Merge remote-tracking branch 'net-next/master' into mac80211-next

Merge back net-next to get wireless driver changes (from Kalle)
to be able to create the API change across all trees properly.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>

5 files changed, 492 insertions, 184 deletions

diff --git a/drivers/net/wireless/iwlwifi/pcie/drv.c b/drivers/net/wireless/iwlwifi/pcie/drv.c
index b18569734922..2ed1e4d2774d 100644
--- a/drivers/net/wireless/iwlwifi/pcie/drv.c
+++ b/drivers/net/wireless/iwlwifi/pcie/drv.c
@@ -6,7 +6,7 @@
  * GPL LICENSE SUMMARY
  *
  * Copyright(c) 2007 - 2014 Intel Corporation. All rights reserved.
- * Copyright(c) 2013 - 2014 Intel Mobile Communications GmbH
+ * Copyright(c) 2013 - 2015 Intel Mobile Communications GmbH
  *
  * This program is free software; you can redistribute it and/or modify
  * it under the terms of version 2 of the GNU General Public License as
@@ -32,7 +32,7 @@
  * BSD LICENSE
  *
  * Copyright(c) 2005 - 2014 Intel Corporation. All rights reserved.
- * Copyright(c) 2013 - 2014 Intel Mobile Communications GmbH
+ * Copyright(c) 2013 - 2015 Intel Mobile Communications GmbH
  * All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without
@@ -629,7 +629,18 @@ static int iwl_pci_resume(struct device *device)
 	if (!trans->op_mode)
 		return 0;
 
-	iwl_enable_rfkill_int(trans);
+	/*
+	 * On suspend, ict is disabled, and the interrupt mask
+	 * gets cleared. Reconfigure them both in case of d0i3
+	 * image. Otherwise, only enable rfkill interrupt (in
+	 * order to keep track of the rfkill status)
+	 */
+	if (trans->wowlan_d0i3) {
+		iwl_pcie_reset_ict(trans);
+		iwl_enable_interrupts(trans);
+	} else {
+		iwl_enable_rfkill_int(trans);
+	}
 
 	hw_rfkill = iwl_is_rfkill_set(trans);
 	iwl_trans_pcie_rf_kill(trans, hw_rfkill);
diff --git a/drivers/net/wireless/iwlwifi/pcie/internal.h b/drivers/net/wireless/iwlwifi/pcie/internal.h
index 01996c9d98a7..31f72a61cc3f 100644
--- a/drivers/net/wireless/iwlwifi/pcie/internal.h
+++ b/drivers/net/wireless/iwlwifi/pcie/internal.h
@@ -1,7 +1,7 @@
 /******************************************************************************
  *
- * Copyright(c) 2003 - 2014 Intel Corporation. All rights reserved.
- * Copyright(c) 2013 - 2014 Intel Mobile Communications GmbH
+ * Copyright(c) 2003 - 2015 Intel Corporation. All rights reserved.
+ * Copyright(c) 2013 - 2015 Intel Mobile Communications GmbH
  *
  * Portions of this file are derived from the ipw3945 project, as well
  * as portions of the ieee80211 subsystem header files.
@@ -44,6 +44,15 @@
 #include "iwl-io.h"
 #include "iwl-op-mode.h"
 
+/*
+ * RX related structures and functions
+ */
+#define RX_NUM_QUEUES 1
+#define RX_POST_REQ_ALLOC 2
+#define RX_CLAIM_REQ_ALLOC 8
+#define RX_POOL_SIZE ((RX_CLAIM_REQ_ALLOC - RX_POST_REQ_ALLOC) * RX_NUM_QUEUES)
+#define RX_LOW_WATERMARK 8
+
 struct iwl_host_cmd;
 
 /*This file includes the declaration that are internal to the
@@ -77,29 +86,29 @@ struct isr_statistics {
  * 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:
- * @queue:
  * @read: Shared index to newest available Rx buffer
  * @write: Shared index to oldest written Rx packet
  * @free_count: Number of pre-allocated buffers in rx_free
+ * @used_count: Number of RBDs handled to allocator to use for allocation
  * @write_actual:
- * @rx_free: list of free SKBs for use
- * @rx_used: List of Rx buffers with no SKB
+ * @rx_free: list of RBDs with allocated RB ready for use
+ * @rx_used: list of RBDs with no RB attached
  * @need_update: flag to indicate we need to update read/write index
  * @rb_stts: driver's pointer to receive buffer status
  * @rb_stts_dma: bus address of receive buffer status
  * @lock:
+ * @pool: initial pool of iwl_rx_mem_buffer for the queue
+ * @queue: actual rx queue
  *
  * NOTE:  rx_free and rx_used are used as a FIFO for iwl_rx_mem_buffers
  */
 struct iwl_rxq {
 	__le32 *bd;
 	dma_addr_t bd_dma;
-	struct iwl_rx_mem_buffer pool[RX_QUEUE_SIZE + RX_FREE_BUFFERS];
-	struct iwl_rx_mem_buffer *queue[RX_QUEUE_SIZE];
 	u32 read;
 	u32 write;
 	u32 free_count;
+	u32 used_count;
 	u32 write_actual;
 	struct list_head rx_free;
 	struct list_head rx_used;
@@ -107,6 +116,32 @@ struct iwl_rxq {
 	struct iwl_rb_status *rb_stts;
 	dma_addr_t rb_stts_dma;
 	spinlock_t lock;
+	struct iwl_rx_mem_buffer pool[RX_QUEUE_SIZE];
+	struct iwl_rx_mem_buffer *queue[RX_QUEUE_SIZE];
+};
+
+/**
+ * struct iwl_rb_allocator - Rx allocator
+ * @pool: initial pool of allocator
+ * @req_pending: number of requests the allcator had not processed yet
+ * @req_ready: number of requests honored and ready for claiming
+ * @rbd_allocated: RBDs with pages allocated and ready to be handled to
+ *	the queue. This is a list of &struct iwl_rx_mem_buffer
+ * @rbd_empty: RBDs with no page attached for allocator use. This is a list
+ *	of &struct iwl_rx_mem_buffer
+ * @lock: protects the rbd_allocated and rbd_empty lists
+ * @alloc_wq: work queue for background calls
+ * @rx_alloc: work struct for background calls
+ */
+struct iwl_rb_allocator {
+	struct iwl_rx_mem_buffer pool[RX_POOL_SIZE];
+	atomic_t req_pending;
+	atomic_t req_ready;
+	struct list_head rbd_allocated;
+	struct list_head rbd_empty;
+	spinlock_t lock;
+	struct workqueue_struct *alloc_wq;
+	struct work_struct rx_alloc;
 };
 
 struct iwl_dma_ptr {
@@ -250,7 +285,7 @@ iwl_pcie_get_scratchbuf_dma(struct iwl_txq *txq, int idx)
 /**
  * struct iwl_trans_pcie - PCIe transport specific data
  * @rxq: all the RX queue data
- * @rx_replenish: work that will be called when buffers need to be allocated
+ * @rba: allocator for RX replenishing
  * @drv - pointer to iwl_drv
  * @trans: pointer to the generic transport area
  * @scd_base_addr: scheduler sram base address in SRAM
@@ -273,7 +308,7 @@ iwl_pcie_get_scratchbuf_dma(struct iwl_txq *txq, int idx)
  */
 struct iwl_trans_pcie {
 	struct iwl_rxq rxq;
-	struct work_struct rx_replenish;
+	struct iwl_rb_allocator rba;
 	struct iwl_trans *trans;
 	struct iwl_drv *drv;
 
@@ -320,7 +355,7 @@ struct iwl_trans_pcie {
 
 	/*protect hw register */
 	spinlock_t reg_lock;
-	bool cmd_in_flight;
+	bool cmd_hold_nic_awake;
 	bool ref_cmd_in_flight;
 
 	/* protect ref counter */
diff --git a/drivers/net/wireless/iwlwifi/pcie/rx.c b/drivers/net/wireless/iwlwifi/pcie/rx.c
index 7ff69c642103..a3fbaa0ef5e0 100644
--- a/drivers/net/wireless/iwlwifi/pcie/rx.c
+++ b/drivers/net/wireless/iwlwifi/pcie/rx.c
@@ -1,7 +1,7 @@
 /******************************************************************************
  *
  * Copyright(c) 2003 - 2014 Intel Corporation. All rights reserved.
- * Copyright(c) 2013 - 2014 Intel Mobile Communications GmbH
+ * Copyright(c) 2013 - 2015 Intel Mobile Communications GmbH
  *
  * Portions of this file are derived from the ipw3945 project, as well
  * as portions of the ieee80211 subsystem header files.
@@ -74,16 +74,29 @@
  * resets the Rx queue buffers with new memory.
  *
  * The management in the driver is as follows:
- * + 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_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 list of pre-allocated RBDs is stored in iwl->rxq->rx_free.
+ *   When the interrupt handler is called, the request is processed.
+ *   The page is either stolen - transferred to the upper layer
+ *   or reused - added immediately to the iwl->rxq->rx_free list.
+ * + When the page is stolen - the driver updates the matching queue's used
+ *   count, detaches the RBD and transfers it to the queue used list.
+ *   When there are two used RBDs - they are transferred to the allocator empty
+ *   list. Work is then scheduled for the allocator to start allocating
+ *   eight buffers.
+ *   When there are another 6 used RBDs - they are transferred to the allocator
+ *   empty list and the driver tries to claim the pre-allocated buffers and
+ *   add them to iwl->rxq->rx_free. If it fails - it continues to claim them
+ *   until ready.
+ *   When there are 8+ buffers in the free list - either from allocation or from
+ *   8 reused unstolen pages - restock is called to update the FW and indexes.
+ * + In order to make sure the allocator always has RBDs to use for allocation
+ *   the allocator has initial pool in the size of num_queues*(8-2) - the
+ *   maximum missing RBDs per allocation request (request posted with 2
+ *    empty RBDs, there is no guarantee when the other 6 RBDs are supplied).
+ *   The queues supplies the recycle of the rest of the RBDs.
  * + A received packet is processed and handed to the kernel network stack,
  *   detached from the iwl->rxq.  The driver 'processed' index is updated.
- * + The Host/Firmware iwl->rxq is replenished at irq thread time from the
- *   rx_free list. If there are no allocated buffers in iwl->rxq->rx_free,
+ * + If there are no allocated buffers in iwl->rxq->rx_free,
  *   the READ INDEX is not incremented and iwl->status(RX_STALLED) is set.
  *   If there were enough free buffers and RX_STALLED is set it is cleared.
  *
@@ -92,18 +105,32 @@
  *
  * 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.
+ *                            Used only during initialization.
  * 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_pcie_rx_replenish
+ *                            the WRITE index.
+ * iwl_pcie_rx_allocator()     Background work for allocating pages.
  *
  * -- enable interrupts --
  * 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.
+ *                            Posts and claims requests to the allocator.
  *                            Calls iwl_pcie_rxq_restock to refill any empty
  *                            slots.
+ *
+ * RBD life-cycle:
+ *
+ * Init:
+ * rxq.pool -> rxq.rx_used -> rxq.rx_free -> rxq.queue
+ *
+ * Regular Receive interrupt:
+ * Page Stolen:
+ * rxq.queue -> rxq.rx_used -> allocator.rbd_empty ->
+ * allocator.rbd_allocated -> rxq.rx_free -> rxq.queue
+ * Page not Stolen:
+ * rxq.queue -> rxq.rx_free -> rxq.queue
  * ...
  *
  */
@@ -240,10 +267,6 @@ static void iwl_pcie_rxq_restock(struct iwl_trans *trans)
 		rxq->free_count--;
 	}
 	spin_unlock(&rxq->lock);
-	/* If the pre-allocated buffer pool is dropping low, schedule to
-	 * refill it */
-	if (rxq->free_count <= RX_LOW_WATERMARK)
-		schedule_work(&trans_pcie->rx_replenish);
 
 	/* If we've added more space for the firmware to place data, tell it.
 	 * Increment device's write pointer in multiples of 8. */
@@ -255,6 +278,44 @@ static void iwl_pcie_rxq_restock(struct iwl_trans *trans)
 }
 
 /*
+ * iwl_pcie_rx_alloc_page - allocates and returns a page.
+ *
+ */
+static struct page *iwl_pcie_rx_alloc_page(struct iwl_trans *trans)
+{
+	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
+	struct iwl_rxq *rxq = &trans_pcie->rxq;
+	struct page *page;
+	gfp_t gfp_mask = GFP_KERNEL;
+
+	if (rxq->free_count > RX_LOW_WATERMARK)
+		gfp_mask |= __GFP_NOWARN;
+
+	if (trans_pcie->rx_page_order > 0)
+		gfp_mask |= __GFP_COMP;
+
+	/* Alloc a new receive buffer */
+	page = alloc_pages(gfp_mask, trans_pcie->rx_page_order);
+	if (!page) {
+		if (net_ratelimit())
+			IWL_DEBUG_INFO(trans, "alloc_pages failed, order: %d\n",
+				       trans_pcie->rx_page_order);
+		/* Issue an error if the hardware has consumed more than half
+		 * of its free buffer list and we don't have enough
+		 * pre-allocated buffers.
+`		 */
+		if (rxq->free_count <= RX_LOW_WATERMARK &&
+		    iwl_rxq_space(rxq) > (RX_QUEUE_SIZE / 2) &&
+		    net_ratelimit())
+			IWL_CRIT(trans,
+				 "Failed to alloc_pages with GFP_KERNEL. Only %u free buffers remaining.\n",
+				 rxq->free_count);
+		return NULL;
+	}
+	return page;
+}
+
+/*
  * 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
@@ -263,13 +324,12 @@ static void iwl_pcie_rxq_restock(struct iwl_trans *trans)
  * iwl_pcie_rxq_restock. The latter function will update the HW to use the newly
  * allocated buffers.
  */
-static void iwl_pcie_rxq_alloc_rbs(struct iwl_trans *trans, gfp_t priority)
+static void iwl_pcie_rxq_alloc_rbs(struct iwl_trans *trans)
 {
 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
 	struct iwl_rxq *rxq = &trans_pcie->rxq;
 	struct iwl_rx_mem_buffer *rxb;
 	struct page *page;
-	gfp_t gfp_mask = priority;
 
 	while (1) {
 		spin_lock(&rxq->lock);
@@ -279,32 +339,10 @@ static void iwl_pcie_rxq_alloc_rbs(struct iwl_trans *trans, gfp_t priority)
 		}
 		spin_unlock(&rxq->lock);
 
-		if (rxq->free_count > RX_LOW_WATERMARK)
-			gfp_mask |= __GFP_NOWARN;
-
-		if (trans_pcie->rx_page_order > 0)
-			gfp_mask |= __GFP_COMP;
-
 		/* Alloc a new receive buffer */
-		page = alloc_pages(gfp_mask, trans_pcie->rx_page_order);
-		if (!page) {
-			if (net_ratelimit())
-				IWL_DEBUG_INFO(trans, "alloc_pages failed, "
-					   "order: %d\n",
-					   trans_pcie->rx_page_order);
-
-			if ((rxq->free_count <= RX_LOW_WATERMARK) &&
-			    net_ratelimit())
-				IWL_CRIT(trans, "Failed to alloc_pages with %s."
-					 "Only %u free buffers remaining.\n",
-					 priority == GFP_ATOMIC ?
-					 "GFP_ATOMIC" : "GFP_KERNEL",
-					 rxq->free_count);
-			/* We don't reschedule replenish work here -- we will
-			 * call the restock method and if it still needs
-			 * more buffers it will schedule replenish */
+		page = iwl_pcie_rx_alloc_page(trans);
+		if (!page)
 			return;
-		}
 
 		spin_lock(&rxq->lock);
 
@@ -355,7 +393,7 @@ static void iwl_pcie_rxq_free_rbs(struct iwl_trans *trans)
 
 	lockdep_assert_held(&rxq->lock);
 
-	for (i = 0; i < RX_FREE_BUFFERS + RX_QUEUE_SIZE; i++) {
+	for (i = 0; i < RX_QUEUE_SIZE; i++) {
 		if (!rxq->pool[i].page)
 			continue;
 		dma_unmap_page(trans->dev, rxq->pool[i].page_dma,
@@ -372,32 +410,144 @@ static void iwl_pcie_rxq_free_rbs(struct iwl_trans *trans)
  * When moving to rx_free an page is allocated for the slot.
  *
  * Also restock the Rx queue via iwl_pcie_rxq_restock.
- * This is called as a scheduled work item (except for during initialization)
+ * This is called only during initialization
  */
-static void iwl_pcie_rx_replenish(struct iwl_trans *trans, gfp_t gfp)
+static void iwl_pcie_rx_replenish(struct iwl_trans *trans)
 {
-	iwl_pcie_rxq_alloc_rbs(trans, gfp);
+	iwl_pcie_rxq_alloc_rbs(trans);
 
 	iwl_pcie_rxq_restock(trans);
 }
 
-static void iwl_pcie_rx_replenish_work(struct work_struct *data)
+/*
+ * iwl_pcie_rx_allocator - Allocates pages in the background for RX queues
+ *
+ * Allocates for each received request 8 pages
+ * Called as a scheduled work item.
+ */
+static void iwl_pcie_rx_allocator(struct iwl_trans *trans)
+{
+	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
+	struct iwl_rb_allocator *rba = &trans_pcie->rba;
+
+	while (atomic_read(&rba->req_pending)) {
+		int i;
+		struct list_head local_empty;
+		struct list_head local_allocated;
+
+		INIT_LIST_HEAD(&local_allocated);
+		spin_lock(&rba->lock);
+		/* swap out the entire rba->rbd_empty to a local list */
+		list_replace_init(&rba->rbd_empty, &local_empty);
+		spin_unlock(&rba->lock);
+
+		for (i = 0; i < RX_CLAIM_REQ_ALLOC;) {
+			struct iwl_rx_mem_buffer *rxb;
+			struct page *page;
+
+			/* List should never be empty - each reused RBD is
+			 * returned to the list, and initial pool covers any
+			 * possible gap between the time the page is allocated
+			 * to the time the RBD is added.
+			 */
+			BUG_ON(list_empty(&local_empty));
+			/* Get the first rxb from the rbd list */
+			rxb = list_first_entry(&local_empty,
+					       struct iwl_rx_mem_buffer, list);
+			BUG_ON(rxb->page);
+
+			/* Alloc a new receive buffer */
+			page = iwl_pcie_rx_alloc_page(trans);
+			if (!page)
+				continue;
+			rxb->page = page;
+
+			/* Get physical address of the RB */
+			rxb->page_dma = dma_map_page(trans->dev, page, 0,
+					PAGE_SIZE << trans_pcie->rx_page_order,
+					DMA_FROM_DEVICE);
+			if (dma_mapping_error(trans->dev, rxb->page_dma)) {
+				rxb->page = NULL;
+				__free_pages(page, trans_pcie->rx_page_order);
+				continue;
+			}
+			/* dma address must be no more than 36 bits */
+			BUG_ON(rxb->page_dma & ~DMA_BIT_MASK(36));
+			/* and also 256 byte aligned! */
+			BUG_ON(rxb->page_dma & DMA_BIT_MASK(8));
+
+			/* move the allocated entry to the out list */
+			list_move(&rxb->list, &local_allocated);
+			i++;
+		}
+
+		spin_lock(&rba->lock);
+		/* add the allocated rbds to the allocator allocated list */
+		list_splice_tail(&local_allocated, &rba->rbd_allocated);
+		/* add the unused rbds back to the allocator empty list */
+		list_splice_tail(&local_empty, &rba->rbd_empty);
+		spin_unlock(&rba->lock);
+
+		atomic_dec(&rba->req_pending);
+		atomic_inc(&rba->req_ready);
+	}
+}
+
+/*
+ * iwl_pcie_rx_allocator_get - Returns the pre-allocated pages
+.*
+.* Called by queue when the queue posted allocation request and
+ * has freed 8 RBDs in order to restock itself.
+ */
+static int iwl_pcie_rx_allocator_get(struct iwl_trans *trans,
+				     struct iwl_rx_mem_buffer
+				     *out[RX_CLAIM_REQ_ALLOC])
+{
+	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
+	struct iwl_rb_allocator *rba = &trans_pcie->rba;
+	int i;
+
+	if (atomic_dec_return(&rba->req_ready) < 0) {
+		atomic_inc(&rba->req_ready);
+		IWL_DEBUG_RX(trans,
+			     "Allocation request not ready, pending requests = %d\n",
+			     atomic_read(&rba->req_pending));
+		return -ENOMEM;
+	}
+
+	spin_lock(&rba->lock);
+	for (i = 0; i < RX_CLAIM_REQ_ALLOC; i++) {
+		/* Get next free Rx buffer, remove it from free list */
+		out[i] = list_first_entry(&rba->rbd_allocated,
+			       struct iwl_rx_mem_buffer, list);
+		list_del(&out[i]->list);
+	}
+	spin_unlock(&rba->lock);
+
+	return 0;
+}
+
+static void iwl_pcie_rx_allocator_work(struct work_struct *data)
 {
+	struct iwl_rb_allocator *rba_p =
+		container_of(data, struct iwl_rb_allocator, rx_alloc);
 	struct iwl_trans_pcie *trans_pcie =
-	    container_of(data, struct iwl_trans_pcie, rx_replenish);
+		container_of(rba_p, struct iwl_trans_pcie, rba);
 
-	iwl_pcie_rx_replenish(trans_pcie->trans, GFP_KERNEL);
+	iwl_pcie_rx_allocator(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 iwl_rb_allocator *rba = &trans_pcie->rba;
 	struct device *dev = trans->dev;
 
 	memset(&trans_pcie->rxq, 0, sizeof(trans_pcie->rxq));
 
 	spin_lock_init(&rxq->lock);
+	spin_lock_init(&rba->lock);
 
 	if (WARN_ON(rxq->bd || rxq->rb_stts))
 		return -EINVAL;
@@ -487,15 +637,49 @@ static void iwl_pcie_rx_init_rxb_lists(struct iwl_rxq *rxq)
 	INIT_LIST_HEAD(&rxq->rx_free);
 	INIT_LIST_HEAD(&rxq->rx_used);
 	rxq->free_count = 0;
+	rxq->used_count = 0;
 
-	for (i = 0; i < RX_FREE_BUFFERS + RX_QUEUE_SIZE; i++)
+	for (i = 0; i < RX_QUEUE_SIZE; i++)
 		list_add(&rxq->pool[i].list, &rxq->rx_used);
 }
 
+static void iwl_pcie_rx_init_rba(struct iwl_rb_allocator *rba)
+{
+	int i;
+
+	lockdep_assert_held(&rba->lock);
+
+	INIT_LIST_HEAD(&rba->rbd_allocated);
+	INIT_LIST_HEAD(&rba->rbd_empty);
+
+	for (i = 0; i < RX_POOL_SIZE; i++)
+		list_add(&rba->pool[i].list, &rba->rbd_empty);
+}
+
+static void iwl_pcie_rx_free_rba(struct iwl_trans *trans)
+{
+	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
+	struct iwl_rb_allocator *rba = &trans_pcie->rba;
+	int i;
+
+	lockdep_assert_held(&rba->lock);
+
+	for (i = 0; i < RX_POOL_SIZE; i++) {
+		if (!rba->pool[i].page)
+			continue;
+		dma_unmap_page(trans->dev, rba->pool[i].page_dma,
+			       PAGE_SIZE << trans_pcie->rx_page_order,
+			       DMA_FROM_DEVICE);
+		__free_pages(rba->pool[i].page, trans_pcie->rx_page_order);
+		rba->pool[i].page = NULL;
+	}
+}
+
 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;
+	struct iwl_rb_allocator *rba = &trans_pcie->rba;
 	int i, err;
 
 	if (!rxq->bd) {
@@ -503,11 +687,21 @@ int iwl_pcie_rx_init(struct iwl_trans *trans)
 		if (err)
 			return err;
 	}
+	if (!rba->alloc_wq)
+		rba->alloc_wq = alloc_workqueue("rb_allocator",
+						WQ_HIGHPRI | WQ_UNBOUND, 1);
+	INIT_WORK(&rba->rx_alloc, iwl_pcie_rx_allocator_work);
+
+	spin_lock(&rba->lock);
+	atomic_set(&rba->req_pending, 0);
+	atomic_set(&rba->req_ready, 0);
+	/* free all first - we might be reconfigured for a different size */
+	iwl_pcie_rx_free_rba(trans);
+	iwl_pcie_rx_init_rba(rba);
+	spin_unlock(&rba->lock);
 
 	spin_lock(&rxq->lock);
 
-	INIT_WORK(&trans_pcie->rx_replenish, iwl_pcie_rx_replenish_work);
-
 	/* free all first - we might be reconfigured for a different size */
 	iwl_pcie_rxq_free_rbs(trans);
 	iwl_pcie_rx_init_rxb_lists(rxq);
@@ -522,7 +716,7 @@ int iwl_pcie_rx_init(struct iwl_trans *trans)
 	memset(rxq->rb_stts, 0, sizeof(*rxq->rb_stts));
 	spin_unlock(&rxq->lock);
 
-	iwl_pcie_rx_replenish(trans, GFP_KERNEL);
+	iwl_pcie_rx_replenish(trans);
 
 	iwl_pcie_rx_hw_init(trans, rxq);
 
@@ -537,6 +731,7 @@ 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;
+	struct iwl_rb_allocator *rba = &trans_pcie->rba;
 
 	/*if rxq->bd is NULL, it means that nothing has been allocated,
 	 * exit now */
@@ -545,7 +740,15 @@ void iwl_pcie_rx_free(struct iwl_trans *trans)
 		return;
 	}
 
-	cancel_work_sync(&trans_pcie->rx_replenish);
+	cancel_work_sync(&rba->rx_alloc);
+	if (rba->alloc_wq) {
+		destroy_workqueue(rba->alloc_wq);
+		rba->alloc_wq = NULL;
+	}
+
+	spin_lock(&rba->lock);
+	iwl_pcie_rx_free_rba(trans);
+	spin_unlock(&rba->lock);
 
 	spin_lock(&rxq->lock);
 	iwl_pcie_rxq_free_rbs(trans);
@@ -566,6 +769,43 @@ void iwl_pcie_rx_free(struct iwl_trans *trans)
 	rxq->rb_stts = NULL;
 }
 
+/*
+ * iwl_pcie_rx_reuse_rbd - Recycle used RBDs
+ *
+ * Called when a RBD can be reused. The RBD is transferred to the allocator.
+ * When there are 2 empty RBDs - a request for allocation is posted
+ */
+static void iwl_pcie_rx_reuse_rbd(struct iwl_trans *trans,
+				  struct iwl_rx_mem_buffer *rxb,
+				  struct iwl_rxq *rxq)
+{
+	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
+	struct iwl_rb_allocator *rba = &trans_pcie->rba;
+
+	/* Count the used RBDs */
+	rxq->used_count++;
+
+	/* Move the RBD to the used list, will be moved to allocator in batches
+	 * before claiming or posting a request*/
+	list_add_tail(&rxb->list, &rxq->rx_used);
+
+	/* If we have RX_POST_REQ_ALLOC new released rx buffers -
+	 * issue a request for allocator. Modulo RX_CLAIM_REQ_ALLOC is
+	 * used for the case we failed to claim RX_CLAIM_REQ_ALLOC,
+	 * after but we still need to post another request.
+	 */
+	if ((rxq->used_count % RX_CLAIM_REQ_ALLOC) == RX_POST_REQ_ALLOC) {
+		/* Move the 2 RBDs to the allocator ownership.
+		 Allocator has another 6 from pool for the request completion*/
+		spin_lock(&rba->lock);
+		list_splice_tail_init(&rxq->rx_used, &rba->rbd_empty);
+		spin_unlock(&rba->lock);
+
+		atomic_inc(&rba->req_pending);
+		queue_work(rba->alloc_wq, &rba->rx_alloc);
+	}
+}
+
 static void iwl_pcie_rx_handle_rb(struct iwl_trans *trans,
 				struct iwl_rx_mem_buffer *rxb)
 {
@@ -688,13 +928,13 @@ static void iwl_pcie_rx_handle_rb(struct iwl_trans *trans,
 			 */
 			__free_pages(rxb->page, trans_pcie->rx_page_order);
 			rxb->page = NULL;
-			list_add_tail(&rxb->list, &rxq->rx_used);
+			iwl_pcie_rx_reuse_rbd(trans, rxb, rxq);
 		} else {
 			list_add_tail(&rxb->list, &rxq->rx_free);
 			rxq->free_count++;
 		}
 	} else
-		list_add_tail(&rxb->list, &rxq->rx_used);
+		iwl_pcie_rx_reuse_rbd(trans, rxb, rxq);
 }
 
 /*
@@ -704,10 +944,7 @@ static void iwl_pcie_rx_handle(struct iwl_trans *trans)
 {
 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
 	struct iwl_rxq *rxq = &trans_pcie->rxq;
-	u32 r, i;
-	u8 fill_rx = 0;
-	u32 count = 8;
-	int total_empty;
+	u32 r, i, j;
 
 restart:
 	spin_lock(&rxq->lock);
@@ -720,14 +957,6 @@ restart:
 	if (i == r)
 		IWL_DEBUG_RX(trans, "HW = SW = %d\n", r);
 
-	/* calculate total frames need to be restock after handling RX */
-	total_empty = r - rxq->write_actual;
-	if (total_empty < 0)
-		total_empty += RX_QUEUE_SIZE;
-
-	if (total_empty > (RX_QUEUE_SIZE / 2))
-		fill_rx = 1;
-
 	while (i != r) {
 		struct iwl_rx_mem_buffer *rxb;
 
@@ -739,29 +968,48 @@ restart:
 		iwl_pcie_rx_handle_rb(trans, rxb);
 
 		i = (i + 1) & RX_QUEUE_MASK;
-		/* If there are a lot of unused frames,
-		 * restock the Rx queue so ucode wont assert. */
-		if (fill_rx) {
-			count++;
-			if (count >= 8) {
-				rxq->read = i;
-				spin_unlock(&rxq->lock);
-				iwl_pcie_rx_replenish(trans, GFP_ATOMIC);
-				count = 0;
-				goto restart;
+
+		/* If we have RX_CLAIM_REQ_ALLOC released rx buffers -
+		 * try to claim the pre-allocated buffers from the allocator */
+		if (rxq->used_count >= RX_CLAIM_REQ_ALLOC) {
+			struct iwl_rb_allocator *rba = &trans_pcie->rba;
+			struct iwl_rx_mem_buffer *out[RX_CLAIM_REQ_ALLOC];
+
+			/* Add the remaining 6 empty RBDs for allocator use */
+			spin_lock(&rba->lock);
+			list_splice_tail_init(&rxq->rx_used, &rba->rbd_empty);
+			spin_unlock(&rba->lock);
+
+			/* If not ready - continue, will try to reclaim later.
+			* No need to reschedule work - allocator exits only on
+			* success */
+			if (!iwl_pcie_rx_allocator_get(trans, out)) {
+				/* If success - then RX_CLAIM_REQ_ALLOC
+				 * buffers were retrieved and should be added
+				 * to free list */
+				rxq->used_count -= RX_CLAIM_REQ_ALLOC;
+				for (j = 0; j < RX_CLAIM_REQ_ALLOC; j++) {
+					list_add_tail(&out[j]->list,
+						      &rxq->rx_free);
+					rxq->free_count++;
+				}
 			}
 		}
+		/* handle restock for two cases:
+		* - we just pulled buffers from the allocator
+		* - we have 8+ unstolen pages accumulated */
+		if (rxq->free_count >=  RX_CLAIM_REQ_ALLOC) {
+			rxq->read = i;
+			spin_unlock(&rxq->lock);
+			iwl_pcie_rxq_restock(trans);
+			goto restart;
+		}
 	}
 
 	/* Backtrack one entry */
 	rxq->read = i;
 	spin_unlock(&rxq->lock);
 
-	if (fill_rx)
-		iwl_pcie_rx_replenish(trans, GFP_ATOMIC);
-	else
-		iwl_pcie_rxq_restock(trans);
-
 	if (trans_pcie->napi.poll)
 		napi_gro_flush(&trans_pcie->napi, false);
 }
@@ -775,6 +1023,7 @@ static void iwl_pcie_irq_handle_error(struct iwl_trans *trans)
 
 	/* W/A for WiFi/WiMAX coex and WiMAX own the RF */
 	if (trans->cfg->internal_wimax_coex &&
+	    !trans->cfg->apmg_not_supported &&
 	    (!(iwl_read_prph(trans, APMG_CLK_CTRL_REG) &
 			     APMS_CLK_VAL_MRB_FUNC_MODE) ||
 	     (iwl_read_prph(trans, APMG_PS_CTRL_REG) &
diff --git a/drivers/net/wireless/iwlwifi/pcie/trans.c b/drivers/net/wireless/iwlwifi/pcie/trans.c
index 2de8fbfe4edf..43ae658af6ec 100644
--- a/drivers/net/wireless/iwlwifi/pcie/trans.c
+++ b/drivers/net/wireless/iwlwifi/pcie/trans.c
@@ -5,8 +5,8 @@
  *
  * GPL LICENSE SUMMARY
  *
- * Copyright(c) 2007 - 2014 Intel Corporation. All rights reserved.
- * Copyright(c) 2013 - 2014 Intel Mobile Communications GmbH
+ * Copyright(c) 2007 - 2015 Intel Corporation. All rights reserved.
+ * Copyright(c) 2013 - 2015 Intel Mobile Communications GmbH
  *
  * This program is free software; you can redistribute it and/or modify
  * it under the terms of version 2 of the GNU General Public License as
@@ -31,8 +31,8 @@
  *
  * BSD LICENSE
  *
- * Copyright(c) 2005 - 2014 Intel Corporation. All rights reserved.
- * Copyright(c) 2013 - 2014 Intel Mobile Communications GmbH
+ * Copyright(c) 2005 - 2015 Intel Corporation. All rights reserved.
+ * Copyright(c) 2013 - 2015 Intel Mobile Communications GmbH
  * All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without
@@ -101,14 +101,26 @@ static void iwl_pcie_free_fw_monitor(struct iwl_trans *trans)
 	trans_pcie->fw_mon_size = 0;
 }
 
-static void iwl_pcie_alloc_fw_monitor(struct iwl_trans *trans)
+static void iwl_pcie_alloc_fw_monitor(struct iwl_trans *trans, u8 max_power)
 {
 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
-	struct page *page;
+	struct page *page = NULL;
 	dma_addr_t phys;
-	u32 size;
+	u32 size = 0;
 	u8 power;
 
+	if (!max_power) {
+		/* default max_power is maximum */
+		max_power = 26;
+	} else {
+		max_power += 11;
+	}
+
+	if (WARN(max_power > 26,
+		 "External buffer size for monitor is too big %d, check the FW TLV\n",
+		 max_power))
+		return;
+
 	if (trans_pcie->fw_mon_page) {
 		dma_sync_single_for_device(trans->dev, trans_pcie->fw_mon_phys,
 					   trans_pcie->fw_mon_size,
@@ -117,7 +129,7 @@ static void iwl_pcie_alloc_fw_monitor(struct iwl_trans *trans)
 	}
 
 	phys = 0;
-	for (power = 26; power >= 11; power--) {
+	for (power = max_power; power >= 11; power--) {
 		int order;
 
 		size = BIT(power);
@@ -131,6 +143,7 @@ static void iwl_pcie_alloc_fw_monitor(struct iwl_trans *trans)
 				    DMA_FROM_DEVICE);
 		if (dma_mapping_error(trans->dev, phys)) {
 			__free_pages(page, order);
+			page = NULL;
 			continue;
 		}
 		IWL_INFO(trans,
@@ -142,6 +155,12 @@ static void iwl_pcie_alloc_fw_monitor(struct iwl_trans *trans)
 	if (WARN_ON_ONCE(!page))
 		return;
 
+	if (power != max_power)
+		IWL_ERR(trans,
+			"Sorry - debug buffer is only %luK while you requested %luK\n",
+			(unsigned long)BIT(power - 10),
+			(unsigned long)BIT(max_power - 10));
+
 	trans_pcie->fw_mon_page = page;
 	trans_pcie->fw_mon_phys = phys;
 	trans_pcie->fw_mon_size = size;
@@ -163,6 +182,9 @@ static void iwl_trans_pcie_write_shr(struct iwl_trans *trans, u32 reg, u32 val)
 
 static void iwl_pcie_set_pwr(struct iwl_trans *trans, bool vaux)
 {
+	if (!trans->cfg->apmg_not_supported)
+		return;
+
 	if (vaux && pci_pme_capable(to_pci_dev(trans->dev), PCI_D3cold))
 		iwl_set_bits_mask_prph(trans, APMG_PS_CTRL_REG,
 				       APMG_PS_CTRL_VAL_PWR_SRC_VAUX,
@@ -296,7 +318,7 @@ static int iwl_pcie_apm_init(struct iwl_trans *trans)
 	 * bits do not disable clocks.  This preserves any hardware
 	 * bits already set by default in "CLK_CTRL_REG" after reset.
 	 */
-	if (trans->cfg->device_family != IWL_DEVICE_FAMILY_8000) {
+	if (!trans->cfg->apmg_not_supported) {
 		iwl_write_prph(trans, APMG_CLK_EN_REG,
 			       APMG_CLK_VAL_DMA_CLK_RQT);
 		udelay(20);
@@ -496,8 +518,7 @@ static int iwl_pcie_nic_init(struct iwl_trans *trans)
 
 	spin_unlock(&trans_pcie->irq_lock);
 
-	if (trans->cfg->device_family != IWL_DEVICE_FAMILY_8000)
-		iwl_pcie_set_pwr(trans, false);
+	iwl_pcie_set_pwr(trans, false);
 
 	iwl_op_mode_nic_config(trans->op_mode);
 
@@ -833,7 +854,7 @@ static void iwl_pcie_apply_destination(struct iwl_trans *trans)
 		 get_fw_dbg_mode_string(dest->monitor_mode));
 
 	if (dest->monitor_mode == EXTERNAL_MODE)
-		iwl_pcie_alloc_fw_monitor(trans);
+		iwl_pcie_alloc_fw_monitor(trans, dest->size_power);
 	else
 		IWL_WARN(trans, "PCI should have external buffer debug\n");
 
@@ -907,7 +928,7 @@ static int iwl_pcie_load_given_ucode(struct iwl_trans *trans,
 	/* supported for 7000 only for the moment */
 	if (iwlwifi_mod_params.fw_monitor &&
 	    trans->cfg->device_family == IWL_DEVICE_FAMILY_7000) {
-		iwl_pcie_alloc_fw_monitor(trans);
+		iwl_pcie_alloc_fw_monitor(trans, 0);
 
 		if (trans_pcie->fw_mon_size) {
 			iwl_write_prph(trans, MON_BUFF_BASE_ADDR,
@@ -954,12 +975,8 @@ static int iwl_pcie_load_given_ucode_8000(struct iwl_trans *trans,
 		return ret;
 
 	/* load to FW the binary sections of CPU2 */
-	ret = iwl_pcie_load_cpu_sections_8000(trans, image, 2,
-					      &first_ucode_section);
-	if (ret)
-		return ret;
-
-	return 0;
+	return iwl_pcie_load_cpu_sections_8000(trans, image, 2,
+					       &first_ucode_section);
 }
 
 static int iwl_trans_pcie_start_fw(struct iwl_trans *trans,
@@ -1020,7 +1037,7 @@ static void iwl_trans_pcie_fw_alive(struct iwl_trans *trans, u32 scd_addr)
 	iwl_pcie_tx_start(trans, scd_addr);
 }
 
-static void iwl_trans_pcie_stop_device(struct iwl_trans *trans)
+static void iwl_trans_pcie_stop_device(struct iwl_trans *trans, bool low_power)
 {
 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
 	bool hw_rfkill, was_hw_rfkill;
@@ -1048,9 +1065,11 @@ static void iwl_trans_pcie_stop_device(struct iwl_trans *trans)
 		iwl_pcie_rx_stop(trans);
 
 		/* Power-down device's busmaster DMA clocks */
-		iwl_write_prph(trans, APMG_CLK_DIS_REG,
-			       APMG_CLK_VAL_DMA_CLK_RQT);
-		udelay(5);
+		if (!trans->cfg->apmg_not_supported) {
+			iwl_write_prph(trans, APMG_CLK_DIS_REG,
+				       APMG_CLK_VAL_DMA_CLK_RQT);
+			udelay(5);
+		}
 	}
 
 	/* Make sure (redundant) we've released our request to stay awake */
@@ -1115,7 +1134,7 @@ static void iwl_trans_pcie_stop_device(struct iwl_trans *trans)
 void iwl_trans_pcie_rf_kill(struct iwl_trans *trans, bool state)
 {
 	if (iwl_op_mode_hw_rf_kill(trans->op_mode, state))
-		iwl_trans_pcie_stop_device(trans);
+		iwl_trans_pcie_stop_device(trans, true);
 }
 
 static void iwl_trans_pcie_d3_suspend(struct iwl_trans *trans, bool test)
@@ -1200,7 +1219,7 @@ static int iwl_trans_pcie_d3_resume(struct iwl_trans *trans,
 	return 0;
 }
 
-static int iwl_trans_pcie_start_hw(struct iwl_trans *trans)
+static int iwl_trans_pcie_start_hw(struct iwl_trans *trans, bool low_power)
 {
 	bool hw_rfkill;
 	int err;
@@ -1343,14 +1362,13 @@ void iwl_trans_pcie_free(struct iwl_trans *trans)
 	iounmap(trans_pcie->hw_base);
 	pci_release_regions(trans_pcie->pci_dev);
 	pci_disable_device(trans_pcie->pci_dev);
-	kmem_cache_destroy(trans->dev_cmd_pool);
 
 	if (trans_pcie->napi.poll)
 		netif_napi_del(&trans_pcie->napi);
 
 	iwl_pcie_free_fw_monitor(trans);
 
-	kfree(trans);
+	iwl_trans_free(trans);
 }
 
 static void iwl_trans_pcie_set_pmi(struct iwl_trans *trans, bool state)
@@ -1369,7 +1387,7 @@ static bool iwl_trans_pcie_grab_nic_access(struct iwl_trans *trans, bool silent,
 
 	spin_lock_irqsave(&trans_pcie->reg_lock, *flags);
 
-	if (trans_pcie->cmd_in_flight)
+	if (trans_pcie->cmd_hold_nic_awake)
 		goto out;
 
 	/* this bit wakes up the NIC */
@@ -1435,7 +1453,7 @@ static void iwl_trans_pcie_release_nic_access(struct iwl_trans *trans,
 	 */
 	__acquire(&trans_pcie->reg_lock);
 
-	if (trans_pcie->cmd_in_flight)
+	if (trans_pcie->cmd_hold_nic_awake)
 		goto out;
 
 	__iwl_trans_pcie_clear_bit(trans, CSR_GP_CNTRL,
@@ -2197,6 +2215,29 @@ static u32 iwl_trans_pcie_fh_regs_dump(struct iwl_trans *trans,
 	return sizeof(**data) + fh_regs_len;
 }
 
+static u32
+iwl_trans_pci_dump_marbh_monitor(struct iwl_trans *trans,
+				 struct iwl_fw_error_dump_fw_mon *fw_mon_data,
+				 u32 monitor_len)
+{
+	u32 buf_size_in_dwords = (monitor_len >> 2);
+	u32 *buffer = (u32 *)fw_mon_data->data;
+	unsigned long flags;
+	u32 i;
+
+	if (!iwl_trans_grab_nic_access(trans, false, &flags))
+		return 0;
+
+	__iwl_write_prph(trans, MON_DMARB_RD_CTL_ADDR, 0x1);
+	for (i = 0; i < buf_size_in_dwords; i++)
+		buffer[i] = __iwl_read_prph(trans, MON_DMARB_RD_DATA_ADDR);
+	__iwl_write_prph(trans, MON_DMARB_RD_CTL_ADDR, 0x0);
+
+	iwl_trans_release_nic_access(trans, &flags);
+
+	return monitor_len;
+}
+
 static
 struct iwl_trans_dump_data *iwl_trans_pcie_dump_data(struct iwl_trans *trans)
 {
@@ -2249,7 +2290,8 @@ struct iwl_trans_dump_data *iwl_trans_pcie_dump_data(struct iwl_trans *trans)
 		      trans->dbg_dest_tlv->end_shift;
 
 		/* Make "end" point to the actual end */
-		if (trans->cfg->device_family == IWL_DEVICE_FAMILY_8000)
+		if (trans->cfg->device_family == IWL_DEVICE_FAMILY_8000 ||
+		    trans->dbg_dest_tlv->monitor_mode == MARBH_MODE)
 			end += (1 << trans->dbg_dest_tlv->end_shift);
 		monitor_len = end - base;
 		len += sizeof(*data) + sizeof(struct iwl_fw_error_dump_fw_mon) +
@@ -2325,9 +2367,6 @@ struct iwl_trans_dump_data *iwl_trans_pcie_dump_data(struct iwl_trans *trans)
 
 		len += sizeof(*data) + sizeof(*fw_mon_data);
 		if (trans_pcie->fw_mon_page) {
-			data->len = cpu_to_le32(trans_pcie->fw_mon_size +
-						sizeof(*fw_mon_data));
-
 			/*
 			 * The firmware is now asserted, it won't write anything
 			 * to the buffer. CPU can take ownership to fetch the
@@ -2342,10 +2381,8 @@ struct iwl_trans_dump_data *iwl_trans_pcie_dump_data(struct iwl_trans *trans)
 			       page_address(trans_pcie->fw_mon_page),
 			       trans_pcie->fw_mon_size);
 
-			len += trans_pcie->fw_mon_size;
-		} else {
-			/* If we are here then the buffer is internal */
-
+			monitor_len = trans_pcie->fw_mon_size;
+		} else if (trans->dbg_dest_tlv->monitor_mode == SMEM_MODE) {
 			/*
 			 * Update pointers to reflect actual values after
 			 * shifting
@@ -2354,10 +2391,18 @@ struct iwl_trans_dump_data *iwl_trans_pcie_dump_data(struct iwl_trans *trans)
 			       trans->dbg_dest_tlv->base_shift;
 			iwl_trans_read_mem(trans, base, fw_mon_data->data,
 					   monitor_len / sizeof(u32));
-			data->len = cpu_to_le32(sizeof(*fw_mon_data) +
-						monitor_len);
-			len += monitor_len;
+		} else if (trans->dbg_dest_tlv->monitor_mode == MARBH_MODE) {
+			monitor_len =
+				iwl_trans_pci_dump_marbh_monitor(trans,
+								 fw_mon_data,
+								 monitor_len);
+		} else {
+			/* Didn't match anything - output no monitor data */
+			monitor_len = 0;
 		}
+
+		len += monitor_len;
+		data->len = cpu_to_le32(monitor_len + sizeof(*fw_mon_data));
 	}
 
 	dump_data->len = len;
@@ -2416,18 +2461,13 @@ struct iwl_trans *iwl_trans_pcie_alloc(struct pci_dev *pdev,
 	u16 pci_cmd;
 	int err;
 
-	trans = kzalloc(sizeof(struct iwl_trans) +
-			sizeof(struct iwl_trans_pcie), GFP_KERNEL);
-	if (!trans) {
-		err = -ENOMEM;
-		goto out;
-	}
+	trans = iwl_trans_alloc(sizeof(struct iwl_trans_pcie),
+				&pdev->dev, cfg, &trans_ops_pcie, 0);
+	if (!trans)
+		return ERR_PTR(-ENOMEM);
 
 	trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
 
-	trans->ops = &trans_ops_pcie;
-	trans->cfg = cfg;
-	trans_lockdep_init(trans);
 	trans_pcie->trans = trans;
 	spin_lock_init(&trans_pcie->irq_lock);
 	spin_lock_init(&trans_pcie->reg_lock);
@@ -2551,25 +2591,8 @@ struct iwl_trans *iwl_trans_pcie_alloc(struct pci_dev *pdev,
 	/* Initialize the wait queue for commands */
 	init_waitqueue_head(&trans_pcie->wait_command_queue);
 
-	snprintf(trans->dev_cmd_pool_name, sizeof(trans->dev_cmd_pool_name),
-		 "iwl_cmd_pool:%s", dev_name(trans->dev));
-
-	trans->dev_cmd_headroom = 0;
-	trans->dev_cmd_pool =
-		kmem_cache_create(trans->dev_cmd_pool_name,
-				  sizeof(struct iwl_device_cmd)
-				  + trans->dev_cmd_headroom,
-				  sizeof(void *),
-				  SLAB_HWCACHE_ALIGN,
-				  NULL);
-
-	if (!trans->dev_cmd_pool) {
-		err = -ENOMEM;
-		goto out_pci_disable_msi;
-	}
-
 	if (iwl_pcie_alloc_ict(trans))
-		goto out_free_cmd_pool;
+		goto out_pci_disable_msi;
 
 	err = request_threaded_irq(pdev->irq, iwl_pcie_isr,
 				   iwl_pcie_irq_handler,
@@ -2586,8 +2609,6 @@ struct iwl_trans *iwl_trans_pcie_alloc(struct pci_dev *pdev,
 
 out_free_ict:
 	iwl_pcie_free_ict(trans);
-out_free_cmd_pool:
-	kmem_cache_destroy(trans->dev_cmd_pool);
 out_pci_disable_msi:
 	pci_disable_msi(pdev);
 out_pci_release_regions:
@@ -2595,7 +2616,6 @@ out_pci_release_regions:
 out_pci_disable_device:
 	pci_disable_device(pdev);
 out_no_pci:
-	kfree(trans);
-out:
+	iwl_trans_free(trans);
 	return ERR_PTR(err);
 }
diff --git a/drivers/net/wireless/iwlwifi/pcie/tx.c b/drivers/net/wireless/iwlwifi/pcie/tx.c
index 06952aadfd7b..2b86c2135de3 100644
--- a/drivers/net/wireless/iwlwifi/pcie/tx.c
+++ b/drivers/net/wireless/iwlwifi/pcie/tx.c
@@ -1039,22 +1039,16 @@ static int iwl_pcie_set_cmd_in_flight(struct iwl_trans *trans,
 		iwl_trans_pcie_ref(trans);
 	}
 
-	if (trans_pcie->cmd_in_flight)
-		return 0;
-
-	trans_pcie->cmd_in_flight = true;
-
 	/*
 	 * wake up the NIC to make sure that the firmware will see the host
 	 * command - we will let the NIC sleep once all the host commands
 	 * returned. This needs to be done only on NICs that have
 	 * apmg_wake_up_wa set.
 	 */
-	if (trans->cfg->base_params->apmg_wake_up_wa) {
+	if (trans->cfg->base_params->apmg_wake_up_wa &&
+	    !trans_pcie->cmd_hold_nic_awake) {
 		__iwl_trans_pcie_set_bit(trans, CSR_GP_CNTRL,
 					 CSR_GP_CNTRL_REG_FLAG_MAC_ACCESS_REQ);
-		if (trans->cfg->device_family == IWL_DEVICE_FAMILY_8000)
-			udelay(2);
 
 		ret = iwl_poll_bit(trans, CSR_GP_CNTRL,
 				   CSR_GP_CNTRL_REG_VAL_MAC_ACCESS_EN,
@@ -1064,10 +1058,10 @@ static int iwl_pcie_set_cmd_in_flight(struct iwl_trans *trans,
 		if (ret < 0) {
 			__iwl_trans_pcie_clear_bit(trans, CSR_GP_CNTRL,
 					CSR_GP_CNTRL_REG_FLAG_MAC_ACCESS_REQ);
-			trans_pcie->cmd_in_flight = false;
 			IWL_ERR(trans, "Failed to wake NIC for hcmd\n");
 			return -EIO;
 		}
+		trans_pcie->cmd_hold_nic_awake = true;
 	}
 
 	return 0;
@@ -1085,15 +1079,14 @@ static int iwl_pcie_clear_cmd_in_flight(struct iwl_trans *trans)
 		iwl_trans_pcie_unref(trans);
 	}
 
-	if (WARN_ON(!trans_pcie->cmd_in_flight))
-		return 0;
-
-	trans_pcie->cmd_in_flight = false;
+	if (trans->cfg->base_params->apmg_wake_up_wa) {
+		if (WARN_ON(!trans_pcie->cmd_hold_nic_awake))
+			return 0;
 
-	if (trans->cfg->base_params->apmg_wake_up_wa)
+		trans_pcie->cmd_hold_nic_awake = false;
 		__iwl_trans_pcie_clear_bit(trans, CSR_GP_CNTRL,
-					CSR_GP_CNTRL_REG_FLAG_MAC_ACCESS_REQ);
-
+					   CSR_GP_CNTRL_REG_FLAG_MAC_ACCESS_REQ);
+	}
 	return 0;
 }