Merge branch 'for-john' of git://git.kernel.org/pub/scm/linux/kernel/git/iwlwifi/iwlwifi-next

Conflicts:
	drivers/net/wireless/iwlwifi/pcie/trans.c

9 files changed, 1299 insertions, 1342 deletions

diff --git a/drivers/net/wireless/iwlwifi/dvm/mac80211.c b/drivers/net/wireless/iwlwifi/dvm/mac80211.c
index bf189f11541..c862c090634 100644
--- a/drivers/net/wireless/iwlwifi/dvm/mac80211.c
+++ b/drivers/net/wireless/iwlwifi/dvm/mac80211.c
@@ -168,8 +168,13 @@ int iwlagn_mac_setup_register(struct iwl_priv *priv,
 		hw->flags |= IEEE80211_HW_SUPPORTS_DYNAMIC_SMPS |
 			     IEEE80211_HW_SUPPORTS_STATIC_SMPS;
 
-	/* enable 11w if the uCode advertise */
-	if (capa->flags & IWL_UCODE_TLV_FLAGS_MFP)
+	/*
+	 * Enable 11w if advertised by firmware and software crypto
+	 * is not enabled (as the firmware will interpret some mgmt
+	 * packets, so enabling it with software crypto isn't safe)
+	 */
+	if (priv->fw->ucode_capa.flags & IWL_UCODE_TLV_FLAGS_MFP &&
+	    !iwlwifi_mod_params.sw_crypto)
 		hw->flags |= IEEE80211_HW_MFP_CAPABLE;
 
 	hw->sta_data_size = sizeof(struct iwl_station_priv);
diff --git a/drivers/net/wireless/iwlwifi/dvm/main.c b/drivers/net/wireless/iwlwifi/dvm/main.c
index 37bb4575ad8..e3a07c91681 100644
--- a/drivers/net/wireless/iwlwifi/dvm/main.c
+++ b/drivers/net/wireless/iwlwifi/dvm/main.c
@@ -1191,8 +1191,6 @@ static void iwl_option_config(struct iwl_priv *priv)
 
 static int iwl_eeprom_init_hw_params(struct iwl_priv *priv)
 {
-	priv->eeprom_data->sku = priv->eeprom_data->sku;
-
 	if (priv->eeprom_data->sku & EEPROM_SKU_CAP_11N_ENABLE &&
 	    !priv->cfg->ht_params) {
 		IWL_ERR(priv, "Invalid 11n configuration\n");
diff --git a/drivers/net/wireless/iwlwifi/iwl-config.h b/drivers/net/wireless/iwlwifi/iwl-config.h
index 87f465a49df..196266aa5a9 100644
--- a/drivers/net/wireless/iwlwifi/iwl-config.h
+++ b/drivers/net/wireless/iwlwifi/iwl-config.h
@@ -150,7 +150,7 @@ enum iwl_led_mode {
 struct iwl_base_params {
 	int eeprom_size;
 	int num_of_queues;	/* def: HW dependent */
-	/* for iwl_apm_init() */
+	/* for iwl_pcie_apm_init() */
 	u32 pll_cfg_val;
 
 	const u16 max_ll_items;
diff --git a/drivers/net/wireless/iwlwifi/iwl-fh.h b/drivers/net/wireless/iwlwifi/iwl-fh.h
index 80604664174..ec48563d3c6 100644
--- a/drivers/net/wireless/iwlwifi/iwl-fh.h
+++ b/drivers/net/wireless/iwlwifi/iwl-fh.h
@@ -267,7 +267,7 @@ static inline unsigned int FH_MEM_CBBC_QUEUE(unsigned int chnl)
 
 #define FH_RCSR_RX_CONFIG_RBDCB_SIZE_POS	(20)
 #define FH_RCSR_RX_CONFIG_REG_IRQ_RBTH_POS	(4)
-#define RX_RB_TIMEOUT	(0x10)
+#define RX_RB_TIMEOUT	(0x11)
 
 #define FH_RCSR_RX_CONFIG_CHNL_EN_PAUSE_VAL         (0x00000000)
 #define FH_RCSR_RX_CONFIG_CHNL_EN_PAUSE_EOF_VAL     (0x40000000)
diff --git a/drivers/net/wireless/iwlwifi/pcie/drv.c b/drivers/net/wireless/iwlwifi/pcie/drv.c
index 2a467539670..956fe6c370b 100644
--- a/drivers/net/wireless/iwlwifi/pcie/drv.c
+++ b/drivers/net/wireless/iwlwifi/pcie/drv.c
@@ -69,7 +69,6 @@
 
 #include "iwl-trans.h"
 #include "iwl-drv.h"
-#include "iwl-trans.h"
 
 #include "cfg.h"
 #include "internal.h"
diff --git a/drivers/net/wireless/iwlwifi/pcie/internal.h b/drivers/net/wireless/iwlwifi/pcie/internal.h
index 1f065c630d4..d91d2e8c62f 100644
--- a/drivers/net/wireless/iwlwifi/pcie/internal.h
+++ b/drivers/net/wireless/iwlwifi/pcie/internal.h
@@ -73,7 +73,7 @@ struct isr_statistics {
 };
 
 /**
- * struct iwl_rx_queue - Rx queue
+ * struct iwl_rxq - Rx queue
  * @bd: driver's pointer to buffer of receive buffer descriptors (rbd)
  * @bd_dma: bus address of buffer of receive buffer descriptors (rbd)
  * @pool:
@@ -91,7 +91,7 @@ struct isr_statistics {
  *
  * NOTE:  rx_free and rx_used are used as a FIFO for iwl_rx_mem_buffers
  */
-struct iwl_rx_queue {
+struct iwl_rxq {
 	__le32 *bd;
 	dma_addr_t bd_dma;
 	struct iwl_rx_mem_buffer pool[RX_QUEUE_SIZE + RX_FREE_BUFFERS];
@@ -157,8 +157,8 @@ struct iwl_cmd_meta {
  * 32 since we don't need so many commands pending. Since the HW
  * still uses 256 BDs for DMA though, n_bd stays 256. As a result,
  * the software buffers (in the variables @meta, @txb in struct
- * iwl_tx_queue) only have 32 entries, while the HW buffers (@tfds
- * in the same struct) have 256.
+ * iwl_txq) only have 32 entries, while the HW buffers (@tfds in
+ * the same struct) have 256.
  * This means that we end up with the following:
  *  HW entries: | 0 | ... | N * 32 | ... | N * 32 + 31 | ... | 255 |
  *  SW entries:           | 0      | ... | 31          |
@@ -182,7 +182,7 @@ struct iwl_queue {
 #define TFD_TX_CMD_SLOTS 256
 #define TFD_CMD_SLOTS 32
 
-struct iwl_pcie_tx_queue_entry {
+struct iwl_pcie_txq_entry {
 	struct iwl_device_cmd *cmd;
 	struct iwl_device_cmd *copy_cmd;
 	struct sk_buff *skb;
@@ -192,7 +192,7 @@ struct iwl_pcie_tx_queue_entry {
 };
 
 /**
- * struct iwl_tx_queue - Tx Queue for DMA
+ * struct iwl_txq - Tx Queue for DMA
  * @q: generic Rx/Tx queue descriptor
  * @tfds: transmit frame descriptors (DMA memory)
  * @entries: transmit entries (driver state)
@@ -205,10 +205,10 @@ struct iwl_pcie_tx_queue_entry {
  * A Tx queue consists of circular buffer of BDs (a.k.a. TFDs, transmit frame
  * descriptors) and required locking structures.
  */
-struct iwl_tx_queue {
+struct iwl_txq {
 	struct iwl_queue q;
 	struct iwl_tfd *tfds;
-	struct iwl_pcie_tx_queue_entry *entries;
+	struct iwl_pcie_txq_entry *entries;
 	spinlock_t lock;
 	struct timer_list stuck_timer;
 	struct iwl_trans_pcie *trans_pcie;
@@ -238,7 +238,7 @@ struct iwl_tx_queue {
  * @wd_timeout: queue watchdog timeout (jiffies)
  */
 struct iwl_trans_pcie {
-	struct iwl_rx_queue rxq;
+	struct iwl_rxq rxq;
 	struct work_struct rx_replenish;
 	struct iwl_trans *trans;
 	struct iwl_drv *drv;
@@ -260,7 +260,7 @@ struct iwl_trans_pcie {
 	struct iwl_dma_ptr scd_bc_tbls;
 	struct iwl_dma_ptr kw;
 
-	struct iwl_tx_queue *txq;
+	struct iwl_txq *txq;
 	unsigned long queue_used[BITS_TO_LONGS(IWL_MAX_HW_QUEUES)];
 	unsigned long queue_stopped[BITS_TO_LONGS(IWL_MAX_HW_QUEUES)];
 
@@ -287,10 +287,16 @@ struct iwl_trans_pcie {
 	unsigned long wd_timeout;
 };
 
-/*****************************************************
-* DRIVER STATUS FUNCTIONS
-******************************************************/
-enum {
+/**
+ * enum iwl_pcie_status: status of the PCIe transport
+ * @STATUS_HCMD_ACTIVE: a SYNC command is being processed
+ * @STATUS_DEVICE_ENABLED: APM is enabled
+ * @STATUS_TPOWER_PMI: the device might be asleep (need to wake it up)
+ * @STATUS_INT_ENABLED: interrupts are enabled
+ * @STATUS_RFKILL: the HW RFkill switch is in KILL position
+ * @STATUS_FW_ERROR: the fw is in error state
+ */
+enum iwl_pcie_status {
 	STATUS_HCMD_ACTIVE,
 	STATUS_DEVICE_ENABLED,
 	STATUS_TPOWER_PMI,
@@ -309,6 +315,10 @@ iwl_trans_pcie_get_trans(struct iwl_trans_pcie *trans_pcie)
 			    trans_specific);
 }
 
+/*
+ * Convention: trans API functions: iwl_trans_pcie_XXX
+ *	Other functions: iwl_pcie_XXX
+ */
 struct iwl_trans *iwl_trans_pcie_alloc(struct pci_dev *pdev,
 				       const struct pci_device_id *ent,
 				       const struct iwl_cfg *cfg);
@@ -317,51 +327,43 @@ void iwl_trans_pcie_free(struct iwl_trans *trans);
 /*****************************************************
 * RX
 ******************************************************/
-void iwl_bg_rx_replenish(struct work_struct *data);
-void iwl_irq_tasklet(struct iwl_trans *trans);
-void iwl_rx_replenish(struct iwl_trans *trans);
-void iwl_rx_queue_update_write_ptr(struct iwl_trans *trans,
-				   struct iwl_rx_queue *q);
+int iwl_pcie_rx_init(struct iwl_trans *trans);
+void iwl_pcie_tasklet(struct iwl_trans *trans);
+int iwl_pcie_rx_stop(struct iwl_trans *trans);
+void iwl_pcie_rx_free(struct iwl_trans *trans);
 
 /*****************************************************
-* ICT
+* ICT - interrupt handling
 ******************************************************/
-void iwl_reset_ict(struct iwl_trans *trans);
-void iwl_disable_ict(struct iwl_trans *trans);
-int iwl_alloc_isr_ict(struct iwl_trans *trans);
-void iwl_free_isr_ict(struct iwl_trans *trans);
-irqreturn_t iwl_isr_ict(int irq, void *data);
+irqreturn_t iwl_pcie_isr_ict(int irq, void *data);
+int iwl_pcie_alloc_ict(struct iwl_trans *trans);
+void iwl_pcie_free_ict(struct iwl_trans *trans);
+void iwl_pcie_reset_ict(struct iwl_trans *trans);
+void iwl_pcie_disable_ict(struct iwl_trans *trans);
 
 /*****************************************************
 * TX / HCMD
 ******************************************************/
-void iwl_txq_update_write_ptr(struct iwl_trans *trans,
-			      struct iwl_tx_queue *txq);
-int iwlagn_txq_attach_buf_to_tfd(struct iwl_trans *trans,
-				 struct iwl_tx_queue *txq,
-				 dma_addr_t addr, u16 len, u8 reset);
-int iwl_queue_init(struct iwl_queue *q, int count, int slots_num, u32 id);
-int iwl_trans_pcie_send_cmd(struct iwl_trans *trans, struct iwl_host_cmd *cmd);
-void iwl_tx_cmd_complete(struct iwl_trans *trans,
-			 struct iwl_rx_cmd_buffer *rxb, int handler_status);
-void iwl_trans_txq_update_byte_cnt_tbl(struct iwl_trans *trans,
-				       struct iwl_tx_queue *txq,
-				       u16 byte_cnt);
+int iwl_pcie_tx_init(struct iwl_trans *trans);
+void iwl_pcie_tx_start(struct iwl_trans *trans, u32 scd_base_addr);
+int iwl_pcie_tx_stop(struct iwl_trans *trans);
+void iwl_pcie_tx_free(struct iwl_trans *trans);
 void iwl_trans_pcie_txq_enable(struct iwl_trans *trans, int txq_id, int fifo,
 			       int sta_id, int tid, int frame_limit, u16 ssn);
 void iwl_trans_pcie_txq_disable(struct iwl_trans *trans, int queue);
-void iwl_txq_free_tfd(struct iwl_trans *trans, struct iwl_tx_queue *txq,
-		      enum dma_data_direction dma_dir);
-int iwl_tx_queue_reclaim(struct iwl_trans *trans, int txq_id, int index,
-			 struct sk_buff_head *skbs);
-void iwl_tx_queue_unmap(struct iwl_trans *trans, int txq_id);
-int iwl_queue_space(const struct iwl_queue *q);
-
+int iwl_trans_pcie_tx(struct iwl_trans *trans, struct sk_buff *skb,
+		      struct iwl_device_cmd *dev_cmd, int txq_id);
+void iwl_pcie_txq_inc_wr_ptr(struct iwl_trans *trans, struct iwl_txq *txq);
+int iwl_trans_pcie_send_hcmd(struct iwl_trans *trans, struct iwl_host_cmd *cmd);
+void iwl_pcie_hcmd_complete(struct iwl_trans *trans,
+			    struct iwl_rx_cmd_buffer *rxb, int handler_status);
+void iwl_trans_pcie_reclaim(struct iwl_trans *trans, int txq_id, int ssn,
+			    struct sk_buff_head *skbs);
 /*****************************************************
 * Error handling
 ******************************************************/
-int iwl_dump_fh(struct iwl_trans *trans, char **buf);
-void iwl_dump_csr(struct iwl_trans *trans);
+int iwl_pcie_dump_fh(struct iwl_trans *trans, char **buf);
+void iwl_pcie_dump_csr(struct iwl_trans *trans);
 
 /*****************************************************
 * Helpers
@@ -397,7 +399,7 @@ static inline void iwl_enable_rfkill_int(struct iwl_trans *trans)
 }
 
 static inline void iwl_wake_queue(struct iwl_trans *trans,
-				  struct iwl_tx_queue *txq)
+				  struct iwl_txq *txq)
 {
 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
 
@@ -408,7 +410,7 @@ static inline void iwl_wake_queue(struct iwl_trans *trans,
 }
 
 static inline void iwl_stop_queue(struct iwl_trans *trans,
-				  struct iwl_tx_queue *txq)
+				  struct iwl_txq *txq)
 {
 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
 
@@ -420,7 +422,7 @@ static inline void iwl_stop_queue(struct iwl_trans *trans,
 				    txq->q.id);
 }
 
-static inline int iwl_queue_used(const struct iwl_queue *q, int i)
+static inline bool iwl_queue_used(const struct iwl_queue *q, int i)
 {
 	return q->write_ptr >= q->read_ptr ?
 		(i >= q->read_ptr && i < q->write_ptr) :
@@ -432,8 +434,8 @@ static inline u8 get_cmd_index(struct iwl_queue *q, u32 index)
 	return index & (q->n_window - 1);
 }
 
-static inline const char *
-trans_pcie_get_cmd_string(struct iwl_trans_pcie *trans_pcie, u8 cmd)
+static inline const char *get_cmd_string(struct iwl_trans_pcie *trans_pcie,
+					 u8 cmd)
 {
 	if (!trans_pcie->command_names || !trans_pcie->command_names[cmd])
 		return "UNKNOWN";
diff --git a/drivers/net/wireless/iwlwifi/pcie/rx.c b/drivers/net/wireless/iwlwifi/pcie/rx.c
index 32307976956..bb32510fdd6 100644
--- a/drivers/net/wireless/iwlwifi/pcie/rx.c
+++ b/drivers/net/wireless/iwlwifi/pcie/rx.c
@@ -76,7 +76,7 @@
  * + A list of pre-allocated SKBs is stored in iwl->rxq->rx_free.  When
  *   iwl->rxq->free_count drops to or below RX_LOW_WATERMARK, work is scheduled
  *   to replenish the iwl->rxq->rx_free.
- * + In iwl_rx_replenish (scheduled) if 'processed' != 'read' then the
+ * + In iwl_pcie_rx_replenish (scheduled) if 'processed' != 'read' then the
  *   iwl->rxq is replenished and the READ INDEX is updated (updating the
  *   'processed' and 'read' driver indexes as well)
  * + A received packet is processed and handed to the kernel network stack,
@@ -89,28 +89,28 @@
  *
  * Driver sequence:
  *
- * iwl_rx_queue_alloc()   Allocates rx_free
- * iwl_rx_replenish()     Replenishes rx_free list from rx_used, and calls
- *                            iwl_rx_queue_restock
- * iwl_rx_queue_restock() Moves available buffers from rx_free into Rx
+ * iwl_rxq_alloc()            Allocates rx_free
+ * iwl_pcie_rx_replenish()    Replenishes rx_free list from rx_used, and calls
+ *                            iwl_pcie_rxq_restock
+ * iwl_pcie_rxq_restock()     Moves available buffers from rx_free into Rx
  *                            queue, updates firmware pointers, and updates
  *                            the WRITE index.  If insufficient rx_free buffers
- *                            are available, schedules iwl_rx_replenish
+ *                            are available, schedules iwl_pcie_rx_replenish
  *
  * -- enable interrupts --
- * ISR - iwl_rx()         Detach iwl_rx_mem_buffers from pool up to the
+ * ISR - iwl_rx()             Detach iwl_rx_mem_buffers from pool up to the
  *                            READ INDEX, detaching the SKB from the pool.
  *                            Moves the packet buffer from queue to rx_used.
- *                            Calls iwl_rx_queue_restock to refill any empty
+ *                            Calls iwl_pcie_rxq_restock to refill any empty
  *                            slots.
  * ...
  *
  */
 
-/**
- * iwl_rx_queue_space - Return number of free slots available in queue.
+/*
+ * iwl_rxq_space - Return number of free slots available in queue.
  */
-static int iwl_rx_queue_space(const struct iwl_rx_queue *q)
+static int iwl_rxq_space(const struct iwl_rxq *q)
 {
 	int s = q->read - q->write;
 	if (s <= 0)
@@ -122,11 +122,28 @@ static int iwl_rx_queue_space(const struct iwl_rx_queue *q)
 	return s;
 }
 
-/**
- * iwl_rx_queue_update_write_ptr - Update the write pointer for the RX queue
+/*
+ * iwl_dma_addr2rbd_ptr - convert a DMA address to a uCode read buffer ptr
+ */
+static inline __le32 iwl_pcie_dma_addr2rbd_ptr(dma_addr_t dma_addr)
+{
+	return cpu_to_le32((u32)(dma_addr >> 8));
+}
+
+/*
+ * iwl_pcie_rx_stop - stops the Rx DMA
+ */
+int iwl_pcie_rx_stop(struct iwl_trans *trans)
+{
+	iwl_write_direct32(trans, FH_MEM_RCSR_CHNL0_CONFIG_REG, 0);
+	return iwl_poll_direct_bit(trans, FH_MEM_RSSR_RX_STATUS_REG,
+				   FH_RSSR_CHNL0_RX_STATUS_CHNL_IDLE, 1000);
+}
+
+/*
+ * iwl_pcie_rxq_inc_wr_ptr - Update the write pointer for the RX queue
  */
-void iwl_rx_queue_update_write_ptr(struct iwl_trans *trans,
-				   struct iwl_rx_queue *q)
+static void iwl_pcie_rxq_inc_wr_ptr(struct iwl_trans *trans, struct iwl_rxq *q)
 {
 	unsigned long flags;
 	u32 reg;
@@ -176,16 +193,8 @@ void iwl_rx_queue_update_write_ptr(struct iwl_trans *trans,
 	spin_unlock_irqrestore(&q->lock, flags);
 }
 
-/**
- * iwl_dma_addr2rbd_ptr - convert a DMA address to a uCode read buffer ptr
- */
-static inline __le32 iwl_dma_addr2rbd_ptr(dma_addr_t dma_addr)
-{
-	return cpu_to_le32((u32)(dma_addr >> 8));
-}
-
-/**
- * iwl_rx_queue_restock - refill RX queue from pre-allocated pool
+/*
+ * iwl_pcie_rxq_restock - refill RX queue from pre-allocated pool
  *
  * If there are slots in the RX queue that need to be restocked,
  * and we have free pre-allocated buffers, fill the ranks as much
@@ -195,10 +204,10 @@ static inline __le32 iwl_dma_addr2rbd_ptr(dma_addr_t dma_addr)
  * also updates the memory address in the firmware to reference the new
  * target buffer.
  */
-static void iwl_rx_queue_restock(struct iwl_trans *trans)
+static void iwl_pcie_rxq_restock(struct iwl_trans *trans)
 {
 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
-	struct iwl_rx_queue *rxq = &trans_pcie->rxq;
+	struct iwl_rxq *rxq = &trans_pcie->rxq;
 	struct iwl_rx_mem_buffer *rxb;
 	unsigned long flags;
 
@@ -214,7 +223,7 @@ static void iwl_rx_queue_restock(struct iwl_trans *trans)
 		return;
 
 	spin_lock_irqsave(&rxq->lock, flags);
-	while ((iwl_rx_queue_space(rxq) > 0) && (rxq->free_count)) {
+	while ((iwl_rxq_space(rxq) > 0) && (rxq->free_count)) {
 		/* The overwritten rxb must be a used one */
 		rxb = rxq->queue[rxq->write];
 		BUG_ON(rxb && rxb->page);
@@ -225,7 +234,7 @@ static void iwl_rx_queue_restock(struct iwl_trans *trans)
 		list_del(&rxb->list);
 
 		/* Point to Rx buffer via next RBD in circular buffer */
-		rxq->bd[rxq->write] = iwl_dma_addr2rbd_ptr(rxb->page_dma);
+		rxq->bd[rxq->write] = iwl_pcie_dma_addr2rbd_ptr(rxb->page_dma);
 		rxq->queue[rxq->write] = rxb;
 		rxq->write = (rxq->write + 1) & RX_QUEUE_MASK;
 		rxq->free_count--;
@@ -242,23 +251,23 @@ static void iwl_rx_queue_restock(struct iwl_trans *trans)
 		spin_lock_irqsave(&rxq->lock, flags);
 		rxq->need_update = 1;
 		spin_unlock_irqrestore(&rxq->lock, flags);
-		iwl_rx_queue_update_write_ptr(trans, rxq);
+		iwl_pcie_rxq_inc_wr_ptr(trans, rxq);
 	}
 }
 
 /*
- * iwl_rx_allocate - allocate a page for each used RBD
+ * iwl_pcie_rxq_alloc_rbs - allocate a page for each used RBD
  *
  * A used RBD is an Rx buffer that has been given to the stack. To use it again
  * a page must be allocated and the RBD must point to the page. This function
  * doesn't change the HW pointer but handles the list of pages that is used by
- * iwl_rx_queue_restock. The latter function will update the HW to use the newly
+ * iwl_pcie_rxq_restock. The latter function will update the HW to use the newly
  * allocated buffers.
  */
-static void iwl_rx_allocate(struct iwl_trans *trans, gfp_t priority)
+static void iwl_pcie_rxq_alloc_rbs(struct iwl_trans *trans, gfp_t priority)
 {
 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
-	struct iwl_rx_queue *rxq = &trans_pcie->rxq;
+	struct iwl_rxq *rxq = &trans_pcie->rxq;
 	struct iwl_rx_mem_buffer *rxb;
 	struct page *page;
 	unsigned long flags;
@@ -340,47 +349,227 @@ static void iwl_rx_allocate(struct iwl_trans *trans, gfp_t priority)
 	}
 }
 
+static void iwl_pcie_rxq_free_rbs(struct iwl_trans *trans)
+{
+	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
+	struct iwl_rxq *rxq = &trans_pcie->rxq;
+	int i;
+
+	/* Fill the rx_used queue with _all_ of the Rx buffers */
+	for (i = 0; i < RX_FREE_BUFFERS + RX_QUEUE_SIZE; i++) {
+		/* In the reset function, these buffers may have been allocated
+		 * to an SKB, so we need to unmap and free potential storage */
+		if (rxq->pool[i].page != NULL) {
+			dma_unmap_page(trans->dev, rxq->pool[i].page_dma,
+				       PAGE_SIZE << trans_pcie->rx_page_order,
+				       DMA_FROM_DEVICE);
+			__free_pages(rxq->pool[i].page,
+				     trans_pcie->rx_page_order);
+			rxq->pool[i].page = NULL;
+		}
+		list_add_tail(&rxq->pool[i].list, &rxq->rx_used);
+	}
+}
+
 /*
- * iwl_rx_replenish - Move all used buffers from rx_used to rx_free
+ * iwl_pcie_rx_replenish - Move all used buffers from rx_used to rx_free
  *
  * When moving to rx_free an page is allocated for the slot.
  *
- * Also restock the Rx queue via iwl_rx_queue_restock.
+ * Also restock the Rx queue via iwl_pcie_rxq_restock.
  * This is called as a scheduled work item (except for during initialization)
  */
-void iwl_rx_replenish(struct iwl_trans *trans)
+static void iwl_pcie_rx_replenish(struct iwl_trans *trans)
 {
 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
 	unsigned long flags;
 
-	iwl_rx_allocate(trans, GFP_KERNEL);
+	iwl_pcie_rxq_alloc_rbs(trans, GFP_KERNEL);
 
 	spin_lock_irqsave(&trans_pcie->irq_lock, flags);
-	iwl_rx_queue_restock(trans);
+	iwl_pcie_rxq_restock(trans);
 	spin_unlock_irqrestore(&trans_pcie->irq_lock, flags);
 }
 
-static void iwl_rx_replenish_now(struct iwl_trans *trans)
+static void iwl_pcie_rx_replenish_now(struct iwl_trans *trans)
 {
-	iwl_rx_allocate(trans, GFP_ATOMIC);
+	iwl_pcie_rxq_alloc_rbs(trans, GFP_ATOMIC);
 
-	iwl_rx_queue_restock(trans);
+	iwl_pcie_rxq_restock(trans);
 }
 
-void iwl_bg_rx_replenish(struct work_struct *data)
+static void iwl_pcie_rx_replenish_work(struct work_struct *data)
 {
 	struct iwl_trans_pcie *trans_pcie =
 	    container_of(data, struct iwl_trans_pcie, rx_replenish);
 
-	iwl_rx_replenish(trans_pcie->trans);
+	iwl_pcie_rx_replenish(trans_pcie->trans);
+}
+
+static int iwl_pcie_rx_alloc(struct iwl_trans *trans)
+{
+	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
+	struct iwl_rxq *rxq = &trans_pcie->rxq;
+	struct device *dev = trans->dev;
+
+	memset(&trans_pcie->rxq, 0, sizeof(trans_pcie->rxq));
+
+	spin_lock_init(&rxq->lock);
+
+	if (WARN_ON(rxq->bd || rxq->rb_stts))
+		return -EINVAL;
+
+	/* Allocate the circular buffer of Read Buffer Descriptors (RBDs) */
+	rxq->bd = dma_zalloc_coherent(dev, sizeof(__le32) * RX_QUEUE_SIZE,
+				      &rxq->bd_dma, GFP_KERNEL);
+	if (!rxq->bd)
+		goto err_bd;
+
+	/*Allocate the driver's pointer to receive buffer status */
+	rxq->rb_stts = dma_zalloc_coherent(dev, sizeof(*rxq->rb_stts),
+					   &rxq->rb_stts_dma, GFP_KERNEL);
+	if (!rxq->rb_stts)
+		goto err_rb_stts;
+
+	return 0;
+
+err_rb_stts:
+	dma_free_coherent(dev, sizeof(__le32) * RX_QUEUE_SIZE,
+			  rxq->bd, rxq->bd_dma);
+	memset(&rxq->bd_dma, 0, sizeof(rxq->bd_dma));
+	rxq->bd = NULL;
+err_bd:
+	return -ENOMEM;
 }
 
-static void iwl_rx_handle_rxbuf(struct iwl_trans *trans,
+static void iwl_pcie_rx_hw_init(struct iwl_trans *trans, struct iwl_rxq *rxq)
+{
+	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
+	u32 rb_size;
+	const u32 rfdnlog = RX_QUEUE_SIZE_LOG; /* 256 RBDs */
+
+	if (trans_pcie->rx_buf_size_8k)
+		rb_size = FH_RCSR_RX_CONFIG_REG_VAL_RB_SIZE_8K;
+	else
+		rb_size = FH_RCSR_RX_CONFIG_REG_VAL_RB_SIZE_4K;
+
+	/* Stop Rx DMA */
+	iwl_write_direct32(trans, FH_MEM_RCSR_CHNL0_CONFIG_REG, 0);
+
+	/* Reset driver's Rx queue write index */
+	iwl_write_direct32(trans, FH_RSCSR_CHNL0_RBDCB_WPTR_REG, 0);
+
+	/* Tell device where to find RBD circular buffer in DRAM */
+	iwl_write_direct32(trans, FH_RSCSR_CHNL0_RBDCB_BASE_REG,
+			   (u32)(rxq->bd_dma >> 8));
+
+	/* Tell device where in DRAM to update its Rx status */
+	iwl_write_direct32(trans, FH_RSCSR_CHNL0_STTS_WPTR_REG,
+			   rxq->rb_stts_dma >> 4);
+
+	/* Enable Rx DMA
+	 * FH_RCSR_CHNL0_RX_IGNORE_RXF_EMPTY is set because of HW bug in
+	 *      the credit mechanism in 5000 HW RX FIFO
+	 * Direct rx interrupts to hosts
+	 * Rx buffer size 4 or 8k
+	 * RB timeout 0x10
+	 * 256 RBDs
+	 */
+	iwl_write_direct32(trans, FH_MEM_RCSR_CHNL0_CONFIG_REG,
+			   FH_RCSR_RX_CONFIG_CHNL_EN_ENABLE_VAL |
+			   FH_RCSR_CHNL0_RX_IGNORE_RXF_EMPTY |
+			   FH_RCSR_CHNL0_RX_CONFIG_IRQ_DEST_INT_HOST_VAL |
+			   rb_size|
+			   (RX_RB_TIMEOUT << FH_RCSR_RX_CONFIG_REG_IRQ_RBTH_POS)|
+			   (rfdnlog << FH_RCSR_RX_CONFIG_RBDCB_SIZE_POS));
+
+	/* Set interrupt coalescing timer to default (2048 usecs) */
+	iwl_write8(trans, CSR_INT_COALESCING, IWL_HOST_INT_TIMEOUT_DEF);
+}
+
+int iwl_pcie_rx_init(struct iwl_trans *trans)
+{
+	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
+	struct iwl_rxq *rxq = &trans_pcie->rxq;
+
+	int i, err;
+	unsigned long flags;
+
+	if (!rxq->bd) {
+		err = iwl_pcie_rx_alloc(trans);
+		if (err)
+			return err;
+	}
+
+	spin_lock_irqsave(&rxq->lock, flags);
+	INIT_LIST_HEAD(&rxq->rx_free);
+	INIT_LIST_HEAD(&rxq->rx_used);
+
+	INIT_WORK(&trans_pcie->rx_replenish,
+		  iwl_pcie_rx_replenish_work);
+
+	iwl_pcie_rxq_free_rbs(trans);
+
+	for (i = 0; i < RX_QUEUE_SIZE; i++)
+		rxq->queue[i] = NULL;
+
+	/* Set us so that we have processed and used all buffers, but have
+	 * not restocked the Rx queue with fresh buffers */
+	rxq->read = rxq->write = 0;
+	rxq->write_actual = 0;
+	rxq->free_count = 0;
+	spin_unlock_irqrestore(&rxq->lock, flags);
+
+	iwl_pcie_rx_replenish(trans);
+
+	iwl_pcie_rx_hw_init(trans, rxq);
+
+	spin_lock_irqsave(&trans_pcie->irq_lock, flags);
+	rxq->need_update = 1;
+	iwl_pcie_rxq_inc_wr_ptr(trans, rxq);
+	spin_unlock_irqrestore(&trans_pcie->irq_lock, flags);
+
+	return 0;
+}
+
+void iwl_pcie_rx_free(struct iwl_trans *trans)
+{
+	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
+	struct iwl_rxq *rxq = &trans_pcie->rxq;
+	unsigned long flags;
+
+	/*if rxq->bd is NULL, it means that nothing has been allocated,
+	 * exit now */
+	if (!rxq->bd) {
+		IWL_DEBUG_INFO(trans, "Free NULL rx context\n");
+		return;
+	}
+
+	spin_lock_irqsave(&rxq->lock, flags);
+	iwl_pcie_rxq_free_rbs(trans);
+	spin_unlock_irqrestore(&rxq->lock, flags);
+
+	dma_free_coherent(trans->dev, sizeof(__le32) * RX_QUEUE_SIZE,
+			  rxq->bd, rxq->bd_dma);
+	memset(&rxq->bd_dma, 0, sizeof(rxq->bd_dma));
+	rxq->bd = NULL;
+
+	if (rxq->rb_stts)
+		dma_free_coherent(trans->dev,
+				  sizeof(struct iwl_rb_status),
+				  rxq->rb_stts, rxq->rb_stts_dma);
+	else
+		IWL_DEBUG_INFO(trans, "Free rxq->rb_stts which is NULL\n");
+	memset(&rxq->rb_stts_dma, 0, sizeof(rxq->rb_stts_dma));
+	rxq->rb_stts = NULL;
+}
+
+static void iwl_pcie_rx_handle_rb(struct iwl_trans *trans,
 				struct iwl_rx_mem_buffer *rxb)
 {
 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
-	struct iwl_rx_queue *rxq = &trans_pcie->rxq;
-	struct iwl_tx_queue *txq = &trans_pcie->txq[trans_pcie->cmd_queue];
+	struct iwl_rxq *rxq = &trans_pcie->rxq;
+	struct iwl_txq *txq = &trans_pcie->txq[trans_pcie->cmd_queue];
 	unsigned long flags;
 	bool page_stolen = false;
 	int max_len = PAGE_SIZE << trans_pcie->rx_page_order;
@@ -410,8 +599,7 @@ static void iwl_rx_handle_rxbuf(struct iwl_trans *trans,
 			break;
 
 		IWL_DEBUG_RX(trans, "cmd at offset %d: %s (0x%.2x)\n",
-			rxcb._offset,
-			trans_pcie_get_cmd_string(trans_pcie, pkt->hdr.cmd),
+			rxcb._offset, get_cmd_string(trans_pcie, pkt->hdr.cmd),
 			pkt->hdr.cmd);
 
 		len = le32_to_cpu(pkt->len_n_flags) & FH_RSCSR_FRAME_SIZE_MSK;
@@ -443,7 +631,7 @@ static void iwl_rx_handle_rxbuf(struct iwl_trans *trans,
 		cmd_index = get_cmd_index(&txq->q, index);
 
 		if (reclaim) {
-			struct iwl_pcie_tx_queue_entry *ent;
+			struct iwl_pcie_txq_entry *ent;
 			ent = &txq->entries[cmd_index];
 			cmd = ent->copy_cmd;
 			WARN_ON_ONCE(!cmd && ent->meta.flags & CMD_WANT_HCMD);
@@ -473,7 +661,7 @@ static void iwl_rx_handle_rxbuf(struct iwl_trans *trans,
 			 * iwl_trans_send_cmd()
 			 * as we reclaim the driver command queue */
 			if (!rxcb._page_stolen)
-				iwl_tx_cmd_complete(trans, &rxcb, err);
+				iwl_pcie_hcmd_complete(trans, &rxcb, err);
 			else
 				IWL_WARN(trans, "Claim null rxb?\n");
 		}
@@ -515,17 +703,13 @@ static void iwl_rx_handle_rxbuf(struct iwl_trans *trans,
 	spin_unlock_irqrestore(&rxq->lock, flags);
 }
 
-/**
- * iwl_rx_handle - Main entry function for receiving responses from uCode
- *
- * Uses the priv->rx_handlers callback function array to invoke
- * the appropriate handlers, including command responses,
- * frame-received notifications, and other notifications.
+/*
+ * iwl_pcie_rx_handle - Main entry function for receiving responses from fw
  */
-static void iwl_rx_handle(struct iwl_trans *trans)
+static void iwl_pcie_rx_handle(struct iwl_trans *trans)
 {
 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
-	struct iwl_rx_queue *rxq = &trans_pcie->rxq;
+	struct iwl_rxq *rxq = &trans_pcie->rxq;
 	u32 r, i;
 	u8 fill_rx = 0;
 	u32 count = 8;
@@ -556,7 +740,7 @@ static void iwl_rx_handle(struct iwl_trans *trans)
 
 		IWL_DEBUG_RX(trans, "rxbuf: HW = %d, SW = %d (%p)\n",
 			     r, i, rxb);
-		iwl_rx_handle_rxbuf(trans, rxb);
+		iwl_pcie_rx_handle_rb(trans, rxb);
 
 		i = (i + 1) & RX_QUEUE_MASK;
 		/* If there are a lot of unused frames,
@@ -565,7 +749,7 @@ static void iwl_rx_handle(struct iwl_trans *trans)
 			count++;
 			if (count >= 8) {
 				rxq->read = i;
-				iwl_rx_replenish_now(trans);
+				iwl_pcie_rx_replenish_now(trans);
 				count = 0;
 			}
 		}
@@ -574,15 +758,15 @@ static void iwl_rx_handle(struct iwl_trans *trans)
 	/* Backtrack one entry */
 	rxq->read = i;
 	if (fill_rx)
-		iwl_rx_replenish_now(trans);
+		iwl_pcie_rx_replenish_now(trans);
 	else
-		iwl_rx_queue_restock(trans);
+		iwl_pcie_rxq_restock(trans);
 }
 
-/**
- * iwl_irq_handle_error - called for HW or SW error interrupt from card
+/*
+ * iwl_pcie_irq_handle_error - called for HW or SW error interrupt from card
  */
-static void iwl_irq_handle_error(struct iwl_trans *trans)
+static void iwl_pcie_irq_handle_error(struct iwl_trans *trans)
 {
 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
 
@@ -598,8 +782,8 @@ static void iwl_irq_handle_error(struct iwl_trans *trans)
 		return;
 	}
 
-	iwl_dump_csr(trans);
-	iwl_dump_fh(trans, NULL);
+	iwl_pcie_dump_csr(trans);
+	iwl_pcie_dump_fh(trans, NULL);
 
 	set_bit(STATUS_FW_ERROR, &trans_pcie->status);
 	clear_bit(STATUS_HCMD_ACTIVE, &trans_pcie->status);
@@ -608,8 +792,7 @@ static void iwl_irq_handle_error(struct iwl_trans *trans)
 	iwl_op_mode_nic_error(trans->op_mode);
 }
 
-/* tasklet for iwlagn interrupt */
-void iwl_irq_tasklet(struct iwl_trans *trans)
+void iwl_pcie_tasklet(struct iwl_trans *trans)
 {
 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
 	struct isr_statistics *isr_stats = &trans_pcie->isr_stats;
@@ -661,7 +844,7 @@ void iwl_irq_tasklet(struct iwl_trans *trans)
 		iwl_disable_interrupts(trans);
 
 		isr_stats->hw++;
-		iwl_irq_handle_error(trans);
+		iwl_pcie_irq_handle_error(trans);
 
 		handled |= CSR_INT_BIT_HW_ERR;
 
@@ -724,17 +907,16 @@ void iwl_irq_tasklet(struct iwl_trans *trans)
 		IWL_ERR(trans, "Microcode SW error detected. "
 			" Restarting 0x%X.\n", inta);
 		isr_stats->sw++;
-		iwl_irq_handle_error(trans);
+		iwl_pcie_irq_handle_error(trans);
 		handled |= CSR_INT_BIT_SW_ERR;
 	}
 
 	/* uCode wakes up after power-down sleep */
 	if (inta & CSR_INT_BIT_WAKEUP) {
 		IWL_DEBUG_ISR(trans, "Wakeup interrupt\n");
-		iwl_rx_queue_update_write_ptr(trans, &trans_pcie->rxq);
+		iwl_pcie_rxq_inc_wr_ptr(trans, &trans_pcie->rxq);
 		for (i = 0; i < trans->cfg->base_params->num_of_queues; i++)
-			iwl_txq_update_write_ptr(trans,
-						 &trans_pcie->txq[i]);
+			iwl_pcie_txq_inc_wr_ptr(trans, &trans_pcie->txq[i]);
 
 		isr_stats->wakeup++;
 
@@ -772,7 +954,7 @@ void iwl_irq_tasklet(struct iwl_trans *trans)
 		iwl_write8(trans, CSR_INT_PERIODIC_REG,
 			    CSR_INT_PERIODIC_DIS);
 
-		iwl_rx_handle(trans);
+		iwl_pcie_rx_handle(trans);
 
 		/*
 		 * Enable periodic interrupt in 8 msec only if we received
@@ -830,7 +1012,7 @@ void iwl_irq_tasklet(struct iwl_trans *trans)
 #define ICT_COUNT	(ICT_SIZE / sizeof(u32))
 
 /* Free dram table */
-void iwl_free_isr_ict(struct iwl_trans *trans)
+void iwl_pcie_free_ict(struct iwl_trans *trans)
 {
 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
 
@@ -843,13 +1025,12 @@ void iwl_free_isr_ict(struct iwl_trans *trans)
 	}
 }
 
-
 /*
  * allocate dram shared table, it is an aligned memory
  * block of ICT_SIZE.
  * also reset all data related to ICT table interrupt.
  */
-int iwl_alloc_isr_ict(struct iwl_trans *trans)
+int iwl_pcie_alloc_ict(struct iwl_trans *trans)
 {
 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
 
@@ -862,7 +1043,7 @@ int iwl_alloc_isr_ict(struct iwl_trans *trans)
 
 	/* just an API sanity check ... it is guaranteed to be aligned */
 	if (WARN_ON(trans_pcie->ict_tbl_dma & (ICT_SIZE - 1))) {
-		iwl_free_isr_ict(trans);
+		iwl_pcie_free_ict(trans);
 		return -EINVAL;
 	}
 
@@ -883,7 +1064,7 @@ int iwl_alloc_isr_ict(struct iwl_trans *trans)
 /* Device is going up inform it about using ICT interrupt table,
  * also we need to tell the driver to start using ICT interrupt.
  */
-void iwl_reset_ict(struct iwl_trans *trans)
+void iwl_pcie_reset_ict(struct iwl_trans *trans)
 {
 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
 	u32 val;
@@ -913,7 +1094,7 @@ void iwl_reset_ict(struct iwl_trans *trans)
 }
 
 /* Device is going down disable ict interrupt usage */
-void iwl_disable_ict(struct iwl_trans *trans)
+void iwl_pcie_disable_ict(struct iwl_trans *trans)
 {
 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
 	unsigned long flags;
@@ -924,7 +1105,7 @@ void iwl_disable_ict(struct iwl_trans *trans)
 }
 
 /* legacy (non-ICT) ISR. Assumes that trans_pcie->irq_lock is held */
-static irqreturn_t iwl_isr(int irq, void *data)
+static irqreturn_t iwl_pcie_isr(int irq, void *data)
 {
 	struct iwl_trans *trans = data;
 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
@@ -971,7 +1152,7 @@ static irqreturn_t iwl_isr(int irq, void *data)
 #endif
 
 	trans_pcie->inta |= inta;
-	/* iwl_irq_tasklet() will service interrupts and re-enable them */
+	/* iwl_pcie_tasklet() will service interrupts and re-enable them */
 	if (likely(inta))
 		tasklet_schedule(&trans_pcie->irq_tasklet);
 	else if (test_bit(STATUS_INT_ENABLED, &trans_pcie->status) &&
@@ -996,7 +1177,7 @@ none:
  * the interrupt we need to service, driver will set the entries back to 0 and
  * set index.
  */
-irqreturn_t iwl_isr_ict(int irq, void *data)
+irqreturn_t iwl_pcie_isr_ict(int irq, void *data)
 {
 	struct iwl_trans *trans = data;
 	struct iwl_trans_pcie *trans_pcie;
@@ -1016,14 +1197,13 @@ irqreturn_t iwl_isr_ict(int irq, void *data)
 	 * use legacy interrupt.
 	 */
 	if (unlikely(!trans_pcie->use_ict)) {
-		irqreturn_t ret = iwl_isr(irq, data);
+		irqreturn_t ret = iwl_pcie_isr(irq, data);
 		spin_unlock_irqrestore(&trans_pcie->irq_lock, flags);
 		return ret;
 	}
 
 	trace_iwlwifi_dev_irq(trans->dev);
 
-
 	/* Disable (but don't clear!) interrupts here to avoid
 	 * back-to-back ISRs and sporadic interrupts from our NIC.
 	 * If we have something to service, the tasklet will re-enable ints.
@@ -1032,7 +1212,6 @@ irqreturn_t iwl_isr_ict(int irq, void *data)
 	inta_mask = iwl_read32(trans, CSR_INT_MASK);  /* just for debug */
 	iwl_write32(trans, CSR_INT_MASK, 0x00000000);
 
-
 	/* Ignore interrupt if there's nothing in NIC to service.
 	 * This may be due to IRQ shared with another device,
 	 * or due to sporadic interrupts thrown from our NIC. */
@@ -1081,7 +1260,7 @@ irqreturn_t iwl_isr_ict(int irq, void *data)
 	inta &= trans_pcie->inta_mask;
 	trans_pcie->inta |= inta;
 
-	/* iwl_irq_tasklet() will service interrupts and re-enable them */
+	/* iwl_pcie_tasklet() will service interrupts and re-enable them */
 	if (likely(inta))
 		tasklet_schedule(&trans_pcie->irq_tasklet);
 	else if (test_bit(STATUS_INT_ENABLED, &trans_pcie->status) &&
diff --git a/drivers/net/wireless/iwlwifi/pcie/trans.c b/drivers/net/wireless/iwlwifi/pcie/trans.c
index f21bf661931..f6c21e7edaf 100644
--- a/drivers/net/wireless/iwlwifi/pcie/trans.c
+++ b/drivers/net/wireless/iwlwifi/pcie/trans.c
@@ -74,585 +74,8 @@
 #include "iwl-prph.h"
 #include "iwl-agn-hw.h"
 #include "internal.h"
-/* FIXME: need to abstract out TX command (once we know what it looks like) */
-#include "dvm/commands.h"
 
-#define SCD_QUEUECHAIN_SEL_ALL(trans, trans_pcie)	\
-	(((1<<trans->cfg->base_params->num_of_queues) - 1) &\
-	(~(1<<(trans_pcie)->cmd_queue)))
-
-static int iwl_trans_rx_alloc(struct iwl_trans *trans)
-{
-	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
-	struct iwl_rx_queue *rxq = &trans_pcie->rxq;
-	struct device *dev = trans->dev;
-
-	memset(&trans_pcie->rxq, 0, sizeof(trans_pcie->rxq));
-
-	spin_lock_init(&rxq->lock);
-
-	if (WARN_ON(rxq->bd || rxq->rb_stts))
-		return -EINVAL;
-
-	/* Allocate the circular buffer of Read Buffer Descriptors (RBDs) */
-	rxq->bd = dma_zalloc_coherent(dev, sizeof(__le32) * RX_QUEUE_SIZE,
-				      &rxq->bd_dma, GFP_KERNEL);
-	if (!rxq->bd)
-		goto err_bd;
-
-	/*Allocate the driver's pointer to receive buffer status */
-	rxq->rb_stts = dma_zalloc_coherent(dev, sizeof(*rxq->rb_stts),
-					   &rxq->rb_stts_dma, GFP_KERNEL);
-	if (!rxq->rb_stts)
-		goto err_rb_stts;
-
-	return 0;
-
-err_rb_stts:
-	dma_free_coherent(dev, sizeof(__le32) * RX_QUEUE_SIZE,
-			  rxq->bd, rxq->bd_dma);
-	memset(&rxq->bd_dma, 0, sizeof(rxq->bd_dma));
-	rxq->bd = NULL;
-err_bd:
-	return -ENOMEM;
-}
-
-static void iwl_trans_rxq_free_rx_bufs(struct iwl_trans *trans)
-{
-	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
-	struct iwl_rx_queue *rxq = &trans_pcie->rxq;
-	int i;
-
-	/* Fill the rx_used queue with _all_ of the Rx buffers */
-	for (i = 0; i < RX_FREE_BUFFERS + RX_QUEUE_SIZE; i++) {
-		/* In the reset function, these buffers may have been allocated
-		 * to an SKB, so we need to unmap and free potential storage */
-		if (rxq->pool[i].page != NULL) {
-			dma_unmap_page(trans->dev, rxq->pool[i].page_dma,
-				       PAGE_SIZE << trans_pcie->rx_page_order,
-				       DMA_FROM_DEVICE);
-			__free_pages(rxq->pool[i].page,
-				     trans_pcie->rx_page_order);
-			rxq->pool[i].page = NULL;
-		}
-		list_add_tail(&rxq->pool[i].list, &rxq->rx_used);
-	}
-}
-
-static void iwl_trans_rx_hw_init(struct iwl_trans *trans,
-				 struct iwl_rx_queue *rxq)
-{
-	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
-	u32 rb_size;
-	const u32 rfdnlog = RX_QUEUE_SIZE_LOG; /* 256 RBDs */
-	u32 rb_timeout = RX_RB_TIMEOUT; /* FIXME: RX_RB_TIMEOUT for all devices? */
-
-	if (trans_pcie->rx_buf_size_8k)
-		rb_size = FH_RCSR_RX_CONFIG_REG_VAL_RB_SIZE_8K;
-	else
-		rb_size = FH_RCSR_RX_CONFIG_REG_VAL_RB_SIZE_4K;
-
-	/* Stop Rx DMA */
-	iwl_write_direct32(trans, FH_MEM_RCSR_CHNL0_CONFIG_REG, 0);
-
-	/* Reset driver's Rx queue write index */
-	iwl_write_direct32(trans, FH_RSCSR_CHNL0_RBDCB_WPTR_REG, 0);
-
-	/* Tell device where to find RBD circular buffer in DRAM */
-	iwl_write_direct32(trans, FH_RSCSR_CHNL0_RBDCB_BASE_REG,
-			   (u32)(rxq->bd_dma >> 8));
-
-	/* Tell device where in DRAM to update its Rx status */
-	iwl_write_direct32(trans, FH_RSCSR_CHNL0_STTS_WPTR_REG,
-			   rxq->rb_stts_dma >> 4);
-
-	/* Enable Rx DMA
-	 * FH_RCSR_CHNL0_RX_IGNORE_RXF_EMPTY is set because of HW bug in
-	 *      the credit mechanism in 5000 HW RX FIFO
-	 * Direct rx interrupts to hosts
-	 * Rx buffer size 4 or 8k
-	 * RB timeout 0x10
-	 * 256 RBDs
-	 */
-	iwl_write_direct32(trans, FH_MEM_RCSR_CHNL0_CONFIG_REG,
-			   FH_RCSR_RX_CONFIG_CHNL_EN_ENABLE_VAL |
-			   FH_RCSR_CHNL0_RX_IGNORE_RXF_EMPTY |
-			   FH_RCSR_CHNL0_RX_CONFIG_IRQ_DEST_INT_HOST_VAL |
-			   rb_size|
-			   (rb_timeout << FH_RCSR_RX_CONFIG_REG_IRQ_RBTH_POS)|
-			   (rfdnlog << FH_RCSR_RX_CONFIG_RBDCB_SIZE_POS));
-
-	/* Set interrupt coalescing timer to default (2048 usecs) */
-	iwl_write8(trans, CSR_INT_COALESCING, IWL_HOST_INT_TIMEOUT_DEF);
-}
-
-static int iwl_rx_init(struct iwl_trans *trans)
-{
-	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
-	struct iwl_rx_queue *rxq = &trans_pcie->rxq;
-
-	int i, err;
-	unsigned long flags;
-
-	if (!rxq->bd) {
-		err = iwl_trans_rx_alloc(trans);
-		if (err)
-			return err;
-	}
-
-	spin_lock_irqsave(&rxq->lock, flags);
-	INIT_LIST_HEAD(&rxq->rx_free);
-	INIT_LIST_HEAD(&rxq->rx_used);
-
-	iwl_trans_rxq_free_rx_bufs(trans);
-
-	for (i = 0; i < RX_QUEUE_SIZE; i++)
-		rxq->queue[i] = NULL;
-
-	/* Set us so that we have processed and used all buffers, but have
-	 * not restocked the Rx queue with fresh buffers */
-	rxq->read = rxq->write = 0;
-	rxq->write_actual = 0;
-	rxq->free_count = 0;
-	spin_unlock_irqrestore(&rxq->lock, flags);
-
-	iwl_rx_replenish(trans);
-
-	iwl_trans_rx_hw_init(trans, rxq);
-
-	spin_lock_irqsave(&trans_pcie->irq_lock, flags);
-	rxq->need_update = 1;
-	iwl_rx_queue_update_write_ptr(trans, rxq);
-	spin_unlock_irqrestore(&trans_pcie->irq_lock, flags);
-
-	return 0;
-}
-
-static void iwl_trans_pcie_rx_free(struct iwl_trans *trans)
-{
-	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
-	struct iwl_rx_queue *rxq = &trans_pcie->rxq;
-	unsigned long flags;
-
-	/*if rxq->bd is NULL, it means that nothing has been allocated,
-	 * exit now */
-	if (!rxq->bd) {
-		IWL_DEBUG_INFO(trans, "Free NULL rx context\n");
-		return;
-	}
-
-	spin_lock_irqsave(&rxq->lock, flags);
-	iwl_trans_rxq_free_rx_bufs(trans);
-	spin_unlock_irqrestore(&rxq->lock, flags);
-
-	dma_free_coherent(trans->dev, sizeof(__le32) * RX_QUEUE_SIZE,
-			  rxq->bd, rxq->bd_dma);
-	memset(&rxq->bd_dma, 0, sizeof(rxq->bd_dma));
-	rxq->bd = NULL;
-
-	if (rxq->rb_stts)
-		dma_free_coherent(trans->dev,
-				  sizeof(struct iwl_rb_status),
-				  rxq->rb_stts, rxq->rb_stts_dma);
-	else
-		IWL_DEBUG_INFO(trans, "Free rxq->rb_stts which is NULL\n");
-	memset(&rxq->rb_stts_dma, 0, sizeof(rxq->rb_stts_dma));
-	rxq->rb_stts = NULL;
-}
-
-static int iwl_trans_rx_stop(struct iwl_trans *trans)
-{
-
-	/* stop Rx DMA */
-	iwl_write_direct32(trans, FH_MEM_RCSR_CHNL0_CONFIG_REG, 0);
-	return iwl_poll_direct_bit(trans, FH_MEM_RSSR_RX_STATUS_REG,
-				   FH_RSSR_CHNL0_RX_STATUS_CHNL_IDLE, 1000);
-}
-
-static int iwlagn_alloc_dma_ptr(struct iwl_trans *trans,
-				struct iwl_dma_ptr *ptr, size_t size)
-{
-	if (WARN_ON(ptr->addr))
-		return -EINVAL;
-
-	ptr->addr = dma_alloc_coherent(trans->dev, size,
-				       &ptr->dma, GFP_KERNEL);
-	if (!ptr->addr)
-		return -ENOMEM;
-	ptr->size = size;
-	return 0;
-}
-
-static void iwlagn_free_dma_ptr(struct iwl_trans *trans,
-				struct iwl_dma_ptr *ptr)
-{
-	if (unlikely(!ptr->addr))
-		return;
-
-	dma_free_coherent(trans->dev, ptr->size, ptr->addr, ptr->dma);
-	memset(ptr, 0, sizeof(*ptr));
-}
-
-static void iwl_trans_pcie_queue_stuck_timer(unsigned long data)
-{
-	struct iwl_tx_queue *txq = (void *)data;
-	struct iwl_queue *q = &txq->q;
-	struct iwl_trans_pcie *trans_pcie = txq->trans_pcie;
-	struct iwl_trans *trans = iwl_trans_pcie_get_trans(trans_pcie);
-	u32 scd_sram_addr = trans_pcie->scd_base_addr +
-				SCD_TX_STTS_QUEUE_OFFSET(txq->q.id);
-	u8 buf[16];
-	int i;
-
-	spin_lock(&txq->lock);
-	/* check if triggered erroneously */
-	if (txq->q.read_ptr == txq->q.write_ptr) {
-		spin_unlock(&txq->lock);
-		return;
-	}
-	spin_unlock(&txq->lock);
-
-	IWL_ERR(trans, "Queue %d stuck for %u ms.\n", txq->q.id,
-		jiffies_to_msecs(trans_pcie->wd_timeout));
-	IWL_ERR(trans, "Current SW read_ptr %d write_ptr %d\n",
-		txq->q.read_ptr, txq->q.write_ptr);
-
-	iwl_read_targ_mem_bytes(trans, scd_sram_addr, buf, sizeof(buf));
-
-	iwl_print_hex_error(trans, buf, sizeof(buf));
-
-	for (i = 0; i < FH_TCSR_CHNL_NUM; i++)
-		IWL_ERR(trans, "FH TRBs(%d) = 0x%08x\n", i,
-			iwl_read_direct32(trans, FH_TX_TRB_REG(i)));
-
-	for (i = 0; i < trans->cfg->base_params->num_of_queues; i++) {
-		u32 status = iwl_read_prph(trans, SCD_QUEUE_STATUS_BITS(i));
-		u8 fifo = (status >> SCD_QUEUE_STTS_REG_POS_TXF) & 0x7;
-		bool active = !!(status & BIT(SCD_QUEUE_STTS_REG_POS_ACTIVE));
-		u32 tbl_dw =
-			iwl_read_targ_mem(trans,
-					  trans_pcie->scd_base_addr +
-					  SCD_TRANS_TBL_OFFSET_QUEUE(i));
-
-		if (i & 0x1)
-			tbl_dw = (tbl_dw & 0xFFFF0000) >> 16;
-		else
-			tbl_dw = tbl_dw & 0x0000FFFF;
-
-		IWL_ERR(trans,
-			"Q %d is %sactive and mapped to fifo %d ra_tid 0x%04x [%d,%d]\n",
-			i, active ? "" : "in", fifo, tbl_dw,
-			iwl_read_prph(trans,
-				      SCD_QUEUE_RDPTR(i)) & (txq->q.n_bd - 1),
-			iwl_read_prph(trans, SCD_QUEUE_WRPTR(i)));
-	}
-
-	for (i = q->read_ptr; i != q->write_ptr;
-	     i = iwl_queue_inc_wrap(i, q->n_bd)) {
-		struct iwl_tx_cmd *tx_cmd =
-			(struct iwl_tx_cmd *)txq->entries[i].cmd->payload;
-		IWL_ERR(trans, "scratch %d = 0x%08x\n", i,
-			get_unaligned_le32(&tx_cmd->scratch));
-	}
-
-	iwl_op_mode_nic_error(trans->op_mode);
-}
-
-static int iwl_trans_txq_alloc(struct iwl_trans *trans,
-			       struct iwl_tx_queue *txq, int slots_num,
-			       u32 txq_id)
-{
-	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
-	size_t tfd_sz = sizeof(struct iwl_tfd) * TFD_QUEUE_SIZE_MAX;
-	int i;
-
-	if (WARN_ON(txq->entries || txq->tfds))
-		return -EINVAL;
-
-	setup_timer(&txq->stuck_timer, iwl_trans_pcie_queue_stuck_timer,
-		    (unsigned long)txq);
-	txq->trans_pcie = trans_pcie;
-
-	txq->q.n_window = slots_num;
-
-	txq->entries = kcalloc(slots_num,
-			       sizeof(struct iwl_pcie_tx_queue_entry),
-			       GFP_KERNEL);
-
-	if (!txq->entries)
-		goto error;
-
-	if (txq_id == trans_pcie->cmd_queue)
-		for (i = 0; i < slots_num; i++) {
-			txq->entries[i].cmd =
-				kmalloc(sizeof(struct iwl_device_cmd),
-					GFP_KERNEL);
-			if (!txq->entries[i].cmd)
-				goto error;
-		}
-
-	/* Circular buffer of transmit frame descriptors (TFDs),
-	 * shared with device */
-	txq->tfds = dma_alloc_coherent(trans->dev, tfd_sz,
-				       &txq->q.dma_addr, GFP_KERNEL);
-	if (!txq->tfds) {
-		IWL_ERR(trans, "dma_alloc_coherent(%zd) failed\n", tfd_sz);
-		goto error;
-	}
-	txq->q.id = txq_id;
-
-	return 0;
-error:
-	if (txq->entries && txq_id == trans_pcie->cmd_queue)
-		for (i = 0; i < slots_num; i++)
-			kfree(txq->entries[i].cmd);
-	kfree(txq->entries);
-	txq->entries = NULL;
-
-	return -ENOMEM;
-
-}
-
-static int iwl_trans_txq_init(struct iwl_trans *trans, struct iwl_tx_queue *txq,
-			      int slots_num, u32 txq_id)
-{
-	int ret;
-
-	txq->need_update = 0;
-
-	/* 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));
-
-	/* Initialize queue's high/low-water marks, and head/tail indexes */
-	ret = iwl_queue_init(&txq->q, TFD_QUEUE_SIZE_MAX, slots_num,
-			txq_id);
-	if (ret)
-		return ret;
-
-	spin_lock_init(&txq->lock);
-
-	/*
-	 * Tell nic where to find circular buffer of Tx Frame Descriptors for
-	 * given Tx queue, and enable the DMA channel used for that queue.
-	 * Circular buffer (TFD queue in DRAM) physical base address */
-	iwl_write_direct32(trans, FH_MEM_CBBC_QUEUE(txq_id),
-			     txq->q.dma_addr >> 8);
-
-	return 0;
-}
-
-/*
- * iwl_tx_queue_unmap -  Unmap any remaining DMA mappings and free skb's
- */
-void iwl_tx_queue_unmap(struct iwl_trans *trans, int txq_id)
-{
-	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
-	struct iwl_tx_queue *txq = &trans_pcie->txq[txq_id];
-	struct iwl_queue *q = &txq->q;
-	enum dma_data_direction dma_dir;
-
-	if (!q->n_bd)
-		return;
-
-	/* In the command queue, all the TBs are mapped as BIDI
-	 * so unmap them as such.
-	 */
-	if (txq_id == trans_pcie->cmd_queue)
-		dma_dir = DMA_BIDIRECTIONAL;
-	else
-		dma_dir = DMA_TO_DEVICE;
-
-	spin_lock_bh(&txq->lock);
-	while (q->write_ptr != q->read_ptr) {
-		iwl_txq_free_tfd(trans, txq, dma_dir);
-		q->read_ptr = iwl_queue_inc_wrap(q->read_ptr, q->n_bd);
-	}
-	spin_unlock_bh(&txq->lock);
-}
-
-/**
- * iwl_tx_queue_free - Deallocate DMA queue.
- * @txq: Transmit queue to deallocate.
- *
- * Empty queue by removing and destroying all BD's.
- * Free all buffers.
- * 0-fill, but do not free "txq" descriptor structure.
- */
-static void iwl_tx_queue_free(struct iwl_trans *trans, int txq_id)
-{
-	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
-	struct iwl_tx_queue *txq = &trans_pcie->txq[txq_id];
-	struct device *dev = trans->dev;
-	int i;
-
-	if (WARN_ON(!txq))
-		return;
-
-	iwl_tx_queue_unmap(trans, txq_id);
-
-	/* De-alloc array of command/tx buffers */
-	if (txq_id == trans_pcie->cmd_queue)
-		for (i = 0; i < txq->q.n_window; i++) {
-			kfree(txq->entries[i].cmd);
-			kfree(txq->entries[i].copy_cmd);
-			kfree(txq->entries[i].free_buf);
-		}
-
-	/* De-alloc circular buffer of TFDs */
-	if (txq->q.n_bd) {
-		dma_free_coherent(dev, sizeof(struct iwl_tfd) *
-				  txq->q.n_bd, txq->tfds, txq->q.dma_addr);
-		memset(&txq->q.dma_addr, 0, sizeof(txq->q.dma_addr));
-	}
-
-	kfree(txq->entries);
-	txq->entries = NULL;
-
-	del_timer_sync(&txq->stuck_timer);
-
-	/* 0-fill queue descriptor structure */
-	memset(txq, 0, sizeof(*txq));
-}
-
-/**
- * iwl_trans_tx_free - Free TXQ Context
- *
- * Destroy all TX DMA queues and structures
- */
-static void iwl_trans_pcie_tx_free(struct iwl_trans *trans)
-{
-	int txq_id;
-	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
-
-	/* Tx queues */
-	if (trans_pcie->txq) {
-		for (txq_id = 0;
-		     txq_id < trans->cfg->base_params->num_of_queues; txq_id++)
-			iwl_tx_queue_free(trans, txq_id);
-	}
-
-	kfree(trans_pcie->txq);
-	trans_pcie->txq = NULL;
-
-	iwlagn_free_dma_ptr(trans, &trans_pcie->kw);
-
-	iwlagn_free_dma_ptr(trans, &trans_pcie->scd_bc_tbls);
-}
-
-/**
- * iwl_trans_tx_alloc - allocate TX context
- * Allocate all Tx DMA structures and initialize them
- *
- * @param priv
- * @return error code
- */
-static int iwl_trans_tx_alloc(struct iwl_trans *trans)
-{
-	int ret;
-	int txq_id, slots_num;
-	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
-
-	u16 scd_bc_tbls_size = trans->cfg->base_params->num_of_queues *
-			sizeof(struct iwlagn_scd_bc_tbl);
-
-	/*It is not allowed to alloc twice, so warn when this happens.
-	 * We cannot rely on the previous allocation, so free and fail */
-	if (WARN_ON(trans_pcie->txq)) {
-		ret = -EINVAL;
-		goto error;
-	}
-
-	ret = iwlagn_alloc_dma_ptr(trans, &trans_pcie->scd_bc_tbls,
-				   scd_bc_tbls_size);
-	if (ret) {
-		IWL_ERR(trans, "Scheduler BC Table allocation failed\n");
-		goto error;
-	}
-
-	/* Alloc keep-warm buffer */
-	ret = iwlagn_alloc_dma_ptr(trans, &trans_pcie->kw, IWL_KW_SIZE);
-	if (ret) {
-		IWL_ERR(trans, "Keep Warm allocation failed\n");
-		goto error;
-	}
-
-	trans_pcie->txq = kcalloc(trans->cfg->base_params->num_of_queues,
-				  sizeof(struct iwl_tx_queue), GFP_KERNEL);
-	if (!trans_pcie->txq) {
-		IWL_ERR(trans, "Not enough memory for txq\n");
-		ret = ENOMEM;
-		goto error;
-	}
-
-	/* Alloc and init all Tx queues, including the command queue (#4/#9) */
-	for (txq_id = 0; txq_id < trans->cfg->base_params->num_of_queues;
-	     txq_id++) {
-		slots_num = (txq_id == trans_pcie->cmd_queue) ?
-					TFD_CMD_SLOTS : TFD_TX_CMD_SLOTS;
-		ret = iwl_trans_txq_alloc(trans, &trans_pcie->txq[txq_id],
-					  slots_num, txq_id);
-		if (ret) {
-			IWL_ERR(trans, "Tx %d queue alloc failed\n", txq_id);
-			goto error;
-		}
-	}
-
-	return 0;
-
-error:
-	iwl_trans_pcie_tx_free(trans);
-
-	return ret;
-}
-static int iwl_tx_init(struct iwl_trans *trans)
-{
-	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
-	int ret;
-	int txq_id, slots_num;
-	unsigned long flags;
-	bool alloc = false;
-
-	if (!trans_pcie->txq) {
-		ret = iwl_trans_tx_alloc(trans);
-		if (ret)
-			goto error;
-		alloc = true;
-	}
-
-	spin_lock_irqsave(&trans_pcie->irq_lock, flags);
-
-	/* Turn off all Tx DMA fifos */
-	iwl_write_prph(trans, SCD_TXFACT, 0);
-
-	/* Tell NIC where to find the "keep warm" buffer */
-	iwl_write_direct32(trans, FH_KW_MEM_ADDR_REG,
-			   trans_pcie->kw.dma >> 4);
-
-	spin_unlock_irqrestore(&trans_pcie->irq_lock, flags);
-
-	/* Alloc and init all Tx queues, including the command queue (#4/#9) */
-	for (txq_id = 0; txq_id < trans->cfg->base_params->num_of_queues;
-	     txq_id++) {
-		slots_num = (txq_id == trans_pcie->cmd_queue) ?
-					TFD_CMD_SLOTS : TFD_TX_CMD_SLOTS;
-		ret = iwl_trans_txq_init(trans, &trans_pcie->txq[txq_id],
-					 slots_num, txq_id);
-		if (ret) {
-			IWL_ERR(trans, "Tx %d queue init failed\n", txq_id);
-			goto error;
-		}
-	}
-
-	return 0;
-error:
-	/*Upon error, free only if we allocated something */
-	if (alloc)
-		iwl_trans_pcie_tx_free(trans);
-	return ret;
-}
-
-static void iwl_set_pwr_vmain(struct iwl_trans *trans)
+static void iwl_pcie_set_pwr_vmain(struct iwl_trans *trans)
 {
 /*
  * (for documentation purposes)
@@ -674,18 +97,11 @@ static void iwl_set_pwr_vmain(struct iwl_trans *trans)
 #define PCI_CFG_LINK_CTRL_VAL_L0S_EN	0x01
 #define PCI_CFG_LINK_CTRL_VAL_L1_EN	0x02
 
-static u16 iwl_pciexp_link_ctrl(struct iwl_trans *trans)
+static void iwl_pcie_apm_config(struct iwl_trans *trans)
 {
 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
-	u16 pci_lnk_ctl;
+	u16 lctl;
 
-	pcie_capability_read_word(trans_pcie->pci_dev, PCI_EXP_LNKCTL,
-				  &pci_lnk_ctl);
-	return pci_lnk_ctl;
-}
-
-static void iwl_apm_config(struct iwl_trans *trans)
-{
 	/*
 	 * HW bug W/A for instability in PCIe bus L0S->L1 transition.
 	 * Check if BIOS (or OS) enabled L1-ASPM on this device.
@@ -694,7 +110,7 @@ static void iwl_apm_config(struct iwl_trans *trans)
 	 * If not (unlikely), enable L0S, so there is at least some
 	 *    power savings, even without L1.
 	 */
-	u16 lctl = iwl_pciexp_link_ctrl(trans);
+	pcie_capability_read_word(trans_pcie->pci_dev, PCI_EXP_LNKCTL, &lctl);
 
 	if ((lctl & PCI_CFG_LINK_CTRL_VAL_L1_EN) ==
 				PCI_CFG_LINK_CTRL_VAL_L1_EN) {
@@ -711,10 +127,10 @@ static void iwl_apm_config(struct iwl_trans *trans)
 
 /*
  * Start up NIC's basic functionality after it has been reset
- * (e.g. after platform boot, or shutdown via iwl_apm_stop())
+ * (e.g. after platform boot, or shutdown via iwl_pcie_apm_stop())
  * NOTE:  This does not load uCode nor start the embedded processor
  */
-static int iwl_apm_init(struct iwl_trans *trans)
+static int iwl_pcie_apm_init(struct iwl_trans *trans)
 {
 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
 	int ret = 0;
@@ -746,7 +162,7 @@ static int iwl_apm_init(struct iwl_trans *trans)
 	iwl_set_bit(trans, CSR_HW_IF_CONFIG_REG,
 		    CSR_HW_IF_CONFIG_REG_BIT_HAP_WAKE_L1A);
 
-	iwl_apm_config(trans);
+	iwl_pcie_apm_config(trans);
 
 	/* Configure analog phase-lock-loop before activating to D0A */
 	if (trans->cfg->base_params->pll_cfg_val)
@@ -792,7 +208,7 @@ out:
 	return ret;
 }
 
-static int iwl_apm_stop_master(struct iwl_trans *trans)
+static int iwl_pcie_apm_stop_master(struct iwl_trans *trans)
 {
 	int ret = 0;
 
@@ -810,7 +226,7 @@ static int iwl_apm_stop_master(struct iwl_trans *trans)
 	return ret;
 }
 
-static void iwl_apm_stop(struct iwl_trans *trans)
+static void iwl_pcie_apm_stop(struct iwl_trans *trans)
 {
 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
 	IWL_DEBUG_INFO(trans, "Stop card, put in low power state\n");
@@ -818,7 +234,7 @@ static void iwl_apm_stop(struct iwl_trans *trans)
 	clear_bit(STATUS_DEVICE_ENABLED, &trans_pcie->status);
 
 	/* Stop device's DMA activity */
-	iwl_apm_stop_master(trans);
+	iwl_pcie_apm_stop_master(trans);
 
 	/* Reset the entire device */
 	iwl_set_bit(trans, CSR_RESET, CSR_RESET_REG_FLAG_SW_RESET);
@@ -833,29 +249,29 @@ static void iwl_apm_stop(struct iwl_trans *trans)
 		      CSR_GP_CNTRL_REG_FLAG_INIT_DONE);
 }
 
-static int iwl_nic_init(struct iwl_trans *trans)
+static int iwl_pcie_nic_init(struct iwl_trans *trans)
 {
 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
 	unsigned long flags;
 
 	/* nic_init */
 	spin_lock_irqsave(&trans_pcie->irq_lock, flags);
-	iwl_apm_init(trans);
+	iwl_pcie_apm_init(trans);
 
 	/* Set interrupt coalescing calibration timer to default (512 usecs) */
 	iwl_write8(trans, CSR_INT_COALESCING, IWL_HOST_INT_CALIB_TIMEOUT_DEF);
 
 	spin_unlock_irqrestore(&trans_pcie->irq_lock, flags);
 
-	iwl_set_pwr_vmain(trans);
+	iwl_pcie_set_pwr_vmain(trans);
 
 	iwl_op_mode_nic_config(trans->op_mode);
 
 	/* Allocate the RX queue, or reset if it is already allocated */
-	iwl_rx_init(trans);
+	iwl_pcie_rx_init(trans);
 
 	/* Allocate or reset and init all Tx and Command queues */
-	if (iwl_tx_init(trans))
+	if (iwl_pcie_tx_init(trans))
 		return -ENOMEM;
 
 	if (trans->cfg->base_params->shadow_reg_enable) {
@@ -870,7 +286,7 @@ static int iwl_nic_init(struct iwl_trans *trans)
 #define HW_READY_TIMEOUT (50)
 
 /* Note: returns poll_bit return value, which is >= 0 if success */
-static int iwl_set_hw_ready(struct iwl_trans *trans)
+static int iwl_pcie_set_hw_ready(struct iwl_trans *trans)
 {
 	int ret;
 
@@ -888,14 +304,14 @@ static int iwl_set_hw_ready(struct iwl_trans *trans)
 }
 
 /* Note: returns standard 0/-ERROR code */
-static int iwl_prepare_card_hw(struct iwl_trans *trans)
+static int iwl_pcie_prepare_card_hw(struct iwl_trans *trans)
 {
 	int ret;
 	int t = 0;
 
 	IWL_DEBUG_INFO(trans, "iwl_trans_prepare_card_hw enter\n");
 
-	ret = iwl_set_hw_ready(trans);
+	ret = iwl_pcie_set_hw_ready(trans);
 	/* If the card is ready, exit 0 */
 	if (ret >= 0)
 		return 0;
@@ -905,7 +321,7 @@ static int iwl_prepare_card_hw(struct iwl_trans *trans)
 		    CSR_HW_IF_CONFIG_REG_PREPARE);
 
 	do {
-		ret = iwl_set_hw_ready(trans);
+		ret = iwl_pcie_set_hw_ready(trans);
 		if (ret >= 0)
 			return 0;
 
@@ -919,7 +335,7 @@ static int iwl_prepare_card_hw(struct iwl_trans *trans)
 /*
  * ucode
  */
-static int iwl_load_firmware_chunk(struct iwl_trans *trans, u32 dst_addr,
+static int iwl_pcie_load_firmware_chunk(struct iwl_trans *trans, u32 dst_addr,
 				   dma_addr_t phy_addr, u32 byte_cnt)
 {
 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
@@ -966,7 +382,7 @@ static int iwl_load_firmware_chunk(struct iwl_trans *trans, u32 dst_addr,
 	return 0;
 }
 
-static int iwl_load_section(struct iwl_trans *trans, u8 section_num,
+static int iwl_pcie_load_section(struct iwl_trans *trans, u8 section_num,
 			    const struct fw_desc *section)
 {
 	u8 *v_addr;
@@ -987,8 +403,9 @@ static int iwl_load_section(struct iwl_trans *trans, u8 section_num,
 		copy_size = min_t(u32, PAGE_SIZE, section->len - offset);
 
 		memcpy(v_addr, (u8 *)section->data + offset, copy_size);
-		ret = iwl_load_firmware_chunk(trans, section->offset + offset,
-					      p_addr, copy_size);
+		ret = iwl_pcie_load_firmware_chunk(trans,
+						   section->offset + offset,
+						   p_addr, copy_size);
 		if (ret) {
 			IWL_ERR(trans,
 				"Could not load the [%d] uCode section\n",
@@ -1001,7 +418,7 @@ static int iwl_load_section(struct iwl_trans *trans, u8 section_num,
 	return ret;
 }
 
-static int iwl_load_given_ucode(struct iwl_trans *trans,
+static int iwl_pcie_load_given_ucode(struct iwl_trans *trans,
 				const struct fw_img *image)
 {
 	int i, ret = 0;
@@ -1010,7 +427,7 @@ static int iwl_load_given_ucode(struct iwl_trans *trans,
 		if (!image->sec[i].data)
 			break;
 
-		ret = iwl_load_section(trans, i, &image->sec[i]);
+		ret = iwl_pcie_load_section(trans, i, &image->sec[i]);
 		if (ret)
 			return ret;
 	}
@@ -1029,7 +446,7 @@ static int iwl_trans_pcie_start_fw(struct iwl_trans *trans,
 	bool hw_rfkill;
 
 	/* This may fail if AMT took ownership of the device */
-	if (iwl_prepare_card_hw(trans)) {
+	if (iwl_pcie_prepare_card_hw(trans)) {
 		IWL_WARN(trans, "Exit HW not ready\n");
 		return -EIO;
 	}
@@ -1046,7 +463,7 @@ static int iwl_trans_pcie_start_fw(struct iwl_trans *trans,
 
 	iwl_write32(trans, CSR_INT, 0xFFFFFFFF);
 
-	ret = iwl_nic_init(trans);
+	ret = iwl_pcie_nic_init(trans);
 	if (ret) {
 		IWL_ERR(trans, "Unable to init nic\n");
 		return ret;
@@ -1066,129 +483,13 @@ static int iwl_trans_pcie_start_fw(struct iwl_trans *trans,
 	iwl_write32(trans, CSR_UCODE_DRV_GP1_CLR, CSR_UCODE_SW_BIT_RFKILL);
 
 	/* Load the given image to the HW */
-	return iwl_load_given_ucode(trans, fw);
-}
-
-/*
- * Activate/Deactivate Tx DMA/FIFO channels according tx fifos mask
- */
-static void iwl_trans_txq_set_sched(struct iwl_trans *trans, u32 mask)
-{
-	struct iwl_trans_pcie __maybe_unused *trans_pcie =
-		IWL_TRANS_GET_PCIE_TRANS(trans);
-
-	iwl_write_prph(trans, SCD_TXFACT, mask);
-}
-
-static void iwl_tx_start(struct iwl_trans *trans, u32 scd_base_addr)
-{
-	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
-	u32 a;
-	int chan;
-	u32 reg_val;
-
-	/* make sure all queue are not stopped/used */
-	memset(trans_pcie->queue_stopped, 0, sizeof(trans_pcie->queue_stopped));
-	memset(trans_pcie->queue_used, 0, sizeof(trans_pcie->queue_used));
-
-	trans_pcie->scd_base_addr =
-		iwl_read_prph(trans, SCD_SRAM_BASE_ADDR);
-
-	WARN_ON(scd_base_addr != 0 &&
-		scd_base_addr != trans_pcie->scd_base_addr);
-
-	a = trans_pcie->scd_base_addr + SCD_CONTEXT_MEM_LOWER_BOUND;
-	/* reset conext data memory */
-	for (; a < trans_pcie->scd_base_addr + SCD_CONTEXT_MEM_UPPER_BOUND;
-		a += 4)
-		iwl_write_targ_mem(trans, a, 0);
-	/* reset tx status memory */
-	for (; a < trans_pcie->scd_base_addr + SCD_TX_STTS_MEM_UPPER_BOUND;
-		a += 4)
-		iwl_write_targ_mem(trans, a, 0);
-	for (; a < trans_pcie->scd_base_addr +
-	       SCD_TRANS_TBL_OFFSET_QUEUE(
-				trans->cfg->base_params->num_of_queues);
-	       a += 4)
-		iwl_write_targ_mem(trans, a, 0);
-
-	iwl_write_prph(trans, SCD_DRAM_BASE_ADDR,
-		       trans_pcie->scd_bc_tbls.dma >> 10);
-
-	/* The chain extension of the SCD doesn't work well. This feature is
-	 * enabled by default by the HW, so we need to disable it manually.
-	 */
-	iwl_write_prph(trans, SCD_CHAINEXT_EN, 0);
-
-	iwl_trans_ac_txq_enable(trans, trans_pcie->cmd_queue,
-				trans_pcie->cmd_fifo);
-
-	/* Activate all Tx DMA/FIFO channels */
-	iwl_trans_txq_set_sched(trans, IWL_MASK(0, 7));
-
-	/* Enable DMA channel */
-	for (chan = 0; chan < FH_TCSR_CHNL_NUM ; chan++)
-		iwl_write_direct32(trans, FH_TCSR_CHNL_TX_CONFIG_REG(chan),
-				   FH_TCSR_TX_CONFIG_REG_VAL_DMA_CHNL_ENABLE |
-				   FH_TCSR_TX_CONFIG_REG_VAL_DMA_CREDIT_ENABLE);
-
-	/* Update FH chicken bits */
-	reg_val = iwl_read_direct32(trans, FH_TX_CHICKEN_BITS_REG);
-	iwl_write_direct32(trans, FH_TX_CHICKEN_BITS_REG,
-			   reg_val | FH_TX_CHICKEN_BITS_SCD_AUTO_RETRY_EN);
-
-	/* Enable L1-Active */
-	iwl_clear_bits_prph(trans, APMG_PCIDEV_STT_REG,
-			    APMG_PCIDEV_STT_VAL_L1_ACT_DIS);
+	return iwl_pcie_load_given_ucode(trans, fw);
 }
 
 static void iwl_trans_pcie_fw_alive(struct iwl_trans *trans, u32 scd_addr)
 {
-	iwl_reset_ict(trans);
-	iwl_tx_start(trans, scd_addr);
-}
-
-/**
- * iwlagn_txq_ctx_stop - Stop all Tx DMA channels
- */
-static int iwl_trans_tx_stop(struct iwl_trans *trans)
-{
-	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
-	int ch, txq_id, ret;
-	unsigned long flags;
-
-	/* Turn off all Tx DMA fifos */
-	spin_lock_irqsave(&trans_pcie->irq_lock, flags);
-
-	iwl_trans_txq_set_sched(trans, 0);
-
-	/* Stop each Tx DMA channel, and wait for it to be idle */
-	for (ch = 0; ch < FH_TCSR_CHNL_NUM; ch++) {
-		iwl_write_direct32(trans,
-				   FH_TCSR_CHNL_TX_CONFIG_REG(ch), 0x0);
-		ret = iwl_poll_direct_bit(trans, FH_TSSR_TX_STATUS_REG,
-			FH_TSSR_TX_STATUS_REG_MSK_CHNL_IDLE(ch), 1000);
-		if (ret < 0)
-			IWL_ERR(trans,
-				"Failing on timeout while stopping DMA channel %d [0x%08x]\n",
-				ch,
-				iwl_read_direct32(trans,
-						  FH_TSSR_TX_STATUS_REG));
-	}
-	spin_unlock_irqrestore(&trans_pcie->irq_lock, flags);
-
-	if (!trans_pcie->txq) {
-		IWL_WARN(trans,
-			 "Stopping tx queues that aren't allocated...\n");
-		return 0;
-	}
-
-	/* Unmap DMA from host system and free skb's */
-	for (txq_id = 0; txq_id < trans->cfg->base_params->num_of_queues;
-	     txq_id++)
-		iwl_tx_queue_unmap(trans, txq_id);
-
-	return 0;
+	iwl_pcie_reset_ict(trans);
+	iwl_pcie_tx_start(trans, scd_addr);
 }
 
 static void iwl_trans_pcie_stop_device(struct iwl_trans *trans)
@@ -1202,7 +503,7 @@ static void iwl_trans_pcie_stop_device(struct iwl_trans *trans)
 	spin_unlock_irqrestore(&trans_pcie->irq_lock, flags);
 
 	/* device going down, Stop using ICT table */
-	iwl_disable_ict(trans);
+	iwl_pcie_disable_ict(trans);
 
 	/*
 	 * If a HW restart happens during firmware loading,
@@ -1212,8 +513,8 @@ static void iwl_trans_pcie_stop_device(struct iwl_trans *trans)
 	 * already dead.
 	 */
 	if (test_bit(STATUS_DEVICE_ENABLED, &trans_pcie->status)) {
-		iwl_trans_tx_stop(trans);
-		iwl_trans_rx_stop(trans);
+		iwl_pcie_tx_stop(trans);
+		iwl_pcie_rx_stop(trans);
 
 		/* Power-down device's busmaster DMA clocks */
 		iwl_write_prph(trans, APMG_CLK_DIS_REG,
@@ -1226,7 +527,7 @@ static void iwl_trans_pcie_stop_device(struct iwl_trans *trans)
 		      CSR_GP_CNTRL_REG_FLAG_MAC_ACCESS_REQ);
 
 	/* Stop the device, and put it in low power state */
-	iwl_apm_stop(trans);
+	iwl_pcie_apm_stop(trans);
 
 	/* Upon stop, the APM issues an interrupt if HW RF kill is set.
 	 * Clean again the interrupt here
@@ -1265,171 +566,6 @@ static void iwl_trans_pcie_wowlan_suspend(struct iwl_trans *trans)
 		      CSR_GP_CNTRL_REG_FLAG_MAC_ACCESS_REQ);
 }
 
-static int iwl_trans_pcie_tx(struct iwl_trans *trans, struct sk_buff *skb,
-			     struct iwl_device_cmd *dev_cmd, int txq_id)
-{
-	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
-	struct ieee80211_hdr *hdr = (struct ieee80211_hdr *)skb->data;
-	struct iwl_tx_cmd *tx_cmd = (struct iwl_tx_cmd *) dev_cmd->payload;
-	struct iwl_cmd_meta *out_meta;
-	struct iwl_tx_queue *txq;
-	struct iwl_queue *q;
-	dma_addr_t phys_addr = 0;
-	dma_addr_t txcmd_phys;
-	dma_addr_t scratch_phys;
-	u16 len, firstlen, secondlen;
-	u8 wait_write_ptr = 0;
-	__le16 fc = hdr->frame_control;
-	u8 hdr_len = ieee80211_hdrlen(fc);
-	u16 __maybe_unused wifi_seq;
-
-	txq = &trans_pcie->txq[txq_id];
-	q = &txq->q;
-
-	if (unlikely(!test_bit(txq_id, trans_pcie->queue_used))) {
-		WARN_ON_ONCE(1);
-		return -EINVAL;
-	}
-
-	spin_lock(&txq->lock);
-
-	/* In AGG mode, the index in the ring must correspond to the WiFi
-	 * sequence number. This is a HW requirements to help the SCD to parse
-	 * the BA.
-	 * Check here that the packets are in the right place on the ring.
-	 */
-#ifdef CONFIG_IWLWIFI_DEBUG
-	wifi_seq = SEQ_TO_SN(le16_to_cpu(hdr->seq_ctrl));
-	WARN_ONCE((iwl_read_prph(trans, SCD_AGGR_SEL) & BIT(txq_id)) &&
-		  ((wifi_seq & 0xff) != q->write_ptr),
-		  "Q: %d WiFi Seq %d tfdNum %d",
-		  txq_id, wifi_seq, q->write_ptr);
-#endif
-
-	/* Set up driver data for this TFD */
-	txq->entries[q->write_ptr].skb = skb;
-	txq->entries[q->write_ptr].cmd = dev_cmd;
-
-	dev_cmd->hdr.cmd = REPLY_TX;
-	dev_cmd->hdr.sequence =
-		cpu_to_le16((u16)(QUEUE_TO_SEQ(txq_id) |
-			    INDEX_TO_SEQ(q->write_ptr)));
-
-	/* Set up first empty entry in queue's array of Tx/cmd buffers */
-	out_meta = &txq->entries[q->write_ptr].meta;
-
-	/*
-	 * Use the first empty entry in this queue's command buffer array
-	 * to contain the Tx command and MAC header concatenated together
-	 * (payload data will be in another buffer).
-	 * Size of this varies, due to varying MAC header length.
-	 * If end is not dword aligned, we'll have 2 extra bytes at the end
-	 * of the MAC header (device reads on dword boundaries).
-	 * We'll tell device about this padding later.
-	 */
-	len = sizeof(struct iwl_tx_cmd) +
-		sizeof(struct iwl_cmd_header) + hdr_len;
-	firstlen = (len + 3) & ~3;
-
-	/* Tell NIC about any 2-byte padding after MAC header */
-	if (firstlen != len)
-		tx_cmd->tx_flags |= TX_CMD_FLG_MH_PAD_MSK;
-
-	/* Physical address of this Tx command's header (not MAC header!),
-	 * within command buffer array. */
-	txcmd_phys = dma_map_single(trans->dev,
-				    &dev_cmd->hdr, firstlen,
-				    DMA_BIDIRECTIONAL);
-	if (unlikely(dma_mapping_error(trans->dev, txcmd_phys)))
-		goto out_err;
-	dma_unmap_addr_set(out_meta, mapping, txcmd_phys);
-	dma_unmap_len_set(out_meta, len, firstlen);
-
-	if (!ieee80211_has_morefrags(fc)) {
-		txq->need_update = 1;
-	} else {
-		wait_write_ptr = 1;
-		txq->need_update = 0;
-	}
-
-	/* Set up TFD's 2nd entry to point directly to remainder of skb,
-	 * if any (802.11 null frames have no payload). */
-	secondlen = skb->len - hdr_len;
-	if (secondlen > 0) {
-		phys_addr = dma_map_single(trans->dev, skb->data + hdr_len,
-					   secondlen, DMA_TO_DEVICE);
-		if (unlikely(dma_mapping_error(trans->dev, phys_addr))) {
-			dma_unmap_single(trans->dev,
-					 dma_unmap_addr(out_meta, mapping),
-					 dma_unmap_len(out_meta, len),
-					 DMA_BIDIRECTIONAL);
-			goto out_err;
-		}
-	}
-
-	/* Attach buffers to TFD */
-	iwlagn_txq_attach_buf_to_tfd(trans, txq, txcmd_phys, firstlen, 1);
-	if (secondlen > 0)
-		iwlagn_txq_attach_buf_to_tfd(trans, txq, phys_addr,
-					     secondlen, 0);
-
-	scratch_phys = txcmd_phys + sizeof(struct iwl_cmd_header) +
-				offsetof(struct iwl_tx_cmd, scratch);
-
-	/* take back ownership of DMA buffer to enable update */
-	dma_sync_single_for_cpu(trans->dev, txcmd_phys, firstlen,
-				DMA_BIDIRECTIONAL);
-	tx_cmd->dram_lsb_ptr = cpu_to_le32(scratch_phys);
-	tx_cmd->dram_msb_ptr = iwl_get_dma_hi_addr(scratch_phys);
-
-	IWL_DEBUG_TX(trans, "sequence nr = 0X%x\n",
-		     le16_to_cpu(dev_cmd->hdr.sequence));
-	IWL_DEBUG_TX(trans, "tx_flags = 0X%x\n", le32_to_cpu(tx_cmd->tx_flags));
-
-	/* Set up entry for this TFD in Tx byte-count array */
-	iwl_trans_txq_update_byte_cnt_tbl(trans, txq, le16_to_cpu(tx_cmd->len));
-
-	dma_sync_single_for_device(trans->dev, txcmd_phys, firstlen,
-				   DMA_BIDIRECTIONAL);
-
-	trace_iwlwifi_dev_tx(trans->dev, skb,
-			     &txq->tfds[txq->q.write_ptr],
-			     sizeof(struct iwl_tfd),
-			     &dev_cmd->hdr, firstlen,
-			     skb->data + hdr_len, secondlen);
-	trace_iwlwifi_dev_tx_data(trans->dev, skb,
-				  skb->data + hdr_len, secondlen);
-
-	/* start timer if queue currently empty */
-	if (txq->need_update && q->read_ptr == q->write_ptr &&
-	    trans_pcie->wd_timeout)
-		mod_timer(&txq->stuck_timer, jiffies + trans_pcie->wd_timeout);
-
-	/* Tell device the write index *just past* this latest filled TFD */
-	q->write_ptr = iwl_queue_inc_wrap(q->write_ptr, q->n_bd);
-	iwl_txq_update_write_ptr(trans, txq);
-
-	/*
-	 * 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.
-	 */
-	if (iwl_queue_space(q) < q->high_mark) {
-		if (wait_write_ptr) {
-			txq->need_update = 1;
-			iwl_txq_update_write_ptr(trans, txq);
-		} else {
-			iwl_stop_queue(trans, txq);
-		}
-	}
-	spin_unlock(&txq->lock);
-	return 0;
- out_err:
-	spin_unlock(&txq->lock);
-	return -1;
-}
-
 static int iwl_trans_pcie_start_hw(struct iwl_trans *trans)
 {
 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
@@ -1440,29 +576,28 @@ static int iwl_trans_pcie_start_hw(struct iwl_trans *trans)
 
 	if (!trans_pcie->irq_requested) {
 		tasklet_init(&trans_pcie->irq_tasklet, (void (*)(unsigned long))
-			iwl_irq_tasklet, (unsigned long)trans);
+			iwl_pcie_tasklet, (unsigned long)trans);
 
-		iwl_alloc_isr_ict(trans);
+		iwl_pcie_alloc_ict(trans);
 
-		err = request_irq(trans_pcie->irq, iwl_isr_ict, IRQF_SHARED,
-				  DRV_NAME, trans);
+		err = request_irq(trans_pcie->irq, iwl_pcie_isr_ict,
+				  IRQF_SHARED, DRV_NAME, trans);
 		if (err) {
 			IWL_ERR(trans, "Error allocating IRQ %d\n",
 				trans_pcie->irq);
 			goto error;
 		}
 
-		INIT_WORK(&trans_pcie->rx_replenish, iwl_bg_rx_replenish);
 		trans_pcie->irq_requested = true;
 	}
 
-	err = iwl_prepare_card_hw(trans);
+	err = iwl_pcie_prepare_card_hw(trans);
 	if (err) {
 		IWL_ERR(trans, "Error while preparing HW: %d\n", err);
 		goto err_free_irq;
 	}
 
-	iwl_apm_init(trans);
+	iwl_pcie_apm_init(trans);
 
 	/* From now on, the op_mode will be kept updated about RF kill state */
 	iwl_enable_rfkill_int(trans);
@@ -1476,7 +611,7 @@ err_free_irq:
 	trans_pcie->irq_requested = false;
 	free_irq(trans_pcie->irq, trans);
 error:
-	iwl_free_isr_ict(trans);
+	iwl_pcie_free_ict(trans);
 	tasklet_kill(&trans_pcie->irq_tasklet);
 	return err;
 }
@@ -1492,7 +627,7 @@ static void iwl_trans_pcie_stop_hw(struct iwl_trans *trans,
 	iwl_disable_interrupts(trans);
 	spin_unlock_irqrestore(&trans_pcie->irq_lock, flags);
 
-	iwl_apm_stop(trans);
+	iwl_pcie_apm_stop(trans);
 
 	spin_lock_irqsave(&trans_pcie->irq_lock, flags);
 	iwl_disable_interrupts(trans);
@@ -1516,27 +651,6 @@ static void iwl_trans_pcie_stop_hw(struct iwl_trans *trans,
 	}
 }
 
-static void iwl_trans_pcie_reclaim(struct iwl_trans *trans, int txq_id, int ssn,
-				   struct sk_buff_head *skbs)
-{
-	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
-	struct iwl_tx_queue *txq = &trans_pcie->txq[txq_id];
-	/* n_bd is usually 256 => n_bd - 1 = 0xff */
-	int tfd_num = ssn & (txq->q.n_bd - 1);
-
-	spin_lock(&txq->lock);
-
-	if (txq->q.read_ptr != tfd_num) {
-		IWL_DEBUG_TX_REPLY(trans, "[Q %d] %d -> %d (%d)\n",
-				   txq_id, txq->q.read_ptr, tfd_num, ssn);
-		iwl_tx_queue_reclaim(trans, txq_id, tfd_num, skbs);
-		if (iwl_queue_space(&txq->q) > txq->q.low_mark)
-			iwl_wake_queue(trans, txq);
-	}
-
-	spin_unlock(&txq->lock);
-}
-
 static void iwl_trans_pcie_write8(struct iwl_trans *trans, u32 ofs, u8 val)
 {
 	writeb(val, IWL_TRANS_GET_PCIE_TRANS(trans)->hw_base + ofs);
@@ -1583,12 +697,12 @@ void iwl_trans_pcie_free(struct iwl_trans *trans)
 {
 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
 
-	iwl_trans_pcie_tx_free(trans);
-	iwl_trans_pcie_rx_free(trans);
+	iwl_pcie_tx_free(trans);
+	iwl_pcie_rx_free(trans);
 
 	if (trans_pcie->irq_requested == true) {
 		free_irq(trans_pcie->irq, trans);
-		iwl_free_isr_ict(trans);
+		iwl_pcie_free_ict(trans);
 	}
 
 	pci_disable_msi(trans_pcie->pci_dev);
@@ -1634,10 +748,10 @@ static int iwl_trans_pcie_resume(struct iwl_trans *trans)
 
 #define IWL_FLUSH_WAIT_MS	2000
 
-static int iwl_trans_pcie_wait_tx_queue_empty(struct iwl_trans *trans)
+static int iwl_trans_pcie_wait_txq_empty(struct iwl_trans *trans)
 {
 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
-	struct iwl_tx_queue *txq;
+	struct iwl_txq *txq;
 	struct iwl_queue *q;
 	int cnt;
 	unsigned long now = jiffies;
@@ -1681,7 +795,7 @@ static const char *get_fh_string(int cmd)
 #undef IWL_CMD
 }
 
-int iwl_dump_fh(struct iwl_trans *trans, char **buf)
+int iwl_pcie_dump_fh(struct iwl_trans *trans, char **buf)
 {
 	int i;
 	static const u32 fh_tbl[] = {
@@ -1760,7 +874,7 @@ static const char *get_csr_string(int cmd)
 #undef IWL_CMD
 }
 
-void iwl_dump_csr(struct iwl_trans *trans)
+void iwl_pcie_dump_csr(struct iwl_trans *trans)
 {
 	int i;
 	static const u32 csr_tbl[] = {
@@ -1817,7 +931,6 @@ static ssize_t iwl_dbgfs_##name##_write(struct file *file,              \
 					const char __user *user_buf,    \
 					size_t count, loff_t *ppos);
 
-
 #define DEBUGFS_READ_FILE_OPS(name)					\
 	DEBUGFS_READ_FUNC(name);					\
 static const struct file_operations iwl_dbgfs_##name##_ops = {		\
@@ -1850,7 +963,7 @@ static ssize_t iwl_dbgfs_tx_queue_read(struct file *file,
 {
 	struct iwl_trans *trans = file->private_data;
 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
-	struct iwl_tx_queue *txq;
+	struct iwl_txq *txq;
 	struct iwl_queue *q;
 	char *buf;
 	int pos = 0;
@@ -1887,7 +1000,7 @@ static ssize_t iwl_dbgfs_rx_queue_read(struct file *file,
 {
 	struct iwl_trans *trans = file->private_data;
 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
-	struct iwl_rx_queue *rxq = &trans_pcie->rxq;
+	struct iwl_rxq *rxq = &trans_pcie->rxq;
 	char buf[256];
 	int pos = 0;
 	const size_t bufsz = sizeof(buf);
@@ -2006,7 +1119,7 @@ static ssize_t iwl_dbgfs_csr_write(struct file *file,
 	if (sscanf(buf, "%d", &csr) != 1)
 		return -EFAULT;
 
-	iwl_dump_csr(trans);
+	iwl_pcie_dump_csr(trans);
 
 	return count;
 }
@@ -2020,7 +1133,7 @@ static ssize_t iwl_dbgfs_fh_reg_read(struct file *file,
 	int pos = 0;
 	ssize_t ret = -EFAULT;
 
-	ret = pos = iwl_dump_fh(trans, &buf);
+	ret = pos = iwl_pcie_dump_fh(trans, &buf);
 	if (buf) {
 		ret = simple_read_from_buffer(user_buf,
 					      count, ppos, buf, pos);
@@ -2089,7 +1202,7 @@ static const struct iwl_trans_ops trans_ops_pcie = {
 
 	.wowlan_suspend = iwl_trans_pcie_wowlan_suspend,
 
-	.send_cmd = iwl_trans_pcie_send_cmd,
+	.send_cmd = iwl_trans_pcie_send_hcmd,
 
 	.tx = iwl_trans_pcie_tx,
 	.reclaim = iwl_trans_pcie_reclaim,
@@ -2099,7 +1212,7 @@ static const struct iwl_trans_ops trans_ops_pcie = {
 
 	.dbgfs_register = iwl_trans_pcie_dbgfs_register,
 
-	.wait_tx_queue_empty = iwl_trans_pcie_wait_tx_queue_empty,
+	.wait_tx_queue_empty = iwl_trans_pcie_wait_txq_empty,
 
 #ifdef CONFIG_PM_SLEEP
 	.suspend = iwl_trans_pcie_suspend,
@@ -2124,7 +1237,7 @@ struct iwl_trans *iwl_trans_pcie_alloc(struct pci_dev *pdev,
 	trans = kzalloc(sizeof(struct iwl_trans) +
 			sizeof(struct iwl_trans_pcie), GFP_KERNEL);
 
-	if (WARN_ON(!trans))
+	if (!trans)
 		return NULL;
 
 	trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
@@ -2180,8 +1293,15 @@ struct iwl_trans *iwl_trans_pcie_alloc(struct pci_dev *pdev,
 	pci_write_config_byte(pdev, PCI_CFG_RETRY_TIMEOUT, 0x00);
 
 	err = pci_enable_msi(pdev);
-	if (err)
+	if (err) {
 		dev_err(&pdev->dev, "pci_enable_msi failed(0X%x)\n", err);
+		/* enable rfkill interrupt: hw bug w/a */
+		pci_read_config_word(pdev, PCI_COMMAND, &pci_cmd);
+		if (pci_cmd & PCI_COMMAND_INTX_DISABLE) {
+			pci_cmd &= ~PCI_COMMAND_INTX_DISABLE;
+			pci_write_config_word(pdev, PCI_COMMAND, pci_cmd);
+		}
+	}
 
 	trans->dev = &pdev->dev;
 	trans_pcie->irq = pdev->irq;
@@ -2191,14 +1311,6 @@ struct iwl_trans *iwl_trans_pcie_alloc(struct pci_dev *pdev,
 	snprintf(trans->hw_id_str, sizeof(trans->hw_id_str),
 		 "PCI ID: 0x%04X:0x%04X", pdev->device, pdev->subsystem_device);
 
-	/* TODO: Move this away, not needed if not MSI */
-	/* enable rfkill interrupt: hw bug w/a */
-	pci_read_config_word(pdev, PCI_COMMAND, &pci_cmd);
-	if (pci_cmd & PCI_COMMAND_INTX_DISABLE) {
-		pci_cmd &= ~PCI_COMMAND_INTX_DISABLE;
-		pci_write_config_word(pdev, PCI_COMMAND, pci_cmd);
-	}
-
 	/* Initialize the wait queue for commands */
 	init_waitqueue_head(&trans_pcie->wait_command_queue);
 	spin_lock_init(&trans->reg_lock);
diff --git a/drivers/net/wireless/iwlwifi/pcie/tx.c b/drivers/net/wireless/iwlwifi/pcie/tx.c
index dcc7e1256e3..6c5b867c353 100644
--- a/drivers/net/wireless/iwlwifi/pcie/tx.c
+++ b/drivers/net/wireless/iwlwifi/pcie/tx.c
@@ -42,12 +42,170 @@
 #define IWL_TX_CRC_SIZE 4
 #define IWL_TX_DELIMITER_SIZE 4
 
-/**
- * iwl_trans_txq_update_byte_cnt_tbl - Set up entry in Tx byte-count array
+/*************** DMA-QUEUE-GENERAL-FUNCTIONS  *****
+ * DMA services
+ *
+ * Theory of operation
+ *
+ * A Tx or Rx queue resides in host DRAM, and is comprised of a circular buffer
+ * of buffer descriptors, each of which points to one or more data buffers for
+ * the device to read from or fill.  Driver and device exchange status of each
+ * queue via "read" and "write" pointers.  Driver keeps minimum of 2 empty
+ * entries in each circular buffer, to protect against confusing empty and full
+ * queue states.
+ *
+ * The device reads or writes the data in the queues via the device's several
+ * DMA/FIFO channels.  Each queue is mapped to a single DMA channel.
+ *
+ * For Tx queue, there are low mark and high mark limits. If, after queuing
+ * the packet for Tx, free space become < low mark, Tx queue stopped. When
+ * reclaiming packets (on 'tx done IRQ), if free space become > high mark,
+ * Tx queue resumed.
+ *
+ ***************************************************/
+static int iwl_queue_space(const struct iwl_queue *q)
+{
+	int s = q->read_ptr - q->write_ptr;
+
+	if (q->read_ptr > q->write_ptr)
+		s -= q->n_bd;
+
+	if (s <= 0)
+		s += q->n_window;
+	/* keep some reserve to not confuse empty and full situations */
+	s -= 2;
+	if (s < 0)
+		s = 0;
+	return s;
+}
+
+/*
+ * iwl_queue_init - Initialize queue's high/low-water and read/write indexes
+ */
+static int iwl_queue_init(struct iwl_queue *q, int count, int slots_num, u32 id)
+{
+	q->n_bd = count;
+	q->n_window = slots_num;
+	q->id = id;
+
+	/* count must be power-of-two size, otherwise iwl_queue_inc_wrap
+	 * and iwl_queue_dec_wrap are broken. */
+	if (WARN_ON(!is_power_of_2(count)))
+		return -EINVAL;
+
+	/* slots_num must be power-of-two size, otherwise
+	 * get_cmd_index is broken. */
+	if (WARN_ON(!is_power_of_2(slots_num)))
+		return -EINVAL;
+
+	q->low_mark = q->n_window / 4;
+	if (q->low_mark < 4)
+		q->low_mark = 4;
+
+	q->high_mark = q->n_window / 8;
+	if (q->high_mark < 2)
+		q->high_mark = 2;
+
+	q->write_ptr = 0;
+	q->read_ptr = 0;
+
+	return 0;
+}
+
+static int iwl_pcie_alloc_dma_ptr(struct iwl_trans *trans,
+				  struct iwl_dma_ptr *ptr, size_t size)
+{
+	if (WARN_ON(ptr->addr))
+		return -EINVAL;
+
+	ptr->addr = dma_alloc_coherent(trans->dev, size,
+				       &ptr->dma, GFP_KERNEL);
+	if (!ptr->addr)
+		return -ENOMEM;
+	ptr->size = size;
+	return 0;
+}
+
+static void iwl_pcie_free_dma_ptr(struct iwl_trans *trans,
+				  struct iwl_dma_ptr *ptr)
+{
+	if (unlikely(!ptr->addr))
+		return;
+
+	dma_free_coherent(trans->dev, ptr->size, ptr->addr, ptr->dma);
+	memset(ptr, 0, sizeof(*ptr));
+}
+
+static void iwl_pcie_txq_stuck_timer(unsigned long data)
+{
+	struct iwl_txq *txq = (void *)data;
+	struct iwl_queue *q = &txq->q;
+	struct iwl_trans_pcie *trans_pcie = txq->trans_pcie;
+	struct iwl_trans *trans = iwl_trans_pcie_get_trans(trans_pcie);
+	u32 scd_sram_addr = trans_pcie->scd_base_addr +
+				SCD_TX_STTS_QUEUE_OFFSET(txq->q.id);
+	u8 buf[16];
+	int i;
+
+	spin_lock(&txq->lock);
+	/* check if triggered erroneously */
+	if (txq->q.read_ptr == txq->q.write_ptr) {
+		spin_unlock(&txq->lock);
+		return;
+	}
+	spin_unlock(&txq->lock);
+
+	IWL_ERR(trans, "Queue %d stuck for %u ms.\n", txq->q.id,
+		jiffies_to_msecs(trans_pcie->wd_timeout));
+	IWL_ERR(trans, "Current SW read_ptr %d write_ptr %d\n",
+		txq->q.read_ptr, txq->q.write_ptr);
+
+	iwl_read_targ_mem_bytes(trans, scd_sram_addr, buf, sizeof(buf));
+
+	iwl_print_hex_error(trans, buf, sizeof(buf));
+
+	for (i = 0; i < FH_TCSR_CHNL_NUM; i++)
+		IWL_ERR(trans, "FH TRBs(%d) = 0x%08x\n", i,
+			iwl_read_direct32(trans, FH_TX_TRB_REG(i)));
+
+	for (i = 0; i < trans->cfg->base_params->num_of_queues; i++) {
+		u32 status = iwl_read_prph(trans, SCD_QUEUE_STATUS_BITS(i));
+		u8 fifo = (status >> SCD_QUEUE_STTS_REG_POS_TXF) & 0x7;
+		bool active = !!(status & BIT(SCD_QUEUE_STTS_REG_POS_ACTIVE));
+		u32 tbl_dw =
+			iwl_read_targ_mem(trans,
+					  trans_pcie->scd_base_addr +
+					  SCD_TRANS_TBL_OFFSET_QUEUE(i));
+
+		if (i & 0x1)
+			tbl_dw = (tbl_dw & 0xFFFF0000) >> 16;
+		else
+			tbl_dw = tbl_dw & 0x0000FFFF;
+
+		IWL_ERR(trans,
+			"Q %d is %sactive and mapped to fifo %d ra_tid 0x%04x [%d,%d]\n",
+			i, active ? "" : "in", fifo, tbl_dw,
+			iwl_read_prph(trans,
+				      SCD_QUEUE_RDPTR(i)) & (txq->q.n_bd - 1),
+			iwl_read_prph(trans, SCD_QUEUE_WRPTR(i)));
+	}
+
+	for (i = q->read_ptr; i != q->write_ptr;
+	     i = iwl_queue_inc_wrap(i, q->n_bd)) {
+		struct iwl_tx_cmd *tx_cmd =
+			(struct iwl_tx_cmd *)txq->entries[i].cmd->payload;
+		IWL_ERR(trans, "scratch %d = 0x%08x\n", i,
+			get_unaligned_le32(&tx_cmd->scratch));
+	}
+
+	iwl_op_mode_nic_error(trans->op_mode);
+}
+
+/*
+ * iwl_pcie_txq_update_byte_cnt_tbl - Set up entry in Tx byte-count array
  */
-void iwl_trans_txq_update_byte_cnt_tbl(struct iwl_trans *trans,
-				       struct iwl_tx_queue *txq,
-				       u16 byte_cnt)
+static void iwl_pcie_txq_update_byte_cnt_tbl(struct iwl_trans *trans,
+					     struct iwl_txq *txq, u16 byte_cnt)
 {
 	struct iwlagn_scd_bc_tbl *scd_bc_tbl;
 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
@@ -88,10 +246,36 @@ void iwl_trans_txq_update_byte_cnt_tbl(struct iwl_trans *trans,
 			tfd_offset[TFD_QUEUE_SIZE_MAX + write_ptr] = bc_ent;
 }
 
-/**
- * iwl_txq_update_write_ptr - Send new write index to hardware
+static void iwl_pcie_txq_inval_byte_cnt_tbl(struct iwl_trans *trans,
+					    struct iwl_txq *txq)
+{
+	struct iwl_trans_pcie *trans_pcie =
+		IWL_TRANS_GET_PCIE_TRANS(trans);
+	struct iwlagn_scd_bc_tbl *scd_bc_tbl = trans_pcie->scd_bc_tbls.addr;
+	int txq_id = txq->q.id;
+	int read_ptr = txq->q.read_ptr;
+	u8 sta_id = 0;
+	__le16 bc_ent;
+	struct iwl_tx_cmd *tx_cmd =
+		(void *)txq->entries[txq->q.read_ptr].cmd->payload;
+
+	WARN_ON(read_ptr >= TFD_QUEUE_SIZE_MAX);
+
+	if (txq_id != trans_pcie->cmd_queue)
+		sta_id = tx_cmd->sta_id;
+
+	bc_ent = cpu_to_le16(1 | (sta_id << 12));
+	scd_bc_tbl[txq_id].tfd_offset[read_ptr] = bc_ent;
+
+	if (read_ptr < TFD_QUEUE_SIZE_BC_DUP)
+		scd_bc_tbl[txq_id].
+			tfd_offset[TFD_QUEUE_SIZE_MAX + read_ptr] = bc_ent;
+}
+
+/*
+ * iwl_pcie_txq_inc_wr_ptr - Send new write index to hardware
  */
-void iwl_txq_update_write_ptr(struct iwl_trans *trans, struct iwl_tx_queue *txq)
+void iwl_pcie_txq_inc_wr_ptr(struct iwl_trans *trans, struct iwl_txq *txq)
 {
 	u32 reg = 0;
 	int txq_id = txq->q.id;
@@ -137,7 +321,7 @@ void iwl_txq_update_write_ptr(struct iwl_trans *trans, struct iwl_tx_queue *txq)
 	txq->need_update = 0;
 }
 
-static inline dma_addr_t iwl_tfd_tb_get_addr(struct iwl_tfd *tfd, u8 idx)
+static inline dma_addr_t iwl_pcie_tfd_tb_get_addr(struct iwl_tfd *tfd, u8 idx)
 {
 	struct iwl_tfd_tb *tb = &tfd->tbs[idx];
 
@@ -149,15 +333,15 @@ static inline dma_addr_t iwl_tfd_tb_get_addr(struct iwl_tfd *tfd, u8 idx)
 	return addr;
 }
 
-static inline u16 iwl_tfd_tb_get_len(struct iwl_tfd *tfd, u8 idx)
+static inline u16 iwl_pcie_tfd_tb_get_len(struct iwl_tfd *tfd, u8 idx)
 {
 	struct iwl_tfd_tb *tb = &tfd->tbs[idx];
 
 	return le16_to_cpu(tb->hi_n_len) >> 4;
 }
 
-static inline void iwl_tfd_set_tb(struct iwl_tfd *tfd, u8 idx,
-				  dma_addr_t addr, u16 len)
+static inline void iwl_pcie_tfd_set_tb(struct iwl_tfd *tfd, u8 idx,
+				       dma_addr_t addr, u16 len)
 {
 	struct iwl_tfd_tb *tb = &tfd->tbs[idx];
 	u16 hi_n_len = len << 4;
@@ -171,19 +355,20 @@ static inline void iwl_tfd_set_tb(struct iwl_tfd *tfd, u8 idx,
 	tfd->num_tbs = idx + 1;
 }
 
-static inline u8 iwl_tfd_get_num_tbs(struct iwl_tfd *tfd)
+static inline u8 iwl_pcie_tfd_get_num_tbs(struct iwl_tfd *tfd)
 {
 	return tfd->num_tbs & 0x1f;
 }
 
-static void iwl_unmap_tfd(struct iwl_trans *trans, struct iwl_cmd_meta *meta,
-			  struct iwl_tfd *tfd, enum dma_data_direction dma_dir)
+static void iwl_pcie_tfd_unmap(struct iwl_trans *trans,
+			       struct iwl_cmd_meta *meta, struct iwl_tfd *tfd,
+			       enum dma_data_direction dma_dir)
 {
 	int i;
 	int num_tbs;
 
 	/* Sanity check on number of chunks */
-	num_tbs = iwl_tfd_get_num_tbs(tfd);
+	num_tbs = iwl_pcie_tfd_get_num_tbs(tfd);
 
 	if (num_tbs >= IWL_NUM_OF_TBS) {
 		IWL_ERR(trans, "Too many chunks: %i\n", num_tbs);
@@ -200,14 +385,14 @@ static void iwl_unmap_tfd(struct iwl_trans *trans, struct iwl_cmd_meta *meta,
 
 	/* Unmap chunks, if any. */
 	for (i = 1; i < num_tbs; i++)
-		dma_unmap_single(trans->dev, iwl_tfd_tb_get_addr(tfd, i),
-				iwl_tfd_tb_get_len(tfd, i), dma_dir);
+		dma_unmap_single(trans->dev, iwl_pcie_tfd_tb_get_addr(tfd, i),
+				 iwl_pcie_tfd_tb_get_len(tfd, i), dma_dir);
 
 	tfd->num_tbs = 0;
 }
 
-/**
- * iwl_txq_free_tfd - Free all chunks referenced by TFD [txq->q.read_ptr]
+/*
+ * iwl_pcie_txq_free_tfd - Free all chunks referenced by TFD [txq->q.read_ptr]
  * @trans - transport private data
  * @txq - tx queue
  * @dma_dir - the direction of the DMA mapping
@@ -215,8 +400,8 @@ static void iwl_unmap_tfd(struct iwl_trans *trans, struct iwl_cmd_meta *meta,
  * Does NOT advance any TFD circular buffer read/write indexes
  * Does NOT free the TFD itself (which is within circular buffer)
  */
-void iwl_txq_free_tfd(struct iwl_trans *trans, struct iwl_tx_queue *txq,
-		      enum dma_data_direction dma_dir)
+static void iwl_pcie_txq_free_tfd(struct iwl_trans *trans, struct iwl_txq *txq,
+				  enum dma_data_direction dma_dir)
 {
 	struct iwl_tfd *tfd_tmp = txq->tfds;
 
@@ -227,8 +412,8 @@ void iwl_txq_free_tfd(struct iwl_trans *trans, struct iwl_tx_queue *txq,
 	lockdep_assert_held(&txq->lock);
 
 	/* We have only q->n_window txq->entries, but we use q->n_bd tfds */
-	iwl_unmap_tfd(trans, &txq->entries[idx].meta, &tfd_tmp[rd_ptr],
-		      dma_dir);
+	iwl_pcie_tfd_unmap(trans, &txq->entries[idx].meta, &tfd_tmp[rd_ptr],
+			   dma_dir);
 
 	/* free SKB */
 	if (txq->entries) {
@@ -247,10 +432,8 @@ void iwl_txq_free_tfd(struct iwl_trans *trans, struct iwl_tx_queue *txq,
 	}
 }
 
-int iwlagn_txq_attach_buf_to_tfd(struct iwl_trans *trans,
-				 struct iwl_tx_queue *txq,
-				 dma_addr_t addr, u16 len,
-				 u8 reset)
+static int iwl_pcie_txq_build_tfd(struct iwl_trans *trans, struct iwl_txq *txq,
+				  dma_addr_t addr, u16 len, u8 reset)
 {
 	struct iwl_queue *q;
 	struct iwl_tfd *tfd, *tfd_tmp;
@@ -263,7 +446,7 @@ int iwlagn_txq_attach_buf_to_tfd(struct iwl_trans *trans,
 	if (reset)
 		memset(tfd, 0, sizeof(*tfd));
 
-	num_tbs = iwl_tfd_get_num_tbs(tfd);
+	num_tbs = iwl_pcie_tfd_get_num_tbs(tfd);
 
 	/* Each TFD can point to a maximum 20 Tx buffers */
 	if (num_tbs >= IWL_NUM_OF_TBS) {
@@ -279,108 +462,534 @@ int iwlagn_txq_attach_buf_to_tfd(struct iwl_trans *trans,
 		IWL_ERR(trans, "Unaligned address = %llx\n",
 			(unsigned long long)addr);
 
-	iwl_tfd_set_tb(tfd, num_tbs, addr, len);
+	iwl_pcie_tfd_set_tb(tfd, num_tbs, addr, len);
 
 	return 0;
 }
 
-/*************** DMA-QUEUE-GENERAL-FUNCTIONS  *****
- * DMA services
- *
- * Theory of operation
- *
- * A Tx or Rx queue resides in host DRAM, and is comprised of a circular buffer
- * of buffer descriptors, each of which points to one or more data buffers for
- * the device to read from or fill.  Driver and device exchange status of each
- * queue via "read" and "write" pointers.  Driver keeps minimum of 2 empty
- * entries in each circular buffer, to protect against confusing empty and full
- * queue states.
- *
- * The device reads or writes the data in the queues via the device's several
- * DMA/FIFO channels.  Each queue is mapped to a single DMA channel.
- *
- * For Tx queue, there are low mark and high mark limits. If, after queuing
- * the packet for Tx, free space become < low mark, Tx queue stopped. When
- * reclaiming packets (on 'tx done IRQ), if free space become > high mark,
- * Tx queue resumed.
+static int iwl_pcie_txq_alloc(struct iwl_trans *trans,
+			       struct iwl_txq *txq, int slots_num,
+			       u32 txq_id)
+{
+	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
+	size_t tfd_sz = sizeof(struct iwl_tfd) * TFD_QUEUE_SIZE_MAX;
+	int i;
+
+	if (WARN_ON(txq->entries || txq->tfds))
+		return -EINVAL;
+
+	setup_timer(&txq->stuck_timer, iwl_pcie_txq_stuck_timer,
+		    (unsigned long)txq);
+	txq->trans_pcie = trans_pcie;
+
+	txq->q.n_window = slots_num;
+
+	txq->entries = kcalloc(slots_num,
+			       sizeof(struct iwl_pcie_txq_entry),
+			       GFP_KERNEL);
+
+	if (!txq->entries)
+		goto error;
+
+	if (txq_id == trans_pcie->cmd_queue)
+		for (i = 0; i < slots_num; i++) {
+			txq->entries[i].cmd =
+				kmalloc(sizeof(struct iwl_device_cmd),
+					GFP_KERNEL);
+			if (!txq->entries[i].cmd)
+				goto error;
+		}
+
+	/* Circular buffer of transmit frame descriptors (TFDs),
+	 * shared with device */
+	txq->tfds = dma_alloc_coherent(trans->dev, tfd_sz,
+				       &txq->q.dma_addr, GFP_KERNEL);
+	if (!txq->tfds) {
+		IWL_ERR(trans, "dma_alloc_coherent(%zd) failed\n", tfd_sz);
+		goto error;
+	}
+	txq->q.id = txq_id;
+
+	return 0;
+error:
+	if (txq->entries && txq_id == trans_pcie->cmd_queue)
+		for (i = 0; i < slots_num; i++)
+			kfree(txq->entries[i].cmd);
+	kfree(txq->entries);
+	txq->entries = NULL;
+
+	return -ENOMEM;
+
+}
+
+static int iwl_pcie_txq_init(struct iwl_trans *trans, struct iwl_txq *txq,
+			      int slots_num, u32 txq_id)
+{
+	int ret;
+
+	txq->need_update = 0;
+
+	/* 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));
+
+	/* Initialize queue's high/low-water marks, and head/tail indexes */
+	ret = iwl_queue_init(&txq->q, TFD_QUEUE_SIZE_MAX, slots_num,
+			txq_id);
+	if (ret)
+		return ret;
+
+	spin_lock_init(&txq->lock);
+
+	/*
+	 * Tell nic where to find circular buffer of Tx Frame Descriptors for
+	 * given Tx queue, and enable the DMA channel used for that queue.
+	 * Circular buffer (TFD queue in DRAM) physical base address */
+	iwl_write_direct32(trans, FH_MEM_CBBC_QUEUE(txq_id),
+			   txq->q.dma_addr >> 8);
+
+	return 0;
+}
+
+/*
+ * iwl_pcie_txq_unmap -  Unmap any remaining DMA mappings and free skb's
+ */
+static void iwl_pcie_txq_unmap(struct iwl_trans *trans, int txq_id)
+{
+	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
+	struct iwl_txq *txq = &trans_pcie->txq[txq_id];
+	struct iwl_queue *q = &txq->q;
+	enum dma_data_direction dma_dir;
+
+	if (!q->n_bd)
+		return;
+
+	/* In the command queue, all the TBs are mapped as BIDI
+	 * so unmap them as such.
+	 */
+	if (txq_id == trans_pcie->cmd_queue)
+		dma_dir = DMA_BIDIRECTIONAL;
+	else
+		dma_dir = DMA_TO_DEVICE;
+
+	spin_lock_bh(&txq->lock);
+	while (q->write_ptr != q->read_ptr) {
+		iwl_pcie_txq_free_tfd(trans, txq, dma_dir);
+		q->read_ptr = iwl_queue_inc_wrap(q->read_ptr, q->n_bd);
+	}
+	spin_unlock_bh(&txq->lock);
+}
+
+/*
+ * iwl_pcie_txq_free - Deallocate DMA queue.
+ * @txq: Transmit queue to deallocate.
  *
- ***************************************************/
+ * Empty queue by removing and destroying all BD's.
+ * Free all buffers.
+ * 0-fill, but do not free "txq" descriptor structure.
+ */
+static void iwl_pcie_txq_free(struct iwl_trans *trans, int txq_id)
+{
+	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
+	struct iwl_txq *txq = &trans_pcie->txq[txq_id];
+	struct device *dev = trans->dev;
+	int i;
+
+	if (WARN_ON(!txq))
+		return;
+
+	iwl_pcie_txq_unmap(trans, txq_id);
+
+	/* De-alloc array of command/tx buffers */
+	if (txq_id == trans_pcie->cmd_queue)
+		for (i = 0; i < txq->q.n_window; i++) {
+			kfree(txq->entries[i].cmd);
+			kfree(txq->entries[i].copy_cmd);
+			kfree(txq->entries[i].free_buf);
+		}
+
+	/* De-alloc circular buffer of TFDs */
+	if (txq->q.n_bd) {
+		dma_free_coherent(dev, sizeof(struct iwl_tfd) *
+				  txq->q.n_bd, txq->tfds, txq->q.dma_addr);
+		memset(&txq->q.dma_addr, 0, sizeof(txq->q.dma_addr));
+	}
+
+	kfree(txq->entries);
+	txq->entries = NULL;
 
-int iwl_queue_space(const struct iwl_queue *q)
+	del_timer_sync(&txq->stuck_timer);
+
+	/* 0-fill queue descriptor structure */
+	memset(txq, 0, sizeof(*txq));
+}
+
+/*
+ * Activate/Deactivate Tx DMA/FIFO channels according tx fifos mask
+ */
+static void iwl_pcie_txq_set_sched(struct iwl_trans *trans, u32 mask)
 {
-	int s = q->read_ptr - q->write_ptr;
+	struct iwl_trans_pcie __maybe_unused *trans_pcie =
+		IWL_TRANS_GET_PCIE_TRANS(trans);
 
-	if (q->read_ptr > q->write_ptr)
-		s -= q->n_bd;
+	iwl_write_prph(trans, SCD_TXFACT, mask);
+}
 
-	if (s <= 0)
-		s += q->n_window;
-	/* keep some reserve to not confuse empty and full situations */
-	s -= 2;
-	if (s < 0)
-		s = 0;
-	return s;
+void iwl_pcie_tx_start(struct iwl_trans *trans, u32 scd_base_addr)
+{
+	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
+	u32 a;
+	int chan;
+	u32 reg_val;
+
+	/* make sure all queue are not stopped/used */
+	memset(trans_pcie->queue_stopped, 0, sizeof(trans_pcie->queue_stopped));
+	memset(trans_pcie->queue_used, 0, sizeof(trans_pcie->queue_used));
+
+	trans_pcie->scd_base_addr =
+		iwl_read_prph(trans, SCD_SRAM_BASE_ADDR);
+
+	WARN_ON(scd_base_addr != 0 &&
+		scd_base_addr != trans_pcie->scd_base_addr);
+
+	a = trans_pcie->scd_base_addr + SCD_CONTEXT_MEM_LOWER_BOUND;
+	/* reset conext data memory */
+	for (; a < trans_pcie->scd_base_addr + SCD_CONTEXT_MEM_UPPER_BOUND;
+		a += 4)
+		iwl_write_targ_mem(trans, a, 0);
+	/* reset tx status memory */
+	for (; a < trans_pcie->scd_base_addr + SCD_TX_STTS_MEM_UPPER_BOUND;
+		a += 4)
+		iwl_write_targ_mem(trans, a, 0);
+	for (; a < trans_pcie->scd_base_addr +
+	       SCD_TRANS_TBL_OFFSET_QUEUE(
+				trans->cfg->base_params->num_of_queues);
+	       a += 4)
+		iwl_write_targ_mem(trans, a, 0);
+
+	iwl_write_prph(trans, SCD_DRAM_BASE_ADDR,
+		       trans_pcie->scd_bc_tbls.dma >> 10);
+
+	/* The chain extension of the SCD doesn't work well. This feature is
+	 * enabled by default by the HW, so we need to disable it manually.
+	 */
+	iwl_write_prph(trans, SCD_CHAINEXT_EN, 0);
+
+	iwl_trans_ac_txq_enable(trans, trans_pcie->cmd_queue,
+				trans_pcie->cmd_fifo);
+
+	/* Activate all Tx DMA/FIFO channels */
+	iwl_pcie_txq_set_sched(trans, IWL_MASK(0, 7));
+
+	/* Enable DMA channel */
+	for (chan = 0; chan < FH_TCSR_CHNL_NUM; chan++)
+		iwl_write_direct32(trans, FH_TCSR_CHNL_TX_CONFIG_REG(chan),
+				   FH_TCSR_TX_CONFIG_REG_VAL_DMA_CHNL_ENABLE |
+				   FH_TCSR_TX_CONFIG_REG_VAL_DMA_CREDIT_ENABLE);
+
+	/* Update FH chicken bits */
+	reg_val = iwl_read_direct32(trans, FH_TX_CHICKEN_BITS_REG);
+	iwl_write_direct32(trans, FH_TX_CHICKEN_BITS_REG,
+			   reg_val | FH_TX_CHICKEN_BITS_SCD_AUTO_RETRY_EN);
+
+	/* Enable L1-Active */
+	iwl_clear_bits_prph(trans, APMG_PCIDEV_STT_REG,
+			    APMG_PCIDEV_STT_VAL_L1_ACT_DIS);
 }
 
-/**
- * iwl_queue_init - Initialize queue's high/low-water and read/write indexes
+/*
+ * iwl_pcie_tx_stop - Stop all Tx DMA channels
  */
-int iwl_queue_init(struct iwl_queue *q, int count, int slots_num, u32 id)
+int iwl_pcie_tx_stop(struct iwl_trans *trans)
 {
-	q->n_bd = count;
-	q->n_window = slots_num;
-	q->id = id;
+	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
+	int ch, txq_id, ret;
+	unsigned long flags;
 
-	/* count must be power-of-two size, otherwise iwl_queue_inc_wrap
-	 * and iwl_queue_dec_wrap are broken. */
-	if (WARN_ON(!is_power_of_2(count)))
-		return -EINVAL;
+	/* Turn off all Tx DMA fifos */
+	spin_lock_irqsave(&trans_pcie->irq_lock, flags);
 
-	/* slots_num must be power-of-two size, otherwise
-	 * get_cmd_index is broken. */
-	if (WARN_ON(!is_power_of_2(slots_num)))
-		return -EINVAL;
+	iwl_pcie_txq_set_sched(trans, 0);
 
-	q->low_mark = q->n_window / 4;
-	if (q->low_mark < 4)
-		q->low_mark = 4;
+	/* Stop each Tx DMA channel, and wait for it to be idle */
+	for (ch = 0; ch < FH_TCSR_CHNL_NUM; ch++) {
+		iwl_write_direct32(trans,
+				   FH_TCSR_CHNL_TX_CONFIG_REG(ch), 0x0);
+		ret = iwl_poll_direct_bit(trans, FH_TSSR_TX_STATUS_REG,
+			FH_TSSR_TX_STATUS_REG_MSK_CHNL_IDLE(ch), 1000);
+		if (ret < 0)
+			IWL_ERR(trans,
+				"Failing on timeout while stopping DMA channel %d [0x%08x]\n",
+				ch,
+				iwl_read_direct32(trans,
+						  FH_TSSR_TX_STATUS_REG));
+	}
+	spin_unlock_irqrestore(&trans_pcie->irq_lock, flags);
 
-	q->high_mark = q->n_window / 8;
-	if (q->high_mark < 2)
-		q->high_mark = 2;
+	if (!trans_pcie->txq) {
+		IWL_WARN(trans,
+			 "Stopping tx queues that aren't allocated...\n");
+		return 0;
+	}
 
-	q->write_ptr = q->read_ptr = 0;
+	/* Unmap DMA from host system and free skb's */
+	for (txq_id = 0; txq_id < trans->cfg->base_params->num_of_queues;
+	     txq_id++)
+		iwl_pcie_txq_unmap(trans, txq_id);
 
 	return 0;
 }
 
-static void iwlagn_txq_inval_byte_cnt_tbl(struct iwl_trans *trans,
-					  struct iwl_tx_queue *txq)
+/*
+ * iwl_trans_tx_free - Free TXQ Context
+ *
+ * Destroy all TX DMA queues and structures
+ */
+void iwl_pcie_tx_free(struct iwl_trans *trans)
 {
-	struct iwl_trans_pcie *trans_pcie =
-		IWL_TRANS_GET_PCIE_TRANS(trans);
-	struct iwlagn_scd_bc_tbl *scd_bc_tbl = trans_pcie->scd_bc_tbls.addr;
-	int txq_id = txq->q.id;
-	int read_ptr = txq->q.read_ptr;
-	u8 sta_id = 0;
-	__le16 bc_ent;
-	struct iwl_tx_cmd *tx_cmd =
-		(void *)txq->entries[txq->q.read_ptr].cmd->payload;
+	int txq_id;
+	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
 
-	WARN_ON(read_ptr >= TFD_QUEUE_SIZE_MAX);
+	/* Tx queues */
+	if (trans_pcie->txq) {
+		for (txq_id = 0;
+		     txq_id < trans->cfg->base_params->num_of_queues; txq_id++)
+			iwl_pcie_txq_free(trans, txq_id);
+	}
 
-	if (txq_id != trans_pcie->cmd_queue)
-		sta_id = tx_cmd->sta_id;
+	kfree(trans_pcie->txq);
+	trans_pcie->txq = NULL;
 
-	bc_ent = cpu_to_le16(1 | (sta_id << 12));
-	scd_bc_tbl[txq_id].tfd_offset[read_ptr] = bc_ent;
+	iwl_pcie_free_dma_ptr(trans, &trans_pcie->kw);
 
-	if (read_ptr < TFD_QUEUE_SIZE_BC_DUP)
-		scd_bc_tbl[txq_id].
-			tfd_offset[TFD_QUEUE_SIZE_MAX + read_ptr] = bc_ent;
+	iwl_pcie_free_dma_ptr(trans, &trans_pcie->scd_bc_tbls);
+}
+
+/*
+ * iwl_pcie_tx_alloc - allocate TX context
+ * Allocate all Tx DMA structures and initialize them
+ */
+static int iwl_pcie_tx_alloc(struct iwl_trans *trans)
+{
+	int ret;
+	int txq_id, slots_num;
+	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
+
+	u16 scd_bc_tbls_size = trans->cfg->base_params->num_of_queues *
+			sizeof(struct iwlagn_scd_bc_tbl);
+
+	/*It is not allowed to alloc twice, so warn when this happens.
+	 * We cannot rely on the previous allocation, so free and fail */
+	if (WARN_ON(trans_pcie->txq)) {
+		ret = -EINVAL;
+		goto error;
+	}
+
+	ret = iwl_pcie_alloc_dma_ptr(trans, &trans_pcie->scd_bc_tbls,
+				   scd_bc_tbls_size);
+	if (ret) {
+		IWL_ERR(trans, "Scheduler BC Table allocation failed\n");
+		goto error;
+	}
+
+	/* Alloc keep-warm buffer */
+	ret = iwl_pcie_alloc_dma_ptr(trans, &trans_pcie->kw, IWL_KW_SIZE);
+	if (ret) {
+		IWL_ERR(trans, "Keep Warm allocation failed\n");
+		goto error;
+	}
+
+	trans_pcie->txq = kcalloc(trans->cfg->base_params->num_of_queues,
+				  sizeof(struct iwl_txq), GFP_KERNEL);
+	if (!trans_pcie->txq) {
+		IWL_ERR(trans, "Not enough memory for txq\n");
+		ret = ENOMEM;
+		goto error;
+	}
+
+	/* Alloc and init all Tx queues, including the command queue (#4/#9) */
+	for (txq_id = 0; txq_id < trans->cfg->base_params->num_of_queues;
+	     txq_id++) {
+		slots_num = (txq_id == trans_pcie->cmd_queue) ?
+					TFD_CMD_SLOTS : TFD_TX_CMD_SLOTS;
+		ret = iwl_pcie_txq_alloc(trans, &trans_pcie->txq[txq_id],
+					  slots_num, txq_id);
+		if (ret) {
+			IWL_ERR(trans, "Tx %d queue alloc failed\n", txq_id);
+			goto error;
+		}
+	}
+
+	return 0;
+
+error:
+	iwl_pcie_tx_free(trans);
+
+	return ret;
+}
+int iwl_pcie_tx_init(struct iwl_trans *trans)
+{
+	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
+	int ret;
+	int txq_id, slots_num;
+	unsigned long flags;
+	bool alloc = false;
+
+	if (!trans_pcie->txq) {
+		ret = iwl_pcie_tx_alloc(trans);
+		if (ret)
+			goto error;
+		alloc = true;
+	}
+
+	spin_lock_irqsave(&trans_pcie->irq_lock, flags);
+
+	/* Turn off all Tx DMA fifos */
+	iwl_write_prph(trans, SCD_TXFACT, 0);
+
+	/* Tell NIC where to find the "keep warm" buffer */
+	iwl_write_direct32(trans, FH_KW_MEM_ADDR_REG,
+			   trans_pcie->kw.dma >> 4);
+
+	spin_unlock_irqrestore(&trans_pcie->irq_lock, flags);
+
+	/* Alloc and init all Tx queues, including the command queue (#4/#9) */
+	for (txq_id = 0; txq_id < trans->cfg->base_params->num_of_queues;
+	     txq_id++) {
+		slots_num = (txq_id == trans_pcie->cmd_queue) ?
+					TFD_CMD_SLOTS : TFD_TX_CMD_SLOTS;
+		ret = iwl_pcie_txq_init(trans, &trans_pcie->txq[txq_id],
+					 slots_num, txq_id);
+		if (ret) {
+			IWL_ERR(trans, "Tx %d queue init failed\n", txq_id);
+			goto error;
+		}
+	}
+
+	return 0;
+error:
+	/*Upon error, free only if we allocated something */
+	if (alloc)
+		iwl_pcie_tx_free(trans);
+	return ret;
+}
+
+static inline void iwl_pcie_txq_progress(struct iwl_trans_pcie *trans_pcie,
+					   struct iwl_txq *txq)
+{
+	if (!trans_pcie->wd_timeout)
+		return;
+
+	/*
+	 * if empty delete timer, otherwise move timer forward
+	 * since we're making progress on this queue
+	 */
+	if (txq->q.read_ptr == txq->q.write_ptr)
+		del_timer(&txq->stuck_timer);
+	else
+		mod_timer(&txq->stuck_timer, jiffies + trans_pcie->wd_timeout);
+}
+
+/* Frees buffers until index _not_ inclusive */
+void iwl_trans_pcie_reclaim(struct iwl_trans *trans, int txq_id, int ssn,
+			    struct sk_buff_head *skbs)
+{
+	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
+	struct iwl_txq *txq = &trans_pcie->txq[txq_id];
+	/* n_bd is usually 256 => n_bd - 1 = 0xff */
+	int tfd_num = ssn & (txq->q.n_bd - 1);
+	struct iwl_queue *q = &txq->q;
+	int last_to_free;
+
+	/* This function is not meant to release cmd queue*/
+	if (WARN_ON(txq_id == trans_pcie->cmd_queue))
+		return;
+
+	spin_lock(&txq->lock);
+
+	if (txq->q.read_ptr == tfd_num)
+		goto out;
+
+	IWL_DEBUG_TX_REPLY(trans, "[Q %d] %d -> %d (%d)\n",
+			   txq_id, txq->q.read_ptr, tfd_num, ssn);
+
+	/*Since we free until index _not_ inclusive, the one before index is
+	 * the last we will free. This one must be used */
+	last_to_free = iwl_queue_dec_wrap(tfd_num, q->n_bd);
+
+	if (!iwl_queue_used(q, last_to_free)) {
+		IWL_ERR(trans,
+			"%s: Read index for DMA queue txq id (%d), last_to_free %d is out of range [0-%d] %d %d.\n",
+			__func__, txq_id, last_to_free, q->n_bd,
+			q->write_ptr, q->read_ptr);
+		goto out;
+	}
+
+	if (WARN_ON(!skb_queue_empty(skbs)))
+		goto out;
+
+	for (;
+	     q->read_ptr != tfd_num;
+	     q->read_ptr = iwl_queue_inc_wrap(q->read_ptr, q->n_bd)) {
+
+		if (WARN_ON_ONCE(txq->entries[txq->q.read_ptr].skb == NULL))
+			continue;
+
+		__skb_queue_tail(skbs, txq->entries[txq->q.read_ptr].skb);
+
+		txq->entries[txq->q.read_ptr].skb = NULL;
+
+		iwl_pcie_txq_inval_byte_cnt_tbl(trans, txq);
+
+		iwl_pcie_txq_free_tfd(trans, txq, DMA_TO_DEVICE);
+	}
+
+	iwl_pcie_txq_progress(trans_pcie, txq);
+
+	if (iwl_queue_space(&txq->q) > txq->q.low_mark)
+		iwl_wake_queue(trans, txq);
+out:
+	spin_unlock(&txq->lock);
 }
 
-static int iwl_txq_set_ratid_map(struct iwl_trans *trans, u16 ra_tid,
+/*
+ * iwl_pcie_cmdq_reclaim - Reclaim TX command queue entries already Tx'd
+ *
+ * When FW advances 'R' index, all entries between old and new 'R' index
+ * need to be reclaimed. As result, some free space forms.  If there is
+ * enough free space (> low mark), wake the stack that feeds us.
+ */
+static void iwl_pcie_cmdq_reclaim(struct iwl_trans *trans, int txq_id, int idx)
+{
+	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
+	struct iwl_txq *txq = &trans_pcie->txq[txq_id];
+	struct iwl_queue *q = &txq->q;
+	int nfreed = 0;
+
+	lockdep_assert_held(&txq->lock);
+
+	if ((idx >= q->n_bd) || (!iwl_queue_used(q, idx))) {
+		IWL_ERR(trans,
+			"%s: Read index for DMA queue txq id (%d), index %d is out of range [0-%d] %d %d.\n",
+			__func__, txq_id, idx, q->n_bd,
+			q->write_ptr, q->read_ptr);
+		return;
+	}
+
+	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)) {
+
+		if (nfreed++ > 0) {
+			IWL_ERR(trans, "HCMD skipped: index (%d) %d %d\n",
+				idx, q->write_ptr, q->read_ptr);
+			iwl_op_mode_nic_error(trans->op_mode);
+		}
+	}
+
+	iwl_pcie_txq_progress(trans_pcie, txq);
+}
+
+static int iwl_pcie_txq_set_ratid_map(struct iwl_trans *trans, u16 ra_tid,
 				 u16 txq_id)
 {
 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
@@ -405,7 +1014,8 @@ static int iwl_txq_set_ratid_map(struct iwl_trans *trans, u16 ra_tid,
 	return 0;
 }
 
-static inline void iwl_txq_set_inactive(struct iwl_trans *trans, u16 txq_id)
+static inline void iwl_pcie_txq_set_inactive(struct iwl_trans *trans,
+					     u16 txq_id)
 {
 	/* Simply stop the queue, but don't change any configuration;
 	 * the SCD_ACT_EN bit is the write-enable mask for the ACTIVE bit. */
@@ -424,7 +1034,7 @@ void iwl_trans_pcie_txq_enable(struct iwl_trans *trans, int txq_id, int fifo,
 		WARN_ONCE(1, "queue %d already used - expect issues", txq_id);
 
 	/* Stop this Tx queue before configuring it */
-	iwl_txq_set_inactive(trans, txq_id);
+	iwl_pcie_txq_set_inactive(trans, txq_id);
 
 	/* Set this queue as a chain-building queue unless it is CMD queue */
 	if (txq_id != trans_pcie->cmd_queue)
@@ -435,7 +1045,7 @@ void iwl_trans_pcie_txq_enable(struct iwl_trans *trans, int txq_id, int fifo,
 		u16 ra_tid = BUILD_RAxTID(sta_id, tid);
 
 		/* Map receiver-address / traffic-ID to this queue */
-		iwl_txq_set_ratid_map(trans, ra_tid, txq_id);
+		iwl_pcie_txq_set_ratid_map(trans, ra_tid, txq_id);
 
 		/* enable aggregations for the queue */
 		iwl_set_bits_prph(trans, SCD_AGGR_SEL, BIT(txq_id));
@@ -489,20 +1099,20 @@ void iwl_trans_pcie_txq_disable(struct iwl_trans *trans, int txq_id)
 		return;
 	}
 
-	iwl_txq_set_inactive(trans, txq_id);
+	iwl_pcie_txq_set_inactive(trans, txq_id);
 
 	_iwl_write_targ_mem_dwords(trans, stts_addr,
 				   zero_val, ARRAY_SIZE(zero_val));
 
-	iwl_tx_queue_unmap(trans, txq_id);
+	iwl_pcie_txq_unmap(trans, txq_id);
 
 	IWL_DEBUG_TX_QUEUES(trans, "Deactivate queue %d\n", txq_id);
 }
 
 /*************** HOST COMMAND QUEUE FUNCTIONS   *****/
 
-/**
- * iwl_enqueue_hcmd - enqueue a uCode command
+/*
+ * iwl_pcie_enqueue_hcmd - enqueue a uCode command
  * @priv: device private data point
  * @cmd: a point to the ucode command structure
  *
@@ -510,10 +1120,11 @@ void iwl_trans_pcie_txq_disable(struct iwl_trans *trans, int txq_id)
  * failed. On success, it turns the index (> 0) of command in the
  * command queue.
  */
-static int iwl_enqueue_hcmd(struct iwl_trans *trans, struct iwl_host_cmd *cmd)
+static int iwl_pcie_enqueue_hcmd(struct iwl_trans *trans,
+				 struct iwl_host_cmd *cmd)
 {
 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
-	struct iwl_tx_queue *txq = &trans_pcie->txq[trans_pcie->cmd_queue];
+	struct iwl_txq *txq = &trans_pcie->txq[trans_pcie->cmd_queue];
 	struct iwl_queue *q = &txq->q;
 	struct iwl_device_cmd *out_cmd;
 	struct iwl_cmd_meta *out_meta;
@@ -576,8 +1187,7 @@ static int iwl_enqueue_hcmd(struct iwl_trans *trans, struct iwl_host_cmd *cmd)
 	 */
 	if (WARN(copy_size > TFD_MAX_PAYLOAD_SIZE,
 		 "Command %s (%#x) is too large (%d bytes)\n",
-		 trans_pcie_get_cmd_string(trans_pcie, cmd->id),
-		 cmd->id, copy_size)) {
+		 get_cmd_string(trans_pcie, cmd->id), cmd->id, copy_size)) {
 		idx = -EINVAL;
 		goto free_dup_buf;
 	}
@@ -640,7 +1250,7 @@ static int iwl_enqueue_hcmd(struct iwl_trans *trans, struct iwl_host_cmd *cmd)
 
 	IWL_DEBUG_HC(trans,
 		     "Sending command %s (#%x), seq: 0x%04X, %d bytes at %d[%d]:%d\n",
-		     trans_pcie_get_cmd_string(trans_pcie, out_cmd->hdr.cmd),
+		     get_cmd_string(trans_pcie, out_cmd->hdr.cmd),
 		     out_cmd->hdr.cmd, le16_to_cpu(out_cmd->hdr.sequence),
 		     cmd_size, q->write_ptr, idx, trans_pcie->cmd_queue);
 
@@ -654,7 +1264,7 @@ static int iwl_enqueue_hcmd(struct iwl_trans *trans, struct iwl_host_cmd *cmd)
 	dma_unmap_addr_set(out_meta, mapping, phys_addr);
 	dma_unmap_len_set(out_meta, len, copy_size);
 
-	iwlagn_txq_attach_buf_to_tfd(trans, txq, phys_addr, copy_size, 1);
+	iwl_pcie_txq_build_tfd(trans, txq, phys_addr, copy_size, 1);
 
 	for (i = 0; i < IWL_MAX_CMD_TFDS; i++) {
 		const void *data = cmd->data[i];
@@ -669,15 +1279,14 @@ static int iwl_enqueue_hcmd(struct iwl_trans *trans, struct iwl_host_cmd *cmd)
 		phys_addr = dma_map_single(trans->dev, (void *)data,
 					   cmd->len[i], DMA_BIDIRECTIONAL);
 		if (dma_mapping_error(trans->dev, phys_addr)) {
-			iwl_unmap_tfd(trans, out_meta,
-				      &txq->tfds[q->write_ptr],
-				      DMA_BIDIRECTIONAL);
+			iwl_pcie_tfd_unmap(trans, out_meta,
+					   &txq->tfds[q->write_ptr],
+					   DMA_BIDIRECTIONAL);
 			idx = -ENOMEM;
 			goto out;
 		}
 
-		iwlagn_txq_attach_buf_to_tfd(trans, txq, phys_addr,
-					     cmd->len[i], 0);
+		iwl_pcie_txq_build_tfd(trans, txq, phys_addr, cmd->len[i], 0);
 	}
 
 	out_meta->flags = cmd->flags;
@@ -696,7 +1305,7 @@ static int iwl_enqueue_hcmd(struct iwl_trans *trans, struct iwl_host_cmd *cmd)
 
 	/* Increment and update queue's write index */
 	q->write_ptr = iwl_queue_inc_wrap(q->write_ptr, q->n_bd);
-	iwl_txq_update_write_ptr(trans, txq);
+	iwl_pcie_txq_inc_wr_ptr(trans, txq);
 
  out:
 	spin_unlock_bh(&txq->lock);
@@ -706,63 +1315,8 @@ static int iwl_enqueue_hcmd(struct iwl_trans *trans, struct iwl_host_cmd *cmd)
 	return idx;
 }
 
-static inline void iwl_queue_progress(struct iwl_trans_pcie *trans_pcie,
-				      struct iwl_tx_queue *txq)
-{
-	if (!trans_pcie->wd_timeout)
-		return;
-
-	/*
-	 * if empty delete timer, otherwise move timer forward
-	 * since we're making progress on this queue
-	 */
-	if (txq->q.read_ptr == txq->q.write_ptr)
-		del_timer(&txq->stuck_timer);
-	else
-		mod_timer(&txq->stuck_timer, jiffies + trans_pcie->wd_timeout);
-}
-
-/**
- * iwl_hcmd_queue_reclaim - Reclaim TX command queue entries already Tx'd
- *
- * When FW advances 'R' index, all entries between old and new 'R' index
- * need to be reclaimed. As result, some free space forms.  If there is
- * enough free space (> low mark), wake the stack that feeds us.
- */
-static void iwl_hcmd_queue_reclaim(struct iwl_trans *trans, int txq_id,
-				   int idx)
-{
-	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
-	struct iwl_tx_queue *txq = &trans_pcie->txq[txq_id];
-	struct iwl_queue *q = &txq->q;
-	int nfreed = 0;
-
-	lockdep_assert_held(&txq->lock);
-
-	if ((idx >= q->n_bd) || (iwl_queue_used(q, idx) == 0)) {
-		IWL_ERR(trans,
-			"%s: Read index for DMA queue txq id (%d), index %d is out of range [0-%d] %d %d.\n",
-			__func__, txq_id, idx, q->n_bd,
-			q->write_ptr, q->read_ptr);
-		return;
-	}
-
-	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)) {
-
-		if (nfreed++ > 0) {
-			IWL_ERR(trans, "HCMD skipped: index (%d) %d %d\n",
-				idx, q->write_ptr, q->read_ptr);
-			iwl_op_mode_nic_error(trans->op_mode);
-		}
-
-	}
-
-	iwl_queue_progress(trans_pcie, txq);
-}
-
-/**
- * iwl_tx_cmd_complete - Pull unused buffers off the queue and reclaim them
+/*
+ * iwl_pcie_hcmd_complete - Pull unused buffers off the queue and reclaim them
  * @rxb: Rx buffer to reclaim
  * @handler_status: return value of the handler of the command
  *	(put in setup_rx_handlers)
@@ -771,8 +1325,8 @@ static void iwl_hcmd_queue_reclaim(struct iwl_trans *trans, int txq_id,
  * will be executed.  The attached skb (if present) will only be freed
  * if the callback returns 1
  */
-void iwl_tx_cmd_complete(struct iwl_trans *trans, struct iwl_rx_cmd_buffer *rxb,
-			 int handler_status)
+void iwl_pcie_hcmd_complete(struct iwl_trans *trans,
+			    struct iwl_rx_cmd_buffer *rxb, int handler_status)
 {
 	struct iwl_rx_packet *pkt = rxb_addr(rxb);
 	u16 sequence = le16_to_cpu(pkt->hdr.sequence);
@@ -782,7 +1336,7 @@ void iwl_tx_cmd_complete(struct iwl_trans *trans, struct iwl_rx_cmd_buffer *rxb,
 	struct iwl_device_cmd *cmd;
 	struct iwl_cmd_meta *meta;
 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
-	struct iwl_tx_queue *txq = &trans_pcie->txq[trans_pcie->cmd_queue];
+	struct iwl_txq *txq = &trans_pcie->txq[trans_pcie->cmd_queue];
 
 	/* 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
@@ -802,7 +1356,7 @@ void iwl_tx_cmd_complete(struct iwl_trans *trans, struct iwl_rx_cmd_buffer *rxb,
 	cmd = txq->entries[cmd_index].cmd;
 	meta = &txq->entries[cmd_index].meta;
 
-	iwl_unmap_tfd(trans, meta, &txq->tfds[index], DMA_BIDIRECTIONAL);
+	iwl_pcie_tfd_unmap(trans, meta, &txq->tfds[index], DMA_BIDIRECTIONAL);
 
 	/* Input error checking is done when commands are added to queue. */
 	if (meta->flags & CMD_WANT_SKB) {
@@ -814,19 +1368,17 @@ void iwl_tx_cmd_complete(struct iwl_trans *trans, struct iwl_rx_cmd_buffer *rxb,
 		meta->source->handler_status = handler_status;
 	}
 
-	iwl_hcmd_queue_reclaim(trans, txq_id, index);
+	iwl_pcie_cmdq_reclaim(trans, txq_id, index);
 
 	if (!(meta->flags & CMD_ASYNC)) {
 		if (!test_bit(STATUS_HCMD_ACTIVE, &trans_pcie->status)) {
 			IWL_WARN(trans,
 				 "HCMD_ACTIVE already clear for command %s\n",
-				 trans_pcie_get_cmd_string(trans_pcie,
-							   cmd->hdr.cmd));
+				 get_cmd_string(trans_pcie, cmd->hdr.cmd));
 		}
 		clear_bit(STATUS_HCMD_ACTIVE, &trans_pcie->status);
 		IWL_DEBUG_INFO(trans, "Clearing HCMD_ACTIVE for command %s\n",
-			       trans_pcie_get_cmd_string(trans_pcie,
-							 cmd->hdr.cmd));
+			       get_cmd_string(trans_pcie, cmd->hdr.cmd));
 		wake_up(&trans_pcie->wait_command_queue);
 	}
 
@@ -837,7 +1389,8 @@ void iwl_tx_cmd_complete(struct iwl_trans *trans, struct iwl_rx_cmd_buffer *rxb,
 
 #define HOST_COMPLETE_TIMEOUT (2 * HZ)
 
-static int iwl_send_cmd_async(struct iwl_trans *trans, struct iwl_host_cmd *cmd)
+static int iwl_pcie_send_hcmd_async(struct iwl_trans *trans,
+				    struct iwl_host_cmd *cmd)
 {
 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
 	int ret;
@@ -846,43 +1399,43 @@ static int iwl_send_cmd_async(struct iwl_trans *trans, struct iwl_host_cmd *cmd)
 	if (WARN_ON(cmd->flags & CMD_WANT_SKB))
 		return -EINVAL;
 
-
-	ret = iwl_enqueue_hcmd(trans, cmd);
+	ret = iwl_pcie_enqueue_hcmd(trans, cmd);
 	if (ret < 0) {
 		IWL_ERR(trans,
 			"Error sending %s: enqueue_hcmd failed: %d\n",
-			trans_pcie_get_cmd_string(trans_pcie, cmd->id), ret);
+			get_cmd_string(trans_pcie, cmd->id), ret);
 		return ret;
 	}
 	return 0;
 }
 
-static int iwl_send_cmd_sync(struct iwl_trans *trans, struct iwl_host_cmd *cmd)
+static int iwl_pcie_send_hcmd_sync(struct iwl_trans *trans,
+				   struct iwl_host_cmd *cmd)
 {
 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
 	int cmd_idx;
 	int ret;
 
 	IWL_DEBUG_INFO(trans, "Attempting to send sync command %s\n",
-		       trans_pcie_get_cmd_string(trans_pcie, cmd->id));
+		       get_cmd_string(trans_pcie, cmd->id));
 
 	if (WARN_ON(test_and_set_bit(STATUS_HCMD_ACTIVE,
 				     &trans_pcie->status))) {
 		IWL_ERR(trans, "Command %s: a command is already active!\n",
-			trans_pcie_get_cmd_string(trans_pcie, cmd->id));
+			get_cmd_string(trans_pcie, cmd->id));
 		return -EIO;
 	}
 
 	IWL_DEBUG_INFO(trans, "Setting HCMD_ACTIVE for command %s\n",
-		       trans_pcie_get_cmd_string(trans_pcie, cmd->id));
+		       get_cmd_string(trans_pcie, cmd->id));
 
-	cmd_idx = iwl_enqueue_hcmd(trans, cmd);
+	cmd_idx = iwl_pcie_enqueue_hcmd(trans, cmd);
 	if (cmd_idx < 0) {
 		ret = cmd_idx;
 		clear_bit(STATUS_HCMD_ACTIVE, &trans_pcie->status);
 		IWL_ERR(trans,
 			"Error sending %s: enqueue_hcmd failed: %d\n",
-			trans_pcie_get_cmd_string(trans_pcie, cmd->id), ret);
+			get_cmd_string(trans_pcie, cmd->id), ret);
 		return ret;
 	}
 
@@ -892,13 +1445,13 @@ static int iwl_send_cmd_sync(struct iwl_trans *trans, struct iwl_host_cmd *cmd)
 				 HOST_COMPLETE_TIMEOUT);
 	if (!ret) {
 		if (test_bit(STATUS_HCMD_ACTIVE, &trans_pcie->status)) {
-			struct iwl_tx_queue *txq =
+			struct iwl_txq *txq =
 				&trans_pcie->txq[trans_pcie->cmd_queue];
 			struct iwl_queue *q = &txq->q;
 
 			IWL_ERR(trans,
 				"Error sending %s: time out after %dms.\n",
-				trans_pcie_get_cmd_string(trans_pcie, cmd->id),
+				get_cmd_string(trans_pcie, cmd->id),
 				jiffies_to_msecs(HOST_COMPLETE_TIMEOUT));
 
 			IWL_ERR(trans,
@@ -908,8 +1461,7 @@ static int iwl_send_cmd_sync(struct iwl_trans *trans, struct iwl_host_cmd *cmd)
 			clear_bit(STATUS_HCMD_ACTIVE, &trans_pcie->status);
 			IWL_DEBUG_INFO(trans,
 				       "Clearing HCMD_ACTIVE for command %s\n",
-				       trans_pcie_get_cmd_string(trans_pcie,
-								 cmd->id));
+				       get_cmd_string(trans_pcie, cmd->id));
 			ret = -ETIMEDOUT;
 			goto cancel;
 		}
@@ -917,7 +1469,7 @@ static int iwl_send_cmd_sync(struct iwl_trans *trans, struct iwl_host_cmd *cmd)
 
 	if (test_bit(STATUS_FW_ERROR, &trans_pcie->status)) {
 		IWL_ERR(trans, "FW error in SYNC CMD %s\n",
-			trans_pcie_get_cmd_string(trans_pcie, cmd->id));
+			get_cmd_string(trans_pcie, cmd->id));
 		ret = -EIO;
 		goto cancel;
 	}
@@ -930,7 +1482,7 @@ static int iwl_send_cmd_sync(struct iwl_trans *trans, struct iwl_host_cmd *cmd)
 
 	if ((cmd->flags & CMD_WANT_SKB) && !cmd->resp_pkt) {
 		IWL_ERR(trans, "Error: Response NULL in '%s'\n",
-			trans_pcie_get_cmd_string(trans_pcie, cmd->id));
+			get_cmd_string(trans_pcie, cmd->id));
 		ret = -EIO;
 		goto cancel;
 	}
@@ -957,7 +1509,7 @@ cancel:
 	return ret;
 }
 
-int iwl_trans_pcie_send_cmd(struct iwl_trans *trans, struct iwl_host_cmd *cmd)
+int iwl_trans_pcie_send_hcmd(struct iwl_trans *trans, struct iwl_host_cmd *cmd)
 {
 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
 
@@ -968,62 +1520,172 @@ int iwl_trans_pcie_send_cmd(struct iwl_trans *trans, struct iwl_host_cmd *cmd)
 		return -ERFKILL;
 
 	if (cmd->flags & CMD_ASYNC)
-		return iwl_send_cmd_async(trans, cmd);
+		return iwl_pcie_send_hcmd_async(trans, cmd);
 
 	/* We still can fail on RFKILL that can be asserted while we wait */
-	return iwl_send_cmd_sync(trans, cmd);
+	return iwl_pcie_send_hcmd_sync(trans, cmd);
 }
 
-/* Frees buffers until index _not_ inclusive */
-int iwl_tx_queue_reclaim(struct iwl_trans *trans, int txq_id, int index,
-			 struct sk_buff_head *skbs)
+int iwl_trans_pcie_tx(struct iwl_trans *trans, struct sk_buff *skb,
+		      struct iwl_device_cmd *dev_cmd, int txq_id)
 {
 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
-	struct iwl_tx_queue *txq = &trans_pcie->txq[txq_id];
-	struct iwl_queue *q = &txq->q;
-	int last_to_free;
-	int freed = 0;
+	struct ieee80211_hdr *hdr = (struct ieee80211_hdr *)skb->data;
+	struct iwl_tx_cmd *tx_cmd = (struct iwl_tx_cmd *)dev_cmd->payload;
+	struct iwl_cmd_meta *out_meta;
+	struct iwl_txq *txq;
+	struct iwl_queue *q;
+	dma_addr_t phys_addr = 0;
+	dma_addr_t txcmd_phys;
+	dma_addr_t scratch_phys;
+	u16 len, firstlen, secondlen;
+	u8 wait_write_ptr = 0;
+	__le16 fc = hdr->frame_control;
+	u8 hdr_len = ieee80211_hdrlen(fc);
+	u16 __maybe_unused wifi_seq;
+
+	txq = &trans_pcie->txq[txq_id];
+	q = &txq->q;
 
-	/* This function is not meant to release cmd queue*/
-	if (WARN_ON(txq_id == trans_pcie->cmd_queue))
-		return 0;
+	if (unlikely(!test_bit(txq_id, trans_pcie->queue_used))) {
+		WARN_ON_ONCE(1);
+		return -EINVAL;
+	}
 
-	lockdep_assert_held(&txq->lock);
+	spin_lock(&txq->lock);
 
-	/*Since we free until index _not_ inclusive, the one before index is
-	 * the last we will free. This one must be used */
-	last_to_free = iwl_queue_dec_wrap(index, q->n_bd);
+	/* In AGG mode, the index in the ring must correspond to the WiFi
+	 * sequence number. This is a HW requirements to help the SCD to parse
+	 * the BA.
+	 * Check here that the packets are in the right place on the ring.
+	 */
+#ifdef CONFIG_IWLWIFI_DEBUG
+	wifi_seq = SEQ_TO_SN(le16_to_cpu(hdr->seq_ctrl));
+	WARN_ONCE((iwl_read_prph(trans, SCD_AGGR_SEL) & BIT(txq_id)) &&
+		  ((wifi_seq & 0xff) != q->write_ptr),
+		  "Q: %d WiFi Seq %d tfdNum %d",
+		  txq_id, wifi_seq, q->write_ptr);
+#endif
+
+	/* Set up driver data for this TFD */
+	txq->entries[q->write_ptr].skb = skb;
+	txq->entries[q->write_ptr].cmd = dev_cmd;
+
+	dev_cmd->hdr.cmd = REPLY_TX;
+	dev_cmd->hdr.sequence =
+		cpu_to_le16((u16)(QUEUE_TO_SEQ(txq_id) |
+			    INDEX_TO_SEQ(q->write_ptr)));
+
+	/* Set up first empty entry in queue's array of Tx/cmd buffers */
+	out_meta = &txq->entries[q->write_ptr].meta;
 
-	if ((index >= q->n_bd) ||
-	   (iwl_queue_used(q, last_to_free) == 0)) {
-		IWL_ERR(trans,
-			"%s: Read index for DMA queue txq id (%d), last_to_free %d is out of range [0-%d] %d %d.\n",
-			__func__, txq_id, last_to_free, q->n_bd,
-			q->write_ptr, q->read_ptr);
-		return 0;
+	/*
+	 * Use the first empty entry in this queue's command buffer array
+	 * to contain the Tx command and MAC header concatenated together
+	 * (payload data will be in another buffer).
+	 * Size of this varies, due to varying MAC header length.
+	 * If end is not dword aligned, we'll have 2 extra bytes at the end
+	 * of the MAC header (device reads on dword boundaries).
+	 * We'll tell device about this padding later.
+	 */
+	len = sizeof(struct iwl_tx_cmd) +
+		sizeof(struct iwl_cmd_header) + hdr_len;
+	firstlen = (len + 3) & ~3;
+
+	/* Tell NIC about any 2-byte padding after MAC header */
+	if (firstlen != len)
+		tx_cmd->tx_flags |= TX_CMD_FLG_MH_PAD_MSK;
+
+	/* Physical address of this Tx command's header (not MAC header!),
+	 * within command buffer array. */
+	txcmd_phys = dma_map_single(trans->dev,
+				    &dev_cmd->hdr, firstlen,
+				    DMA_BIDIRECTIONAL);
+	if (unlikely(dma_mapping_error(trans->dev, txcmd_phys)))
+		goto out_err;
+	dma_unmap_addr_set(out_meta, mapping, txcmd_phys);
+	dma_unmap_len_set(out_meta, len, firstlen);
+
+	if (!ieee80211_has_morefrags(fc)) {
+		txq->need_update = 1;
+	} else {
+		wait_write_ptr = 1;
+		txq->need_update = 0;
 	}
 
-	if (WARN_ON(!skb_queue_empty(skbs)))
-		return 0;
+	/* Set up TFD's 2nd entry to point directly to remainder of skb,
+	 * if any (802.11 null frames have no payload). */
+	secondlen = skb->len - hdr_len;
+	if (secondlen > 0) {
+		phys_addr = dma_map_single(trans->dev, skb->data + hdr_len,
+					   secondlen, DMA_TO_DEVICE);
+		if (unlikely(dma_mapping_error(trans->dev, phys_addr))) {
+			dma_unmap_single(trans->dev,
+					 dma_unmap_addr(out_meta, mapping),
+					 dma_unmap_len(out_meta, len),
+					 DMA_BIDIRECTIONAL);
+			goto out_err;
+		}
+	}
 
-	for (;
-	     q->read_ptr != index;
-	     q->read_ptr = iwl_queue_inc_wrap(q->read_ptr, q->n_bd)) {
+	/* Attach buffers to TFD */
+	iwl_pcie_txq_build_tfd(trans, txq, txcmd_phys, firstlen, 1);
+	if (secondlen > 0)
+		iwl_pcie_txq_build_tfd(trans, txq, phys_addr, secondlen, 0);
 
-		if (WARN_ON_ONCE(txq->entries[txq->q.read_ptr].skb == NULL))
-			continue;
+	scratch_phys = txcmd_phys + sizeof(struct iwl_cmd_header) +
+				offsetof(struct iwl_tx_cmd, scratch);
 
-		__skb_queue_tail(skbs, txq->entries[txq->q.read_ptr].skb);
+	/* take back ownership of DMA buffer to enable update */
+	dma_sync_single_for_cpu(trans->dev, txcmd_phys, firstlen,
+				DMA_BIDIRECTIONAL);
+	tx_cmd->dram_lsb_ptr = cpu_to_le32(scratch_phys);
+	tx_cmd->dram_msb_ptr = iwl_get_dma_hi_addr(scratch_phys);
 
-		txq->entries[txq->q.read_ptr].skb = NULL;
+	IWL_DEBUG_TX(trans, "sequence nr = 0X%x\n",
+		     le16_to_cpu(dev_cmd->hdr.sequence));
+	IWL_DEBUG_TX(trans, "tx_flags = 0X%x\n", le32_to_cpu(tx_cmd->tx_flags));
 
-		iwlagn_txq_inval_byte_cnt_tbl(trans, txq);
+	/* Set up entry for this TFD in Tx byte-count array */
+	iwl_pcie_txq_update_byte_cnt_tbl(trans, txq, le16_to_cpu(tx_cmd->len));
 
-		iwl_txq_free_tfd(trans, txq, DMA_TO_DEVICE);
-		freed++;
-	}
+	dma_sync_single_for_device(trans->dev, txcmd_phys, firstlen,
+				   DMA_BIDIRECTIONAL);
+
+	trace_iwlwifi_dev_tx(trans->dev, skb,
+			     &txq->tfds[txq->q.write_ptr],
+			     sizeof(struct iwl_tfd),
+			     &dev_cmd->hdr, firstlen,
+			     skb->data + hdr_len, secondlen);
+	trace_iwlwifi_dev_tx_data(trans->dev, skb,
+				  skb->data + hdr_len, secondlen);
+
+	/* start timer if queue currently empty */
+	if (txq->need_update && q->read_ptr == q->write_ptr &&
+	    trans_pcie->wd_timeout)
+		mod_timer(&txq->stuck_timer, jiffies + trans_pcie->wd_timeout);
 
-	iwl_queue_progress(trans_pcie, txq);
+	/* Tell device the write index *just past* this latest filled TFD */
+	q->write_ptr = iwl_queue_inc_wrap(q->write_ptr, q->n_bd);
+	iwl_pcie_txq_inc_wr_ptr(trans, txq);
 
-	return freed;
+	/*
+	 * 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.
+	 */
+	if (iwl_queue_space(q) < q->high_mark) {
+		if (wait_write_ptr) {
+			txq->need_update = 1;
+			iwl_pcie_txq_inc_wr_ptr(trans, txq);
+		} else {
+			iwl_stop_queue(trans, txq);
+		}
+	}
+	spin_unlock(&txq->lock);
+	return 0;
+out_err:
+	spin_unlock(&txq->lock);
+	return -1;
 }