iwlwifi: sanity check before counting number of tfds can be free
[safe/jmp/linux-2.6] / drivers / net / wireless / iwlwifi / iwl-tx.c
index 5febb31..8f40715 100644 (file)
@@ -28,6 +28,7 @@
  *****************************************************************************/
 
 #include <linux/etherdevice.h>
+#include <linux/sched.h>
 #include <net/mac80211.h>
 #include "iwl-eeprom.h"
 #include "iwl-dev.h"
@@ -96,7 +97,8 @@ int iwl_txq_update_write_ptr(struct iwl_priv *priv, struct iwl_tx_queue *txq)
                reg = iwl_read32(priv, CSR_UCODE_DRV_GP1);
 
                if (reg & CSR_UCODE_DRV_GP1_BIT_MAC_SLEEP) {
-                       IWL_DEBUG_INFO(priv, "Requesting wakeup, GP1 = 0x%x\n", reg);
+                       IWL_DEBUG_INFO(priv, "Tx queue %d requesting wakeup, GP1 = 0x%x\n",
+                                     txq_id, reg);
                        iwl_set_bit(priv, CSR_GP_CNTRL,
                                    CSR_GP_CNTRL_REG_FLAG_MAC_ACCESS_REQ);
                        return ret;
@@ -118,6 +120,20 @@ int iwl_txq_update_write_ptr(struct iwl_priv *priv, struct iwl_tx_queue *txq)
 EXPORT_SYMBOL(iwl_txq_update_write_ptr);
 
 
+void iwl_free_tfds_in_queue(struct iwl_priv *priv,
+                           int sta_id, int tid, int freed)
+{
+       if (priv->stations[sta_id].tid[tid].tfds_in_queue >= freed)
+               priv->stations[sta_id].tid[tid].tfds_in_queue -= freed;
+       else {
+               IWL_ERR(priv, "free more than tfds_in_queue (%u:%d)\n",
+                       priv->stations[sta_id].tid[tid].tfds_in_queue,
+                       freed);
+               priv->stations[sta_id].tid[tid].tfds_in_queue = 0;
+       }
+}
+EXPORT_SYMBOL(iwl_free_tfds_in_queue);
+
 /**
  * iwl_tx_queue_free - Deallocate DMA queue.
  * @txq: Transmit queue to deallocate.
@@ -131,7 +147,7 @@ void iwl_tx_queue_free(struct iwl_priv *priv, int txq_id)
        struct iwl_tx_queue *txq = &priv->txq[txq_id];
        struct iwl_queue *q = &txq->q;
        struct pci_dev *dev = priv->pci_dev;
-       int i, len;
+       int i;
 
        if (q->n_bd == 0)
                return;
@@ -141,8 +157,6 @@ void iwl_tx_queue_free(struct iwl_priv *priv, int txq_id)
             q->read_ptr = iwl_queue_inc_wrap(q->read_ptr, q->n_bd))
                priv->cfg->ops->lib->txq_free_tfd(priv, txq);
 
-       len = sizeof(struct iwl_cmd) * q->n_window;
-
        /* De-alloc array of command/tx buffers */
        for (i = 0; i < TFD_TX_CMD_SLOTS; i++)
                kfree(txq->cmd[i]);
@@ -156,6 +170,12 @@ void iwl_tx_queue_free(struct iwl_priv *priv, int txq_id)
        kfree(txq->txb);
        txq->txb = NULL;
 
+       /* deallocate arrays */
+       kfree(txq->cmd);
+       kfree(txq->meta);
+       txq->cmd = NULL;
+       txq->meta = NULL;
+
        /* 0-fill queue descriptor structure */
        memset(txq, 0, sizeof(*txq));
 }
@@ -174,14 +194,11 @@ void iwl_cmd_queue_free(struct iwl_priv *priv)
        struct iwl_tx_queue *txq = &priv->txq[IWL_CMD_QUEUE_NUM];
        struct iwl_queue *q = &txq->q;
        struct pci_dev *dev = priv->pci_dev;
-       int i, len;
+       int i;
 
        if (q->n_bd == 0)
                return;
 
-       len = sizeof(struct iwl_cmd) * q->n_window;
-       len += IWL_MAX_SCAN_SIZE;
-
        /* De-alloc array of command/tx buffers */
        for (i = 0; i <= TFD_CMD_SLOTS; i++)
                kfree(txq->cmd[i]);
@@ -191,6 +208,12 @@ void iwl_cmd_queue_free(struct iwl_priv *priv)
                pci_free_consistent(dev, priv->hw_params.tfd_size *
                                    txq->q.n_bd, txq->tfds, txq->q.dma_addr);
 
+       /* deallocate arrays */
+       kfree(txq->cmd);
+       kfree(txq->meta);
+       txq->cmd = NULL;
+       txq->meta = NULL;
+
        /* 0-fill queue descriptor structure */
        memset(txq, 0, sizeof(*txq));
 }
@@ -318,6 +341,7 @@ int iwl_tx_queue_init(struct iwl_priv *priv, struct iwl_tx_queue *txq,
 {
        int i, len;
        int ret;
+       int actual_slots = slots_num;
 
        /*
         * Alloc buffer array for commands (Tx or other types of commands).
@@ -327,14 +351,22 @@ int iwl_tx_queue_init(struct iwl_priv *priv, struct iwl_tx_queue *txq,
         * For normal Tx queues (all other queues), no super-size command
         * space is needed.
         */
-       len = sizeof(struct iwl_cmd);
-       for (i = 0; i <= slots_num; i++) {
-               if (i == slots_num) {
-                       if (txq_id == IWL_CMD_QUEUE_NUM)
-                               len += IWL_MAX_SCAN_SIZE;
-                       else
-                               continue;
-               }
+       if (txq_id == IWL_CMD_QUEUE_NUM)
+               actual_slots++;
+
+       txq->meta = kzalloc(sizeof(struct iwl_cmd_meta) * actual_slots,
+                           GFP_KERNEL);
+       txq->cmd = kzalloc(sizeof(struct iwl_device_cmd *) * actual_slots,
+                          GFP_KERNEL);
+
+       if (!txq->meta || !txq->cmd)
+               goto out_free_arrays;
+
+       len = sizeof(struct iwl_device_cmd);
+       for (i = 0; i < actual_slots; i++) {
+               /* only happens for cmd queue */
+               if (i == slots_num)
+                       len += IWL_MAX_SCAN_SIZE;
 
                txq->cmd[i] = kmalloc(len, GFP_KERNEL);
                if (!txq->cmd[i])
@@ -348,6 +380,15 @@ int iwl_tx_queue_init(struct iwl_priv *priv, struct iwl_tx_queue *txq,
 
        txq->need_update = 0;
 
+       /*
+        * Aggregation TX queues will get their ID when aggregation begins;
+        * they overwrite the setting done here. The command FIFO doesn't
+        * need an swq_id so don't set one to catch errors, all others can
+        * be set up to the identity mapping.
+        */
+       if (txq_id != IWL_CMD_QUEUE_NUM)
+               txq->swq_id = txq_id;
+
        /* TFD_QUEUE_SIZE_MAX must be power-of-two size, otherwise
         * iwl_queue_inc_wrap and iwl_queue_dec_wrap are broken. */
        BUILD_BUG_ON(TFD_QUEUE_SIZE_MAX & (TFD_QUEUE_SIZE_MAX - 1));
@@ -360,15 +401,12 @@ int iwl_tx_queue_init(struct iwl_priv *priv, struct iwl_tx_queue *txq,
 
        return 0;
 err:
-       for (i = 0; i < slots_num; i++) {
+       for (i = 0; i < actual_slots; i++)
                kfree(txq->cmd[i]);
-               txq->cmd[i] = NULL;
-       }
+out_free_arrays:
+       kfree(txq->meta);
+       kfree(txq->cmd);
 
-       if (txq_id == IWL_CMD_QUEUE_NUM) {
-               kfree(txq->cmd[slots_num]);
-               txq->cmd[slots_num] = NULL;
-       }
        return -ENOMEM;
 }
 EXPORT_SYMBOL(iwl_tx_queue_init);
@@ -383,15 +421,20 @@ void iwl_hw_txq_ctx_free(struct iwl_priv *priv)
        int txq_id;
 
        /* Tx queues */
-       for (txq_id = 0; txq_id < priv->hw_params.max_txq_num; txq_id++)
-               if (txq_id == IWL_CMD_QUEUE_NUM)
-                       iwl_cmd_queue_free(priv);
-               else
-                       iwl_tx_queue_free(priv, txq_id);
-
+       if (priv->txq) {
+               for (txq_id = 0; txq_id < priv->hw_params.max_txq_num;
+                    txq_id++)
+                       if (txq_id == IWL_CMD_QUEUE_NUM)
+                               iwl_cmd_queue_free(priv);
+                       else
+                               iwl_tx_queue_free(priv, txq_id);
+       }
        iwl_free_dma_ptr(priv, &priv->kw);
 
        iwl_free_dma_ptr(priv, &priv->scd_bc_tbls);
+
+       /* free tx queue structure */
+       iwl_free_txq_mem(priv);
 }
 EXPORT_SYMBOL(iwl_hw_txq_ctx_free);
 
@@ -423,6 +466,12 @@ int iwl_txq_ctx_reset(struct iwl_priv *priv)
                IWL_ERR(priv, "Keep Warm allocation failed\n");
                goto error_kw;
        }
+
+       /* allocate tx queue structure */
+       ret = iwl_alloc_txq_mem(priv);
+       if (ret)
+               goto error;
+
        spin_lock_irqsave(&priv->lock, flags);
 
        /* Turn off all Tx DMA fifos */
@@ -550,62 +599,79 @@ static void iwl_tx_cmd_build_basic(struct iwl_priv *priv,
 static void iwl_tx_cmd_build_rate(struct iwl_priv *priv,
                              struct iwl_tx_cmd *tx_cmd,
                              struct ieee80211_tx_info *info,
-                             __le16 fc, int sta_id,
-                             int is_hcca)
+                             __le16 fc, int is_hcca)
 {
-       u32 rate_flags = 0;
+       u32 rate_flags;
        int rate_idx;
-       u8 rts_retry_limit = 0;
-       u8 data_retry_limit = 0;
+       u8 rts_retry_limit;
+       u8 data_retry_limit;
        u8 rate_plcp;
 
-       rate_idx = min(ieee80211_get_tx_rate(priv->hw, info)->hw_value & 0xffff,
-                       IWL_RATE_COUNT - 1);
-
-       rate_plcp = iwl_rates[rate_idx].plcp;
-
-       rts_retry_limit = (is_hcca) ?
-           RTS_HCCA_RETRY_LIMIT : RTS_DFAULT_RETRY_LIMIT;
-
-       if ((rate_idx >= IWL_FIRST_CCK_RATE) && (rate_idx <= IWL_LAST_CCK_RATE))
-               rate_flags |= RATE_MCS_CCK_MSK;
-
-
-       if (ieee80211_is_probe_resp(fc)) {
+       /* Set retry limit on DATA packets and Probe Responses*/
+       if (ieee80211_is_probe_resp(fc))
                data_retry_limit = 3;
-               if (data_retry_limit < rts_retry_limit)
-                       rts_retry_limit = data_retry_limit;
-       } else
+       else
                data_retry_limit = IWL_DEFAULT_TX_RETRY;
+       tx_cmd->data_retry_limit = data_retry_limit;
 
-       if (priv->data_retry_limit != -1)
-               data_retry_limit = priv->data_retry_limit;
-
+       /* Set retry limit on RTS packets */
+       rts_retry_limit = (is_hcca) ?  RTS_HCCA_RETRY_LIMIT :
+               RTS_DFAULT_RETRY_LIMIT;
+       if (data_retry_limit < rts_retry_limit)
+               rts_retry_limit = data_retry_limit;
+       tx_cmd->rts_retry_limit = rts_retry_limit;
 
+       /* DATA packets will use the uCode station table for rate/antenna
+        * selection */
        if (ieee80211_is_data(fc)) {
                tx_cmd->initial_rate_index = 0;
                tx_cmd->tx_flags |= TX_CMD_FLG_STA_RATE_MSK;
-       } else {
-               switch (fc & cpu_to_le16(IEEE80211_FCTL_STYPE)) {
-               case cpu_to_le16(IEEE80211_STYPE_AUTH):
-               case cpu_to_le16(IEEE80211_STYPE_DEAUTH):
-               case cpu_to_le16(IEEE80211_STYPE_ASSOC_REQ):
-               case cpu_to_le16(IEEE80211_STYPE_REASSOC_REQ):
-                       if (tx_cmd->tx_flags & TX_CMD_FLG_RTS_MSK) {
-                               tx_cmd->tx_flags &= ~TX_CMD_FLG_RTS_MSK;
-                               tx_cmd->tx_flags |= TX_CMD_FLG_CTS_MSK;
-                       }
-                       break;
-               default:
-                       break;
-               }
+               return;
+       }
+
+       /**
+        * If the current TX rate stored in mac80211 has the MCS bit set, it's
+        * not really a TX rate.  Thus, we use the lowest supported rate for
+        * this band.  Also use the lowest supported rate if the stored rate
+        * index is invalid.
+        */
+       rate_idx = info->control.rates[0].idx;
+       if (info->control.rates[0].flags & IEEE80211_TX_RC_MCS ||
+                       (rate_idx < 0) || (rate_idx > IWL_RATE_COUNT_LEGACY))
+               rate_idx = rate_lowest_index(&priv->bands[info->band],
+                               info->control.sta);
+       /* For 5 GHZ band, remap mac80211 rate indices into driver indices */
+       if (info->band == IEEE80211_BAND_5GHZ)
+               rate_idx += IWL_FIRST_OFDM_RATE;
+       /* Get PLCP rate for tx_cmd->rate_n_flags */
+       rate_plcp = iwl_rates[rate_idx].plcp;
+       /* Zero out flags for this packet */
+       rate_flags = 0;
 
-               priv->mgmt_tx_ant = iwl_toggle_tx_ant(priv, priv->mgmt_tx_ant);
-               rate_flags |= iwl_ant_idx_to_flags(priv->mgmt_tx_ant);
+       /* Set CCK flag as needed */
+       if ((rate_idx >= IWL_FIRST_CCK_RATE) && (rate_idx <= IWL_LAST_CCK_RATE))
+               rate_flags |= RATE_MCS_CCK_MSK;
+
+       /* Set up RTS and CTS flags for certain packets */
+       switch (fc & cpu_to_le16(IEEE80211_FCTL_STYPE)) {
+       case cpu_to_le16(IEEE80211_STYPE_AUTH):
+       case cpu_to_le16(IEEE80211_STYPE_DEAUTH):
+       case cpu_to_le16(IEEE80211_STYPE_ASSOC_REQ):
+       case cpu_to_le16(IEEE80211_STYPE_REASSOC_REQ):
+               if (tx_cmd->tx_flags & TX_CMD_FLG_RTS_MSK) {
+                       tx_cmd->tx_flags &= ~TX_CMD_FLG_RTS_MSK;
+                       tx_cmd->tx_flags |= TX_CMD_FLG_CTS_MSK;
+               }
+               break;
+       default:
+               break;
        }
 
-       tx_cmd->rts_retry_limit = rts_retry_limit;
-       tx_cmd->data_retry_limit = data_retry_limit;
+       /* Set up antennas */
+       priv->mgmt_tx_ant = iwl_toggle_tx_ant(priv, priv->mgmt_tx_ant);
+       rate_flags |= iwl_ant_idx_to_flags(priv->mgmt_tx_ant);
+
+       /* Set the rate in the TX cmd */
        tx_cmd->rate_n_flags = iwl_hw_set_rate_n_flags(rate_plcp, rate_flags);
 }
 
@@ -652,14 +718,6 @@ static void iwl_tx_cmd_build_hwcrypto(struct iwl_priv *priv,
        }
 }
 
-static void iwl_update_tx_stats(struct iwl_priv *priv, u16 fc, u16 len)
-{
-       /* 0 - mgmt, 1 - cnt, 2 - data */
-       int idx = (fc & IEEE80211_FCTL_FTYPE) >> 2;
-       priv->tx_stats[idx].cnt++;
-       priv->tx_stats[idx].bytes += len;
-}
-
 /*
  * start REPLY_TX command process
  */
@@ -667,15 +725,18 @@ int iwl_tx_skb(struct iwl_priv *priv, struct sk_buff *skb)
 {
        struct ieee80211_hdr *hdr = (struct ieee80211_hdr *)skb->data;
        struct ieee80211_tx_info *info = IEEE80211_SKB_CB(skb);
+       struct ieee80211_sta *sta = info->control.sta;
+       struct iwl_station_priv *sta_priv = NULL;
        struct iwl_tx_queue *txq;
        struct iwl_queue *q;
-       struct iwl_cmd *out_cmd;
+       struct iwl_device_cmd *out_cmd;
+       struct iwl_cmd_meta *out_meta;
        struct iwl_tx_cmd *tx_cmd;
        int swq_id, txq_id;
        dma_addr_t phys_addr;
        dma_addr_t txcmd_phys;
        dma_addr_t scratch_phys;
-       u16 len, len_org;
+       u16 len, len_org, firstlen, secondlen;
        u16 seq_number = 0;
        __le16 fc;
        u8 hdr_len;
@@ -692,12 +753,6 @@ int iwl_tx_skb(struct iwl_priv *priv, struct sk_buff *skb)
                goto drop_unlock;
        }
 
-       if ((ieee80211_get_tx_rate(priv->hw, info)->hw_value & 0xFF) ==
-            IWL_INVALID_RATE) {
-               IWL_ERR(priv, "ERROR: No TX rate available.\n");
-               goto drop_unlock;
-       }
-
        fc = hdr->frame_control;
 
 #ifdef CONFIG_IWLWIFI_DEBUG
@@ -709,10 +764,9 @@ int iwl_tx_skb(struct iwl_priv *priv, struct sk_buff *skb)
                IWL_DEBUG_TX(priv, "Sending REASSOC frame\n");
 #endif
 
-       /* drop all data frame if we are not associated */
+       /* drop all non-injected data frame if we are not associated */
        if (ieee80211_is_data(fc) &&
-           (!iwl_is_monitor_mode(priv) ||
-           !(info->flags & IEEE80211_TX_CTL_INJECTED)) && /* packet injection */
+           !(info->flags & IEEE80211_TX_CTL_INJECTED) &&
            (!iwl_is_associated(priv) ||
             ((priv->iw_mode == NL80211_IFTYPE_STATION) && !priv->assoc_id) ||
             !priv->assoc_station_added)) {
@@ -723,7 +777,10 @@ int iwl_tx_skb(struct iwl_priv *priv, struct sk_buff *skb)
        hdr_len = ieee80211_hdrlen(fc);
 
        /* Find (or create) index into station table for destination station */
-       sta_id = iwl_get_sta_id(priv, hdr);
+       if (info->flags & IEEE80211_TX_CTL_INJECTED)
+               sta_id = priv->hw_params.bcast_sta_id;
+       else
+               sta_id = iwl_get_sta_id(priv, hdr);
        if (sta_id == IWL_INVALID_STATION) {
                IWL_DEBUG_DROP(priv, "Dropping - INVALID STATION: %pM\n",
                               hdr->addr1);
@@ -732,11 +789,30 @@ int iwl_tx_skb(struct iwl_priv *priv, struct sk_buff *skb)
 
        IWL_DEBUG_TX(priv, "station Id %d\n", sta_id);
 
-       swq_id = skb_get_queue_mapping(skb);
-       txq_id = swq_id;
+       if (sta)
+               sta_priv = (void *)sta->drv_priv;
+
+       if (sta_priv && sta_id != priv->hw_params.bcast_sta_id &&
+           sta_priv->asleep) {
+               WARN_ON(!(info->flags & IEEE80211_TX_CTL_PSPOLL_RESPONSE));
+               /*
+                * This sends an asynchronous command to the device,
+                * but we can rely on it being processed before the
+                * next frame is processed -- and the next frame to
+                * this station is the one that will consume this
+                * counter.
+                * For now set the counter to just 1 since we do not
+                * support uAPSD yet.
+                */
+               iwl_sta_modify_sleep_tx_count(priv, sta_id, 1);
+       }
+
+       txq_id = skb_get_queue_mapping(skb);
        if (ieee80211_is_data_qos(fc)) {
                qc = ieee80211_get_qos_ctl(hdr);
                tid = qc[0] & IEEE80211_QOS_CTL_TID_MASK;
+               if (unlikely(tid >= MAX_TID_COUNT))
+                       goto drop_unlock;
                seq_number = priv->stations[sta_id].tid[tid].seq_number;
                seq_number &= IEEE80211_SCTL_SEQ;
                hdr->seq_ctrl = hdr->seq_ctrl &
@@ -744,15 +820,13 @@ int iwl_tx_skb(struct iwl_priv *priv, struct sk_buff *skb)
                hdr->seq_ctrl |= cpu_to_le16(seq_number);
                seq_number += 0x10;
                /* aggregation is on for this <sta,tid> */
-               if (info->flags & IEEE80211_TX_CTL_AMPDU) {
+               if (info->flags & IEEE80211_TX_CTL_AMPDU)
                        txq_id = priv->stations[sta_id].tid[tid].agg.txq_id;
-                       swq_id = iwl_virtual_agg_queue_num(swq_id, txq_id);
-               }
        }
 
        txq = &priv->txq[txq_id];
+       swq_id = txq->swq_id;
        q = &txq->q;
-       txq->swq_id = swq_id;
 
        if (unlikely(iwl_queue_space(q) < q->high_mark))
                goto drop_unlock;
@@ -766,6 +840,7 @@ int iwl_tx_skb(struct iwl_priv *priv, struct sk_buff *skb)
 
        /* Set up first empty entry in queue's array of Tx/cmd buffers */
        out_cmd = txq->cmd[q->write_ptr];
+       out_meta = &txq->meta[q->write_ptr];
        tx_cmd = &out_cmd->cmd.tx;
        memset(&out_cmd->hdr, 0, sizeof(out_cmd->hdr));
        memset(tx_cmd, 0, sizeof(struct iwl_tx_cmd));
@@ -793,12 +868,12 @@ int iwl_tx_skb(struct iwl_priv *priv, struct sk_buff *skb)
 
        /* TODO need this for burst mode later on */
        iwl_tx_cmd_build_basic(priv, tx_cmd, info, hdr, sta_id);
+       iwl_dbg_log_tx_data_frame(priv, len, hdr);
 
        /* set is_hcca to 0; it probably will never be implemented */
-       iwl_tx_cmd_build_rate(priv, tx_cmd, info, fc, sta_id, 0);
-
-       iwl_update_tx_stats(priv, le16_to_cpu(fc), len);
+       iwl_tx_cmd_build_rate(priv, tx_cmd, info, fc, 0);
 
+       iwl_update_stats(priv, true, fc, len);
        /*
         * Use the first empty entry in this queue's command buffer array
         * to contain the Tx command and MAC header concatenated together
@@ -812,7 +887,7 @@ int iwl_tx_skb(struct iwl_priv *priv, struct sk_buff *skb)
                sizeof(struct iwl_cmd_header) + hdr_len;
 
        len_org = len;
-       len = (len + 3) & ~3;
+       firstlen = len = (len + 3) & ~3;
 
        if (len_org != len)
                len_org = 1;
@@ -828,8 +903,8 @@ int iwl_tx_skb(struct iwl_priv *priv, struct sk_buff *skb)
        txcmd_phys = pci_map_single(priv->pci_dev,
                                    &out_cmd->hdr, len,
                                    PCI_DMA_BIDIRECTIONAL);
-       pci_unmap_addr_set(&out_cmd->meta, mapping, txcmd_phys);
-       pci_unmap_len_set(&out_cmd->meta, len, len);
+       pci_unmap_addr_set(out_meta, mapping, txcmd_phys);
+       pci_unmap_len_set(out_meta, len, len);
        /* Add buffer containing Tx command and MAC(!) header to TFD's
         * first entry */
        priv->cfg->ops->lib->txq_attach_buf_to_tfd(priv, txq,
@@ -846,7 +921,7 @@ int iwl_tx_skb(struct iwl_priv *priv, struct sk_buff *skb)
 
        /* Set up TFD's 2nd entry to point directly to remainder of skb,
         * if any (802.11 null frames have no payload). */
-       len = skb->len - hdr_len;
+       secondlen = len = skb->len - hdr_len;
        if (len) {
                phys_addr = pci_map_single(priv->pci_dev, skb->data + hdr_len,
                                           len, PCI_DMA_TODEVICE);
@@ -880,11 +955,28 @@ int iwl_tx_skb(struct iwl_priv *priv, struct sk_buff *skb)
        pci_dma_sync_single_for_device(priv->pci_dev, txcmd_phys,
                                       len, PCI_DMA_BIDIRECTIONAL);
 
+       trace_iwlwifi_dev_tx(priv,
+                            &((struct iwl_tfd *)txq->tfds)[txq->q.write_ptr],
+                            sizeof(struct iwl_tfd),
+                            &out_cmd->hdr, firstlen,
+                            skb->data + hdr_len, secondlen);
+
        /* Tell device the write index *just past* this latest filled TFD */
        q->write_ptr = iwl_queue_inc_wrap(q->write_ptr, q->n_bd);
        ret = iwl_txq_update_write_ptr(priv, txq);
        spin_unlock_irqrestore(&priv->lock, flags);
 
+       /*
+        * At this point the frame is "transmitted" successfully
+        * and we will get a TX status notification eventually,
+        * regardless of the value of ret. "ret" only indicates
+        * whether or not we should update the write pointer.
+        */
+
+       /* avoid atomic ops if it isn't an associated client */
+       if (sta_priv && sta_priv->client)
+               atomic_inc(&sta_priv->pending_frames);
+
        if (ret)
                return ret;
 
@@ -922,7 +1014,8 @@ int iwl_enqueue_hcmd(struct iwl_priv *priv, struct iwl_host_cmd *cmd)
 {
        struct iwl_tx_queue *txq = &priv->txq[IWL_CMD_QUEUE_NUM];
        struct iwl_queue *q = &txq->q;
-       struct iwl_cmd *out_cmd;
+       struct iwl_device_cmd *out_cmd;
+       struct iwl_cmd_meta *out_meta;
        dma_addr_t phys_addr;
        unsigned long flags;
        int len, ret;
@@ -936,25 +1029,39 @@ int iwl_enqueue_hcmd(struct iwl_priv *priv, struct iwl_host_cmd *cmd)
         * the TFD_MAX_PAYLOAD_SIZE, and it sent as a 'small' command then
         * we will need to increase the size of the TFD entries */
        BUG_ON((fix_size > TFD_MAX_PAYLOAD_SIZE) &&
-              !(cmd->meta.flags & CMD_SIZE_HUGE));
+              !(cmd->flags & CMD_SIZE_HUGE));
 
-       if (iwl_is_rfkill(priv)) {
-               IWL_DEBUG_INFO(priv, "Not sending command - RF KILL");
+       if (iwl_is_rfkill(priv) || iwl_is_ctkill(priv)) {
+               IWL_WARN(priv, "Not sending command - %s KILL\n",
+                        iwl_is_rfkill(priv) ? "RF" : "CT");
                return -EIO;
        }
 
-       if (iwl_queue_space(q) < ((cmd->meta.flags & CMD_ASYNC) ? 2 : 1)) {
-               IWL_ERR(priv, "No space for Tx\n");
+       if (iwl_queue_space(q) < ((cmd->flags & CMD_ASYNC) ? 2 : 1)) {
+               IWL_ERR(priv, "No space in command queue\n");
+               if (iwl_within_ct_kill_margin(priv))
+                       iwl_tt_enter_ct_kill(priv);
+               else {
+                       IWL_ERR(priv, "Restarting adapter due to queue full\n");
+                       queue_work(priv->workqueue, &priv->restart);
+               }
                return -ENOSPC;
        }
 
        spin_lock_irqsave(&priv->hcmd_lock, flags);
 
-       idx = get_cmd_index(q, q->write_ptr, cmd->meta.flags & CMD_SIZE_HUGE);
+       idx = get_cmd_index(q, q->write_ptr, cmd->flags & CMD_SIZE_HUGE);
        out_cmd = txq->cmd[idx];
+       out_meta = &txq->meta[idx];
+
+       memset(out_meta, 0, sizeof(*out_meta)); /* re-initialize to NULL */
+       out_meta->flags = cmd->flags;
+       if (cmd->flags & CMD_WANT_SKB)
+               out_meta->source = cmd;
+       if (cmd->flags & CMD_ASYNC)
+               out_meta->callback = cmd->callback;
 
        out_cmd->hdr.cmd = cmd->id;
-       memcpy(&out_cmd->meta, &cmd->meta, sizeof(cmd->meta));
        memcpy(&out_cmd->cmd.payload, cmd->data, cmd->len);
 
        /* At this point, the out_cmd now has all of the incoming cmd
@@ -963,9 +1070,9 @@ int iwl_enqueue_hcmd(struct iwl_priv *priv, struct iwl_host_cmd *cmd)
        out_cmd->hdr.flags = 0;
        out_cmd->hdr.sequence = cpu_to_le16(QUEUE_TO_SEQ(IWL_CMD_QUEUE_NUM) |
                        INDEX_TO_SEQ(q->write_ptr));
-       if (out_cmd->meta.flags & CMD_SIZE_HUGE)
+       if (cmd->flags & CMD_SIZE_HUGE)
                out_cmd->hdr.sequence |= SEQ_HUGE_FRAME;
-       len = sizeof(struct iwl_cmd) - sizeof(struct iwl_cmd_meta);
+       len = sizeof(struct iwl_device_cmd);
        len += (idx == TFD_CMD_SLOTS) ?  IWL_MAX_SCAN_SIZE : 0;
 
 
@@ -997,8 +1104,10 @@ int iwl_enqueue_hcmd(struct iwl_priv *priv, struct iwl_host_cmd *cmd)
 
        phys_addr = pci_map_single(priv->pci_dev, &out_cmd->hdr,
                                   fix_size, PCI_DMA_BIDIRECTIONAL);
-       pci_unmap_addr_set(&out_cmd->meta, mapping, phys_addr);
-       pci_unmap_len_set(&out_cmd->meta, len, fix_size);
+       pci_unmap_addr_set(out_meta, mapping, phys_addr);
+       pci_unmap_len_set(out_meta, len, fix_size);
+
+       trace_iwlwifi_dev_hcmd(priv, &out_cmd->hdr, fix_size, cmd->flags);
 
        priv->cfg->ops->lib->txq_attach_buf_to_tfd(priv, txq,
                                                   phys_addr, fix_size, 1,
@@ -1012,12 +1121,31 @@ int iwl_enqueue_hcmd(struct iwl_priv *priv, struct iwl_host_cmd *cmd)
        return ret ? ret : idx;
 }
 
+static void iwl_tx_status(struct iwl_priv *priv, struct sk_buff *skb)
+{
+       struct ieee80211_hdr *hdr = (struct ieee80211_hdr *) skb->data;
+       struct ieee80211_sta *sta;
+       struct iwl_station_priv *sta_priv;
+
+       sta = ieee80211_find_sta(priv->vif, hdr->addr1);
+       if (sta) {
+               sta_priv = (void *)sta->drv_priv;
+               /* avoid atomic ops if this isn't a client */
+               if (sta_priv->client &&
+                   atomic_dec_return(&sta_priv->pending_frames) == 0)
+                       ieee80211_sta_block_awake(priv->hw, sta, false);
+       }
+
+       ieee80211_tx_status_irqsafe(priv->hw, skb);
+}
+
 int iwl_tx_queue_reclaim(struct iwl_priv *priv, int txq_id, int index)
 {
        struct iwl_tx_queue *txq = &priv->txq[txq_id];
        struct iwl_queue *q = &txq->q;
        struct iwl_tx_info *tx_info;
        int nfreed = 0;
+       struct ieee80211_hdr *hdr;
 
        if ((index >= q->n_bd) || (iwl_queue_used(q, index) == 0)) {
                IWL_ERR(priv, "Read index for DMA queue txq id (%d), index %d, "
@@ -1031,14 +1159,17 @@ int iwl_tx_queue_reclaim(struct iwl_priv *priv, int txq_id, int index)
             q->read_ptr = iwl_queue_inc_wrap(q->read_ptr, q->n_bd)) {
 
                tx_info = &txq->txb[txq->q.read_ptr];
-               ieee80211_tx_status_irqsafe(priv->hw, tx_info->skb[0]);
+               iwl_tx_status(priv, tx_info->skb[0]);
+
+               hdr = (struct ieee80211_hdr *)tx_info->skb[0]->data;
+               if (hdr && ieee80211_is_data_qos(hdr->frame_control))
+                       nfreed++;
                tx_info->skb[0] = NULL;
 
                if (priv->cfg->ops->lib->txq_inval_byte_cnt_tbl)
                        priv->cfg->ops->lib->txq_inval_byte_cnt_tbl(priv, txq);
 
                priv->cfg->ops->lib->txq_free_tfd(priv, txq);
-               nfreed++;
        }
        return nfreed;
 }
@@ -1066,11 +1197,6 @@ static void iwl_hcmd_queue_reclaim(struct iwl_priv *priv, int txq_id,
                return;
        }
 
-       pci_unmap_single(priv->pci_dev,
-               pci_unmap_addr(&txq->cmd[cmd_idx]->meta, mapping),
-               pci_unmap_len(&txq->cmd[cmd_idx]->meta, len),
-               PCI_DMA_BIDIRECTIONAL);
-
        for (idx = iwl_queue_inc_wrap(idx, q->n_bd); q->read_ptr != idx;
             q->read_ptr = iwl_queue_inc_wrap(q->read_ptr, q->n_bd)) {
 
@@ -1093,13 +1219,14 @@ static void iwl_hcmd_queue_reclaim(struct iwl_priv *priv, int txq_id,
  */
 void iwl_tx_cmd_complete(struct iwl_priv *priv, struct iwl_rx_mem_buffer *rxb)
 {
-       struct iwl_rx_packet *pkt = (struct iwl_rx_packet *)rxb->skb->data;
+       struct iwl_rx_packet *pkt = rxb_addr(rxb);
        u16 sequence = le16_to_cpu(pkt->hdr.sequence);
        int txq_id = SEQ_TO_QUEUE(sequence);
        int index = SEQ_TO_INDEX(sequence);
        int cmd_index;
        bool huge = !!(pkt->hdr.sequence & SEQ_HUGE_FRAME);
-       struct iwl_cmd *cmd;
+       struct iwl_device_cmd *cmd;
+       struct iwl_cmd_meta *meta;
 
        /* If a Tx command is being handled and it isn't in the actual
         * command queue then there a command routing bug has been introduced
@@ -1109,24 +1236,29 @@ void iwl_tx_cmd_complete(struct iwl_priv *priv, struct iwl_rx_mem_buffer *rxb)
                  txq_id, sequence,
                  priv->txq[IWL_CMD_QUEUE_NUM].q.read_ptr,
                  priv->txq[IWL_CMD_QUEUE_NUM].q.write_ptr)) {
-               iwl_print_hex_dump(priv, IWL_DL_INFO , rxb, 32);
+               iwl_print_hex_error(priv, pkt, 32);
                return;
        }
 
        cmd_index = get_cmd_index(&priv->txq[IWL_CMD_QUEUE_NUM].q, index, huge);
        cmd = priv->txq[IWL_CMD_QUEUE_NUM].cmd[cmd_index];
+       meta = &priv->txq[IWL_CMD_QUEUE_NUM].meta[cmd_index];
+
+       pci_unmap_single(priv->pci_dev,
+                        pci_unmap_addr(meta, mapping),
+                        pci_unmap_len(meta, len),
+                        PCI_DMA_BIDIRECTIONAL);
 
        /* Input error checking is done when commands are added to queue. */
-       if (cmd->meta.flags & CMD_WANT_SKB) {
-               cmd->meta.source->u.skb = rxb->skb;
-               rxb->skb = NULL;
-       } else if (cmd->meta.u.callback &&
-                  !cmd->meta.u.callback(priv, cmd, rxb->skb))
-               rxb->skb = NULL;
+       if (meta->flags & CMD_WANT_SKB) {
+               meta->source->reply_page = (unsigned long)rxb_addr(rxb);
+               rxb->page = NULL;
+       } else if (meta->callback)
+               meta->callback(priv, cmd, pkt);
 
        iwl_hcmd_queue_reclaim(priv, txq_id, index, cmd_index);
 
-       if (!(cmd->meta.flags & CMD_ASYNC)) {
+       if (!(meta->flags & CMD_ASYNC)) {
                clear_bit(STATUS_HCMD_ACTIVE, &priv->status);
                wake_up_interruptible(&priv->wait_command_queue);
        }
@@ -1171,6 +1303,8 @@ int iwl_tx_agg_start(struct iwl_priv *priv, const u8 *ra, u16 tid, u16 *ssn)
                IWL_ERR(priv, "Start AGG on invalid station\n");
                return -ENXIO;
        }
+       if (unlikely(tid >= MAX_TID_COUNT))
+               return -EINVAL;
 
        if (priv->stations[sta_id].tid[tid].agg.state != IWL_AGG_OFF) {
                IWL_ERR(priv, "Start AGG when state is not IWL_AGG_OFF !\n");
@@ -1187,6 +1321,7 @@ int iwl_tx_agg_start(struct iwl_priv *priv, const u8 *ra, u16 tid, u16 *ssn)
        tid_data = &priv->stations[sta_id].tid[tid];
        *ssn = SEQ_TO_SN(tid_data->seq_number);
        tid_data->agg.txq_id = txq_id;
+       priv->txq[txq_id].swq_id = iwl_virtual_agg_queue_num(tx_fifo, txq_id);
        spin_unlock_irqrestore(&priv->sta_lock, flags);
 
        ret = priv->cfg->ops->lib->txq_agg_enable(priv, txq_id, tx_fifo,
@@ -1197,7 +1332,7 @@ int iwl_tx_agg_start(struct iwl_priv *priv, const u8 *ra, u16 tid, u16 *ssn)
        if (tid_data->tfds_in_queue == 0) {
                IWL_DEBUG_HT(priv, "HW queue is empty\n");
                tid_data->agg.state = IWL_AGG_ON;
-               ieee80211_start_tx_ba_cb_irqsafe(priv->hw, ra, tid);
+               ieee80211_start_tx_ba_cb_irqsafe(priv->vif, ra, tid);
        } else {
                IWL_DEBUG_HT(priv, "HW queue is NOT empty: %d packets in HW queue\n",
                             tid_data->tfds_in_queue);
@@ -1219,6 +1354,9 @@ int iwl_tx_agg_stop(struct iwl_priv *priv , const u8 *ra, u16 tid)
                return -EINVAL;
        }
 
+       if (unlikely(tid >= MAX_TID_COUNT))
+               return -EINVAL;
+
        if (likely(tid < ARRAY_SIZE(default_tid_to_tx_fifo)))
                tx_fifo_id = default_tid_to_tx_fifo[tid];
        else
@@ -1231,8 +1369,16 @@ int iwl_tx_agg_stop(struct iwl_priv *priv , const u8 *ra, u16 tid)
                return -ENXIO;
        }
 
+       if (priv->stations[sta_id].tid[tid].agg.state ==
+                               IWL_EMPTYING_HW_QUEUE_ADDBA) {
+               IWL_DEBUG_HT(priv, "AGG stop before setup done\n");
+               ieee80211_stop_tx_ba_cb_irqsafe(priv->vif, ra, tid);
+               priv->stations[sta_id].tid[tid].agg.state = IWL_AGG_OFF;
+               return 0;
+       }
+
        if (priv->stations[sta_id].tid[tid].agg.state != IWL_AGG_ON)
-               IWL_WARN(priv, "Stopping AGG while state not IWL_AGG_ON\n");
+               IWL_WARN(priv, "Stopping AGG while state not ON or starting\n");
 
        tid_data = &priv->stations[sta_id].tid[tid];
        ssn = (tid_data->seq_number & IEEE80211_SCTL_SEQ) >> 4;
@@ -1259,7 +1405,7 @@ int iwl_tx_agg_stop(struct iwl_priv *priv , const u8 *ra, u16 tid)
        if (ret)
                return ret;
 
-       ieee80211_stop_tx_ba_cb_irqsafe(priv->hw, ra, tid);
+       ieee80211_stop_tx_ba_cb_irqsafe(priv->vif, ra, tid);
 
        return 0;
 }
@@ -1283,7 +1429,7 @@ int iwl_txq_check_empty(struct iwl_priv *priv, int sta_id, u8 tid, int txq_id)
                        priv->cfg->ops->lib->txq_agg_disable(priv, txq_id,
                                                             ssn, tx_fifo);
                        tid_data->agg.state = IWL_AGG_OFF;
-                       ieee80211_stop_tx_ba_cb_irqsafe(priv->hw, addr, tid);
+                       ieee80211_stop_tx_ba_cb_irqsafe(priv->vif, addr, tid);
                }
                break;
        case IWL_EMPTYING_HW_QUEUE_ADDBA:
@@ -1291,7 +1437,7 @@ int iwl_txq_check_empty(struct iwl_priv *priv, int sta_id, u8 tid, int txq_id)
                if (tid_data->tfds_in_queue == 0) {
                        IWL_DEBUG_HT(priv, "HW queue empty: continue ADDBA flow\n");
                        tid_data->agg.state = IWL_AGG_ON;
-                       ieee80211_start_tx_ba_cb_irqsafe(priv->hw, addr, tid);
+                       ieee80211_start_tx_ba_cb_irqsafe(priv->vif, addr, tid);
                }
                break;
        }
@@ -1355,7 +1501,7 @@ static int iwl_tx_status_reply_compressed_ba(struct iwl_priv *priv,
 
        info = IEEE80211_SKB_CB(priv->txq[scd_flow].txb[agg->start_idx].skb[0]);
        memset(&info->status, 0, sizeof(info->status));
-       info->flags = IEEE80211_TX_STAT_ACK;
+       info->flags |= IEEE80211_TX_STAT_ACK;
        info->flags |= IEEE80211_TX_STAT_AMPDU;
        info->status.ampdu_ack_map = successes;
        info->status.ampdu_ack_len = agg->frame_count;
@@ -1375,7 +1521,7 @@ static int iwl_tx_status_reply_compressed_ba(struct iwl_priv *priv,
 void iwl_rx_reply_compressed_ba(struct iwl_priv *priv,
                                           struct iwl_rx_mem_buffer *rxb)
 {
-       struct iwl_rx_packet *pkt = (struct iwl_rx_packet *)rxb->skb->data;
+       struct iwl_rx_packet *pkt = rxb_addr(rxb);
        struct iwl_compressed_ba_resp *ba_resp = &pkt->u.compressed_ba;
        struct iwl_tx_queue *txq = NULL;
        struct iwl_ht_agg *agg;
@@ -1431,7 +1577,7 @@ void iwl_rx_reply_compressed_ba(struct iwl_priv *priv,
        if (txq->q.read_ptr != (ba_resp_scd_ssn & 0xff)) {
                /* calculate mac80211 ampdu sw queue to wake */
                int freed = iwl_tx_queue_reclaim(priv, scd_flow, index);
-               priv->stations[sta_id].tid[tid].tfds_in_queue -= freed;
+               iwl_free_tfds_in_queue(priv, sta_id, tid, freed);
 
                if ((iwl_queue_space(&txq->q) > txq->q.low_mark) &&
                    priv->mac80211_registered &&