bit(REQUIRE_DMA, &rt2x00dev->cap_flags)) {
+		dma_addr_t skb_dma;
+
+		skb_dma = dma_map_single(rt2x00dev->dev, skb->data, skb->len,
+					 DMA_FROM_DEVICE);
+		if (unlikely(dma_mapping_error(rt2x00dev->dev, skb_dma))) {
+			dev_kfree_skb_any(skb);
+			return NULL;
+		}
+
+		skbdesc->skb_dma = skb_dma;
 		skbdesc->flags |= SKBDESC_DMA_MAPPED_RX;
 	}
 
 	return skb;
 }
 
-void rt2x00queue_map_txskb(struct rt2x00_dev *rt2x00dev, struct sk_buff *skb)
+int rt2x00queue_map_txskb(struct queue_entry *entry)
 {
-	struct skb_frame_desc *skbdesc = get_skb_frame_desc(skb);
+	struct device *dev = entry->queue->rt2x00dev->dev;
+	struct skb_frame_desc *skbdesc = get_skb_frame_desc(entry->skb);
+
+	skbdesc->skb_dma =
+	    dma_map_single(dev, entry->skb->data, entry->skb->len, DMA_TO_DEVICE);
+
+	if (unlikely(dma_mapping_error(dev, skbdesc->skb_dma)))
+		return -ENOMEM;
 
-	skbdesc->skb_dma = dma_map_single(rt2x00dev->dev, skb->data, skb->len,
-					  DMA_TO_DEVICE);
 	skbdesc->flags |= SKBDESC_DMA_MAPPED_TX;
+	return 0;
 }
 EXPORT_SYMBOL_GPL(rt2x00queue_map_txskb);
 
-void rt2x00queue_unmap_skb(struct rt2x00_dev *rt2x00dev, struct sk_buff *skb)
+void rt2x00queue_unmap_skb(struct queue_entry *entry)
 {
-	struct skb_frame_desc *skbdesc = get_skb_frame_desc(skb);
+	struct device *dev = entry->queue->rt2x00dev->dev;
+	struct skb_frame_desc *skbdesc = get_skb_frame_desc(entry->skb);
 
 	if (skbdesc->flags & SKBDESC_DMA_MAPPED_RX) {
-		dma_unmap_single(rt2x00dev->dev, skbdesc->skb_dma, skb->len,
+		dma_unmap_single(dev, skbdesc->skb_dma, entry->skb->len,
 				 DMA_FROM_DEVICE);
 		skbdesc->flags &= ~SKBDESC_DMA_MAPPED_RX;
-	}
-
-	if (skbdesc->flags & SKBDESC_DMA_MAPPED_TX) {
-		dma_unmap_single(rt2x00dev->dev, skbdesc->skb_dma, skb->len,
+	} else if (skbdesc->flags & SKBDESC_DMA_MAPPED_TX) {
+		dma_unmap_single(dev, skbdesc->skb_dma, entry->skb->len,
 				 DMA_TO_DEVICE);
 		skbdesc->flags &= ~SKBDESC_DMA_MAPPED_TX;
 	}
 }
+EXPORT_SYMBOL_GPL(rt2x00queue_unmap_skb);
 
-void rt2x00queue_free_skb(struct rt2x00_dev *rt2x00dev, struct sk_buff *skb)
+void rt2x00queue_free_skb(struct queue_entry *entry)
 {
-	if (!skb)
+	if (!entry->skb)
 		return;
 
-	rt2x00queue_unmap_skb(rt2x00dev, skb);
-	dev_kfree_skb_any(skb);
+	rt2x00queue_unmap_skb(entry);
+	dev_kfree_skb_any(entry->skb);
+	entry->skb = NULL;
 }
 
-static void rt2x00queue_create_tx_descriptor(struct queue_entry *entry,
-					     struct txentry_desc *txdesc)
+void rt2x00queue_align_frame(struct sk_buff *skb)
 {
-	struct rt2x00_dev *rt2x00dev = entry->queue->rt2x00dev;
-	struct ieee80211_tx_info *tx_info = IEEE80211_SKB_CB(entry->skb);
+	unsigned int frame_length = skb->len;
+	unsigned int align = ALIGN_SIZE(skb, 0);
+
+	if (!align)
+		return;
+
+	skb_push(skb, align);
+	memmove(skb->data, skb->data + align, frame_length);
+	skb_trim(skb, frame_length);
+}
+
+void rt2x00queue_insert_l2pad(struct sk_buff *skb, unsigned int header_length)
+{
+	unsigned int payload_length = skb->len - header_length;
+	unsigned int header_align = ALIGN_SIZE(skb, 0);
+	unsigned int payload_align = ALIGN_SIZE(skb, header_length);
+	unsigned int l2pad = payload_length ? L2PAD_SIZE(header_length) : 0;
+
+	/*
+	 * Adjust the header alignment if the payload needs to be moved more
+	 * than the header.
+	 */
+	if (payload_align > header_align)
+		header_align += 4;
+
+	/* There is nothing to do if no alignment is needed */
+	if (!header_align)
+		return;
+
+	/* Reserve the amount of space needed in front of the frame */
+	skb_push(skb, header_align);
+
+	/*
+	 * Move the header.
+	 */
+	memmove(skb->data, skb->data + header_align, header_length);
+
+	/* Move the payload, if present and if required */
+	if (payload_length && payload_align)
+		memmove(skb->data + header_length + l2pad,
+			skb->data + header_length + l2pad + payload_align,
+			payload_length);
+
+	/* Trim the skb to the correct size */
+	skb_trim(skb, header_length + l2pad + payload_length);
+}
+
+void rt2x00queue_remove_l2pad(struct sk_buff *skb, unsigned int header_length)
+{
+	/*
+	 * L2 padding is only present if the skb contains more than just the
+	 * IEEE 802.11 header.
+	 */
+	unsigned int l2pad = (skb->len > header_length) ?
+				L2PAD_SIZE(header_length) : 0;
+
+	if (!l2pad)
+		return;
+
+	memmove(skb->data + l2pad, skb->data, header_length);
+	skb_pull(skb, l2pad);
+}
+
+static void rt2x00queue_create_tx_descriptor_seq(struct rt2x00_dev *rt2x00dev,
+						 struct sk_buff *skb,
+						 struct txentry_desc *txdesc)
+{
+	struct ieee80211_tx_info *tx_info = IEEE80211_SKB_CB(skb);
+	struct ieee80211_hdr *hdr = (struct ieee80211_hdr *)skb->data;
 	struct rt2x00_intf *intf = vif_to_intf(tx_info->control.vif);
-	struct ieee80211_hdr *hdr = (struct ieee80211_hdr *)entry->skb->data;
-	struct ieee80211_rate *rate =
-	    ieee80211_get_tx_rate(rt2x00dev->hw, tx_info);
-	const struct rt2x00_rate *hwrate;
+	u16 seqno;
+
+	if (!(tx_info->flags & IEEE80211_TX_CTL_ASSIGN_SEQ))
+		return;
+
+	__set_bit(ENTRY_TXD_GENERATE_SEQ, &txdesc->flags);
+
+	if (!test_bit(REQUIRE_SW_SEQNO, &rt2x00dev->cap_flags)) {
+		/*
+		 * rt2800 has a H/W (or F/W) bug, device incorrectly increase
+		 * seqno on retransmited data (non-QOS) frames. To workaround
+		 * the problem let's generate seqno in software if QOS is
+		 * disabled.
+		 */
+		if (test_bit(CONFIG_QOS_DISABLED, &rt2x00dev->flags))
+			__clear_bit(ENTRY_TXD_GENERATE_SEQ, &txdesc->flags);
+		else
+			/* H/W will generate sequence number */
+			return;
+	}
+
+	/*
+	 * The hardware is not able to insert a sequence number. Assign a
+	 * software generated one here.
+	 *
+	 * This is wrong because beacons are not getting sequence
+	 * numbers assigned properly.
+	 *
+	 * A secondary problem exists for drivers that cannot toggle
+	 * sequence counting per-frame, since those will override the
+	 * sequence counter given by mac80211.
+	 */
+	if (test_bit(ENTRY_TXD_FIRST_FRAGMENT, &txdesc->flags))
+		seqno = atomic_add_return(0x10, &intf->seqno);
+	else
+		seqno = atomic_read(&intf->seqno);
+
+	hdr->seq_ctrl &= cpu_to_le16(IEEE80211_SCTL_FRAG);
+	hdr->seq_ctrl |= cpu_to_le16(seqno);
+}
+
+static void rt2x00queue_create_tx_descriptor_plcp(struct rt2x00_dev *rt2x00dev,
+						  struct sk_buff *skb,
+						  struct txentry_desc *txdesc,
+						  const struct rt2x00_rate *hwrate)
+{
+	struct ieee80211_tx_info *tx_info = IEEE80211_SKB_CB(skb);
+	struct ieee80211_tx_rate *txrate = &tx_info->control.rates[0];
 	unsigned int data_length;
 	unsigned int duration;
 	unsigned int residual;
-	unsigned long irqflags;
 
-	memset(txdesc, 0, sizeof(*txdesc));
+	/*
+	 * Determine with what IFS priority this frame should be send.
+	 * Set ifs to IFS_SIFS when the this is not the first fragment,
+	 * or this fragment came after RTS/CTS.
+	 */
+	if (test_bit(ENTRY_TXD_FIRST_FRAGMENT, &txdesc->flags))
+		txdesc->u.plcp.ifs = IFS_BACKOFF;
+	else
+		txdesc->u.plcp.ifs = IFS_SIFS;
+
+	/* Data length + CRC + Crypto overhead (IV/EIV/ICV/MIC) */
+	data_length = skb->len + 4;
+	data_length += rt2x00crypto_tx_overhead(rt2x00dev, skb);
+
+	/*
+	 * PLCP setup
+	 * Length calculation depends on OFDM/CCK rate.
+	 */
+	txdesc->u.plcp.signal = hwrate->plcp;
+	txdesc->u.plcp.service = 0x04;
+
+	if (hwrate->flags & DEV_RATE_OFDM) {
+		txdesc->u.plcp.length_high = (data_length >> 6) & 0x3f;
+		txdesc->u.plcp.length_low = data_length & 0x3f;
+	} else {
+		/*
+		 * Convert length to microseconds.
+		 */
+		residual = GET_DURATION_RES(data_length, hwrate->bitrate);
+		duration = GET_DURATION(data_length, hwrate->bitrate);
+
+		if (residual != 0) {
+			duration++;
+
+			/*
+			 * Check if we need to set the Length Extension
+			 */
+			if (hwrate->bitrate == 110 && residual <= 30)
+				txdesc->u.plcp.service |= 0x80;
+		}
+
+		txdesc->u.plcp.length_high = (duration >> 8) & 0xff;
+		txdesc->u.plcp.length_low = duration & 0xff;
+
+		/*
+		 * When preamble is enabled we should set the
+		 * preamble bit for the signal.
+		 */
+		if (txrate->flags & IEEE80211_TX_RC_USE_SHORT_PREAMBLE)
+			txdesc->u.plcp.signal |= 0x08;
+	}
+}
+
+static void rt2x00queue_create_tx_descriptor_ht(struct rt2x00_dev *rt2x00dev,
+						struct sk_buff *skb,
+						struct txentry_desc *txdesc,
+						struct ieee80211_sta *sta,
+						const struct rt2x00_rate *hwrate)
+{
+	struct ieee80211_tx_info *tx_info = IEEE80211_SKB_CB(skb);
+	struct ieee80211_tx_rate *txrate = &tx_info->control.rates[0];
+	struct ieee80211_hdr *hdr = (struct ieee80211_hdr *)skb->data;
+	struct rt2x00_sta *sta_priv = NULL;
+
+	if (sta) {
+		txdesc->u.ht.mpdu_density =
+		    sta->ht_cap.ampdu_density;
+
+		sta_priv = sta_to_rt2x00_sta(sta);
+		txdesc->u.ht.wcid = sta_priv->wcid;
+	}
 
 	/*
-	 * Initialize information from queue
+	 * If IEEE80211_TX_RC_MCS is set txrate->idx just contains the
+	 * mcs rate to be used
 	 */
-	txdesc->queue = entry->queue->qid;
-	txdesc->cw_min = entry->queue->cw_min;
-	txdesc->cw_max = entry->queue->cw_max;
-	txdesc->aifs = entry->queue->aifs;
+	if (txrate->flags & IEEE80211_TX_RC_MCS) {
+		txdesc->u.ht.mcs = txrate->idx;
+
+		/*
+		 * MIMO PS should be set to 1 for STA's using dynamic SM PS
+		 * when using more then one tx stream (>MCS7).
+		 */
+		if (sta && txdesc->u.ht.mcs > 7 &&
+		    sta->smps_mode == IEEE80211_SMPS_DYNAMIC)
+			__set_bit(ENTRY_TXD_HT_MIMO_PS, &txdesc->flags);
+	} else {
+		txdesc->u.ht.mcs = rt2x00_get_rate_mcs(hwrate->mcs);
+		if (txrate->flags & IEEE80211_TX_RC_USE_SHORT_PREAMBLE)
+			txdesc->u.ht.mcs |= 0x08;
+	}
 
-	/* Data length should be extended with 4 bytes for CRC */
-	data_length = entry->skb->len + 4;
+	if (test_bit(CONFIG_HT_DISABLED, &rt2x00dev->flags)) {
+		if (!(tx_info->flags & IEEE80211_TX_CTL_FIRST_FRAGMENT))
+			txdesc->u.ht.txop = TXOP_SIFS;
+		else
+			txdesc->u.ht.txop = TXOP_BACKOFF;
+
+		/* Left zero on all other settings. */
+		return;
+	}
+
+	txdesc->u.ht.ba_size = 7;	/* FIXME: What value is needed? */
+
+	/*
+	 * Only one STBC stream is supported for now.
+	 */
+	if (tx_info->flags & IEEE80211_TX_CTL_STBC)
+		txdesc->u.ht.stbc = 1;
+
+	/*
+	 * This frame is eligible for an AMPDU, however, don't aggregate
+	 * frames that are intended to probe a specific tx rate.
+	 */
+	if (tx_info->flags & IEEE80211_TX_CTL_AMPDU &&
+	    !(tx_info->flags & IEEE80211_TX_CTL_RATE_CTRL_PROBE))
+		__set_bit(ENTRY_TXD_HT_AMPDU, &txdesc->flags);
+
+	/*
+	 * Set 40Mhz mode if necessary (for legacy rates this will
+	 * duplicate the frame to both channels).
+	 */
+	if (txrate->flags & IEEE80211_TX_RC_40_MHZ_WIDTH ||
+	    txrate->flags & IEEE80211_TX_RC_DUP_DATA)
+		__set_bit(ENTRY_TXD_HT_BW_40, &txdesc->flags);
+	if (txrate->flags & IEEE80211_TX_RC_SHORT_GI)
+		__set_bit(ENTRY_TXD_HT_SHORT_GI, &txdesc->flags);
+
+	/*
+	 * Determine IFS values
+	 * - Use TXOP_BACKOFF for management frames except beacons
+	 * - Use TXOP_SIFS for fragment bursts
+	 * - Use TXOP_HTTXOP for everything else
+	 *
+	 * Note: rt2800 devices won't use CTS protection (if used)
+	 * for frames not transmitted with TXOP_HTTXOP
+	 */
+	if (ieee80211_is_mgmt(hdr->frame_control) &&
+	    !ieee80211_is_beacon(hdr->frame_control))
+		txdesc->u.ht.txop = TXOP_BACKOFF;
+	else if (!(tx_info->flags & IEEE80211_TX_CTL_FIRST_FRAGMENT))
+		txdesc->u.ht.txop = TXOP_SIFS;
+	else
+		txdesc->u.ht.txop = TXOP_HTTXOP;
+}
+
+static void rt2x00queue_create_tx_descriptor(struct rt2x00_dev *rt2x00dev,
+					     struct sk_buff *skb,
+					     struct txentry_desc *txdesc,
+					     struct ieee80211_sta *sta)
+{
+	struct ieee80211_tx_info *tx_info = IEEE80211_SKB_CB(skb);
+	struct ieee80211_hdr *hdr = (struct ieee80211_hdr *)skb->data;
+	struct ieee80211_tx_rate *txrate = &tx_info->control.rates[0];
+	struct ieee80211_rate *rate;
+	const struct rt2x00_rate *hwrate = NULL;
+
+	memset(txdesc, 0, sizeof(*txdesc));
+
+	/*
+	 * Header and frame information.
+	 */
+	txdesc->length = skb->len;
+	txdesc->header_length = ieee80211_get_hdrlen_from_skb(skb);
 
 	/*
 	 * Check whether this frame is to be acked.
@@ -167,8 +458,8 @@ static void rt2x00queue_create_tx_descriptor(struct queue_entry *entry,
 	/*
 	 * Determine retry information.
 	 */
-	txdesc->retry_limit = tx_info->control.retry_limit;
-	if (tx_info->flags & IEEE80211_TX_CTL_LONG_RETRY_LIMIT)
+	txdesc->retry_limit = tx_info->control.rates[0].count - 1;
+	if (txdesc->retry_limit >= rt2x00dev->long_retry)
 		__set_bit(ENTRY_TXD_RETRY_MODE, &txdesc->flags);
 
 	/*
@@ -180,6 +471,12 @@ static void rt2x00queue_create_tx_descriptor(struct queue_entry *entry,
 	}
 
 	/*
+	 * Check if more frames (!= fragments) are pending
+	 */
+	if (tx_info->flags & IEEE80211_TX_CTL_MORE_FRAMES)
+		__set_bit(ENTRY_TXD_BURST, &txdesc->flags);
+
+	/*
 	 * Beacons and probe responses require the tsf timestamp
 	 * to be inserted into the frame.
 	 */
@@ -187,178 +484,310 @@ static void rt2x00queue_create_tx_descriptor(struct queue_entry *entry,
 	    ieee80211_is_probe_resp(hdr->frame_control))
 		__set_bit(ENTRY_TXD_REQ_TIMESTAMP, &txdesc->flags);
 
+	if ((tx_info->flags & IEEE80211_TX_CTL_FIRST_FRAGMENT) &&
+	    !test_bit(ENTRY_TXD_RTS_FRAME, &txdesc->flags))
+		__set_bit(ENTRY_TXD_FIRST_FRAGMENT, &txdesc->flags);
+
 	/*
-	 * Determine with what IFS priority this frame should be send.
-	 * Set ifs to IFS_SIFS when the this is not the first fragment,
-	 * or this fragment came after RTS/CTS.
+	 * Determine rate modulation.
 	 */
-	if (test_bit(ENTRY_TXD_RTS_FRAME, &txdesc->flags)) {
-		txdesc->ifs = IFS_SIFS;
-	} else if (tx_info->flags & IEEE80211_TX_CTL_FIRST_FRAGMENT) {
-		__set_bit(ENTRY_TXD_FIRST_FRAGMENT, &txdesc->flags);
-		txdesc->ifs = IFS_BACKOFF;
-	} else {
-		txdesc->ifs = IFS_SIFS;
+	if (txrate->flags & IEEE80211_TX_RC_GREEN_FIELD)
+		txdesc->rate_mode = RATE_MODE_HT_GREENFIELD;
+	else if (txrate->flags & IEEE80211_TX_RC_MCS)
+		txdesc->rate_mode = RATE_MODE_HT_MIX;
+	else {
+		rate = ieee80211_get_tx_rate(rt2x00dev->hw, tx_info);
+		hwrate = rt2x00_get_rate(rate->hw_value);
+		if (hwrate->flags & DEV_RATE_OFDM)
+			txdesc->rate_mode = RATE_MODE_OFDM;
+		else
+			txdesc->rate_mode = RATE_MODE_CCK;
 	}
 
 	/*
-	 * Hardware should insert sequence counter.
-	 * FIXME: We insert a software sequence counter first for
-	 * hardware that doesn't support hardware sequence counting.
-	 *
-	 * This is wrong because beacons are not getting sequence
-	 * numbers assigned properly.
-	 *
-	 * A secondary problem exists for drivers that cannot toggle
-	 * sequence counting per-frame, since those will override the
-	 * sequence counter given by mac80211.
+	 * Apply TX descriptor handling by components
 	 */
-	if (tx_info->flags & IEEE80211_TX_CTL_ASSIGN_SEQ) {
-		spin_lock_irqsave(&intf->seqlock, irqflags);
-
-		if (test_bit(ENTRY_TXD_FIRST_FRAGMENT, &txdesc->flags))
-			intf->seqno += 0x10;
-		hdr->seq_ctrl &= cpu_to_le16(IEEE80211_SCTL_FRAG);
-		hdr->seq_ctrl |= cpu_to_le16(intf->seqno);
+	rt2x00crypto_create_tx_descriptor(rt2x00dev, skb, txdesc);
+	rt2x00queue_create_tx_descriptor_seq(rt2x00dev, skb, txdesc);
+
+	if (test_bit(REQUIRE_HT_TX_DESC, &rt2x00dev->cap_flags))
+		rt2x00queue_create_tx_descriptor_ht(rt2x00dev, skb, txdesc,
+						   sta, hwrate);
+	else
+		rt2x00queue_create_tx_descriptor_plcp(rt2x00dev, skb, txdesc,
+						      hwrate);
+}
 
-		spin_unlock_irqrestore(&intf->seqlock, irqflags);
+static int rt2x00queue_write_tx_data(struct queue_entry *entry,
+				     struct txentry_desc *txdesc)
+{
+	struct rt2x00_dev *rt2x00dev = entry->queue->rt2x00dev;
 
-		__set_bit(ENTRY_TXD_GENERATE_SEQ, &txdesc->flags);
+	/*
+	 * This should not happen, we already checked the entry
+	 * was ours. When the hardware disagrees there has been
+	 * a queue corruption!
+	 */
+	if (unlikely(rt2x00dev->ops->lib->get_entry_state &&
+		     rt2x00dev->ops->lib->get_entry_state(entry))) {
+		rt2x00_err(rt2x00dev,
+			   "Corrupt queue %d, accessing entry which is not ours\n"
+			   "Please file bug report to %s\n",
+			   entry->queue->qid, DRV_PROJECT);
+		return -EINVAL;
 	}
 
 	/*
-	 * PLCP setup
-	 * Length calculation depends on OFDM/CCK rate.
+	 * Add the requested extra tx headroom in front of the skb.
 	 */
-	hwrate = rt2x00_get_rate(rate->hw_value);
-	txdesc->signal = hwrate->plcp;
-	txdesc->service = 0x04;
+	skb_push(entry->skb, rt2x00dev->extra_tx_headroom);
+	memset(entry->skb->data, 0, rt2x00dev->extra_tx_headroom);
 
-	if (hwrate->flags & DEV_RATE_OFDM) {
-		__set_bit(ENTRY_TXD_OFDM_RATE, &txdesc->flags);
-
-		txdesc->length_high = (data_length >> 6) & 0x3f;
-		txdesc->length_low = data_length & 0x3f;
-	} else {
-		/*
-		 * Convert length to microseconds.
-		 */
-		residual = get_duration_res(data_length, hwrate->bitrate);
-		duration = get_duration(data_length, hwrate->bitrate);
-
-		if (residual != 0) {
-			duration++;
-
-			/*
-			 * Check if we need to set the Length Extension
-			 */
-			if (hwrate->bitrate == 110 && residual <= 30)
-				txdesc->service |= 0x80;
-		}
+	/*
+	 * Call the driver's write_tx_data function, if it exists.
+	 */
+	if (rt2x00dev->ops->lib->write_tx_data)
+		rt2x00dev->ops->lib->write_tx_data(entry, txdesc);
 
-		txdesc->length_high = (duration >> 8) & 0xff;
-		txdesc->length_low = duration & 0xff;
+	/*
+	 * Map the skb to DMA.
+	 */
+	if (test_bit(REQUIRE_DMA, &rt2x00dev->cap_flags) &&
+	    rt2x00queue_map_txskb(entry))
+		return -ENOMEM;
 
-		/*
-		 * When preamble is enabled we should set the
-		 * preamble bit for the signal.
-		 */
-		if (rt2x00_get_rate_preamble(rate->hw_value))
-			txdesc->signal |= 0x08;
-	}
+	return 0;
 }
 
 static void rt2x00queue_write_tx_descriptor(struct queue_entry *entry,
 					    struct txentry_desc *txdesc)
 {
 	struct data_queue *queue = entry->queue;
-	struct rt2x00_dev *rt2x00dev = queue->rt2x00dev;
 
-	rt2x00dev->ops->lib->write_tx_desc(rt2x00dev, entry->skb, txdesc);
+	queue->rt2x00dev->ops->lib->write_tx_desc(entry, txdesc);
 
 	/*
 	 * All processing on the frame has been completed, this means
 	 * it is now ready to be dumped to userspace through debugfs.
 	 */
-	rt2x00debug_dump_frame(rt2x00dev, DUMP_FRAME_TX, entry->skb);
+	rt2x00debug_dump_frame(queue->rt2x00dev, DUMP_FRAME_TX, entry->skb);
+}
 
+static void rt2x00queue_kick_tx_queue(struct data_queue *queue,
+				      struct txentry_desc *txdesc)
+{
 	/*
 	 * Check if we need to kick the queue, there are however a few rules
-	 *	1) Don't kick beacon queue
-	 *	2) Don't kick unless this is the last in frame in a burst.
+	 *	1) Don't kick unless this is the last in frame in a burst.
 	 *	   When the burst flag is set, this frame is always followed
 	 *	   by another frame which in some way are related to eachother.
 	 *	   This is true for fragments, RTS or CTS-to-self frames.
-	 *	3) Rule 2 can be broken when the available entries
+	 *	2) Rule 1 can be broken when the available entries
 	 *	   in the queue are less then a certain threshold.
 	 */
-	if (entry->queue->qid == QID_BEACON)
-		return;
-
 	if (rt2x00queue_threshold(queue) ||
 	    !test_bit(ENTRY_TXD_BURST, &txdesc->flags))
-		rt2x00dev->ops->lib->kick_tx_queue(rt2x00dev, queue->qid);
+		queue->rt2x00dev->ops->lib->kick_queue(queue);
 }
 
-int rt2x00queue_write_tx_frame(struct data_queue *queue, struct sk_buff *skb)
+static void rt2x00queue_bar_check(struct queue_entry *entry)
 {
-	struct queue_entry *entry = rt2x00queue_get_entry(queue, Q_INDEX);
-	struct txentry_desc txdesc;
-	struct skb_frame_desc *skbdesc;
+	struct rt2x00_dev *rt2x00dev = entry->queue->rt2x00dev;
+	struct ieee80211_bar *bar = (void *) (entry->skb->data +
+				    rt2x00dev->extra_tx_headroom);
+	struct rt2x00_bar_list_entry *bar_entry;
 
-	if (unlikely(rt2x00queue_full(queue)))
-		return -EINVAL;
+	if (likely(!ieee80211_is_back_req(bar->frame_control)))
+		return;
 
-	if (__test_and_set_bit(ENTRY_OWNER_DEVICE_DATA, &entry->flags)) {
-		ERROR(queue->rt2x00dev,
-		      "Arrived at non-free entry in the non-full queue %d.\n"
-		      "Please file bug report to %s.\n",
-		      queue->qid, DRV_PROJECT);
-		return -EINVAL;
-	}
+	bar_entry = kmalloc(sizeof(*bar_entry), GFP_ATOMIC);
+
+	/*
+	 * If the alloc fails we still send the BAR out but just don't track
+	 * it in our bar list. And as a result we will report it to mac80211
+	 * back as failed.
+	 */
+	if (!bar_entry)
+		return;
+
+	bar_entry->entry = entry;
+	bar_entry->block_acked = 0;
+
+	/*
+	 * Copy the relevant parts of the 802.11 BAR into out check list
+	 * such that we can use RCU for less-overhead in the RX path since
+	 * sending BARs and processing the according BlockAck should be
+	 * the exception.
+	 */
+	memcpy(bar_entry->ra, bar->ra, sizeof(bar->ra));
+	memcpy(bar_entry->ta, bar->ta, sizeof(bar->ta));
+	bar_entry->control = bar->control;
+	bar_entry->start_seq_num = bar->start_seq_num;
+
+	/*
+	 * Insert BAR into our BAR check list.
+	 */
+	spin_lock_bh(&rt2x00dev->bar_list_lock);
+	list_add_tail_rcu(&bar_entry->list, &rt2x00dev->bar_list);
+	spin_unlock_bh(&rt2x00dev->bar_list_lock);
+}
+
+int rt2x00queue_write_tx_frame(struct data_queue *queue, struct sk_buff *skb,
+			       struct ieee80211_sta *sta, bool local)
+{
+	struct ieee80211_tx_info *tx_info;
+	struct queue_entry *entry;
+	struct txentry_desc txdesc;
+	struct skb_frame_desc *skbdesc;
+	u8 rate_idx, rate_flags;
+	int ret = 0;
 
 	/*
 	 * Copy all TX descriptor information into txdesc,
 	 * after that we are free to use the skb->cb array
 	 * for our information.
 	 */
-	entry->skb = skb;
-	rt2x00queue_create_tx_descriptor(entry, &txdesc);
+	rt2x00queue_create_tx_descriptor(queue->rt2x00dev, skb, &txdesc, sta);
 
 	/*
-	 * skb->cb array is now ours and we are free to use it.
+	 * All information is retrieved from the skb->cb array,
+	 * now we should claim ownership of the driver part of that
+	 * array, preserving the bitrate index and flags.
 	 */
-	skbdesc = get_skb_frame_desc(entry->skb);
+	tx_info = IEEE80211_SKB_CB(skb);
+	rate_idx = tx_info->control.rates[0].idx;
+	rate_flags = tx_info->control.rates[0].flags;
+	skbdesc = get_skb_frame_desc(skb);
 	memset(skbdesc, 0, sizeof(*skbdesc));
+	skbdesc->tx_rate_idx = rate_idx;
+	skbdesc->tx_rate_flags = rate_flags;
+
+	if (local)
+		skbdesc->flags |= SKBDESC_NOT_MAC80211;
+
+	/*
+	 * When hardware encryption is supported, and this frame
+	 * is to be encrypted, we should strip the IV/EIV data from
+	 * the frame so we can provide it to the driver separately.
+	 */
+	if (test_bit(ENTRY_TXD_ENCRYPT, &txdesc.flags) &&
+	    !test_bit(ENTRY_TXD_ENCRYPT_IV, &txdesc.flags)) {
+		if (test_bit(REQUIRE_COPY_IV, &queue->rt2x00dev->cap_flags))
+			rt2x00crypto_tx_copy_iv(skb, &txdesc);
+		else
+			rt2x00crypto_tx_remove_iv(skb, &txdesc);
+	}
+
+	/*
+	 * When DMA allocation is required we should guarantee to the
+	 * driver that the DMA is aligned to a 4-byte boundary.
+	 * However some drivers require L2 padding to pad the payload
+	 * rather then the header. This could be a requirement for
+	 * PCI and USB devices, while header alignment only is valid
+	 * for PCI devices.
+	 */
+	if (test_bit(REQUIRE_L2PAD, &queue->rt2x00dev->cap_flags))
+		rt2x00queue_insert_l2pad(skb, txdesc.header_length);
+	else if (test_bit(REQUIRE_DMA, &queue->rt2x00dev->cap_flags))
+		rt2x00queue_align_frame(skb);
+
+	/*
+	 * That function must be called with bh disabled.
+	 */
+	spin_lock(&queue->tx_lock);
+
+	if (unlikely(rt2x00queue_full(queue))) {
+		rt2x00_err(queue->rt2x00dev, "Dropping frame due to full tx queue %d\n",
+			   queue->qid);
+		ret = -ENOBUFS;
+		goto out;
+	}
+
+	entry = rt2x00queue_get_entry(queue, Q_INDEX);
+
+	if (unlikely(test_and_set_bit(ENTRY_OWNER_DEVICE_DATA,
+				      &entry->flags))) {
+		rt2x00_err(queue->rt2x00dev,
+			   "Arrived at non-free entry in the non-full queue %d\n"
+			   "Please file bug report to %s\n",
+			   queue->qid, DRV_PROJECT);
+		ret = -EINVAL;
+		goto out;
+	}
+
 	skbdesc->entry = entry;
+	entry->skb = skb;
 
-	if (unlikely(queue->rt2x00dev->ops->lib->write_tx_data(entry))) {
-		__clear_bit(ENTRY_OWNER_DEVICE_DATA, &entry->flags);
-		return -EIO;
+	/*
+	 * It could be possible that the queue was corrupted and this
+	 * call failed. Since we always return NETDEV_TX_OK to mac80211,
+	 * this frame will simply be dropped.
+	 */
+	if (unlikely(rt2x00queue_write_tx_data(entry, &txdesc))) {
+		clear_bit(ENTRY_OWNER_DEVICE_DATA, &entry->flags);
+		entry->skb = NULL;
+		ret = -EIO;
+		goto out;
 	}
 
-	if (test_bit(DRIVER_REQUIRE_DMA, &queue->rt2x00dev->flags))
-		rt2x00queue_map_txskb(queue->rt2x00dev, skb);
+	/*
+	 * Put BlockAckReqs into our check list for driver BA processing.
+	 */
+	rt2x00queue_bar_check(entry);
 
-	__set_bit(ENTRY_DATA_PENDING, &entry->flags);
+	set_bit(ENTRY_DATA_PENDING, &entry->flags);
 
-	rt2x00queue_index_inc(queue, Q_INDEX);
+	rt2x00queue_index_inc(entry, Q_INDEX);
 	rt2x00queue_write_tx_descriptor(entry, &txdesc);
+	rt2x00queue_kick_tx_queue(queue, &txdesc);
+
+out:
+	spin_unlock(&queue->tx_lock);
+	return ret;
+}
+
+int rt2x00queue_clear_beacon(struct rt2x00_dev *rt2x00dev,
+			     struct ieee80211_vif *vif)
+{
+	struct rt2x00_intf *intf = vif_to_intf(vif);
+
+	if (unlikely(!intf->beacon))
+		return -ENOBUFS;
+
+	mutex_lock(&intf->beacon_skb_mutex);
+
+	/*
+	 * Clean up the beacon skb.
+	 */
+	rt2x00queue_free_skb(intf->beacon);
+
+	/*
+	 * Clear beacon (single bssid devices don't need to clear the beacon
+	 * since the beacon queue will get stopped anyway).
+	 */
+	if (rt2x00dev->ops->lib->clear_beacon)
+		rt2x00dev->ops->lib->clear_beacon(intf->beacon);
+
+	mutex_unlock(&intf->beacon_skb_mutex);
 
 	return 0;
 }
 
-int rt2x00queue_update_beacon(struct rt2x00_dev *rt2x00dev,
-			      struct ieee80211_vif *vif)
+int rt2x00queue_update_beacon_locked(struct rt2x00_dev *rt2x00dev,
+				     struct ieee80211_vif *vif)
 {
 	struct rt2x00_intf *intf = vif_to_intf(vif);
 	struct skb_frame_desc *skbdesc;
 	struct txentry_desc txdesc;
-	__le32 desc[16];
 
 	if (unlikely(!intf->beacon))
 		return -ENOBUFS;
 
+	/*
+	 * Clean up the beacon skb.
+	 */
+	rt2x00queue_free_skb(intf->beacon);
+
 	intf->beacon->skb = ieee80211_beacon_get(rt2x00dev->hw, vif);
 	if (!intf->beacon->skb)
 		return -ENOMEM;
@@ -368,59 +797,91 @@ int rt2x00queue_update_beacon(struct rt2x00_dev *rt2x00dev,
 	 * after that we are free to use the skb->cb array
 	 * for our information.
 	 */
-	rt2x00queue_create_tx_descriptor(intf->beacon, &txdesc);
-
-	/*
-	 * For the descriptor we use a local array from where the
-	 * driver can move it to the correct location required for
-	 * the hardware.
-	 */
-	memset(desc, 0, sizeof(desc));
+	rt2x00queue_create_tx_descriptor(rt2x00dev, intf->beacon->skb, &txdesc, NULL);
 
 	/*
 	 * Fill in skb descriptor
 	 */
 	skbdesc = get_skb_frame_desc(intf->beacon->skb);
 	memset(skbdesc, 0, sizeof(*skbdesc));
-	skbdesc->desc = desc;
-	skbdesc->desc_len = intf->beacon->queue->desc_size;
 	skbdesc->entry = intf->beacon;
 
 	/*
-	 * Write TX descriptor into reserved room in front of the beacon.
-	 */
-	rt2x00queue_write_tx_descriptor(intf->beacon, &txdesc);
-
-	/*
 	 * Send beacon to hardware.
-	 * Also enable beacon generation, which might have been disabled
-	 * by the driver during the config_beacon() callback function.
 	 */
-	rt2x00dev->ops->lib->write_beacon(intf->beacon);
-	rt2x00dev->ops->lib->kick_tx_queue(rt2x00dev, QID_BEACON);
+	rt2x00dev->ops->lib->write_beacon(intf->beacon, &txdesc);
 
 	return 0;
+
+}
+
+int rt2x00queue_update_beacon(struct rt2x00_dev *rt2x00dev,
+			      struct ieee80211_vif *vif)
+{
+	struct rt2x00_intf *intf = vif_to_intf(vif);
+	int ret;
+
+	mutex_lock(&intf->beacon_skb_mutex);
+	ret = rt2x00queue_update_beacon_locked(rt2x00dev, vif);
+	mutex_unlock(&intf->beacon_skb_mutex);
+
+	return ret;
 }
 
-struct data_queue *rt2x00queue_get_queue(struct rt2x00_dev *rt2x00dev,
-					 const enum data_queue_qid queue)
+bool rt2x00queue_for_each_entry(struct data_queue *queue,
+				enum queue_index start,
+				enum queue_index end,
+				void *data,
+				bool (*fn)(struct queue_entry *entry,
+					   void *data))
 {
-	int atim = test_bit(DRIVER_REQUIRE_ATIM_QUEUE, &rt2x00dev->flags);
+	unsigned long irqflags;
+	unsigned int index_start;
+	unsigned int index_end;
+	unsigned int i;
 
-	if (queue < rt2x00dev->ops->tx_queues && rt2x00dev->tx)
-		return &rt2x00dev->tx[queue];
+	if (unlikely(start >= Q_INDEX_MAX || end >= Q_INDEX_MAX)) {
+		rt2x00_err(queue->rt2x00dev,
+			   "Entry requested from invalid index range (%d - %d)\n",
+			   start, end);
+		return true;
+	}
 
-	if (!rt2x00dev->bcn)
-		return NULL;
+	/*
+	 * Only protect the range we are going to loop over,
+	 * if during our loop a extra entry is set to pending
+	 * it should not be kicked during this run, since it
+	 * is part of another TX operation.
+	 */
+	spin_lock_irqsave(&queue->index_lock, irqflags);
+	index_start = queue->index[start];
+	index_end = queue->index[end];
+	spin_unlock_irqrestore(&queue->index_lock, irqflags);
+
+	/*
+	 * Start from the TX done pointer, this guarantees that we will
+	 * send out all frames in the correct order.
+	 */
+	if (index_start < index_end) {
+		for (i = index_start; i < index_end; i++) {
+			if (fn(&queue->entries[i], data))
+				return true;
+		}
+	} else {
+		for (i = index_start; i < queue->limit; i++) {
+			if (fn(&queue->entries[i], data))
+				return true;
+		}
 
-	if (queue == QID_BEACON)
-		return &rt2x00dev->bcn[0];
-	else if (queue == QID_ATIM && atim)
-		return &rt2x00dev->bcn[1];
+		for (i = 0; i < index_end; i++) {
+			if (fn(&queue->entries[i], data))
+				return true;
+		}
+	}
 
-	return NULL;
+	return false;
 }
-EXPORT_SYMBOL_GPL(rt2x00queue_get_queue);
+EXPORT_SYMBOL_GPL(rt2x00queue_for_each_entry);
 
 struct queue_entry *rt2x00queue_get_entry(struct data_queue *queue,
 					  enum queue_index index)
@@ -429,100 +890,255 @@ struct queue_entry *rt2x00queue_get_entry(struct data_queue *queue,
 	unsigned long irqflags;
 
 	if (unlikely(index >= Q_INDEX_MAX)) {
-		ERROR(queue->rt2x00dev,
-		      "Entry requested from invalid index type (%d)\n", index);
+		rt2x00_err(queue->rt2x00dev, "Entry requested from invalid index type (%d)\n",
+			   index);
 		return NULL;
 	}
 
-	spin_lock_irqsave(&queue->lock, irqflags);
+	spin_lock_irqsave(&queue->index_lock, irqflags);
 
 	entry = &queue->entries[queue->index[index]];
 
-	spin_unlock_irqrestore(&queue->lock, irqflags);
+	spin_unlock_irqrestore(&queue->index_lock, irqflags);
 
 	return entry;
 }
 EXPORT_SYMBOL_GPL(rt2x00queue_get_entry);
 
-void rt2x00queue_index_inc(struct data_queue *queue, enum queue_index index)
+void rt2x00queue_index_inc(struct queue_entry *entry, enum queue_index index)
 {
+	struct data_queue *queue = entry->queue;
 	unsigned long irqflags;
 
 	if (unlikely(index >= Q_INDEX_MAX)) {
-		ERROR(queue->rt2x00dev,
-		      "Index change on invalid index type (%d)\n", index);
+		rt2x00_err(queue->rt2x00dev,
+			   "Index change on invalid index type (%d)\n", index);
 		return;
 	}
 
-	spin_lock_irqsave(&queue->lock, irqflags);
+	spin_lock_irqsave(&queue->index_lock, irqflags);
 
 	queue->index[index]++;
 	if (queue->index[index] >= queue->limit)
 		queue->index[index] = 0;
 
+	entry->last_action = jiffies;
+
 	if (index == Q_INDEX) {
 		queue->length++;
 	} else if (index == Q_INDEX_DONE) {
 		queue->length--;
-		queue->count ++;
+		queue->count++;
 	}
 
-	spin_unlock_irqrestore(&queue->lock, irqflags);
+	spin_unlock_irqrestore(&queue->index_lock, irqflags);
 }
 
-static void rt2x00queue_reset(struct data_queue *queue)
+static void rt2x00queue_pause_queue_nocheck(struct data_queue *queue)
 {
-	unsigned long irqflags;
+	switch (queue->qid) {
+	case QID_AC_VO:
+	case QID_AC_VI:
+	case QID_AC_BE:
+	case QID_AC_BK:
+		/*
+		 * For TX queues, we have to disable the queue
+		 * inside mac80211.
+		 */
+		ieee80211_stop_queue(queue->rt2x00dev->hw, queue->qid);
+		break;
+	default:
+		break;
+	}
+}
+void rt2x00queue_pause_queue(struct data_queue *queue)
+{
+	if (!test_bit(DEVICE_STATE_PRESENT, &queue->rt2x00dev->flags) ||
+	    !test_bit(QUEUE_STARTED, &queue->flags) ||
+	    test_and_set_bit(QUEUE_PAUSED, &queue->flags))
+		return;
 
-	spin_lock_irqsave(&queue->lock, irqflags);
+	rt2x00queue_pause_queue_nocheck(queue);
+}
+EXPORT_SYMBOL_GPL(rt2x00queue_pause_queue);
 
-	queue->count = 0;
-	queue->length = 0;
-	memset(queue->index, 0, sizeof(queue->index));
+void rt2x00queue_unpause_queue(struct data_queue *queue)
+{
+	if (!test_bit(DEVICE_STATE_PRESENT, &queue->rt2x00dev->flags) ||
+	    !test_bit(QUEUE_STARTED, &queue->flags) ||
+	    !test_and_clear_bit(QUEUE_PAUSED, &queue->flags))
+		return;
 
-	spin_unlock_irqrestore(&queue->lock, irqflags);
+	switch (queue->qid) {
+	case QID_AC_VO:
+	case QID_AC_VI:
+	case QID_AC_BE:
+	case QID_AC_BK:
+		/*
+		 * For TX queues, we have to enable the queue
+		 * inside mac80211.
+		 */
+		ieee80211_wake_queue(queue->rt2x00dev->hw, queue->qid);
+		break;
+	case QID_RX:
+		/*
+		 * For RX we need to kick the queue now in order to
+		 * receive frames.
+		 */
+		queue->rt2x00dev->ops->lib->kick_queue(queue);
+	default:
+		break;
+	}
 }
+EXPORT_SYMBOL_GPL(rt2x00queue_unpause_queue);
 
-void rt2x00queue_init_rx(struct rt2x00_dev *rt2x00dev)
+void rt2x00queue_start_queue(struct data_queue *queue)
 {
-	struct data_queue *queue = rt2x00dev->rx;
-	unsigned int i;
+	mutex_lock(&queue->status_lock);
 
-	rt2x00queue_reset(queue);
-
-	if (!rt2x00dev->ops->lib->init_rxentry)
+	if (!test_bit(DEVICE_STATE_PRESENT, &queue->rt2x00dev->flags) ||
+	    test_and_set_bit(QUEUE_STARTED, &queue->flags)) {
+		mutex_unlock(&queue->status_lock);
 		return;
+	}
 
-	for (i = 0; i < queue->limit; i++) {
-		queue->entries[i].flags = 0;
+	set_bit(QUEUE_PAUSED, &queue->flags);
+
+	queue->rt2x00dev->ops->lib->start_queue(queue);
+
+	rt2x00queue_unpause_queue(queue);
+
+	mutex_unlock(&queue->status_lock);
+}
+EXPORT_SYMBOL_GPL(rt2x00queue_start_queue);
+
+void rt2x00queue_stop_queue(struct data_queue *queue)
+{
+	mutex_lock(&queue->status_lock);
 
-		rt2x00dev->ops->lib->init_rxentry(rt2x00dev,
-						  &queue->entries[i]);
+	if (!test_and_clear_bit(QUEUE_STARTED, &queue->flags)) {
+		mutex_unlock(&queue->status_lock);
+		return;
 	}
+
+	rt2x00queue_pause_queue_nocheck(queue);
+
+	queue->rt2x00dev->ops->lib->stop_queue(queue);
+
+	mutex_unlock(&queue->status_lock);
+}
+EXPORT_SYMBOL_GPL(rt2x00queue_stop_queue);
+
+void rt2x00queue_flush_queue(struct data_queue *queue, bool drop)
+{
+	bool tx_queue =
+		(queue->qid == QID_AC_VO) ||
+		(queue->qid == QID_AC_VI) ||
+		(queue->qid == QID_AC_BE) ||
+		(queue->qid == QID_AC_BK);
+
+
+	/*
+	 * If we are not supposed to drop any pending
+	 * frames, this means we must force a start (=kick)
+	 * to the queue to make sure the hardware will
+	 * start transmitting.
+	 */
+	if (!drop && tx_queue)
+		queue->rt2x00dev->ops->lib->kick_queue(queue);
+
+	/*
+	 * Check if driver supports flushing, if that is the case we can
+	 * defer the flushing to the driver. Otherwise we must use the
+	 * alternative which just waits for the queue to become empty.
+	 */
+	if (likely(queue->rt2x00dev->ops->lib->flush_queue))
+		queue->rt2x00dev->ops->lib->flush_queue(queue, drop);
+
+	/*
+	 * The queue flush has failed...
+	 */
+	if (unlikely(!rt2x00queue_empty(queue)))
+		rt2x00_warn(queue->rt2x00dev, "Queue %d failed to flush\n",
+			    queue->qid);
+}
+EXPORT_SYMBOL_GPL(rt2x00queue_flush_queue);
+
+void rt2x00queue_start_queues(struct rt2x00_dev *rt2x00dev)
+{
+	struct data_queue *queue;
+
+	/*
+	 * rt2x00queue_start_queue will call ieee80211_wake_queue
+	 * for each queue after is has been properly initialized.
+	 */
+	tx_queue_for_each(rt2x00dev, queue)
+		rt2x00queue_start_queue(queue);
+
+	rt2x00queue_start_queue(rt2x00dev->rx);
+}
+EXPORT_SYMBOL_GPL(rt2x00queue_start_queues);
+
+void rt2x00queue_stop_queues(struct rt2x00_dev *rt2x00dev)
+{
+	struct data_queue *queue;
+
+	/*
+	 * rt2x00queue_stop_queue will call ieee80211_stop_queue
+	 * as well, but we are completely shutting doing everything
+	 * now, so it is much safer to stop all TX queues at once,
+	 * and use rt2x00queue_stop_queue for cleaning up.
+	 */
+	ieee80211_stop_queues(rt2x00dev->hw);
+
+	tx_queue_for_each(rt2x00dev, queue)
+		rt2x00queue_stop_queue(queue);
+
+	rt2x00queue_stop_queue(rt2x00dev->rx);
 }
+EXPORT_SYMBOL_GPL(rt2x00queue_stop_queues);
 
-void rt2x00queue_init_tx(struct rt2x00_dev *rt2x00dev)
+void rt2x00queue_flush_queues(struct rt2x00_dev *rt2x00dev, bool drop)
 {
 	struct data_queue *queue;
+
+	tx_queue_for_each(rt2x00dev, queue)
+		rt2x00queue_flush_queue(queue, drop);
+
+	rt2x00queue_flush_queue(rt2x00dev->rx, drop);
+}
+EXPORT_SYMBOL_GPL(rt2x00queue_flush_queues);
+
+static void rt2x00queue_reset(struct data_queue *queue)
+{
+	unsigned long irqflags;
 	unsigned int i;
 
-	txall_queue_for_each(rt2x00dev, queue) {
-		rt2x00queue_reset(queue);
+	spin_lock_irqsave(&queue->index_lock, irqflags);
 
-		if (!rt2x00dev->ops->lib->init_txentry)
-			continue;
+	queue->count = 0;
+	queue->length = 0;
 
-		for (i = 0; i < queue->limit; i++) {
-			queue->entries[i].flags = 0;
+	for (i = 0; i < Q_INDEX_MAX; i++)
+		queue->index[i] = 0;
 
-			rt2x00dev->ops->lib->init_txentry(rt2x00dev,
-							  &queue->entries[i]);
-		}
+	spin_unlock_irqrestore(&queue->index_lock, irqflags);
+}
+
+void rt2x00queue_init_queues(struct rt2x00_dev *rt2x00dev)
+{
+	struct data_queue *queue;
+	unsigned int i;
+
+	queue_for_each(rt2x00dev, queue) {
+		rt2x00queue_reset(queue);
+
+		for (i = 0; i < queue->limit; i++)
+			rt2x00dev->ops->lib->clear_entry(&queue->entries[i]);
 	}
 }
 
-static int rt2x00queue_alloc_entries(struct data_queue *queue,
-				     const struct data_queue_desc *qdesc)
+static int rt2x00queue_alloc_entries(struct data_queue *queue)
 {
 	struct queue_entry *entries;
 	unsigned int entry_size;
@@ -530,22 +1146,17 @@ static int rt2x00queue_alloc_entries(struct data_queue *queue,
 
 	rt2x00queue_reset(queue);
 
-	queue->limit = qdesc->entry_num;
-	queue->threshold = DIV_ROUND_UP(qdesc->entry_num, 10);
-	queue->data_size = qdesc->data_size;
-	queue->desc_size = qdesc->desc_size;
-
 	/*
 	 * Allocate all queue entries.
 	 */
-	entry_size = sizeof(*entries) + qdesc->priv_size;
-	entries = kzalloc(queue->limit * entry_size, GFP_KERNEL);
+	entry_size = sizeof(*entries) + queue->priv_size;
+	entries = kcalloc(queue->limit, entry_size, GFP_KERNEL);
 	if (!entries)
 		return -ENOMEM;
 
 #define QUEUE_ENTRY_PRIV_OFFSET(__base, __index, __limit, __esize, __psize) \
-	( ((char *)(__base)) + ((__limit) * (__esize)) + \
-	    ((__index) * (__psize)) )
+	(((char *)(__base)) + ((__limit) * (__esize)) + \
+	    ((__index) * (__psize)))
 
 	for (i = 0; i < queue->limit; i++) {
 		entries[i].flags = 0;
@@ -554,7 +1165,7 @@ static int rt2x00queue_alloc_entries(struct data_queue *queue,
 		entries[i].entry_idx = i;
 		entries[i].priv_data =
 		    QUEUE_ENTRY_PRIV_OFFSET(entries, i, queue->limit,
-					    sizeof(*entries), qdesc->priv_size);
+					    sizeof(*entries), queue->priv_size);
 	}
 
 #undef QUEUE_ENTRY_PRIV_OFFSET
@@ -564,8 +1175,7 @@ static int rt2x00queue_alloc_entries(struct data_queue *queue,
 	return 0;
 }
 
-static void rt2x00queue_free_skbs(struct rt2x00_dev *rt2x00dev,
-				  struct data_queue *queue)
+static void rt2x00queue_free_skbs(struct data_queue *queue)
 {
 	unsigned int i;
 
@@ -573,19 +1183,17 @@ static void rt2x00queue_free_skbs(struct rt2x00_dev *rt2x00dev,
 		return;
 
 	for (i = 0; i < queue->limit; i++) {
-		if (queue->entries[i].skb)
-			rt2x00queue_free_skb(rt2x00dev, queue->entries[i].skb);
+		rt2x00queue_free_skb(&queue->entries[i]);
 	}
 }
 
-static int rt2x00queue_alloc_rxskbs(struct rt2x00_dev *rt2x00dev,
-				    struct data_queue *queue)
+static int rt2x00queue_alloc_rxskbs(struct data_queue *queue)
 {
 	unsigned int i;
 	struct sk_buff *skb;
 
 	for (i = 0; i < queue->limit; i++) {
-		skb = rt2x00queue_alloc_rxskb(rt2x00dev, &queue->entries[i]);
+		skb = rt2x00queue_alloc_rxskb(&queue->entries[i], GFP_KERNEL);
 		if (!skb)
 			return -ENOMEM;
 		queue->entries[i].skb = skb;
@@ -599,35 +1207,34 @@ int rt2x00queue_initialize(struct rt2x00_dev *rt2x00dev)
 	struct data_queue *queue;
 	int status;
 
-	status = rt2x00queue_alloc_entries(rt2x00dev->rx, rt2x00dev->ops->rx);
+	status = rt2x00queue_alloc_entries(rt2x00dev->rx);
 	if (status)
 		goto exit;
 
 	tx_queue_for_each(rt2x00dev, queue) {
-		status = rt2x00queue_alloc_entries(queue, rt2x00dev->ops->tx);
+		status = rt2x00queue_alloc_entries(queue);
 		if (status)
 			goto exit;
 	}
 
-	status = rt2x00queue_alloc_entries(rt2x00dev->bcn, rt2x00dev->ops->bcn);
+	status = rt2x00queue_alloc_entries(rt2x00dev->bcn);
 	if (status)
 		goto exit;
 
-	if (test_bit(DRIVER_REQUIRE_ATIM_QUEUE, &rt2x00dev->flags)) {
-		status = rt2x00queue_alloc_entries(&rt2x00dev->bcn[1],
-						   rt2x00dev->ops->atim);
+	if (test_bit(REQUIRE_ATIM_QUEUE, &rt2x00dev->cap_flags)) {
+		status = rt2x00queue_alloc_entries(rt2x00dev->atim);
 		if (status)
 			goto exit;
 	}
 
-	status = rt2x00queue_alloc_rxskbs(rt2x00dev, rt2x00dev->rx);
+	status = rt2x00queue_alloc_rxskbs(rt2x00dev->rx);
 	if (status)
 		goto exit;
 
 	return 0;
 
 exit:
-	ERROR(rt2x00dev, "Queue entries allocation failed.\n");
+	rt2x00_err(rt2x00dev, "Queue entries allocation failed\n");
 
 	rt2x00queue_uninitialize(rt2x00dev);
 
@@ -638,7 +1245,7 @@ void rt2x00queue_uninitialize(struct rt2x00_dev *rt2x00dev)
 {
 	struct data_queue *queue;
 
-	rt2x00queue_free_skbs(rt2x00dev, rt2x00dev->rx);
+	rt2x00queue_free_skbs(rt2x00dev->rx);
 
 	queue_for_each(rt2x00dev, queue) {
 		kfree(queue->entries);
@@ -649,13 +1256,20 @@ void rt2x00queue_uninitialize(struct rt2x00_dev *rt2x00dev)
 static void rt2x00queue_init(struct rt2x00_dev *rt2x00dev,
 			     struct data_queue *queue, enum data_queue_qid qid)
 {
-	spin_lock_init(&queue->lock);
+	mutex_init(&queue->status_lock);
+	spin_lock_init(&queue->tx_lock);
+	spin_lock_init(&queue->index_lock);
 
 	queue->rt2x00dev = rt2x00dev;
 	queue->qid = qid;
+	queue->txop = 0;
 	queue->aifs = 2;
 	queue->cw_min = 5;
 	queue->cw_max = 10;
+
+	rt2x00dev->ops->queue_init(queue);
+
+	queue->threshold = DIV_ROUND_UP(queue->limit, 10);
 }
 
 int rt2x00queue_allocate(struct rt2x00_dev *rt2x00dev)
@@ -663,7 +1277,7 @@ int rt2x00queue_allocate(struct rt2x00_dev *rt2x00dev)
 	struct data_queue *queue;
 	enum data_queue_qid qid;
 	unsigned int req_atim =
-	    !!test_bit(DRIVER_REQUIRE_ATIM_QUEUE, &rt2x00dev->flags);
+	    !!test_bit(REQUIRE_ATIM_QUEUE, &rt2x00dev->cap_flags);
 
 	/*
 	 * We need the following queues:
@@ -674,9 +1288,9 @@ int rt2x00queue_allocate(struct rt2x00_dev *rt2x00dev)
 	 */
 	rt2x00dev->data_queues = 2 + rt2x00dev->ops->tx_queues + req_atim;
 
-	queue = kzalloc(rt2x00dev->data_queues * sizeof(*queue), GFP_KERNEL);
+	queue = kcalloc(rt2x00dev->data_queues, sizeof(*queue), GFP_KERNEL);
 	if (!queue) {
-		ERROR(rt2x00dev, "Queue allocation failed.\n");
+		rt2x00_err(rt2x00dev, "Queue allocation failed\n");
 		return -ENOMEM;
 	}
 
@@ -686,11 +1300,12 @@ int rt2x00queue_allocate(struct rt2x00_dev *rt2x00dev)
 	rt2x00dev->rx = queue;
 	rt2x00dev->tx = &queue[1];
 	rt2x00dev->bcn = &queue[1 + rt2x00dev->ops->tx_queues];
+	rt2x00dev->atim = req_atim ? &queue[2 + rt2x00dev->ops->tx_queues] : NULL;
 
 	/*
 	 * Initialize queue parameters.
 	 * RX: qid = QID_RX
-	 * TX: qid = QID_AC_BE + index
+	 * TX: qid = QID_AC_VO + index
 	 * TX: cw_min: 2^5 = 32.
 	 * TX: cw_max: 2^10 = 1024.
 	 * BCN: qid = QID_BEACON
@@ -698,13 +1313,13 @@ int rt2x00queue_allocate(struct rt2x00_dev *rt2x00dev)
 	 */
 	rt2x00queue_init(rt2x00dev, rt2x00dev->rx, QID_RX);
 
-	qid = QID_AC_BE;
+	qid = QID_AC_VO;
 	tx_queue_for_each(rt2x00dev, queue)
 		rt2x00queue_init(rt2x00dev, queue, qid++);
 
-	rt2x00queue_init(rt2x00dev, &rt2x00dev->bcn[0], QID_BEACON);
+	rt2x00queue_init(rt2x00dev, rt2x00dev->bcn, QID_BEACON);
 	if (req_atim)
-		rt2x00queue_init(rt2x00dev, &rt2x00dev->bcn[1], QID_ATIM);
+		rt2x00queue_init(rt2x00dev, rt2x00dev->atim, QID_ATIM);
 
 	return 0;
 }