carl9170: 802.11 rx/tx processing and usb backend

Signed-off-by: Christian Lamparter <chunkeey@googlemail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

drivers/net/wireless/ath/carl9170/rx.c

+/*
+ * Atheros CARL9170 driver
+ *
+ * 802.11 & command trap routines
+ *
+ * Copyright 2008, Johannes Berg <johannes@sipsolutions.net>
+ * Copyright 2009, 2010, Christian Lamparter <chunkeey@googlemail.com>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; see the file COPYING.  If not, see
+ * http://www.gnu.org/licenses/.
+ *
+ * This file incorporates work covered by the following copyright and
+ * permission notice:
+ *    Copyright (c) 2007-2008 Atheros Communications, Inc.
+ *
+ *    Permission to use, copy, modify, and/or distribute this software for any
+ *    purpose with or without fee is hereby granted, provided that the above
+ *    copyright notice and this permission notice appear in all copies.
+ *
+ *    THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+ *    WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+ *    MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+ *    ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ *    WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+ *    ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+ *    OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ */
+
+#include <linux/init.h>
+#include <linux/slab.h>
+#include <linux/module.h>
+#include <linux/etherdevice.h>
+#include <linux/crc32.h>
+#include <net/mac80211.h>
+#include "carl9170.h"
+#include "hw.h"
+#include "cmd.h"
+
+static void carl9170_dbg_message(struct ar9170 *ar, const char *buf, u32 len)
+{
+	bool restart = false;
+	enum carl9170_restart_reasons reason = CARL9170_RR_NO_REASON;
+
+	if (len > 3) {
+		if (memcmp(buf, CARL9170_ERR_MAGIC, 3) == 0) {
+			ar->fw.err_counter++;
+			if (ar->fw.err_counter > 3) {
+				restart = true;
+				reason = CARL9170_RR_TOO_MANY_FIRMWARE_ERRORS;
+			}
+		}
+
+		if (memcmp(buf, CARL9170_BUG_MAGIC, 3) == 0) {
+			ar->fw.bug_counter++;
+			restart = true;
+			reason = CARL9170_RR_FATAL_FIRMWARE_ERROR;
+		}
+	}
+
+	wiphy_info(ar->hw->wiphy, "FW: %.*s\n", len, buf);
+
+	if (restart)
+		carl9170_restart(ar, reason);
+}
+
+static void carl9170_handle_ps(struct ar9170 *ar, struct carl9170_rsp *rsp)
+{
+	u32 ps;
+	bool new_ps;
+
+	ps = le32_to_cpu(rsp->psm.state);
+
+	new_ps = (ps & CARL9170_PSM_COUNTER) != CARL9170_PSM_WAKE;
+	if (ar->ps.state != new_ps) {
+		if (!new_ps) {
+			ar->ps.sleep_ms = jiffies_to_msecs(jiffies -
+				ar->ps.last_action);
+		}
+
+		ar->ps.last_action = jiffies;
+
+		ar->ps.state = new_ps;
+	}
+}
+
+static int carl9170_check_sequence(struct ar9170 *ar, unsigned int seq)
+{
+	if (ar->cmd_seq < -1)
+		return 0;
+
+	/*
+	 * Initialize Counter
+	 */
+	if (ar->cmd_seq < 0)
+		ar->cmd_seq = seq;
+
+	/*
+	 * The sequence is strictly monotonic increasing and it never skips!
+	 *
+	 * Therefore we can safely assume that whenever we received an
+	 * unexpected sequence we have lost some valuable data.
+	 */
+	if (seq != ar->cmd_seq) {
+		int count;
+
+		count = (seq - ar->cmd_seq) % ar->fw.cmd_bufs;
+
+		wiphy_err(ar->hw->wiphy, "lost %d command responses/traps! "
+			  "w:%d g:%d\n", count, ar->cmd_seq, seq);
+
+		carl9170_restart(ar, CARL9170_RR_LOST_RSP);
+		return -EIO;
+	}
+
+	ar->cmd_seq = (ar->cmd_seq + 1) % ar->fw.cmd_bufs;
+	return 0;
+}
+
+static void carl9170_cmd_callback(struct ar9170 *ar, u32 len, void *buffer)
+{
+	/*
+	 * Some commands may have a variable response length
+	 * and we cannot predict the correct length in advance.
+	 * So we only check if we provided enough space for the data.
+	 */
+	if (unlikely(ar->readlen != (len - 4))) {
+		dev_warn(&ar->udev->dev, "received invalid command response:"
+			 "got %d, instead of %d\n", len - 4, ar->readlen);
+		print_hex_dump_bytes("carl9170 cmd:", DUMP_PREFIX_OFFSET,
+			ar->cmd_buf, (ar->cmd.hdr.len + 4) & 0x3f);
+		print_hex_dump_bytes("carl9170 rsp:", DUMP_PREFIX_OFFSET,
+			buffer, len);
+		/*
+		 * Do not complete. The command times out,
+		 * and we get a stack trace from there.
+		 */
+		carl9170_restart(ar, CARL9170_RR_INVALID_RSP);
+	}
+
+	spin_lock(&ar->cmd_lock);
+	if (ar->readbuf) {
+		if (len >= 4)
+			memcpy(ar->readbuf, buffer + 4, len - 4);
+
+		ar->readbuf = NULL;
+	}
+	complete(&ar->cmd_wait);
+	spin_unlock(&ar->cmd_lock);
+}
+
+void carl9170_handle_command_response(struct ar9170 *ar, void *buf, u32 len)
+{
+	struct carl9170_rsp *cmd = (void *) buf;
+	struct ieee80211_vif *vif;
+
+	if (carl9170_check_sequence(ar, cmd->hdr.seq))
+		return;
+
+	if ((cmd->hdr.cmd & CARL9170_RSP_FLAG) != CARL9170_RSP_FLAG) {
+		if (!(cmd->hdr.cmd & CARL9170_CMD_ASYNC_FLAG))
+			carl9170_cmd_callback(ar, len, buf);
+
+		return;
+	}
+
+	if (unlikely(cmd->hdr.len != (len - 4))) {
+		if (net_ratelimit()) {
+			wiphy_err(ar->hw->wiphy, "FW: received over-/under"
+				"sized event %x (%d, but should be %d).\n",
+			       cmd->hdr.cmd, cmd->hdr.len, len - 4);
+
+			print_hex_dump_bytes("dump:", DUMP_PREFIX_NONE,
+					     buf, len);
+		}
+
+		return;
+	}
+
+	/* hardware event handlers */
+	switch (cmd->hdr.cmd) {
+	case CARL9170_RSP_PRETBTT:
+		/* pre-TBTT event */
+		rcu_read_lock();
+		vif = carl9170_get_main_vif(ar);
+
+		if (!vif) {
+			rcu_read_unlock();
+			break;
+		}
+
+		switch (vif->type) {
+		case NL80211_IFTYPE_STATION:
+			carl9170_handle_ps(ar, cmd);
+			break;
+
+		case NL80211_IFTYPE_AP:
+		case NL80211_IFTYPE_ADHOC:
+			carl9170_update_beacon(ar, true);
+			break;
+
+		default:
+			break;
+		}
+		rcu_read_unlock();
+
+		break;
+
+
+	case CARL9170_RSP_TXCOMP:
+		/* TX status notification */
+		carl9170_tx_process_status(ar, cmd);
+		break;
+
+	case CARL9170_RSP_BEACON_CONFIG:
+		/*
+		 * (IBSS) beacon send notification
+		 * bytes: 04 c2 XX YY B4 B3 B2 B1
+		 *
+		 * XX always 80
+		 * YY always 00
+		 * B1-B4 "should" be the number of send out beacons.
+		 */
+		break;
+
+	case CARL9170_RSP_ATIM:
+		/* End of Atim Window */
+		break;
+
+	case CARL9170_RSP_WATCHDOG:
+		/* Watchdog Interrupt */
+		carl9170_restart(ar, CARL9170_RR_WATCHDOG);
+		break;
+
+	case CARL9170_RSP_TEXT:
+		/* firmware debug */
+		carl9170_dbg_message(ar, (char *)buf + 4, len - 4);
+		break;
+
+	case CARL9170_RSP_HEXDUMP:
+		wiphy_dbg(ar->hw->wiphy, "FW: HD %d\n", len - 4);
+		print_hex_dump_bytes("FW:", DUMP_PREFIX_NONE,
+				     (char *)buf + 4, len - 4);
+		break;
+
+	case CARL9170_RSP_RADAR:
+		if (!net_ratelimit())
+			break;
+
+		wiphy_info(ar->hw->wiphy, "FW: RADAR! Please report this "
+		       "incident to linux-wireless@vger.kernel.org !\n");
+		break;
+
+	case CARL9170_RSP_GPIO:
+#ifdef CONFIG_CARL9170_WPC
+		if (ar->wps.pbc) {
+			bool state = !!(cmd->gpio.gpio & cpu_to_le32(
+				AR9170_GPIO_PORT_WPS_BUTTON_PRESSED));
+
+			if (state != ar->wps.pbc_state) {
+				ar->wps.pbc_state = state;
+				input_report_key(ar->wps.pbc, KEY_WPS_BUTTON,
+						 state);
+				input_sync(ar->wps.pbc);
+			}
+		}
+#endif /* CONFIG_CARL9170_WPC */
+		break;
+
+	case CARL9170_RSP_BOOT:
+		complete(&ar->fw_boot_wait);
+		break;
+
+	default:
+		wiphy_err(ar->hw->wiphy, "FW: received unhandled event %x\n",
+			cmd->hdr.cmd);
+		print_hex_dump_bytes("dump:", DUMP_PREFIX_NONE, buf, len);
+		break;
+	}
+}
+
+static int carl9170_rx_mac_status(struct ar9170 *ar,
+	struct ar9170_rx_head *head, struct ar9170_rx_macstatus *mac,
+	struct ieee80211_rx_status *status)
+{
+	struct ieee80211_channel *chan;
+	u8 error, decrypt;
+
+	BUILD_BUG_ON(sizeof(struct ar9170_rx_head) != 12);
+	BUILD_BUG_ON(sizeof(struct ar9170_rx_macstatus) != 4);
+
+	error = mac->error;
+
+	if (error & AR9170_RX_ERROR_WRONG_RA) {
+		if (!ar->sniffer_enabled)
+			return -EINVAL;
+	}
+
+	if (error & AR9170_RX_ERROR_PLCP) {
+		if (!(ar->filter_state & FIF_PLCPFAIL))
+			return -EINVAL;
+
+		status->flag |= RX_FLAG_FAILED_PLCP_CRC;
+	}
+
+	if (error & AR9170_RX_ERROR_FCS) {
+		ar->tx_fcs_errors++;
+
+		if (!(ar->filter_state & FIF_FCSFAIL))
+			return -EINVAL;
+
+		status->flag |= RX_FLAG_FAILED_FCS_CRC;
+	}
+
+	decrypt = ar9170_get_decrypt_type(mac);
+	if (!(decrypt & AR9170_RX_ENC_SOFTWARE) &&
+	    decrypt != AR9170_ENC_ALG_NONE) {
+		if ((decrypt == AR9170_ENC_ALG_TKIP) &&
+		    (error & AR9170_RX_ERROR_MMIC))
+			status->flag |= RX_FLAG_MMIC_ERROR;
+
+		status->flag |= RX_FLAG_DECRYPTED;
+	}
+
+	if (error & AR9170_RX_ERROR_DECRYPT && !ar->sniffer_enabled)
+		return -ENODATA;
+
+	error &= ~(AR9170_RX_ERROR_MMIC |
+		   AR9170_RX_ERROR_FCS |
+		   AR9170_RX_ERROR_WRONG_RA |
+		   AR9170_RX_ERROR_DECRYPT |
+		   AR9170_RX_ERROR_PLCP);
+
+	/* drop any other error frames */
+	if (unlikely(error)) {
+		/* TODO: update netdevice's RX dropped/errors statistics */
+
+		if (net_ratelimit())
+			wiphy_dbg(ar->hw->wiphy, "received frame with "
+			       "suspicious error code (%#x).\n", error);
+
+		return -EINVAL;
+	}
+
+	chan = ar->channel;
+	if (chan) {
+		status->band = chan->band;
+		status->freq = chan->center_freq;
+	}
+
+	switch (mac->status & AR9170_RX_STATUS_MODULATION) {
+	case AR9170_RX_STATUS_MODULATION_CCK:
+		if (mac->status & AR9170_RX_STATUS_SHORT_PREAMBLE)
+			status->flag |= RX_FLAG_SHORTPRE;
+		switch (head->plcp[0]) {
+		case AR9170_RX_PHY_RATE_CCK_1M:
+			status->rate_idx = 0;
+			break;
+		case AR9170_RX_PHY_RATE_CCK_2M:
+			status->rate_idx = 1;
+			break;
+		case AR9170_RX_PHY_RATE_CCK_5M:
+			status->rate_idx = 2;
+			break;
+		case AR9170_RX_PHY_RATE_CCK_11M:
+			status->rate_idx = 3;
+			break;
+		default:
+			if (net_ratelimit()) {
+				wiphy_err(ar->hw->wiphy, "invalid plcp cck "
+				       "rate (%x).\n", head->plcp[0]);
+			}
+
+			return -EINVAL;
+		}
+		break;
+
+	case AR9170_RX_STATUS_MODULATION_DUPOFDM:
+	case AR9170_RX_STATUS_MODULATION_OFDM:
+		switch (head->plcp[0] & 0xf) {
+		case AR9170_TXRX_PHY_RATE_OFDM_6M:
+			status->rate_idx = 0;
+			break;
+		case AR9170_TXRX_PHY_RATE_OFDM_9M:
+			status->rate_idx = 1;
+			break;
+		case AR9170_TXRX_PHY_RATE_OFDM_12M:
+			status->rate_idx = 2;
+			break;
+		case AR9170_TXRX_PHY_RATE_OFDM_18M:
+			status->rate_idx = 3;
+			break;
+		case AR9170_TXRX_PHY_RATE_OFDM_24M:
+			status->rate_idx = 4;
+			break;
+		case AR9170_TXRX_PHY_RATE_OFDM_36M:
+			status->rate_idx = 5;
+			break;
+		case AR9170_TXRX_PHY_RATE_OFDM_48M:
+			status->rate_idx = 6;
+			break;
+		case AR9170_TXRX_PHY_RATE_OFDM_54M:
+			status->rate_idx = 7;
+			break;
+		default:
+			if (net_ratelimit()) {
+				wiphy_err(ar->hw->wiphy, "invalid plcp ofdm "
+					"rate (%x).\n", head->plcp[0]);
+			}
+
+			return -EINVAL;
+		}
+		if (status->band == IEEE80211_BAND_2GHZ)
+			status->rate_idx += 4;
+		break;
+
+	case AR9170_RX_STATUS_MODULATION_HT:
+		if (head->plcp[3] & 0x80)
+			status->flag |= RX_FLAG_40MHZ;
+		if (head->plcp[6] & 0x80)
+			status->flag |= RX_FLAG_SHORT_GI;
+
+		status->rate_idx = clamp(0, 75, head->plcp[3] & 0x7f);
+		status->flag |= RX_FLAG_HT;
+		break;
+
+	default:
+		BUG();
+		return -ENOSYS;
+	}
+
+	return 0;
+}
+
+static void carl9170_rx_phy_status(struct ar9170 *ar,
+	struct ar9170_rx_phystatus *phy, struct ieee80211_rx_status *status)
+{
+	int i;
+
+	BUILD_BUG_ON(sizeof(struct ar9170_rx_phystatus) != 20);
+
+	for (i = 0; i < 3; i++)
+		if (phy->rssi[i] != 0x80)
+			status->antenna |= BIT(i);
+
+	/* post-process RSSI */
+	for (i = 0; i < 7; i++)
+		if (phy->rssi[i] & 0x80)
+			phy->rssi[i] = ((phy->rssi[i] & 0x7f) + 1) & 0x7f;
+
+	/* TODO: we could do something with phy_errors */
+	status->signal = ar->noise[0] + phy->rssi_combined;
+}
+
+static struct sk_buff *carl9170_rx_copy_data(u8 *buf, int len)
+{
+	struct sk_buff *skb;
+	int reserved = 0;
+	struct ieee80211_hdr *hdr = (void *) buf;
+
+	if (ieee80211_is_data_qos(hdr->frame_control)) {
+		u8 *qc = ieee80211_get_qos_ctl(hdr);
+		reserved += NET_IP_ALIGN;
+
+		if (*qc & IEEE80211_QOS_CONTROL_A_MSDU_PRESENT)
+			reserved += NET_IP_ALIGN;
+	}
+
+	if (ieee80211_has_a4(hdr->frame_control))
+		reserved += NET_IP_ALIGN;
+
+	reserved = 32 + (reserved & NET_IP_ALIGN);
+
+	skb = dev_alloc_skb(len + reserved);
+	if (likely(skb)) {
+		skb_reserve(skb, reserved);
+		memcpy(skb_put(skb, len), buf, len);
+	}
+
+	return skb;
+}
+
+static u8 *carl9170_find_ie(u8 *data, unsigned int len, u8 ie)
+{
+	struct ieee80211_mgmt *mgmt = (void *)data;
+	u8 *pos, *end;
+
+	pos = (u8 *)mgmt->u.beacon.variable;
+	end = data + len;
+	while (pos < end) {
+		if (pos + 2 + pos[1] > end)
+			return NULL;
+
+		if (pos[0] == ie)
+			return pos;
+
+		pos += 2 + pos[1];
+	}
+	return NULL;
+}
+
+/*
+ * NOTE:
+ *
+ * The firmware is in charge of waking up the device just before
+ * the AP is expected to transmit the next beacon.
+ *
+ * This leaves the driver with the important task of deciding when
+ * to set the PHY back to bed again.
+ */
+static void carl9170_ps_beacon(struct ar9170 *ar, void *data, unsigned int len)
+{
+	struct ieee80211_hdr *hdr = (void *) data;
+	struct ieee80211_tim_ie *tim_ie;
+	u8 *tim;
+	u8 tim_len;
+	bool cam;
+
+	if (likely(!(ar->hw->conf.flags & IEEE80211_CONF_PS)))
+		return;
+
+	/* check if this really is a beacon */
+	if (!ieee80211_is_beacon(hdr->frame_control))
+		return;
+
+	/* min. beacon length + FCS_LEN */
+	if (len <= 40 + FCS_LEN)
+		return;
+
+	/* and only beacons from the associated BSSID, please */
+	if (compare_ether_addr(hdr->addr3, ar->common.curbssid) ||
+	    !ar->common.curaid)
+		return;
+
+	ar->ps.last_beacon = jiffies;
+
+	tim = carl9170_find_ie(data, len - FCS_LEN, WLAN_EID_TIM);
+	if (!tim)
+		return;
+
+	if (tim[1] < sizeof(*tim_ie))
+		return;
+
+	tim_len = tim[1];
+	tim_ie = (struct ieee80211_tim_ie *) &tim[2];
+
+	if (!WARN_ON_ONCE(!ar->hw->conf.ps_dtim_period))
+		ar->ps.dtim_counter = (tim_ie->dtim_count - 1) %
+			ar->hw->conf.ps_dtim_period;
+
+	/* Check whenever the PHY can be turned off again. */
+
+	/* 1. What about buffered unicast traffic for our AID? */
+	cam = ieee80211_check_tim(tim_ie, tim_len, ar->common.curaid);
+
+	/* 2. Maybe the AP wants to send multicast/broadcast data? */
+	cam = !!(tim_ie->bitmap_ctrl & 0x01);
+
+	if (!cam) {
+		/* back to low-power land. */
+		ar->ps.off_override &= ~PS_OFF_BCN;
+		carl9170_ps_check(ar);
+	} else {
+		/* force CAM */
+		ar->ps.off_override |= PS_OFF_BCN;
+	}
+}
+
+/*
+ * If the frame alignment is right (or the kernel has
+ * CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS), and there
+ * is only a single MPDU in the USB frame, then we could
+ * submit to mac80211 the SKB directly. However, since
+ * there may be multiple packets in one SKB in stream
+ * mode, and we need to observe the proper ordering,
+ * this is non-trivial.
+ */
+
+static void carl9170_handle_mpdu(struct ar9170 *ar, u8 *buf, int len)
+{
+	struct ar9170_rx_head *head;
+	struct ar9170_rx_macstatus *mac;
+	struct ar9170_rx_phystatus *phy = NULL;
+	struct ieee80211_rx_status status;
+	struct sk_buff *skb;
+	int mpdu_len;
+
+	if (!IS_STARTED(ar))
+		return;
+
+	if (unlikely(len < sizeof(*mac))) {
+		ar->rx_dropped++;
+		return;
+	}
+
+	mpdu_len = len - sizeof(*mac);
+
+	mac = (void *)(buf + mpdu_len);
+	if (unlikely(mac->error & AR9170_RX_ERROR_FATAL)) {
+		ar->rx_dropped++;
+		return;
+	}
+
+	switch (mac->status & AR9170_RX_STATUS_MPDU) {
+	case AR9170_RX_STATUS_MPDU_FIRST:
+		/* Aggregated MPDUs start with an PLCP header */
+		if (likely(mpdu_len >= sizeof(struct ar9170_rx_head))) {
+			head = (void *) buf;
+
+			/*
+			 * The PLCP header needs to be cached for the
+			 * following MIDDLE + LAST A-MPDU packets.
+			 *
+			 * So, if you are wondering why all frames seem
+			 * to share a common RX status information,
+			 * then you have the answer right here...
+			 */
+			memcpy(&ar->rx_plcp, (void *) buf,
+			       sizeof(struct ar9170_rx_head));
+
+			mpdu_len -= sizeof(struct ar9170_rx_head);
+			buf += sizeof(struct ar9170_rx_head);
+
+			ar->rx_has_plcp = true;
+		} else {
+			if (net_ratelimit()) {
+				wiphy_err(ar->hw->wiphy, "plcp info "
+					"is clipped.\n");
+			}
+
+			ar->rx_dropped++;
+			return;
+		}
+		break;
+
+	case AR9170_RX_STATUS_MPDU_LAST:
+		/*
+		 * The last frame of an A-MPDU has an extra tail
+		 * which does contain the phy status of the whole
+		 * aggregate.
+		 */
+
+		if (likely(mpdu_len >= sizeof(struct ar9170_rx_phystatus))) {
+			mpdu_len -= sizeof(struct ar9170_rx_phystatus);
+			phy = (void *)(buf + mpdu_len);
+		} else {
+			if (net_ratelimit()) {
+				wiphy_err(ar->hw->wiphy, "frame tail "
+					"is clipped.\n");
+			}
+
+			ar->rx_dropped++;
+			return;
+		}
+
+	case AR9170_RX_STATUS_MPDU_MIDDLE:
+		/*  These are just data + mac status */
+		if (unlikely(!ar->rx_has_plcp)) {
+			if (!net_ratelimit())
+				return;
+
+			wiphy_err(ar->hw->wiphy, "rx stream does not start "
+					"with a first_mpdu frame tag.\n");
+
+			ar->rx_dropped++;
+			return;
+		}
+
+		head = &ar->rx_plcp;
+		break;
+
+	case AR9170_RX_STATUS_MPDU_SINGLE:
+		/* single mpdu has both: plcp (head) and phy status (tail) */
+		head = (void *) buf;
+
+		mpdu_len -= sizeof(struct ar9170_rx_head);
+		mpdu_len -= sizeof(struct ar9170_rx_phystatus);
+
+		buf += sizeof(struct ar9170_rx_head);
+		phy = (void *)(buf + mpdu_len);
+		break;
+
+	default:
+		BUG_ON(1);
+		break;
+	}
+
+	/* FC + DU + RA + FCS */
+	if (unlikely(mpdu_len < (2 + 2 + 6 + FCS_LEN))) {
+		ar->rx_dropped++;
+		return;
+	}
+
+	memset(&status, 0, sizeof(status));
+	if (unlikely(carl9170_rx_mac_status(ar, head, mac, &status))) {
+		ar->rx_dropped++;
+		return;
+	}
+
+	if (phy)
+		carl9170_rx_phy_status(ar, phy, &status);
+
+	carl9170_ps_beacon(ar, buf, mpdu_len);
+
+	skb = carl9170_rx_copy_data(buf, mpdu_len);
+	if (likely(skb)) {
+		memcpy(IEEE80211_SKB_RXCB(skb), &status, sizeof(status));
+		ieee80211_rx(ar->hw, skb);
+	} else {
+		ar->rx_dropped++;
+	}
+}
+
+static void carl9170_rx_untie_cmds(struct ar9170 *ar, const u8 *respbuf,
+				   const unsigned int resplen)
+{
+	struct carl9170_rsp *cmd;
+	int i = 0;
+
+	while (i < resplen) {
+		cmd = (void *) &respbuf[i];
+
+		i += cmd->hdr.len + 4;
+		if (unlikely(i > resplen))
+			break;
+
+		carl9170_handle_command_response(ar, cmd, cmd->hdr.len + 4);
+	}
+
+	if (unlikely(i != resplen)) {
+		if (!net_ratelimit())
+			return;
+
+		wiphy_err(ar->hw->wiphy, "malformed firmware trap:\n");
+		print_hex_dump_bytes("rxcmd:", DUMP_PREFIX_OFFSET,
+				     respbuf, resplen);
+	}
+}
+
+static void __carl9170_rx(struct ar9170 *ar, u8 *buf, unsigned int len)
+{
+	unsigned int i = 0;
+
+	/* weird thing, but this is the same in the original driver */
+	while (len > 2 && i < 12 && buf[0] == 0xff && buf[1] == 0xff) {
+		i += 2;
+		len -= 2;
+		buf += 2;
+	}
+
+	if (unlikely(len < 4))
+		return;
+
+	/* found the 6 * 0xffff marker? */
+	if (i == 12)
+		carl9170_rx_untie_cmds(ar, buf, len);
+	else
+		carl9170_handle_mpdu(ar, buf, len);
+}
+
+static void carl9170_rx_stream(struct ar9170 *ar, void *buf, unsigned int len)
+{
+	unsigned int tlen, wlen = 0, clen = 0;
+	struct ar9170_stream *rx_stream;
+	u8 *tbuf;
+
+	tbuf = buf;
+	tlen = len;
+
+	while (tlen >= 4) {
+		rx_stream = (void *) tbuf;
+		clen = le16_to_cpu(rx_stream->length);
+		wlen = ALIGN(clen, 4);
+
+		/* check if this is stream has a valid tag.*/
+		if (rx_stream->tag != cpu_to_le16(AR9170_RX_STREAM_TAG)) {
+			/*
+			 * TODO: handle the highly unlikely event that the
+			 * corrupted stream has the TAG at the right position.
+			 */
+
+			/* check if the frame can be repaired. */
+			if (!ar->rx_failover_missing) {
+
+				/* this is not "short read". */
+				if (net_ratelimit()) {
+					wiphy_err(ar->hw->wiphy,
+						"missing tag!\n");
+				}
+
+				__carl9170_rx(ar, tbuf, tlen);
+				return;
+			}
+
+			if (ar->rx_failover_missing > tlen) {
+				if (net_ratelimit()) {
+					wiphy_err(ar->hw->wiphy,
+						"possible multi "
+						"stream corruption!\n");
+					goto err_telluser;
+				} else {
+					goto err_silent;
+				}
+			}
+
+			memcpy(skb_put(ar->rx_failover, tlen), tbuf, tlen);
+			ar->rx_failover_missing -= tlen;
+
+			if (ar->rx_failover_missing <= 0) {
+				/*
+				 * nested carl9170_rx_stream call!
+				 *
+				 * termination is guranteed, even when the
+				 * combined frame also have an element with
+				 * a bad tag.
+				 */
+
+				ar->rx_failover_missing = 0;
+				carl9170_rx_stream(ar, ar->rx_failover->data,
+						   ar->rx_failover->len);
+
+				skb_reset_tail_pointer(ar->rx_failover);
+				skb_trim(ar->rx_failover, 0);
+			}
+
+			return;
+		}
+
+		/* check if stream is clipped */
+		if (wlen > tlen - 4) {
+			if (ar->rx_failover_missing) {
+				/* TODO: handle double stream corruption. */
+				if (net_ratelimit()) {
+					wiphy_err(ar->hw->wiphy, "double rx "
+						"stream corruption!\n");
+					goto err_telluser;
+				} else {
+					goto err_silent;
+				}
+			}
+
+			/*
+			 * save incomplete data set.
+			 * the firmware will resend the missing bits when
+			 * the rx - descriptor comes round again.
+			 */
+
+			memcpy(skb_put(ar->rx_failover, tlen), tbuf, tlen);
+			ar->rx_failover_missing = clen - tlen;
+			return;
+		}
+		__carl9170_rx(ar, rx_stream->payload, clen);
+
+		tbuf += wlen + 4;
+		tlen -= wlen + 4;
+	}
+
+	if (tlen) {
+		if (net_ratelimit()) {
+			wiphy_err(ar->hw->wiphy, "%d bytes of unprocessed "
+				"data left in rx stream!\n", tlen);
+		}
+
+		goto err_telluser;
+	}
+
+	return;
+
+err_telluser:
+	wiphy_err(ar->hw->wiphy, "damaged RX stream data [want:%d, "
+		"data:%d, rx:%d, pending:%d ]\n", clen, wlen, tlen,
+		ar->rx_failover_missing);
+
+	if (ar->rx_failover_missing)
+		print_hex_dump_bytes("rxbuf:", DUMP_PREFIX_OFFSET,
+				     ar->rx_failover->data,
+				     ar->rx_failover->len);
+
+	print_hex_dump_bytes("stream:", DUMP_PREFIX_OFFSET,
+			     buf, len);
+
+	wiphy_err(ar->hw->wiphy, "please check your hardware and cables, if "
+		"you see this message frequently.\n");
+
+err_silent:
+	if (ar->rx_failover_missing) {
+		skb_reset_tail_pointer(ar->rx_failover);
+		skb_trim(ar->rx_failover, 0);
+		ar->rx_failover_missing = 0;
+	}
+}
+
+void carl9170_rx(struct ar9170 *ar, void *buf, unsigned int len)
+{
+	if (ar->fw.rx_stream)
+		carl9170_rx_stream(ar, buf, len);
+	else
+		__carl9170_rx(ar, buf, len);
+}

drivers/net/wireless/ath/carl9170/tx.c

+/*
+ * Atheros CARL9170 driver
+ *
+ * 802.11 xmit & status routines
+ *
+ * Copyright 2008, Johannes Berg <johannes@sipsolutions.net>
+ * Copyright 2009, 2010, Christian Lamparter <chunkeey@googlemail.com>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; see the file COPYING.  If not, see
+ * http://www.gnu.org/licenses/.
+ *
+ * This file incorporates work covered by the following copyright and
+ * permission notice:
+ *    Copyright (c) 2007-2008 Atheros Communications, Inc.
+ *
+ *    Permission to use, copy, modify, and/or distribute this software for any
+ *    purpose with or without fee is hereby granted, provided that the above
+ *    copyright notice and this permission notice appear in all copies.
+ *
+ *    THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+ *    WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+ *    MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+ *    ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ *    WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+ *    ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+ *    OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ */
+
+#include <linux/init.h>
+#include <linux/slab.h>
+#include <linux/module.h>
+#include <linux/etherdevice.h>
+#include <net/mac80211.h>
+#include "carl9170.h"
+#include "hw.h"
+#include "cmd.h"
+
+static inline unsigned int __carl9170_get_queue(struct ar9170 *ar,
+						unsigned int queue)
+{
+	if (unlikely(modparam_noht)) {
+		return queue;
+	} else {
+		/*
+		 * This is just another workaround, until
+		 * someone figures out how to get QoS and
+		 * AMPDU to play nicely together.
+		 */
+
+		return 2;		/* AC_BE */
+	}
+}
+
+static inline unsigned int carl9170_get_queue(struct ar9170 *ar,
+					      struct sk_buff *skb)
+{
+	return __carl9170_get_queue(ar, skb_get_queue_mapping(skb));
+}
+
+static bool is_mem_full(struct ar9170 *ar)
+{
+	return (DIV_ROUND_UP(IEEE80211_MAX_FRAME_LEN, ar->fw.mem_block_size) >
+		atomic_read(&ar->mem_free_blocks));
+}
+
+static void carl9170_tx_accounting(struct ar9170 *ar, struct sk_buff *skb)
+{
+	int queue, i;
+	bool mem_full;
+
+	atomic_inc(&ar->tx_total_queued);
+
+	queue = skb_get_queue_mapping(skb);
+	spin_lock_bh(&ar->tx_stats_lock);
+
+	/*
+	 * The driver has to accept the frame, regardless if the queue is
+	 * full to the brim, or not. We have to do the queuing internally,
+	 * since mac80211 assumes that a driver which can operate with
+	 * aggregated frames does not reject frames for this reason.
+	 */
+	ar->tx_stats[queue].len++;
+	ar->tx_stats[queue].count++;
+
+	mem_full = is_mem_full(ar);
+	for (i = 0; i < ar->hw->queues; i++) {
+		if (mem_full || ar->tx_stats[i].len >= ar->tx_stats[i].limit) {
+			ieee80211_stop_queue(ar->hw, i);
+			ar->queue_stop_timeout[i] = jiffies;
+		}
+	}
+
+	spin_unlock_bh(&ar->tx_stats_lock);
+}
+
+static void carl9170_tx_accounting_free(struct ar9170 *ar, struct sk_buff *skb)
+{
+	struct ieee80211_tx_info *txinfo;
+	int queue;
+
+	txinfo = IEEE80211_SKB_CB(skb);
+	queue = skb_get_queue_mapping(skb);
+
+	spin_lock_bh(&ar->tx_stats_lock);
+
+	ar->tx_stats[queue].len--;
+
+	if (!is_mem_full(ar)) {
+		unsigned int i;
+		for (i = 0; i < ar->hw->queues; i++) {
+			if (ar->tx_stats[i].len >= CARL9170_NUM_TX_LIMIT_SOFT)
+				continue;
+
+			if (ieee80211_queue_stopped(ar->hw, i)) {
+				unsigned long tmp;
+
+				tmp = jiffies - ar->queue_stop_timeout[i];
+				if (tmp > ar->max_queue_stop_timeout[i])
+					ar->max_queue_stop_timeout[i] = tmp;
+			}
+
+			ieee80211_wake_queue(ar->hw, i);
+		}
+	}
+
+	spin_unlock_bh(&ar->tx_stats_lock);
+	if (atomic_dec_and_test(&ar->tx_total_queued))
+		complete(&ar->tx_flush);
+}
+
+static int carl9170_alloc_dev_space(struct ar9170 *ar, struct sk_buff *skb)
+{
+	struct _carl9170_tx_superframe *super = (void *) skb->data;
+	unsigned int chunks;
+	int cookie = -1;
+
+	atomic_inc(&ar->mem_allocs);
+
+	chunks = DIV_ROUND_UP(skb->len, ar->fw.mem_block_size);
+	if (unlikely(atomic_sub_return(chunks, &ar->mem_free_blocks) < 0)) {
+		atomic_add(chunks, &ar->mem_free_blocks);
+		return -ENOSPC;
+	}
+
+	spin_lock_bh(&ar->mem_lock);
+	cookie = bitmap_find_free_region(ar->mem_bitmap, ar->fw.mem_blocks, 0);
+	spin_unlock_bh(&ar->mem_lock);
+
+	if (unlikely(cookie < 0)) {
+		atomic_add(chunks, &ar->mem_free_blocks);
+		return -ENOSPC;
+	}
+
+	super = (void *) skb->data;
+
+	/*
+	 * Cookie #0 serves two special purposes:
+	 *  1. The firmware might use it generate BlockACK frames
+	 *     in responds of an incoming BlockAckReqs.
+	 *
+	 *  2. Prevent double-free bugs.
+	 */
+	super->s.cookie = (u8) cookie + 1;
+	return 0;
+}
+
+static void carl9170_release_dev_space(struct ar9170 *ar, struct sk_buff *skb)
+{
+	struct _carl9170_tx_superframe *super = (void *) skb->data;
+	int cookie;
+
+	/* make a local copy of the cookie */
+	cookie = super->s.cookie;
+	/* invalidate cookie */
+	super->s.cookie = 0;
+
+	/*
+	 * Do a out-of-bounds check on the cookie:
+	 *
+	 *  * cookie "0" is reserved and won't be assigned to any
+	 *    out-going frame. Internally however, it is used to
+	 *    mark no longer/un-accounted frames and serves as a
+	 *    cheap way of preventing frames from being freed
+	 *    twice by _accident_. NB: There is a tiny race...
+	 *
+	 *  * obviously, cookie number is limited by the amount
+	 *    of available memory blocks, so the number can
+	 *    never execeed the mem_blocks count.
+	 */
+	if (unlikely(WARN_ON_ONCE(cookie == 0) ||
+	    WARN_ON_ONCE(cookie > ar->fw.mem_blocks)))
+		return;
+
+	atomic_add(DIV_ROUND_UP(skb->len, ar->fw.mem_block_size),
+		   &ar->mem_free_blocks);
+
+	spin_lock_bh(&ar->mem_lock);
+	bitmap_release_region(ar->mem_bitmap, cookie - 1, 0);
+	spin_unlock_bh(&ar->mem_lock);
+}
+
+/* Called from any context */
+static void carl9170_tx_release(struct kref *ref)
+{
+	struct ar9170 *ar;
+	struct carl9170_tx_info *arinfo;
+	struct ieee80211_tx_info *txinfo;
+	struct sk_buff *skb;
+
+	arinfo = container_of(ref, struct carl9170_tx_info, ref);
+	txinfo = container_of((void *) arinfo, struct ieee80211_tx_info,
+			      rate_driver_data);
+	skb = container_of((void *) txinfo, struct sk_buff, cb);
+
+	ar = arinfo->ar;
+	if (WARN_ON_ONCE(!ar))
+		return;
+
+	BUILD_BUG_ON(
+	    offsetof(struct ieee80211_tx_info, status.ampdu_ack_len) != 23);
+
+	memset(&txinfo->status.ampdu_ack_len, 0,
+	       sizeof(struct ieee80211_tx_info) -
+	       offsetof(struct ieee80211_tx_info, status.ampdu_ack_len));
+
+	if (atomic_read(&ar->tx_total_queued))
+		ar->tx_schedule = true;
+
+	if (txinfo->flags & IEEE80211_TX_CTL_AMPDU) {
+		if (!atomic_read(&ar->tx_ampdu_upload))
+			ar->tx_ampdu_schedule = true;
+
+		if (txinfo->flags & IEEE80211_TX_STAT_AMPDU) {
+			txinfo->status.ampdu_len = txinfo->pad[0];
+			txinfo->status.ampdu_ack_len = txinfo->pad[1];
+			txinfo->pad[0] = txinfo->pad[1] = 0;
+		} else if (txinfo->flags & IEEE80211_TX_STAT_ACK) {
+			/*
+			 * drop redundant tx_status reports:
+			 *
+			 * 1. ampdu_ack_len of the final tx_status does
+			 *    include the feedback of this particular frame.
+			 *
+			 * 2. tx_status_irqsafe only queues up to 128
+			 *    tx feedback reports and discards the rest.
+			 *
+			 * 3. minstrel_ht is picky, it only accepts
+			 *    reports of frames with the TX_STATUS_AMPDU flag.
+			 */
+
+			dev_kfree_skb_any(skb);
+			return;
+		} else {
+			/*
+			 * Frame has failed, but we want to keep it in
+			 * case it was lost due to a power-state
+			 * transition.
+			 */
+		}
+	}
+
+	skb_pull(skb, sizeof(struct _carl9170_tx_superframe));
+	ieee80211_tx_status_irqsafe(ar->hw, skb);
+}
+
+void carl9170_tx_get_skb(struct sk_buff *skb)
+{
+	struct carl9170_tx_info *arinfo = (void *)
+		(IEEE80211_SKB_CB(skb))->rate_driver_data;
+	kref_get(&arinfo->ref);
+}
+
+int carl9170_tx_put_skb(struct sk_buff *skb)
+{
+	struct carl9170_tx_info *arinfo = (void *)
+		(IEEE80211_SKB_CB(skb))->rate_driver_data;
+
+	return kref_put(&arinfo->ref, carl9170_tx_release);
+}
+
+/* Caller must hold the tid_info->lock & rcu_read_lock */
+static void carl9170_tx_shift_bm(struct ar9170 *ar,
+	struct carl9170_sta_tid *tid_info, u16 seq)
+{
+	u16 off;
+
+	off = SEQ_DIFF(seq, tid_info->bsn);
+
+	if (WARN_ON_ONCE(off >= CARL9170_BAW_BITS))
+		return;
+
+	/*
+	 * Sanity check. For each MPDU we set the bit in bitmap and
+	 * clear it once we received the tx_status.
+	 * But if the bit is already cleared then we've been bitten
+	 * by a bug.
+	 */
+	WARN_ON_ONCE(!test_and_clear_bit(off, tid_info->bitmap));
+
+	off = SEQ_DIFF(tid_info->snx, tid_info->bsn);
+	if (WARN_ON_ONCE(off >= CARL9170_BAW_BITS))
+		return;
+
+	if (!bitmap_empty(tid_info->bitmap, off))
+		off = find_first_bit(tid_info->bitmap, off);
+
+	tid_info->bsn += off;
+	tid_info->bsn &= 0x0fff;
+
+	bitmap_shift_right(tid_info->bitmap, tid_info->bitmap,
+			   off, CARL9170_BAW_BITS);
+}
+
+static void carl9170_tx_status_process_ampdu(struct ar9170 *ar,
+	struct sk_buff *skb, struct ieee80211_tx_info *txinfo)
+{
+	struct _carl9170_tx_superframe *super = (void *) skb->data;
+	struct ieee80211_hdr *hdr = (void *) super->frame_data;
+	struct ieee80211_tx_info *tx_info;
+	struct carl9170_tx_info *ar_info;
+	struct carl9170_sta_info *sta_info;
+	struct ieee80211_sta *sta;
+	struct carl9170_sta_tid *tid_info;
+	struct ieee80211_vif *vif;
+	unsigned int vif_id;
+	u8 tid;
+
+	if (!(txinfo->flags & IEEE80211_TX_CTL_AMPDU) ||
+	    txinfo->flags & IEEE80211_TX_CTL_INJECTED)
+		return;
+
+	tx_info = IEEE80211_SKB_CB(skb);
+	ar_info = (void *) tx_info->rate_driver_data;
+
+	vif_id = (super->s.misc & CARL9170_TX_SUPER_MISC_VIF_ID) >>
+		 CARL9170_TX_SUPER_MISC_VIF_ID_S;
+
+	if (WARN_ON_ONCE(vif_id >= AR9170_MAX_VIRTUAL_MAC))
+		return;
+
+	rcu_read_lock();
+	vif = rcu_dereference(ar->vif_priv[vif_id].vif);
+	if (unlikely(!vif))
+		goto out_rcu;
+
+	/*
+	 * Normally we should use wrappers like ieee80211_get_DA to get
+	 * the correct peer ieee80211_sta.
+	 *
+	 * But there is a problem with indirect traffic (broadcasts, or
+	 * data which is designated for other stations) in station mode.
+	 * The frame will be directed to the AP for distribution and not
+	 * to the actual destination.
+	 */
+	sta = ieee80211_find_sta(vif, hdr->addr1);
+	if (unlikely(!sta))
+		goto out_rcu;
+
+	tid = get_tid_h(hdr);
+
+	sta_info = (void *) sta->drv_priv;
+	tid_info = rcu_dereference(sta_info->agg[tid]);
+	if (!tid_info)
+		goto out_rcu;
+
+	spin_lock_bh(&tid_info->lock);
+	if (likely(tid_info->state >= CARL9170_TID_STATE_IDLE))
+		carl9170_tx_shift_bm(ar, tid_info, get_seq_h(hdr));
+
+	if (sta_info->stats[tid].clear) {
+		sta_info->stats[tid].clear = false;
+		sta_info->stats[tid].ampdu_len = 0;
+		sta_info->stats[tid].ampdu_ack_len = 0;
+	}
+
+	sta_info->stats[tid].ampdu_len++;
+	if (txinfo->status.rates[0].count == 1)
+		sta_info->stats[tid].ampdu_ack_len++;
+
+	if (super->f.mac_control & cpu_to_le16(AR9170_TX_MAC_IMM_BA)) {
+		txinfo->pad[0] = sta_info->stats[tid].ampdu_len;
+		txinfo->pad[1] = sta_info->stats[tid].ampdu_ack_len;
+		txinfo->flags |= IEEE80211_TX_STAT_AMPDU;
+		sta_info->stats[tid].clear = true;
+	}
+	spin_unlock_bh(&tid_info->lock);
+
+out_rcu:
+	rcu_read_unlock();
+}
+
+void carl9170_tx_status(struct ar9170 *ar, struct sk_buff *skb,
+			const bool success)
+{
+	struct ieee80211_tx_info *txinfo;
+
+	carl9170_tx_accounting_free(ar, skb);
+
+	txinfo = IEEE80211_SKB_CB(skb);
+
+	if (success)
+		txinfo->flags |= IEEE80211_TX_STAT_ACK;
+	else
+		ar->tx_ack_failures++;
+
+	if (txinfo->flags & IEEE80211_TX_CTL_AMPDU)
+		carl9170_tx_status_process_ampdu(ar, skb, txinfo);
+
+	carl9170_tx_put_skb(skb);
+}
+
+/* This function may be called form any context */
+void carl9170_tx_callback(struct ar9170 *ar, struct sk_buff *skb)
+{
+	struct ieee80211_tx_info *txinfo = IEEE80211_SKB_CB(skb);
+
+	atomic_dec(&ar->tx_total_pending);
+
+	if (txinfo->flags & IEEE80211_TX_CTL_AMPDU)
+		atomic_dec(&ar->tx_ampdu_upload);
+
+	if (carl9170_tx_put_skb(skb))
+		tasklet_hi_schedule(&ar->usb_tasklet);
+}
+
+static struct sk_buff *carl9170_get_queued_skb(struct ar9170 *ar, u8 cookie,
+					       struct sk_buff_head *queue)
+{
+	struct sk_buff *skb;
+
+	spin_lock_bh(&queue->lock);
+	skb_queue_walk(queue, skb) {
+		struct _carl9170_tx_superframe *txc = (void *) skb->data;
+
+		if (txc->s.cookie != cookie)
+			continue;
+
+		__skb_unlink(skb, queue);
+		spin_unlock_bh(&queue->lock);
+
+		carl9170_release_dev_space(ar, skb);
+		return skb;
+	}
+	spin_unlock_bh(&queue->lock);
+
+	return NULL;
+}
+
+static void carl9170_tx_fill_rateinfo(struct ar9170 *ar, unsigned int rix,
+	unsigned int tries, struct ieee80211_tx_info *txinfo)
+{
+	unsigned int i;
+
+	for (i = 0; i < IEEE80211_TX_MAX_RATES; i++) {
+		if (txinfo->status.rates[i].idx < 0)
+			break;
+
+		if (i == rix) {
+			txinfo->status.rates[i].count = tries;
+			i++;
+			break;
+		}
+	}
+
+	for (; i < IEEE80211_TX_MAX_RATES; i++) {
+		txinfo->status.rates[i].idx = -1;
+		txinfo->status.rates[i].count = 0;
+	}
+}
+
+static void carl9170_check_queue_stop_timeout(struct ar9170 *ar)
+{
+	int i;
+	struct sk_buff *skb;
+	struct ieee80211_tx_info *txinfo;
+	struct carl9170_tx_info *arinfo;
+	bool restart = false;
+
+	for (i = 0; i < ar->hw->queues; i++) {
+		spin_lock_bh(&ar->tx_status[i].lock);
+
+		skb = skb_peek(&ar->tx_status[i]);
+
+		if (!skb)
+			goto next;
+
+		txinfo = IEEE80211_SKB_CB(skb);
+		arinfo = (void *) txinfo->rate_driver_data;
+
+		if (time_is_before_jiffies(arinfo->timeout +
+		    msecs_to_jiffies(CARL9170_QUEUE_STUCK_TIMEOUT)) == true)
+			restart = true;
+
+next:
+		spin_unlock_bh(&ar->tx_status[i].lock);
+	}
+
+	if (restart) {
+		/*
+		 * At least one queue has been stuck for long enough.
+		 * Give the device a kick and hope it gets back to
+		 * work.
+		 *
+		 * possible reasons may include:
+		 *  - frames got lost/corrupted (bad connection to the device)
+		 *  - stalled rx processing/usb controller hiccups
+		 *  - firmware errors/bugs
+		 *  - every bug you can think of.
+		 *  - all bugs you can't...
+		 *  - ...
+		 */
+		carl9170_restart(ar, CARL9170_RR_STUCK_TX);
+	}
+}
+
+void carl9170_tx_janitor(struct work_struct *work)
+{
+	struct ar9170 *ar = container_of(work, struct ar9170,
+					 tx_janitor.work);
+	if (!IS_STARTED(ar))
+		return;
+
+	ar->tx_janitor_last_run = jiffies;
+
+	carl9170_check_queue_stop_timeout(ar);
+
+	if (!atomic_read(&ar->tx_total_queued))
+		return;
+
+	ieee80211_queue_delayed_work(ar->hw, &ar->tx_janitor,
+		msecs_to_jiffies(CARL9170_TX_TIMEOUT));
+}
+
+static void __carl9170_tx_process_status(struct ar9170 *ar,
+	const uint8_t cookie, const uint8_t info)
+{
+	struct sk_buff *skb;
+	struct ieee80211_tx_info *txinfo;
+	struct carl9170_tx_info *arinfo;
+	unsigned int r, t, q;
+	bool success = true;
+
+	q = ar9170_qmap[info & CARL9170_TX_STATUS_QUEUE];
+
+	skb = carl9170_get_queued_skb(ar, cookie, &ar->tx_status[q]);
+	if (!skb) {
+		/*
+		 * We have lost the race to another thread.
+		 */
+
+		return ;
+	}
+
+	txinfo = IEEE80211_SKB_CB(skb);
+	arinfo = (void *) txinfo->rate_driver_data;
+
+	if (!(info & CARL9170_TX_STATUS_SUCCESS))
+		success = false;
+
+	r = (info & CARL9170_TX_STATUS_RIX) >> CARL9170_TX_STATUS_RIX_S;
+	t = (info & CARL9170_TX_STATUS_TRIES) >> CARL9170_TX_STATUS_TRIES_S;
+
+	carl9170_tx_fill_rateinfo(ar, r, t, txinfo);
+	carl9170_tx_status(ar, skb, success);
+}
+
+void carl9170_tx_process_status(struct ar9170 *ar,
+				const struct carl9170_rsp *cmd)
+{
+	unsigned int i;
+
+	for (i = 0;  i < cmd->hdr.ext; i++) {
+		if (WARN_ON(i > ((cmd->hdr.len / 2) + 1))) {
+			print_hex_dump_bytes("UU:", DUMP_PREFIX_NONE,
+					     (void *) cmd, cmd->hdr.len + 4);
+			break;
+		}
+
+		__carl9170_tx_process_status(ar, cmd->_tx_status[i].cookie,
+					     cmd->_tx_status[i].info);
+	}
+}
+
+static __le32 carl9170_tx_physet(struct ar9170 *ar,
+	struct ieee80211_tx_info *info, struct ieee80211_tx_rate *txrate)
+{
+	struct ieee80211_rate *rate = NULL;
+	u32 power, chains;
+	__le32 tmp;
+
+	tmp = cpu_to_le32(0);
+
+	if (txrate->flags & IEEE80211_TX_RC_40_MHZ_WIDTH)
+		tmp |= cpu_to_le32(AR9170_TX_PHY_BW_40MHZ <<
+			AR9170_TX_PHY_BW_S);
+	/* this works because 40 MHz is 2 and dup is 3 */
+	if (txrate->flags & IEEE80211_TX_RC_DUP_DATA)
+		tmp |= cpu_to_le32(AR9170_TX_PHY_BW_40MHZ_DUP <<
+			AR9170_TX_PHY_BW_S);
+
+	if (txrate->flags & IEEE80211_TX_RC_SHORT_GI)
+		tmp |= cpu_to_le32(AR9170_TX_PHY_SHORT_GI);
+
+	if (txrate->flags & IEEE80211_TX_RC_MCS) {
+		u32 r = txrate->idx;
+		u8 *txpower;
+
+		/* heavy clip control */
+		tmp |= cpu_to_le32((r & 0x7) <<
+			AR9170_TX_PHY_TX_HEAVY_CLIP_S);
+
+		if (txrate->flags & IEEE80211_TX_RC_40_MHZ_WIDTH) {
+			if (info->band == IEEE80211_BAND_5GHZ)
+				txpower = ar->power_5G_ht40;
+			else
+				txpower = ar->power_2G_ht40;
+		} else {
+			if (info->band == IEEE80211_BAND_5GHZ)
+				txpower = ar->power_5G_ht20;
+			else
+				txpower = ar->power_2G_ht20;
+		}
+
+		power = txpower[r & 7];
+
+		/* +1 dBm for HT40 */
+		if (txrate->flags & IEEE80211_TX_RC_40_MHZ_WIDTH)
+			power += 2;
+
+		r <<= AR9170_TX_PHY_MCS_S;
+		BUG_ON(r & ~AR9170_TX_PHY_MCS);
+
+		tmp |= cpu_to_le32(r & AR9170_TX_PHY_MCS);
+		tmp |= cpu_to_le32(AR9170_TX_PHY_MOD_HT);
+
+		/*
+		 * green field preamble does not work.
+		 *
+		 * if (txrate->flags & IEEE80211_TX_RC_GREEN_FIELD)
+		 * tmp |= cpu_to_le32(AR9170_TX_PHY_GREENFIELD);
+		 */
+	} else {
+		u8 *txpower;
+		u32 mod;
+		u32 phyrate;
+		u8 idx = txrate->idx;
+
+		if (info->band != IEEE80211_BAND_2GHZ) {
+			idx += 4;
+			txpower = ar->power_5G_leg;
+			mod = AR9170_TX_PHY_MOD_OFDM;
+		} else {
+			if (idx < 4) {
+				txpower = ar->power_2G_cck;
+				mod = AR9170_TX_PHY_MOD_CCK;
+			} else {
+				mod = AR9170_TX_PHY_MOD_OFDM;
+				txpower = ar->power_2G_ofdm;
+			}
+		}
+
+		rate = &__carl9170_ratetable[idx];
+
+		phyrate = rate->hw_value & 0xF;
+		power = txpower[(rate->hw_value & 0x30) >> 4];
+		phyrate <<= AR9170_TX_PHY_MCS_S;
+
+		tmp |= cpu_to_le32(mod);
+		tmp |= cpu_to_le32(phyrate);
+
+		/*
+		 * short preamble seems to be broken too.
+		 *
+		 * if (txrate->flags & IEEE80211_TX_RC_USE_SHORT_PREAMBLE)
+		 *	tmp |= cpu_to_le32(AR9170_TX_PHY_SHORT_PREAMBLE);
+		 */
+	}
+	power <<= AR9170_TX_PHY_TX_PWR_S;
+	power &= AR9170_TX_PHY_TX_PWR;
+	tmp |= cpu_to_le32(power);
+
+	/* set TX chains */
+	if (ar->eeprom.tx_mask == 1) {
+		chains = AR9170_TX_PHY_TXCHAIN_1;
+	} else {
+		chains = AR9170_TX_PHY_TXCHAIN_2;
+
+		/* >= 36M legacy OFDM - use only one chain */
+		if (rate && rate->bitrate >= 360 &&
+		    !(txrate->flags & IEEE80211_TX_RC_MCS))
+			chains = AR9170_TX_PHY_TXCHAIN_1;
+	}
+	tmp |= cpu_to_le32(chains << AR9170_TX_PHY_TXCHAIN_S);
+
+	return tmp;
+}
+
+static bool carl9170_tx_rts_check(struct ar9170 *ar,
+				  struct ieee80211_tx_rate *rate,
+				  bool ampdu, bool multi)
+{
+	switch (ar->erp_mode) {
+	case CARL9170_ERP_AUTO:
+		if (ampdu)
+			break;
+
+	case CARL9170_ERP_MAC80211:
+		if (!(rate->flags & IEEE80211_TX_RC_USE_RTS_CTS))
+			break;
+
+	case CARL9170_ERP_RTS:
+		if (likely(!multi))
+			return true;
+
+	default:
+		break;
+	}
+
+	return false;
+}
+
+static bool carl9170_tx_cts_check(struct ar9170 *ar,
+				  struct ieee80211_tx_rate *rate)
+{
+	switch (ar->erp_mode) {
+	case CARL9170_ERP_AUTO:
+	case CARL9170_ERP_MAC80211:
+		if (!(rate->flags & IEEE80211_TX_RC_USE_CTS_PROTECT))
+			break;
+
+	case CARL9170_ERP_CTS:
+		return true;
+
+	default:
+		break;
+	}
+
+	return false;
+}
+
+static int carl9170_tx_prepare(struct ar9170 *ar, struct sk_buff *skb)
+{
+	struct ieee80211_hdr *hdr;
+	struct _carl9170_tx_superframe *txc;
+	struct carl9170_vif_info *cvif;
+	struct ieee80211_tx_info *info;
+	struct ieee80211_tx_rate *txrate;
+	struct ieee80211_sta *sta;
+	struct carl9170_tx_info *arinfo;
+	unsigned int hw_queue;
+	int i;
+	u16 keytype = 0;
+	u16 len, icv = 0;
+	bool ampdu, no_ack;
+
+	BUILD_BUG_ON(sizeof(*arinfo) > sizeof(info->rate_driver_data));
+	BUILD_BUG_ON(sizeof(struct _carl9170_tx_superdesc) !=
+		     CARL9170_TX_SUPERDESC_LEN);
+
+	BUILD_BUG_ON(sizeof(struct _ar9170_tx_hwdesc) !=
+		     AR9170_TX_HWDESC_LEN);
+
+	BUILD_BUG_ON(IEEE80211_TX_MAX_RATES < CARL9170_TX_MAX_RATES);
+
+	hw_queue = ar9170_qmap[carl9170_get_queue(ar, skb)];
+
+	hdr = (void *)skb->data;
+	info = IEEE80211_SKB_CB(skb);
+	len = skb->len;
+
+	/*
+	 * Note: If the frame was sent through a monitor interface,
+	 * the ieee80211_vif pointer can be NULL.
+	 */
+	if (likely(info->control.vif))
+		cvif = (void *) info->control.vif->drv_priv;
+	else
+		cvif = NULL;
+
+	sta = info->control.sta;
+
+	txc = (void *)skb_push(skb, sizeof(*txc));
+	memset(txc, 0, sizeof(*txc));
+
+	ampdu = !!(info->flags & IEEE80211_TX_CTL_AMPDU);
+	no_ack = !!(info->flags & IEEE80211_TX_CTL_NO_ACK);
+
+	if (info->control.hw_key) {
+		icv = info->control.hw_key->icv_len;
+
+		switch (info->control.hw_key->cipher) {
+		case WLAN_CIPHER_SUITE_WEP40:
+		case WLAN_CIPHER_SUITE_WEP104:
+		case WLAN_CIPHER_SUITE_TKIP:
+			keytype = AR9170_TX_MAC_ENCR_RC4;
+			break;
+		case WLAN_CIPHER_SUITE_CCMP:
+			keytype = AR9170_TX_MAC_ENCR_AES;
+			break;
+		default:
+			WARN_ON(1);
+			goto err_out;
+		}
+	}
+
+	BUILD_BUG_ON(AR9170_MAX_VIRTUAL_MAC >
+		((CARL9170_TX_SUPER_MISC_VIF_ID >>
+		 CARL9170_TX_SUPER_MISC_VIF_ID_S) + 1));
+
+	txc->s.len = cpu_to_le16(len + sizeof(*txc));
+	txc->f.length = cpu_to_le16(len + icv + 4);
+	SET_VAL(CARL9170_TX_SUPER_MISC_VIF_ID, txc->s.misc,
+		cvif ? cvif->id : 0);
+
+	txc->f.mac_control = cpu_to_le16(AR9170_TX_MAC_HW_DURATION |
+					 AR9170_TX_MAC_BACKOFF);
+
+	SET_VAL(CARL9170_TX_SUPER_MISC_QUEUE, txc->s.misc, hw_queue);
+
+	txc->f.mac_control |= cpu_to_le16(hw_queue << AR9170_TX_MAC_QOS_S);
+	txc->f.mac_control |= cpu_to_le16(keytype);
+	txc->f.phy_control = cpu_to_le32(0);
+
+	if (no_ack)
+		txc->f.mac_control |= cpu_to_le16(AR9170_TX_MAC_NO_ACK);
+
+	if (info->flags & IEEE80211_TX_CTL_SEND_AFTER_DTIM)
+		txc->s.misc |= CARL9170_TX_SUPER_MISC_CAB;
+
+	txrate = &info->control.rates[0];
+	if (carl9170_tx_rts_check(ar, txrate, ampdu, no_ack))
+		txc->f.mac_control |= cpu_to_le16(AR9170_TX_MAC_PROT_RTS);
+	else if (carl9170_tx_cts_check(ar, txrate))
+		txc->f.mac_control |= cpu_to_le16(AR9170_TX_MAC_PROT_CTS);
+
+	SET_VAL(CARL9170_TX_SUPER_RI_TRIES, txc->s.ri[0], txrate->count);
+	txc->f.phy_control |= carl9170_tx_physet(ar, info, txrate);
+
+	if (info->flags & IEEE80211_TX_CTL_AMPDU) {
+		for (i = 1; i < CARL9170_TX_MAX_RATES; i++) {
+			txrate = &info->control.rates[i];
+			if (txrate->idx >= 0)
+				continue;
+
+			txrate->idx = 0;
+			txrate->count = ar->hw->max_rate_tries;
+		}
+	}
+
+	/*
+	 * NOTE: For the first rate, the ERP & AMPDU flags are directly
+	 * taken from mac_control. For all fallback rate, the firmware
+	 * updates the mac_control flags from the rate info field.
+	 */
+	for (i = 1; i < CARL9170_TX_MAX_RATES; i++) {
+		txrate = &info->control.rates[i];
+		if (txrate->idx < 0)
+			break;
+
+		SET_VAL(CARL9170_TX_SUPER_RI_TRIES, txc->s.ri[i],
+			txrate->count);
+
+		if (carl9170_tx_rts_check(ar, txrate, ampdu, no_ack))
+			txc->s.ri[i] |= (AR9170_TX_MAC_PROT_RTS <<
+				CARL9170_TX_SUPER_RI_ERP_PROT_S);
+		else if (carl9170_tx_cts_check(ar, txrate))
+			txc->s.ri[i] |= (AR9170_TX_MAC_PROT_CTS <<
+				CARL9170_TX_SUPER_RI_ERP_PROT_S);
+
+		/*
+		 * unaggregated fallback, in case aggregation
+		 * proves to be unsuccessful and unreliable.
+		 */
+		if (ampdu && i < 3)
+			txc->s.ri[i] |= CARL9170_TX_SUPER_RI_AMPDU;
+
+		txc->s.rr[i - 1] = carl9170_tx_physet(ar, info, txrate);
+	}
+
+	if (ieee80211_is_probe_resp(hdr->frame_control))
+		txc->s.misc |= CARL9170_TX_SUPER_MISC_FILL_IN_TSF;
+
+	if (ampdu) {
+		unsigned int density, factor;
+
+		if (unlikely(!sta || !cvif))
+			goto err_out;
+
+		density = info->control.sta->ht_cap.ampdu_density;
+		factor = info->control.sta->ht_cap.ampdu_factor;
+
+		if (density) {
+			/*
+			 * Watch out!
+			 *
+			 * Otus uses slightly different density values than
+			 * those from the 802.11n spec.
+			 */
+
+			density = max_t(unsigned int, density + 1, 7u);
+		}
+
+		factor = min_t(unsigned int, 1u, factor);
+
+		SET_VAL(CARL9170_TX_SUPER_AMPDU_DENSITY,
+			txc->s.ampdu_settings, density);
+
+		SET_VAL(CARL9170_TX_SUPER_AMPDU_FACTOR,
+			txc->s.ampdu_settings, factor);
+
+		if (info->control.rates[0].flags & IEEE80211_TX_RC_MCS) {
+			txc->f.mac_control |= cpu_to_le16(AR9170_TX_MAC_AGGR);
+		} else {
+			/*
+			 * Not sure if it's even possible to aggregate
+			 * non-ht rates with this HW.
+			 */
+			WARN_ON_ONCE(1);
+		}
+	}
+
+	arinfo = (void *)info->rate_driver_data;
+	arinfo->timeout = jiffies;
+	arinfo->ar = ar;
+	kref_init(&arinfo->ref);
+	return 0;
+
+err_out:
+	skb_pull(skb, sizeof(*txc));
+	return -EINVAL;
+}
+
+static void carl9170_set_immba(struct ar9170 *ar, struct sk_buff *skb)
+{
+	struct _carl9170_tx_superframe *super;
+
+	super = (void *) skb->data;
+	super->f.mac_control |= cpu_to_le16(AR9170_TX_MAC_IMM_BA);
+}
+
+static void carl9170_set_ampdu_params(struct ar9170 *ar, struct sk_buff *skb)
+{
+	struct _carl9170_tx_superframe *super;
+	int tmp;
+
+	super = (void *) skb->data;
+
+	tmp = (super->s.ampdu_settings & CARL9170_TX_SUPER_AMPDU_DENSITY) <<
+		CARL9170_TX_SUPER_AMPDU_DENSITY_S;
+
+	/*
+	 * If you haven't noticed carl9170_tx_prepare has already filled
+	 * in all ampdu spacing & factor parameters.
+	 * Now it's the time to check whenever the settings have to be
+	 * updated by the firmware, or if everything is still the same.
+	 *
+	 * There's no sane way to handle different density values with
+	 * this hardware, so we may as well just do the compare in the
+	 * driver.
+	 */
+
+	if (tmp != ar->current_density) {
+		ar->current_density = tmp;
+		super->s.ampdu_settings |=
+			CARL9170_TX_SUPER_AMPDU_COMMIT_DENSITY;
+	}
+
+	tmp = (super->s.ampdu_settings & CARL9170_TX_SUPER_AMPDU_FACTOR) <<
+		CARL9170_TX_SUPER_AMPDU_FACTOR_S;
+
+	if (tmp != ar->current_factor) {
+		ar->current_factor = tmp;
+		super->s.ampdu_settings |=
+			CARL9170_TX_SUPER_AMPDU_COMMIT_FACTOR;
+	}
+}
+
+static bool carl9170_tx_rate_check(struct ar9170 *ar, struct sk_buff *_dest,
+				   struct sk_buff *_src)
+{
+	struct _carl9170_tx_superframe *dest, *src;
+
+	dest = (void *) _dest->data;
+	src = (void *) _src->data;
+
+	/*
+	 * The mac80211 rate control algorithm expects that all MPDUs in
+	 * an AMPDU share the same tx vectors.
+	 * This is not really obvious right now, because the hardware
+	 * does the AMPDU setup according to its own rulebook.
+	 * Our nicely assembled, strictly monotonic increasing mpdu
+	 * chains will be broken up, mashed back together...
+	 */
+
+	return (dest->f.phy_control == src->f.phy_control);
+}
+
+static void carl9170_tx_ampdu(struct ar9170 *ar)
+{
+	struct sk_buff_head agg;
+	struct carl9170_sta_tid *tid_info;
+	struct sk_buff *skb, *first;
+	unsigned int i = 0, done_ampdus = 0;
+	u16 seq, queue, tmpssn;
+
+	atomic_inc(&ar->tx_ampdu_scheduler);
+	ar->tx_ampdu_schedule = false;
+
+	if (atomic_read(&ar->tx_ampdu_upload))
+		return;
+
+	if (!ar->tx_ampdu_list_len)
+		return;
+
+	__skb_queue_head_init(&agg);
+
+	rcu_read_lock();
+	tid_info = rcu_dereference(ar->tx_ampdu_iter);
+	if (WARN_ON_ONCE(!tid_info)) {
+		rcu_read_unlock();
+		return;
+	}
+
+retry:
+	list_for_each_entry_continue_rcu(tid_info, &ar->tx_ampdu_list, list) {
+		i++;
+
+		if (tid_info->state < CARL9170_TID_STATE_PROGRESS)
+			continue;
+
+		queue = TID_TO_WME_AC(tid_info->tid);
+
+		spin_lock_bh(&tid_info->lock);
+		if (tid_info->state != CARL9170_TID_STATE_XMIT) {
+			first = skb_peek(&tid_info->queue);
+			if (first) {
+				struct ieee80211_tx_info *txinfo;
+				struct carl9170_tx_info *arinfo;
+
+				txinfo = IEEE80211_SKB_CB(first);
+				arinfo = (void *) txinfo->rate_driver_data;
+
+				if (time_is_after_jiffies(arinfo->timeout +
+				    msecs_to_jiffies(CARL9170_QUEUE_TIMEOUT))
+				    == true)
+					goto processed;
+
+				/*
+				 * We've been waiting for the frame which
+				 * matches "snx" (start sequence of the
+				 * next aggregate) for some time now.
+				 *
+				 * But it never arrived. Therefore
+				 * jump to the next available frame
+				 * and kick-start the transmission.
+				 *
+				 * Note: This might induce odd latency
+				 * spikes because the receiver will be
+				 * waiting for the lost frame too.
+				 */
+				ar->tx_ampdu_timeout++;
+
+				tid_info->snx = carl9170_get_seq(first);
+				tid_info->state = CARL9170_TID_STATE_XMIT;
+			} else {
+				goto processed;
+			}
+		}
+
+		tid_info->counter++;
+		first = skb_peek(&tid_info->queue);
+		tmpssn = carl9170_get_seq(first);
+		seq = tid_info->snx;
+
+		if (unlikely(tmpssn != seq)) {
+			tid_info->state = CARL9170_TID_STATE_IDLE;
+
+			goto processed;
+		}
+
+		while ((skb = skb_peek(&tid_info->queue))) {
+			/* strict 0, 1, ..., n - 1, n frame sequence order */
+			if (unlikely(carl9170_get_seq(skb) != seq))
+				break;
+
+			/* don't upload more than AMPDU FACTOR allows. */
+			if (unlikely(SEQ_DIFF(tid_info->snx, tid_info->bsn) >=
+			    (tid_info->max - 1)))
+				break;
+
+			if (!carl9170_tx_rate_check(ar, skb, first))
+				break;
+
+			atomic_inc(&ar->tx_ampdu_upload);
+			tid_info->snx = seq = SEQ_NEXT(seq);
+			__skb_unlink(skb, &tid_info->queue);
+
+			__skb_queue_tail(&agg, skb);
+
+			if (skb_queue_len(&agg) >= CARL9170_NUM_TX_AGG_MAX)
+				break;
+		}
+
+		if (skb_queue_empty(&tid_info->queue) ||
+		    carl9170_get_seq(skb_peek(&tid_info->queue)) !=
+		    tid_info->snx) {
+			/*
+			 * stop TID, if A-MPDU frames are still missing,
+			 * or whenever the queue is empty.
+			 */
+
+			tid_info->state = CARL9170_TID_STATE_IDLE;
+		}
+		done_ampdus++;
+
+processed:
+		spin_unlock_bh(&tid_info->lock);
+
+		if (skb_queue_empty(&agg))
+			continue;
+
+		/* apply ampdu spacing & factor settings */
+		carl9170_set_ampdu_params(ar, skb_peek(&agg));
+
+		/* set aggregation push bit */
+		carl9170_set_immba(ar, skb_peek_tail(&agg));
+
+		spin_lock_bh(&ar->tx_pending[queue].lock);
+		skb_queue_splice_tail_init(&agg, &ar->tx_pending[queue]);
+		spin_unlock_bh(&ar->tx_pending[queue].lock);
+		ar->tx_schedule = true;
+	}
+	if ((done_ampdus++ == 0) && (i++ == 0))
+		goto retry;
+
+	rcu_assign_pointer(ar->tx_ampdu_iter, tid_info);
+	rcu_read_unlock();
+}
+
+static struct sk_buff *carl9170_tx_pick_skb(struct ar9170 *ar,
+					    struct sk_buff_head *queue)
+{
+	struct sk_buff *skb;
+	struct ieee80211_tx_info *info;
+	struct carl9170_tx_info *arinfo;
+
+	BUILD_BUG_ON(sizeof(*arinfo) > sizeof(info->rate_driver_data));
+
+	spin_lock_bh(&queue->lock);
+	skb = skb_peek(queue);
+	if (unlikely(!skb))
+		goto err_unlock;
+
+	if (carl9170_alloc_dev_space(ar, skb))
+		goto err_unlock;
+
+	__skb_unlink(skb, queue);
+	spin_unlock_bh(&queue->lock);
+
+	info = IEEE80211_SKB_CB(skb);
+	arinfo = (void *) info->rate_driver_data;
+
+	arinfo->timeout = jiffies;
+
+	/*
+	 * increase ref count to "2".
+	 * Ref counting is the easiest way to solve the race between
+	 * the the urb's completion routine: carl9170_tx_callback and
+	 * wlan tx status functions: carl9170_tx_status/janitor.
+	 */
+	carl9170_tx_get_skb(skb);
+
+	return skb;
+
+err_unlock:
+	spin_unlock_bh(&queue->lock);
+	return NULL;
+}
+
+void carl9170_tx_drop(struct ar9170 *ar, struct sk_buff *skb)
+{
+	struct _carl9170_tx_superframe *super;
+	uint8_t q = 0;
+
+	ar->tx_dropped++;
+
+	super = (void *)skb->data;
+	SET_VAL(CARL9170_TX_SUPER_MISC_QUEUE, q,
+		ar9170_qmap[carl9170_get_queue(ar, skb)]);
+	__carl9170_tx_process_status(ar, super->s.cookie, q);
+}
+
+static void carl9170_tx(struct ar9170 *ar)
+{
+	struct sk_buff *skb;
+	unsigned int i, q;
+	bool schedule_garbagecollector = false;
+
+	ar->tx_schedule = false;
+
+	if (unlikely(!IS_STARTED(ar)))
+		return;
+
+	carl9170_usb_handle_tx_err(ar);
+
+	for (i = 0; i < ar->hw->queues; i++) {
+		while (!skb_queue_empty(&ar->tx_pending[i])) {
+			skb = carl9170_tx_pick_skb(ar, &ar->tx_pending[i]);
+			if (unlikely(!skb))
+				break;
+
+			atomic_inc(&ar->tx_total_pending);
+
+			q = __carl9170_get_queue(ar, i);
+			/*
+			 * NB: tx_status[i] vs. tx_status[q],
+			 * TODO: Move into pick_skb or alloc_dev_space.
+			 */
+			skb_queue_tail(&ar->tx_status[q], skb);
+
+			carl9170_usb_tx(ar, skb);
+			schedule_garbagecollector = true;
+		}
+	}
+
+	if (!schedule_garbagecollector)
+		return;
+
+	ieee80211_queue_delayed_work(ar->hw, &ar->tx_janitor,
+		msecs_to_jiffies(CARL9170_TX_TIMEOUT));
+}
+
+static bool carl9170_tx_ampdu_queue(struct ar9170 *ar,
+	struct ieee80211_sta *sta, struct sk_buff *skb)
+{
+	struct carl9170_sta_info *sta_info;
+	struct carl9170_sta_tid *agg;
+	struct sk_buff *iter;
+	unsigned int max;
+	u16 tid, seq, qseq, off;
+	bool run = false;
+
+	tid = carl9170_get_tid(skb);
+	seq = carl9170_get_seq(skb);
+	sta_info = (void *) sta->drv_priv;
+
+	rcu_read_lock();
+	agg = rcu_dereference(sta_info->agg[tid]);
+	max = sta_info->ampdu_max_len;
+
+	if (!agg)
+		goto err_unlock_rcu;
+
+	spin_lock_bh(&agg->lock);
+	if (unlikely(agg->state < CARL9170_TID_STATE_IDLE))
+		goto err_unlock;
+
+	/* check if sequence is within the BA window */
+	if (unlikely(!BAW_WITHIN(agg->bsn, CARL9170_BAW_BITS, seq)))
+		goto err_unlock;
+
+	if (WARN_ON_ONCE(!BAW_WITHIN(agg->snx, CARL9170_BAW_BITS, seq)))
+		goto err_unlock;
+
+	off = SEQ_DIFF(seq, agg->bsn);
+	if (WARN_ON_ONCE(test_and_set_bit(off, agg->bitmap)))
+		goto err_unlock;
+
+	if (likely(BAW_WITHIN(agg->hsn, CARL9170_BAW_BITS, seq))) {
+		__skb_queue_tail(&agg->queue, skb);
+		agg->hsn = seq;
+		goto queued;
+	}
+
+	skb_queue_reverse_walk(&agg->queue, iter) {
+		qseq = carl9170_get_seq(iter);
+
+		if (BAW_WITHIN(qseq, CARL9170_BAW_BITS, seq)) {
+			__skb_queue_after(&agg->queue, iter, skb);
+			goto queued;
+		}
+	}
+
+	__skb_queue_head(&agg->queue, skb);
+queued:
+
+	if (unlikely(agg->state != CARL9170_TID_STATE_XMIT)) {
+		if (agg->snx == carl9170_get_seq(skb_peek(&agg->queue))) {
+			agg->state = CARL9170_TID_STATE_XMIT;
+			run = true;
+		}
+	}
+
+	spin_unlock_bh(&agg->lock);
+	rcu_read_unlock();
+
+	return run;
+
+err_unlock:
+	spin_unlock_bh(&agg->lock);
+
+err_unlock_rcu:
+	rcu_read_unlock();
+	carl9170_tx_status(ar, skb, false);
+	ar->tx_dropped++;
+	return false;
+}
+
+int carl9170_op_tx(struct ieee80211_hw *hw, struct sk_buff *skb)
+{
+	struct ar9170 *ar = hw->priv;
+	struct ieee80211_tx_info *info;
+	struct ieee80211_sta *sta;
+	bool run;
+
+	if (unlikely(!IS_STARTED(ar)))
+		goto err_free;
+
+	info = IEEE80211_SKB_CB(skb);
+	sta = info->control.sta;
+
+	if (unlikely(carl9170_tx_prepare(ar, skb)))
+		goto err_free;
+
+	carl9170_tx_accounting(ar, skb);
+	/*
+	 * from now on, one has to use carl9170_tx_status to free
+	 * all ressouces which are associated with the frame.
+	 */
+
+	if (info->flags & IEEE80211_TX_CTL_AMPDU) {
+		if (WARN_ON_ONCE(!sta))
+			goto err_free;
+
+		run = carl9170_tx_ampdu_queue(ar, sta, skb);
+		if (run)
+			carl9170_tx_ampdu(ar);
+
+	} else {
+		unsigned int queue = skb_get_queue_mapping(skb);
+
+		skb_queue_tail(&ar->tx_pending[queue], skb);
+	}
+
+	carl9170_tx(ar);
+	return NETDEV_TX_OK;
+
+err_free:
+	ar->tx_dropped++;
+	dev_kfree_skb_any(skb);
+	return NETDEV_TX_OK;
+}
+
+void carl9170_tx_scheduler(struct ar9170 *ar)
+{
+
+	if (ar->tx_ampdu_schedule)
+		carl9170_tx_ampdu(ar);
+
+	if (ar->tx_schedule)
+		carl9170_tx(ar);
+}

drivers/net/wireless/ath/carl9170/usb.c

+/*
+ * Atheros CARL9170 driver
+ *
+ * USB - frontend
+ *
+ * Copyright 2008, Johannes Berg <johannes@sipsolutions.net>
+ * Copyright 2009, 2010, Christian Lamparter <chunkeey@googlemail.com>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; see the file COPYING.  If not, see
+ * http://www.gnu.org/licenses/.
+ *
+ * This file incorporates work covered by the following copyright and
+ * permission notice:
+ *    Copyright (c) 2007-2008 Atheros Communications, Inc.
+ *
+ *    Permission to use, copy, modify, and/or distribute this software for any
+ *    purpose with or without fee is hereby granted, provided that the above
+ *    copyright notice and this permission notice appear in all copies.
+ *
+ *    THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+ *    WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+ *    MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+ *    ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ *    WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+ *    ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+ *    OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ */
+
+#include <linux/module.h>
+#include <linux/slab.h>
+#include <linux/usb.h>
+#include <linux/firmware.h>
+#include <linux/etherdevice.h>
+#include <linux/device.h>
+#include <net/mac80211.h>
+#include "carl9170.h"
+#include "cmd.h"
+#include "hw.h"
+#include "fwcmd.h"
+
+MODULE_AUTHOR("Johannes Berg <johannes@sipsolutions.net>");
+MODULE_AUTHOR("Christian Lamparter <chunkeey@googlemail.com>");
+MODULE_LICENSE("GPL");
+MODULE_DESCRIPTION("Atheros AR9170 802.11n USB wireless");
+MODULE_FIRMWARE(CARL9170FW_NAME);
+MODULE_ALIAS("ar9170usb");
+MODULE_ALIAS("arusb_lnx");
+
+/*
+ * Note:
+ *
+ * Always update our wiki's device list (located at:
+ * http://wireless.kernel.org/en/users/Drivers/ar9170/devices ),
+ * whenever you add a new device.
+ */
+static struct usb_device_id carl9170_usb_ids[] = {
+	/* Atheros 9170 */
+	{ USB_DEVICE(0x0cf3, 0x9170) },
+	/* Atheros TG121N */
+	{ USB_DEVICE(0x0cf3, 0x1001) },
+	/* TP-Link TL-WN821N v2 */
+	{ USB_DEVICE(0x0cf3, 0x1002), .driver_info = CARL9170_WPS_BUTTON |
+		 CARL9170_ONE_LED },
+	/* 3Com Dual Band 802.11n USB Adapter */
+	{ USB_DEVICE(0x0cf3, 0x1010) },
+	/* H3C Dual Band 802.11n USB Adapter */
+	{ USB_DEVICE(0x0cf3, 0x1011) },
+	/* Cace Airpcap NX */
+	{ USB_DEVICE(0xcace, 0x0300) },
+	/* D-Link DWA 160 A1 */
+	{ USB_DEVICE(0x07d1, 0x3c10) },
+	/* D-Link DWA 160 A2 */
+	{ USB_DEVICE(0x07d1, 0x3a09) },
+	/* Netgear WNA1000 */
+	{ USB_DEVICE(0x0846, 0x9040) },
+	/* Netgear WNDA3100 */
+	{ USB_DEVICE(0x0846, 0x9010) },
+	/* Netgear WN111 v2 */
+	{ USB_DEVICE(0x0846, 0x9001), .driver_info = CARL9170_ONE_LED },
+	/* Zydas ZD1221 */
+	{ USB_DEVICE(0x0ace, 0x1221) },
+	/* Proxim ORiNOCO 802.11n USB */
+	{ USB_DEVICE(0x1435, 0x0804) },
+	/* WNC Generic 11n USB Dongle */
+	{ USB_DEVICE(0x1435, 0x0326) },
+	/* ZyXEL NWD271N */
+	{ USB_DEVICE(0x0586, 0x3417) },
+	/* Z-Com UB81 BG */
+	{ USB_DEVICE(0x0cde, 0x0023) },
+	/* Z-Com UB82 ABG */
+	{ USB_DEVICE(0x0cde, 0x0026) },
+	/* Sphairon Homelink 1202 */
+	{ USB_DEVICE(0x0cde, 0x0027) },
+	/* Arcadyan WN7512 */
+	{ USB_DEVICE(0x083a, 0xf522) },
+	/* Planex GWUS300 */
+	{ USB_DEVICE(0x2019, 0x5304) },
+	/* IO-Data WNGDNUS2 */
+	{ USB_DEVICE(0x04bb, 0x093f) },
+	/* NEC WL300NU-G */
+	{ USB_DEVICE(0x0409, 0x0249) },
+	/* AVM FRITZ!WLAN USB Stick N */
+	{ USB_DEVICE(0x057c, 0x8401) },
+	/* AVM FRITZ!WLAN USB Stick N 2.4 */
+	{ USB_DEVICE(0x057c, 0x8402) },
+	/* Qwest/Actiontec 802AIN Wireless N USB Network Adapter */
+	{ USB_DEVICE(0x1668, 0x1200) },
+
+	/* terminate */
+	{}
+};
+MODULE_DEVICE_TABLE(usb, carl9170_usb_ids);
+
+static void carl9170_usb_submit_data_urb(struct ar9170 *ar)
+{
+	struct urb *urb;
+	int err;
+
+	if (atomic_inc_return(&ar->tx_anch_urbs) > AR9170_NUM_TX_URBS)
+		goto err_acc;
+
+	urb = usb_get_from_anchor(&ar->tx_wait);
+	if (!urb)
+		goto err_acc;
+
+	usb_anchor_urb(urb, &ar->tx_anch);
+
+	err = usb_submit_urb(urb, GFP_ATOMIC);
+	if (unlikely(err)) {
+		if (net_ratelimit()) {
+			dev_err(&ar->udev->dev, "tx submit failed (%d)\n",
+				urb->status);
+		}
+
+		usb_unanchor_urb(urb);
+		usb_anchor_urb(urb, &ar->tx_err);
+	}
+
+	usb_free_urb(urb);
+
+	if (likely(err == 0))
+		return;
+
+err_acc:
+	atomic_dec(&ar->tx_anch_urbs);
+}
+
+static void carl9170_usb_tx_data_complete(struct urb *urb)
+{
+	struct ar9170 *ar = (struct ar9170 *)
+	      usb_get_intfdata(usb_ifnum_to_if(urb->dev, 0));
+
+	if (WARN_ON_ONCE(!ar)) {
+		dev_kfree_skb_irq(urb->context);
+		return;
+	}
+
+	atomic_dec(&ar->tx_anch_urbs);
+
+	switch (urb->status) {
+	/* everything is fine */
+	case 0:
+		carl9170_tx_callback(ar, (void *)urb->context);
+		break;
+
+	/* disconnect */
+	case -ENOENT:
+	case -ECONNRESET:
+	case -ENODEV:
+	case -ESHUTDOWN:
+		/*
+		 * Defer the frame clean-up to the tasklet worker.
+		 * This is necessary, because carl9170_tx_drop
+		 * does not work in an irqsave context.
+		 */
+		usb_anchor_urb(urb, &ar->tx_err);
+		return;
+
+	/* a random transmission error has occurred? */
+	default:
+		if (net_ratelimit()) {
+			dev_err(&ar->udev->dev, "tx failed (%d)\n",
+				urb->status);
+		}
+
+		usb_anchor_urb(urb, &ar->tx_err);
+		break;
+	}
+
+	if (likely(IS_STARTED(ar)))
+		carl9170_usb_submit_data_urb(ar);
+}
+
+static int carl9170_usb_submit_cmd_urb(struct ar9170 *ar)
+{
+	struct urb *urb;
+	int err;
+
+	if (atomic_inc_return(&ar->tx_cmd_urbs) != 1) {
+		atomic_dec(&ar->tx_cmd_urbs);
+		return 0;
+	}
+
+	urb = usb_get_from_anchor(&ar->tx_cmd);
+	if (!urb) {
+		atomic_dec(&ar->tx_cmd_urbs);
+		return 0;
+	}
+
+	usb_anchor_urb(urb, &ar->tx_anch);
+	err = usb_submit_urb(urb, GFP_ATOMIC);
+	if (unlikely(err)) {
+		usb_unanchor_urb(urb);
+		atomic_dec(&ar->tx_cmd_urbs);
+	}
+	usb_free_urb(urb);
+
+	return err;
+}
+
+static void carl9170_usb_cmd_complete(struct urb *urb)
+{
+	struct ar9170 *ar = urb->context;
+	int err = 0;
+
+	if (WARN_ON_ONCE(!ar))
+		return;
+
+	atomic_dec(&ar->tx_cmd_urbs);
+
+	switch (urb->status) {
+	/* everything is fine */
+	case 0:
+		break;
+
+	/* disconnect */
+	case -ENOENT:
+	case -ECONNRESET:
+	case -ENODEV:
+	case -ESHUTDOWN:
+		return;
+
+	default:
+		err = urb->status;
+		break;
+	}
+
+	if (!IS_INITIALIZED(ar))
+		return;
+
+	if (err)
+		dev_err(&ar->udev->dev, "submit cmd cb failed (%d).\n", err);
+
+	err = carl9170_usb_submit_cmd_urb(ar);
+	if (err)
+		dev_err(&ar->udev->dev, "submit cmd failed (%d).\n", err);
+}
+
+static void carl9170_usb_rx_irq_complete(struct urb *urb)
+{
+	struct ar9170 *ar = urb->context;
+
+	if (WARN_ON_ONCE(!ar))
+		return;
+
+	switch (urb->status) {
+	/* everything is fine */
+	case 0:
+		break;
+
+	/* disconnect */
+	case -ENOENT:
+	case -ECONNRESET:
+	case -ENODEV:
+	case -ESHUTDOWN:
+		return;
+
+	default:
+		goto resubmit;
+	}
+
+	carl9170_handle_command_response(ar, urb->transfer_buffer,
+					 urb->actual_length);
+
+resubmit:
+	usb_anchor_urb(urb, &ar->rx_anch);
+	if (unlikely(usb_submit_urb(urb, GFP_ATOMIC)))
+		usb_unanchor_urb(urb);
+}
+
+static int carl9170_usb_submit_rx_urb(struct ar9170 *ar, gfp_t gfp)
+{
+	struct urb *urb;
+	int err = 0, runs = 0;
+
+	while ((atomic_read(&ar->rx_anch_urbs) < AR9170_NUM_RX_URBS) &&
+		(runs++ < AR9170_NUM_RX_URBS)) {
+		err = -ENOSPC;
+		urb = usb_get_from_anchor(&ar->rx_pool);
+		if (urb) {
+			usb_anchor_urb(urb, &ar->rx_anch);
+			err = usb_submit_urb(urb, gfp);
+			if (unlikely(err)) {
+				usb_unanchor_urb(urb);
+				usb_anchor_urb(urb, &ar->rx_pool);
+			} else {
+				atomic_dec(&ar->rx_pool_urbs);
+				atomic_inc(&ar->rx_anch_urbs);
+			}
+			usb_free_urb(urb);
+		}
+	}
+
+	return err;
+}
+
+static void carl9170_usb_rx_work(struct ar9170 *ar)
+{
+	struct urb *urb;
+	int i;
+
+	for (i = 0; i < AR9170_NUM_RX_URBS_POOL; i++) {
+		urb = usb_get_from_anchor(&ar->rx_work);
+		if (!urb)
+			break;
+
+		atomic_dec(&ar->rx_work_urbs);
+		if (IS_INITIALIZED(ar)) {
+			carl9170_rx(ar, urb->transfer_buffer,
+				    urb->actual_length);
+		}
+
+		usb_anchor_urb(urb, &ar->rx_pool);
+		atomic_inc(&ar->rx_pool_urbs);
+
+		usb_free_urb(urb);
+
+		carl9170_usb_submit_rx_urb(ar, GFP_ATOMIC);
+	}
+}
+
+void carl9170_usb_handle_tx_err(struct ar9170 *ar)
+{
+	struct urb *urb;
+
+	while ((urb = usb_get_from_anchor(&ar->tx_err))) {
+		struct sk_buff *skb = (void *)urb->context;
+
+		carl9170_tx_drop(ar, skb);
+		carl9170_tx_callback(ar, skb);
+		usb_free_urb(urb);
+	}
+}
+
+static void carl9170_usb_tasklet(unsigned long data)
+{
+	struct ar9170 *ar = (struct ar9170 *) data;
+
+	carl9170_usb_rx_work(ar);
+
+	/*
+	 * Strictly speaking: The tx scheduler is not part of the USB system.
+	 * But the rx worker returns frames back to the mac80211-stack and
+	 * this is the _perfect_ place to generate the next transmissions.
+	 */
+	if (IS_STARTED(ar))
+		carl9170_tx_scheduler(ar);
+}
+
+static void carl9170_usb_rx_complete(struct urb *urb)
+{
+	struct ar9170 *ar = (struct ar9170 *)urb->context;
+	int err;
+
+	if (WARN_ON_ONCE(!ar))
+		return;
+
+	atomic_dec(&ar->rx_anch_urbs);
+
+	switch (urb->status) {
+	case 0:
+		/* rx path */
+		usb_anchor_urb(urb, &ar->rx_work);
+		atomic_inc(&ar->rx_work_urbs);
+		break;
+
+	case -ENOENT:
+	case -ECONNRESET:
+	case -ENODEV:
+	case -ESHUTDOWN:
+		/* handle disconnect events*/
+		return;
+
+	default:
+		/* handle all other errors */
+		usb_anchor_urb(urb, &ar->rx_pool);
+		atomic_inc(&ar->rx_pool_urbs);
+		break;
+	}
+
+	err = carl9170_usb_submit_rx_urb(ar, GFP_ATOMIC);
+	if (unlikely(err)) {
+		/*
+		 * usb_submit_rx_urb reported a problem.
+		 * In case this is due to a rx buffer shortage,
+		 * elevate the tasklet worker priority to
+		 * the highest available level.
+		 */
+		tasklet_hi_schedule(&ar->usb_tasklet);
+
+		if (atomic_read(&ar->rx_anch_urbs) == 0) {
+			/*
+			 * The system is too slow to cope with
+			 * the enormous workload. We have simply
+			 * run out of active rx urbs and this
+			 * unfortunatly leads to an unpredictable
+			 * device.
+			 */
+
+			carl9170_restart(ar, CARL9170_RR_SLOW_SYSTEM);
+		}
+	} else {
+		/*
+		 * Using anything less than _high_ priority absolutely
+		 * kills the rx performance my UP-System...
+		 */
+		tasklet_hi_schedule(&ar->usb_tasklet);
+	}
+}
+
+static struct urb *carl9170_usb_alloc_rx_urb(struct ar9170 *ar, gfp_t gfp)
+{
+	struct urb *urb;
+	void *buf;
+
+	buf = kmalloc(ar->fw.rx_size, gfp);
+	if (!buf)
+		return NULL;
+
+	urb = usb_alloc_urb(0, gfp);
+	if (!urb) {
+		kfree(buf);
+		return NULL;
+	}
+
+	usb_fill_bulk_urb(urb, ar->udev, usb_rcvbulkpipe(ar->udev,
+			  AR9170_USB_EP_RX), buf, ar->fw.rx_size,
+			  carl9170_usb_rx_complete, ar);
+
+	urb->transfer_flags |= URB_FREE_BUFFER;
+
+	return urb;
+}
+
+static int carl9170_usb_send_rx_irq_urb(struct ar9170 *ar)
+{
+	struct urb *urb = NULL;
+	void *ibuf;
+	int err = -ENOMEM;
+
+	urb = usb_alloc_urb(0, GFP_KERNEL);
+	if (!urb)
+		goto out;
+
+	ibuf = kmalloc(AR9170_USB_EP_CTRL_MAX, GFP_KERNEL);
+	if (!ibuf)
+		goto out;
+
+	usb_fill_int_urb(urb, ar->udev, usb_rcvintpipe(ar->udev,
+			 AR9170_USB_EP_IRQ), ibuf, AR9170_USB_EP_CTRL_MAX,
+			 carl9170_usb_rx_irq_complete, ar, 1);
+
+	urb->transfer_flags |= URB_FREE_BUFFER;
+
+	usb_anchor_urb(urb, &ar->rx_anch);
+	err = usb_submit_urb(urb, GFP_KERNEL);
+	if (err)
+		usb_unanchor_urb(urb);
+
+out:
+	usb_free_urb(urb);
+	return err;
+}
+
+static int carl9170_usb_init_rx_bulk_urbs(struct ar9170 *ar)
+{
+	struct urb *urb;
+	int i, err = -EINVAL;
+
+	/*
+	 * The driver actively maintains a second shadow
+	 * pool for inactive, but fully-prepared rx urbs.
+	 *
+	 * The pool should help the driver to master huge
+	 * workload spikes without running the risk of
+	 * undersupplying the hardware or wasting time by
+	 * processing rx data (streams) inside the urb
+	 * completion (hardirq context).
+	 */
+	for (i = 0; i < AR9170_NUM_RX_URBS_POOL; i++) {
+		urb = carl9170_usb_alloc_rx_urb(ar, GFP_KERNEL);
+		if (!urb) {
+			err = -ENOMEM;
+			goto err_out;
+		}
+
+		usb_anchor_urb(urb, &ar->rx_pool);
+		atomic_inc(&ar->rx_pool_urbs);
+		usb_free_urb(urb);
+	}
+
+	err = carl9170_usb_submit_rx_urb(ar, GFP_KERNEL);
+	if (err)
+		goto err_out;
+
+	/* the device now waiting for the firmware. */
+	carl9170_set_state_when(ar, CARL9170_STOPPED, CARL9170_IDLE);
+	return 0;
+
+err_out:
+
+	usb_scuttle_anchored_urbs(&ar->rx_pool);
+	usb_scuttle_anchored_urbs(&ar->rx_work);
+	usb_kill_anchored_urbs(&ar->rx_anch);
+	return err;
+}
+
+static int carl9170_usb_flush(struct ar9170 *ar)
+{
+	struct urb *urb;
+	int ret, err = 0;
+
+	while ((urb = usb_get_from_anchor(&ar->tx_wait))) {
+		struct sk_buff *skb = (void *)urb->context;
+		carl9170_tx_drop(ar, skb);
+		carl9170_tx_callback(ar, skb);
+		usb_free_urb(urb);
+	}
+
+	ret = usb_wait_anchor_empty_timeout(&ar->tx_cmd, HZ);
+	if (ret == 0)
+		err = -ETIMEDOUT;
+
+	/* lets wait a while until the tx - queues are dried out */
+	ret = usb_wait_anchor_empty_timeout(&ar->tx_anch, HZ);
+	if (ret == 0)
+		err = -ETIMEDOUT;
+
+	usb_kill_anchored_urbs(&ar->tx_anch);
+	carl9170_usb_handle_tx_err(ar);
+
+	return err;
+}
+
+static void carl9170_usb_cancel_urbs(struct ar9170 *ar)
+{
+	int err;
+
+	carl9170_set_state(ar, CARL9170_UNKNOWN_STATE);
+
+	err = carl9170_usb_flush(ar);
+	if (err)
+		dev_err(&ar->udev->dev, "stuck tx urbs!\n");
+
+	usb_poison_anchored_urbs(&ar->tx_anch);
+	carl9170_usb_handle_tx_err(ar);
+	usb_poison_anchored_urbs(&ar->rx_anch);
+
+	tasklet_kill(&ar->usb_tasklet);
+
+	usb_scuttle_anchored_urbs(&ar->rx_work);
+	usb_scuttle_anchored_urbs(&ar->rx_pool);
+	usb_scuttle_anchored_urbs(&ar->tx_cmd);
+}
+
+int __carl9170_exec_cmd(struct ar9170 *ar, struct carl9170_cmd *cmd,
+			const bool free_buf)
+{
+	struct urb *urb;
+
+	if (!IS_INITIALIZED(ar))
+		return -EPERM;
+
+	if (WARN_ON(cmd->hdr.len > CARL9170_MAX_CMD_LEN - 4))
+		return -EINVAL;
+
+	urb = usb_alloc_urb(0, GFP_ATOMIC);
+	if (!urb)
+		return -ENOMEM;
+
+	usb_fill_int_urb(urb, ar->udev, usb_sndintpipe(ar->udev,
+		AR9170_USB_EP_CMD), cmd, cmd->hdr.len + 4,
+		carl9170_usb_cmd_complete, ar, 1);
+
+	urb->transfer_flags |= URB_ZERO_PACKET;
+
+	if (free_buf)
+		urb->transfer_flags |= URB_FREE_BUFFER;
+
+	usb_anchor_urb(urb, &ar->tx_cmd);
+	usb_free_urb(urb);
+
+	return carl9170_usb_submit_cmd_urb(ar);
+}
+
+int carl9170_exec_cmd(struct ar9170 *ar, const enum carl9170_cmd_oids cmd,
+	unsigned int plen, void *payload, unsigned int outlen, void *out)
+{
+	int err = -ENOMEM;
+
+	if (!IS_ACCEPTING_CMD(ar))
+		return -EIO;
+
+	if (!(cmd & CARL9170_CMD_ASYNC_FLAG))
+		might_sleep();
+
+	ar->cmd.hdr.len = plen;
+	ar->cmd.hdr.cmd = cmd;
+	/* writing multiple regs fills this buffer already */
+	if (plen && payload != (u8 *)(ar->cmd.data))
+		memcpy(ar->cmd.data, payload, plen);
+
+	spin_lock_bh(&ar->cmd_lock);
+	ar->readbuf = (u8 *)out;
+	ar->readlen = outlen;
+	spin_unlock_bh(&ar->cmd_lock);
+
+	err = __carl9170_exec_cmd(ar, &ar->cmd, false);
+
+	if (!(cmd & CARL9170_CMD_ASYNC_FLAG)) {
+		err = wait_for_completion_timeout(&ar->cmd_wait, HZ);
+		if (err == 0) {
+			err = -ETIMEDOUT;
+			goto err_unbuf;
+		}
+
+		if (ar->readlen != outlen) {
+			err = -EMSGSIZE;
+			goto err_unbuf;
+		}
+	}
+
+	return 0;
+
+err_unbuf:
+	/* Maybe the device was removed in the moment we were waiting? */
+	if (IS_STARTED(ar)) {
+		dev_err(&ar->udev->dev, "no command feedback "
+			"received (%d).\n", err);
+
+		/* provide some maybe useful debug information */
+		print_hex_dump_bytes("carl9170 cmd: ", DUMP_PREFIX_NONE,
+				     &ar->cmd, plen + 4);
+
+		carl9170_restart(ar, CARL9170_RR_COMMAND_TIMEOUT);
+	}
+
+	/* invalidate to avoid completing the next command prematurely */
+	spin_lock_bh(&ar->cmd_lock);
+	ar->readbuf = NULL;
+	ar->readlen = 0;
+	spin_unlock_bh(&ar->cmd_lock);
+
+	return err;
+}
+
+void carl9170_usb_tx(struct ar9170 *ar, struct sk_buff *skb)
+{
+	struct urb *urb;
+	struct ar9170_stream *tx_stream;
+	void *data;
+	unsigned int len;
+
+	if (!IS_STARTED(ar))
+		goto err_drop;
+
+	urb = usb_alloc_urb(0, GFP_ATOMIC);
+	if (!urb)
+		goto err_drop;
+
+	if (ar->fw.tx_stream) {
+		tx_stream = (void *) (skb->data - sizeof(*tx_stream));
+
+		len = skb->len + sizeof(*tx_stream);
+		tx_stream->length = cpu_to_le16(len);
+		tx_stream->tag = cpu_to_le16(AR9170_TX_STREAM_TAG);
+		data = tx_stream;
+	} else {
+		data = skb->data;
+		len = skb->len;
+	}
+
+	usb_fill_bulk_urb(urb, ar->udev, usb_sndbulkpipe(ar->udev,
+		AR9170_USB_EP_TX), data, len,
+		carl9170_usb_tx_data_complete, skb);
+
+	urb->transfer_flags |= URB_ZERO_PACKET;
+
+	usb_anchor_urb(urb, &ar->tx_wait);
+
+	usb_free_urb(urb);
+
+	carl9170_usb_submit_data_urb(ar);
+	return;
+
+err_drop:
+	carl9170_tx_drop(ar, skb);
+	carl9170_tx_callback(ar, skb);
+}
+
+static void carl9170_release_firmware(struct ar9170 *ar)
+{
+	if (ar->fw.fw) {
+		release_firmware(ar->fw.fw);
+		memset(&ar->fw, 0, sizeof(ar->fw));
+	}
+}
+
+void carl9170_usb_stop(struct ar9170 *ar)
+{
+	int ret;
+
+	carl9170_set_state_when(ar, CARL9170_IDLE, CARL9170_STOPPED);
+
+	ret = carl9170_usb_flush(ar);
+	if (ret)
+		dev_err(&ar->udev->dev, "kill pending tx urbs.\n");
+
+	usb_poison_anchored_urbs(&ar->tx_anch);
+	carl9170_usb_handle_tx_err(ar);
+
+	/* kill any pending command */
+	spin_lock_bh(&ar->cmd_lock);
+	ar->readlen = 0;
+	spin_unlock_bh(&ar->cmd_lock);
+	complete_all(&ar->cmd_wait);
+
+	/* This is required to prevent an early completion on _start */
+	INIT_COMPLETION(ar->cmd_wait);
+
+	/*
+	 * Note:
+	 * So far we freed all tx urbs, but we won't dare to touch any rx urbs.
+	 * Else we would end up with a unresponsive device...
+	 */
+}
+
+int carl9170_usb_open(struct ar9170 *ar)
+{
+	usb_unpoison_anchored_urbs(&ar->tx_anch);
+
+	carl9170_set_state_when(ar, CARL9170_STOPPED, CARL9170_IDLE);
+	return 0;
+}
+
+static int carl9170_usb_load_firmware(struct ar9170 *ar)
+{
+	const u8 *data;
+	u8 *buf;
+	unsigned int transfer;
+	size_t len;
+	u32 addr;
+	int err = 0;
+
+	buf = kmalloc(4096, GFP_KERNEL);
+	if (!buf) {
+		err = -ENOMEM;
+		goto err_out;
+	}
+
+	data = ar->fw.fw->data;
+	len = ar->fw.fw->size;
+	addr = ar->fw.address;
+
+	/* this removes the miniboot image */
+	data += ar->fw.offset;
+	len -= ar->fw.offset;
+
+	while (len) {
+		transfer = min_t(unsigned int, len, 4096u);
+		memcpy(buf, data, transfer);
+
+		err = usb_control_msg(ar->udev, usb_sndctrlpipe(ar->udev, 0),
+				      0x30 /* FW DL */, 0x40 | USB_DIR_OUT,
+				      addr >> 8, 0, buf, transfer, 100);
+
+		if (err < 0) {
+			kfree(buf);
+			goto err_out;
+		}
+
+		len -= transfer;
+		data += transfer;
+		addr += transfer;
+	}
+	kfree(buf);
+
+	err = usb_control_msg(ar->udev, usb_sndctrlpipe(ar->udev, 0),
+			      0x31 /* FW DL COMPLETE */,
+			      0x40 | USB_DIR_OUT, 0, 0, NULL, 0, 200);
+
+	if (wait_for_completion_timeout(&ar->fw_boot_wait, HZ) == 0) {
+		err = -ETIMEDOUT;
+		goto err_out;
+	}
+
+	err = carl9170_echo_test(ar, 0x4a110123);
+	if (err)
+		goto err_out;
+
+	/* firmware restarts cmd counter */
+	ar->cmd_seq = -1;
+
+	return 0;
+
+err_out:
+	dev_err(&ar->udev->dev, "firmware upload failed (%d).\n", err);
+	return err;
+}
+
+int carl9170_usb_restart(struct ar9170 *ar)
+{
+	int err = 0;
+
+	if (ar->intf->condition != USB_INTERFACE_BOUND)
+		return 0;
+
+	/* Disable command response sequence counter. */
+	ar->cmd_seq = -2;
+
+	err = carl9170_reboot(ar);
+
+	carl9170_usb_stop(ar);
+
+	carl9170_set_state(ar, CARL9170_UNKNOWN_STATE);
+
+	if (err)
+		goto err_out;
+
+	tasklet_schedule(&ar->usb_tasklet);
+
+	/* The reboot procedure can take quite a while to complete. */
+	msleep(1100);
+
+	err = carl9170_usb_open(ar);
+	if (err)
+		goto err_out;
+
+	err = carl9170_usb_load_firmware(ar);
+	if (err)
+		goto err_out;
+
+	return 0;
+
+err_out:
+	carl9170_usb_cancel_urbs(ar);
+	return err;
+}
+
+void carl9170_usb_reset(struct ar9170 *ar)
+{
+	/*
+	 * This is the last resort to get the device going again
+	 * without any *user replugging action*.
+	 *
+	 * But there is a catch: usb_reset really is like a physical
+	 * *reconnect*. The mac80211 state will be lost in the process.
+	 * Therefore a userspace application, which is monitoring
+	 * the link must step in.
+	 */
+	carl9170_usb_cancel_urbs(ar);
+
+	carl9170_usb_stop(ar);
+
+	usb_queue_reset_device(ar->intf);
+}
+
+static int carl9170_usb_init_device(struct ar9170 *ar)
+{
+	int err;
+
+	err = carl9170_usb_send_rx_irq_urb(ar);
+	if (err)
+		goto err_out;
+
+	err = carl9170_usb_init_rx_bulk_urbs(ar);
+	if (err)
+		goto err_unrx;
+
+	mutex_lock(&ar->mutex);
+	err = carl9170_usb_load_firmware(ar);
+	mutex_unlock(&ar->mutex);
+	if (err)
+		goto err_unrx;
+
+	return 0;
+
+err_unrx:
+	carl9170_usb_cancel_urbs(ar);
+
+err_out:
+	return err;
+}
+
+static void carl9170_usb_firmware_failed(struct ar9170 *ar)
+{
+	struct device *parent = ar->udev->dev.parent;
+	struct usb_device *udev;
+
+	/*
+	 * Store a copy of the usb_device pointer locally.
+	 * This is because device_release_driver initiates
+	 * carl9170_usb_disconnect, which in turn frees our
+	 * driver context (ar).
+	 */
+	udev = ar->udev;
+
+	complete(&ar->fw_load_wait);
+
+	/* unbind anything failed */
+	if (parent)
+		device_lock(parent);
+
+	device_release_driver(&udev->dev);
+	if (parent)
+		device_unlock(parent);
+
+	usb_put_dev(udev);
+}
+
+static void carl9170_usb_firmware_finish(struct ar9170 *ar)
+{
+	int err;
+
+	err = carl9170_parse_firmware(ar);
+	if (err)
+		goto err_freefw;
+
+	err = carl9170_usb_init_device(ar);
+	if (err)
+		goto err_freefw;
+
+	err = carl9170_usb_open(ar);
+	if (err)
+		goto err_unrx;
+
+	err = carl9170_register(ar);
+
+	carl9170_usb_stop(ar);
+	if (err)
+		goto err_unrx;
+
+	complete(&ar->fw_load_wait);
+	usb_put_dev(ar->udev);
+	return;
+
+err_unrx:
+	carl9170_usb_cancel_urbs(ar);
+
+err_freefw:
+	carl9170_release_firmware(ar);
+	carl9170_usb_firmware_failed(ar);
+}
+
+static void carl9170_usb_firmware_step2(const struct firmware *fw,
+					void *context)
+{
+	struct ar9170 *ar = context;
+
+	if (fw) {
+		ar->fw.fw = fw;
+		carl9170_usb_firmware_finish(ar);
+		return;
+	}
+
+	dev_err(&ar->udev->dev, "firmware not found.\n");
+	carl9170_usb_firmware_failed(ar);
+}
+
+static int carl9170_usb_probe(struct usb_interface *intf,
+			      const struct usb_device_id *id)
+{
+	struct ar9170 *ar;
+	struct usb_device *udev;
+	int err;
+
+	err = usb_reset_device(interface_to_usbdev(intf));
+	if (err)
+		return err;
+
+	ar = carl9170_alloc(sizeof(*ar));
+	if (IS_ERR(ar))
+		return PTR_ERR(ar);
+
+	udev = interface_to_usbdev(intf);
+	usb_get_dev(udev);
+	ar->udev = udev;
+	ar->intf = intf;
+	ar->features = id->driver_info;
+
+	usb_set_intfdata(intf, ar);
+	SET_IEEE80211_DEV(ar->hw, &intf->dev);
+
+	init_usb_anchor(&ar->rx_anch);
+	init_usb_anchor(&ar->rx_pool);
+	init_usb_anchor(&ar->rx_work);
+	init_usb_anchor(&ar->tx_wait);
+	init_usb_anchor(&ar->tx_anch);
+	init_usb_anchor(&ar->tx_cmd);
+	init_usb_anchor(&ar->tx_err);
+	init_completion(&ar->cmd_wait);
+	init_completion(&ar->fw_boot_wait);
+	init_completion(&ar->fw_load_wait);
+	tasklet_init(&ar->usb_tasklet, carl9170_usb_tasklet,
+		     (unsigned long)ar);
+
+	atomic_set(&ar->tx_cmd_urbs, 0);
+	atomic_set(&ar->tx_anch_urbs, 0);
+	atomic_set(&ar->rx_work_urbs, 0);
+	atomic_set(&ar->rx_anch_urbs, 0);
+	atomic_set(&ar->rx_pool_urbs, 0);
+	ar->cmd_seq = -2;
+
+	usb_get_dev(ar->udev);
+
+	carl9170_set_state(ar, CARL9170_STOPPED);
+
+	return request_firmware_nowait(THIS_MODULE, 1, CARL9170FW_NAME,
+		&ar->udev->dev, GFP_KERNEL, ar, carl9170_usb_firmware_step2);
+}
+
+static void carl9170_usb_disconnect(struct usb_interface *intf)
+{
+	struct ar9170 *ar = usb_get_intfdata(intf);
+	struct usb_device *udev;
+
+	if (WARN_ON(!ar))
+		return;
+
+	udev = ar->udev;
+	wait_for_completion(&ar->fw_load_wait);
+
+	if (IS_INITIALIZED(ar)) {
+		carl9170_reboot(ar);
+		carl9170_usb_stop(ar);
+	}
+
+	carl9170_usb_cancel_urbs(ar);
+	carl9170_unregister(ar);
+
+	usb_set_intfdata(intf, NULL);
+
+	carl9170_release_firmware(ar);
+	carl9170_free(ar);
+	usb_put_dev(udev);
+}
+
+#ifdef CONFIG_PM
+static int carl9170_usb_suspend(struct usb_interface *intf,
+				pm_message_t message)
+{
+	struct ar9170 *ar = usb_get_intfdata(intf);
+
+	if (!ar)
+		return -ENODEV;
+
+	carl9170_usb_cancel_urbs(ar);
+
+	/*
+	 * firmware automatically reboots for usb suspend.
+	 */
+
+	return 0;
+}
+
+static int carl9170_usb_resume(struct usb_interface *intf)
+{
+	struct ar9170 *ar = usb_get_intfdata(intf);
+	int err;
+
+	if (!ar)
+		return -ENODEV;
+
+	usb_unpoison_anchored_urbs(&ar->rx_anch);
+
+	err = carl9170_usb_init_device(ar);
+	if (err)
+		goto err_unrx;
+
+	err = carl9170_usb_open(ar);
+	if (err)
+		goto err_unrx;
+
+	return 0;
+
+err_unrx:
+	carl9170_usb_cancel_urbs(ar);
+
+	return err;
+}
+#endif /* CONFIG_PM */
+
+static struct usb_driver carl9170_driver = {
+	.name = KBUILD_MODNAME,
+	.probe = carl9170_usb_probe,
+	.disconnect = carl9170_usb_disconnect,
+	.id_table = carl9170_usb_ids,
+	.soft_unbind = 1,
+#ifdef CONFIG_PM
+	.suspend = carl9170_usb_suspend,
+	.resume = carl9170_usb_resume,
+#endif /* CONFIG_PM */
+};
+
+static int __init carl9170_usb_init(void)
+{
+	return usb_register(&carl9170_driver);
+}
+
+static void __exit carl9170_usb_exit(void)
+{
+	usb_deregister(&carl9170_driver);
+}
+
+module_init(carl9170_usb_init);
+module_exit(carl9170_usb_exit);