Linux-libre 4.7.1-gnupck-4.7.1-gnu

5 files changed, 186 insertions, 136 deletions

diff --git a/drivers/net/wireless/intel/iwlwifi/pcie/drv.c b/drivers/net/wireless/intel/iwlwifi/pcie/drv.c
index 79d7cd7d4..a588b05e3 100644
--- a/drivers/net/wireless/intel/iwlwifi/pcie/drv.c
+++ b/drivers/net/wireless/intel/iwlwifi/pcie/drv.c
@@ -493,17 +493,20 @@ static const struct pci_device_id iwl_hw_card_ids[] = {
 	{IWL_PCI_DEVICE(0x24FD, 0x8130, iwl8265_2ac_cfg)},
 
 /* 9000 Series */
-	{IWL_PCI_DEVICE(0x9DF0, 0x2A10, iwl5165_2ac_cfg)},
-	{IWL_PCI_DEVICE(0x9DF0, 0x2010, iwl5165_2ac_cfg)},
+	{IWL_PCI_DEVICE(0x2526, 0x0000, iwl9260_2ac_cfg)},
+	{IWL_PCI_DEVICE(0x2526, 0x0010, iwl9260_2ac_cfg)},
 	{IWL_PCI_DEVICE(0x9DF0, 0x0A10, iwl9260_2ac_cfg)},
 	{IWL_PCI_DEVICE(0x9DF0, 0x0010, iwl9260_2ac_cfg)},
-	{IWL_PCI_DEVICE(0x9DF0, 0x0000, iwl5165_2ac_cfg)},
-	{IWL_PCI_DEVICE(0x9DF0, 0x0310, iwl5165_2ac_cfg)},
-	{IWL_PCI_DEVICE(0x9DF0, 0x0510, iwl5165_2ac_cfg)},
-	{IWL_PCI_DEVICE(0x9DF0, 0x0710, iwl5165_2ac_cfg)},
 	{IWL_PCI_DEVICE(0x9DF0, 0x0210, iwl9260_2ac_cfg)},
 	{IWL_PCI_DEVICE(0x9DF0, 0x0410, iwl9260_2ac_cfg)},
 	{IWL_PCI_DEVICE(0x9DF0, 0x0610, iwl9260_2ac_cfg)},
+	{IWL_PCI_DEVICE(0x9DF0, 0x0310, iwl5165_2ac_cfg)},
+	{IWL_PCI_DEVICE(0x9DF0, 0x0000, iwl5165_2ac_cfg)},
+	{IWL_PCI_DEVICE(0x9DF0, 0x0510, iwl5165_2ac_cfg)},
+	{IWL_PCI_DEVICE(0x9DF0, 0x2010, iwl5165_2ac_cfg)},
+	{IWL_PCI_DEVICE(0x2526, 0x1420, iwl5165_2ac_cfg)},
+	{IWL_PCI_DEVICE(0x9DF0, 0x0710, iwl5165_2ac_cfg)},
+	{IWL_PCI_DEVICE(0x9DF0, 0x2A10, iwl5165_2ac_cfg)},
 #endif /* CONFIG_IWLMVM */
 
 	{0}
@@ -593,6 +596,7 @@ static int iwl_pci_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
 {
 	const struct iwl_cfg *cfg = (struct iwl_cfg *)(ent->driver_data);
 	const struct iwl_cfg *cfg_7265d __maybe_unused = NULL;
+	const struct iwl_cfg *cfg_9260lc __maybe_unused = NULL;
 	struct iwl_trans *iwl_trans;
 	struct iwl_trans_pcie *trans_pcie;
 	int ret;
@@ -620,6 +624,15 @@ static int iwl_pci_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
 		cfg = cfg_7265d;
 		iwl_trans->cfg = cfg_7265d;
 	}
+
+	if (iwl_trans->cfg->rf_id) {
+		if (cfg == &iwl9260_2ac_cfg)
+			cfg_9260lc = &iwl9260lc_2ac_cfg;
+		if (cfg_9260lc && iwl_trans->hw_rf_id == CSR_HW_RF_ID_TYPE_LC) {
+			cfg = cfg_9260lc;
+			iwl_trans->cfg = cfg_9260lc;
+		}
+	}
 #endif
 
 	pci_set_drvdata(pdev, iwl_trans);
@@ -661,10 +674,8 @@ static int iwl_pci_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
 	/* The PCI device starts with a reference taken and we are
 	 * supposed to release it here.  But to simplify the
 	 * interaction with the opmode, we don't do it now, but let
-	 * the opmode release it when it's ready.  To account for this
-	 * reference, we start with ref_count set to 1.
+	 * the opmode release it when it's ready.
 	 */
-	trans_pcie->ref_count = 1;
 
 	return 0;
 
diff --git a/drivers/net/wireless/intel/iwlwifi/pcie/internal.h b/drivers/net/wireless/intel/iwlwifi/pcie/internal.h
index dadafbdef..de6974f9c 100644
--- a/drivers/net/wireless/intel/iwlwifi/pcie/internal.h
+++ b/drivers/net/wireless/intel/iwlwifi/pcie/internal.h
@@ -348,7 +348,7 @@ struct iwl_tso_hdr_page {
 struct iwl_trans_pcie {
 	struct iwl_rxq *rxq;
 	struct iwl_rx_mem_buffer rx_pool[RX_POOL_SIZE];
-	struct iwl_rx_mem_buffer *global_table[MQ_RX_TABLE_SIZE];
+	struct iwl_rx_mem_buffer *global_table[RX_POOL_SIZE];
 	struct iwl_rb_allocator rba;
 	struct iwl_trans *trans;
 	struct iwl_drv *drv;
@@ -403,10 +403,6 @@ struct iwl_trans_pcie {
 	bool cmd_hold_nic_awake;
 	bool ref_cmd_in_flight;
 
-	/* protect ref counter */
-	spinlock_t ref_lock;
-	u32 ref_count;
-
 	dma_addr_t fw_mon_phys;
 	struct page *fw_mon_page;
 	u32 fw_mon_size;
@@ -485,9 +481,6 @@ void iwl_trans_pcie_reclaim(struct iwl_trans *trans, int txq_id, int ssn,
 			    struct sk_buff_head *skbs);
 void iwl_trans_pcie_tx_reset(struct iwl_trans *trans);
 
-void iwl_trans_pcie_ref(struct iwl_trans *trans);
-void iwl_trans_pcie_unref(struct iwl_trans *trans);
-
 static inline u16 iwl_pcie_tfd_tb_get_len(struct iwl_tfd *tfd, u8 idx)
 {
 	struct iwl_tfd_tb *tb = &tfd->tbs[idx];
diff --git a/drivers/net/wireless/intel/iwlwifi/pcie/rx.c b/drivers/net/wireless/intel/iwlwifi/pcie/rx.c
index 4be3c35af..0a4a3c502 100644
--- a/drivers/net/wireless/intel/iwlwifi/pcie/rx.c
+++ b/drivers/net/wireless/intel/iwlwifi/pcie/rx.c
@@ -161,10 +161,11 @@ static inline __le32 iwl_pcie_dma_addr2rbd_ptr(dma_addr_t dma_addr)
 	return cpu_to_le32((u32)(dma_addr >> 8));
 }
 
-static void iwl_pcie_write_prph_64(struct iwl_trans *trans, u64 ofs, u64 val)
+static void iwl_pcie_write_prph_64_no_grab(struct iwl_trans *trans, u64 ofs,
+					   u64 val)
 {
-	iwl_write_prph(trans, ofs, val & 0xffffffff);
-	iwl_write_prph(trans, ofs + 4, val >> 32);
+	iwl_write_prph_no_grab(trans, ofs, val & 0xffffffff);
+	iwl_write_prph_no_grab(trans, ofs + 4, val >> 32);
 }
 
 /*
@@ -208,10 +209,14 @@ static void iwl_pcie_rxq_inc_wr_ptr(struct iwl_trans *trans,
 
 	rxq->write_actual = round_down(rxq->write, 8);
 	if (trans->cfg->mq_rx_supported)
-		iwl_write_prph(trans, RFH_Q_FRBDCB_WIDX(rxq->id),
-			       rxq->write_actual);
-	else
-		iwl_write32(trans, FH_RSCSR_CHNL0_WPTR, rxq->write_actual);
+		iwl_write32(trans, RFH_Q_FRBDCB_WIDX_TRG(rxq->id),
+			    rxq->write_actual);
+	/*
+	 * write to FH_RSCSR_CHNL0_WPTR register even in MQ as a W/A to
+	 * hardware shadow registers bug - writing to RFH_Q_FRBDCB_WIDX will
+	 * not wake the NIC.
+	 */
+	iwl_write32(trans, FH_RSCSR_CHNL0_WPTR, rxq->write_actual);
 }
 
 static void iwl_pcie_rxq_check_wrptr(struct iwl_trans *trans)
@@ -694,6 +699,7 @@ 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;
+	unsigned long flags;
 	const u32 rfdnlog = RX_QUEUE_SIZE_LOG; /* 256 RBDs */
 
 	switch (trans_pcie->rx_buf_size) {
@@ -711,23 +717,26 @@ static void iwl_pcie_rx_hw_init(struct iwl_trans *trans, struct iwl_rxq *rxq)
 		rb_size = FH_RCSR_RX_CONFIG_REG_VAL_RB_SIZE_4K;
 	}
 
+	if (!iwl_trans_grab_nic_access(trans, &flags))
+		return;
+
 	/* Stop Rx DMA */
-	iwl_write_direct32(trans, FH_MEM_RCSR_CHNL0_CONFIG_REG, 0);
+	iwl_write32(trans, FH_MEM_RCSR_CHNL0_CONFIG_REG, 0);
 	/* reset and flush pointers */
-	iwl_write_direct32(trans, FH_MEM_RCSR_CHNL0_RBDCB_WPTR, 0);
-	iwl_write_direct32(trans, FH_MEM_RCSR_CHNL0_FLUSH_RB_REQ, 0);
-	iwl_write_direct32(trans, FH_RSCSR_CHNL0_RDPTR, 0);
+	iwl_write32(trans, FH_MEM_RCSR_CHNL0_RBDCB_WPTR, 0);
+	iwl_write32(trans, FH_MEM_RCSR_CHNL0_FLUSH_RB_REQ, 0);
+	iwl_write32(trans, FH_RSCSR_CHNL0_RDPTR, 0);
 
 	/* Reset driver's Rx queue write index */
-	iwl_write_direct32(trans, FH_RSCSR_CHNL0_RBDCB_WPTR_REG, 0);
+	iwl_write32(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));
+	iwl_write32(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);
+	iwl_write32(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
@@ -737,13 +746,15 @@ static void iwl_pcie_rx_hw_init(struct iwl_trans *trans, struct iwl_rxq *rxq)
 	 * 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));
+	iwl_write32(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));
+
+	iwl_trans_release_nic_access(trans, &flags);
 
 	/* Set interrupt coalescing timer to default (2048 usecs) */
 	iwl_write8(trans, CSR_INT_COALESCING, IWL_HOST_INT_TIMEOUT_DEF);
@@ -757,6 +768,7 @@ static void iwl_pcie_rx_mq_hw_init(struct iwl_trans *trans)
 {
 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
 	u32 rb_size, enabled = 0;
+	unsigned long flags;
 	int i;
 
 	switch (trans_pcie->rx_buf_size) {
@@ -774,25 +786,31 @@ static void iwl_pcie_rx_mq_hw_init(struct iwl_trans *trans)
 		rb_size = RFH_RXF_DMA_RB_SIZE_4K;
 	}
 
+	if (!iwl_trans_grab_nic_access(trans, &flags))
+		return;
+
 	/* Stop Rx DMA */
-	iwl_write_prph(trans, RFH_RXF_DMA_CFG, 0);
+	iwl_write_prph_no_grab(trans, RFH_RXF_DMA_CFG, 0);
 	/* disable free amd used rx queue operation */
-	iwl_write_prph(trans, RFH_RXF_RXQ_ACTIVE, 0);
+	iwl_write_prph_no_grab(trans, RFH_RXF_RXQ_ACTIVE, 0);
 
 	for (i = 0; i < trans->num_rx_queues; i++) {
 		/* Tell device where to find RBD free table in DRAM */
-		iwl_pcie_write_prph_64(trans, RFH_Q_FRBDCB_BA_LSB(i),
-				       (u64)(trans_pcie->rxq[i].bd_dma));
+		iwl_pcie_write_prph_64_no_grab(trans,
+					       RFH_Q_FRBDCB_BA_LSB(i),
+					       trans_pcie->rxq[i].bd_dma);
 		/* Tell device where to find RBD used table in DRAM */
-		iwl_pcie_write_prph_64(trans, RFH_Q_URBDCB_BA_LSB(i),
-				       (u64)(trans_pcie->rxq[i].used_bd_dma));
+		iwl_pcie_write_prph_64_no_grab(trans,
+					       RFH_Q_URBDCB_BA_LSB(i),
+					       trans_pcie->rxq[i].used_bd_dma);
 		/* Tell device where in DRAM to update its Rx status */
-		iwl_pcie_write_prph_64(trans, RFH_Q_URBD_STTS_WPTR_LSB(i),
-				       trans_pcie->rxq[i].rb_stts_dma);
+		iwl_pcie_write_prph_64_no_grab(trans,
+					       RFH_Q_URBD_STTS_WPTR_LSB(i),
+					       trans_pcie->rxq[i].rb_stts_dma);
 		/* Reset device indice tables */
-		iwl_write_prph(trans, RFH_Q_FRBDCB_WIDX(i), 0);
-		iwl_write_prph(trans, RFH_Q_FRBDCB_RIDX(i), 0);
-		iwl_write_prph(trans, RFH_Q_URBDCB_WIDX(i), 0);
+		iwl_write_prph_no_grab(trans, RFH_Q_FRBDCB_WIDX(i), 0);
+		iwl_write_prph_no_grab(trans, RFH_Q_FRBDCB_RIDX(i), 0);
+		iwl_write_prph_no_grab(trans, RFH_Q_URBDCB_WIDX(i), 0);
 
 		enabled |= BIT(i) | BIT(i + 16);
 	}
@@ -808,23 +826,26 @@ static void iwl_pcie_rx_mq_hw_init(struct iwl_trans *trans)
 	 * Drop frames that exceed RB size
 	 * 512 RBDs
 	 */
-	iwl_write_prph(trans, RFH_RXF_DMA_CFG,
-		       RFH_DMA_EN_ENABLE_VAL |
-		       rb_size | RFH_RXF_DMA_SINGLE_FRAME_MASK |
-		       RFH_RXF_DMA_MIN_RB_4_8 |
-		       RFH_RXF_DMA_DROP_TOO_LARGE_MASK |
-		       RFH_RXF_DMA_RBDCB_SIZE_512);
+	iwl_write_prph_no_grab(trans, RFH_RXF_DMA_CFG,
+			       RFH_DMA_EN_ENABLE_VAL |
+			       rb_size | RFH_RXF_DMA_SINGLE_FRAME_MASK |
+			       RFH_RXF_DMA_MIN_RB_4_8 |
+			       RFH_RXF_DMA_DROP_TOO_LARGE_MASK |
+			       RFH_RXF_DMA_RBDCB_SIZE_512);
 
 	/*
 	 * Activate DMA snooping.
 	 * Set RX DMA chunk size to 64B
 	 * Default queue is 0
 	 */
-	iwl_write_prph(trans, RFH_GEN_CFG, RFH_GEN_CFG_RFH_DMA_SNOOP |
-		       (DEFAULT_RXQ_NUM << RFH_GEN_CFG_DEFAULT_RXQ_NUM_POS) |
-		       RFH_GEN_CFG_SERVICE_DMA_SNOOP);
+	iwl_write_prph_no_grab(trans, RFH_GEN_CFG, RFH_GEN_CFG_RFH_DMA_SNOOP |
+			       (DEFAULT_RXQ_NUM <<
+				RFH_GEN_CFG_DEFAULT_RXQ_NUM_POS) |
+			       RFH_GEN_CFG_SERVICE_DMA_SNOOP);
 	/* Enable the relevant rx queues */
-	iwl_write_prph(trans, RFH_RXF_RXQ_ACTIVE, enabled);
+	iwl_write_prph_no_grab(trans, RFH_RXF_RXQ_ACTIVE, enabled);
+
+	iwl_trans_release_nic_access(trans, &flags);
 
 	/* Set interrupt coalescing timer to default (2048 usecs) */
 	iwl_write8(trans, CSR_INT_COALESCING, IWL_HOST_INT_TIMEOUT_DEF);
@@ -908,6 +929,8 @@ int iwl_pcie_rx_init(struct iwl_trans *trans)
 	allocator_pool_size = trans->num_rx_queues *
 		(RX_CLAIM_REQ_ALLOC - RX_POST_REQ_ALLOC);
 	num_alloc = queue_size + allocator_pool_size;
+	BUILD_BUG_ON(ARRAY_SIZE(trans_pcie->global_table) !=
+		     ARRAY_SIZE(trans_pcie->rx_pool));
 	for (i = 0; i < num_alloc; i++) {
 		struct iwl_rx_mem_buffer *rxb = &trans_pcie->rx_pool[i];
 
@@ -1292,7 +1315,7 @@ static inline void iwl_pcie_clear_irq(struct iwl_trans *trans,
 	 * write 1 clear (W1C) register, meaning that it's being clear
 	 * by writing 1 to the bit.
 	 */
-	iwl_write_direct32(trans, CSR_MSIX_AUTOMASK_ST_AD, BIT(entry->entry));
+	iwl_write32(trans, CSR_MSIX_AUTOMASK_ST_AD, BIT(entry->entry));
 }
 
 /*
@@ -1805,19 +1828,19 @@ irqreturn_t iwl_pcie_irq_msix_handler(int irq, void *dev_id)
 	struct msix_entry *entry = dev_id;
 	struct iwl_trans_pcie *trans_pcie = iwl_pcie_get_trans_pcie(entry);
 	struct iwl_trans *trans = trans_pcie->trans;
-	struct isr_statistics *isr_stats = isr_stats = &trans_pcie->isr_stats;
+	struct isr_statistics *isr_stats = &trans_pcie->isr_stats;
 	u32 inta_fh, inta_hw;
 
 	lock_map_acquire(&trans->sync_cmd_lockdep_map);
 
 	spin_lock(&trans_pcie->irq_lock);
-	inta_fh = iwl_read_direct32(trans, CSR_MSIX_FH_INT_CAUSES_AD);
-	inta_hw = iwl_read_direct32(trans, CSR_MSIX_HW_INT_CAUSES_AD);
+	inta_fh = iwl_read32(trans, CSR_MSIX_FH_INT_CAUSES_AD);
+	inta_hw = iwl_read32(trans, CSR_MSIX_HW_INT_CAUSES_AD);
 	/*
 	 * Clear causes registers to avoid being handling the same cause.
 	 */
-	iwl_write_direct32(trans, CSR_MSIX_FH_INT_CAUSES_AD, inta_fh);
-	iwl_write_direct32(trans, CSR_MSIX_HW_INT_CAUSES_AD, inta_hw);
+	iwl_write32(trans, CSR_MSIX_FH_INT_CAUSES_AD, inta_fh);
+	iwl_write32(trans, CSR_MSIX_HW_INT_CAUSES_AD, inta_hw);
 	spin_unlock(&trans_pcie->irq_lock);
 
 	if (unlikely(!(inta_fh | inta_hw))) {
diff --git a/drivers/net/wireless/intel/iwlwifi/pcie/trans.c b/drivers/net/wireless/intel/iwlwifi/pcie/trans.c
index b2b79354d..f603d7830 100644
--- a/drivers/net/wireless/intel/iwlwifi/pcie/trans.c
+++ b/drivers/net/wireless/intel/iwlwifi/pcie/trans.c
@@ -269,9 +269,8 @@ static int iwl_pcie_apm_init(struct iwl_trans *trans)
 	iwl_pcie_apm_config(trans);
 
 	/* Configure analog phase-lock-loop before activating to D0A */
-	if (trans->cfg->base_params->pll_cfg_val)
-		iwl_set_bit(trans, CSR_ANA_PLL_CFG,
-			    trans->cfg->base_params->pll_cfg_val);
+	if (trans->cfg->base_params->pll_cfg)
+		iwl_set_bit(trans, CSR_ANA_PLL_CFG, CSR50_ANA_PLL_CFG_VAL);
 
 	/*
 	 * Set "initialization complete" bit to move adapter from
@@ -361,8 +360,7 @@ static void iwl_pcie_apm_lp_xtal_enable(struct iwl_trans *trans)
 
 	/* Reset entire device - do controller reset (results in SHRD_HW_RST) */
 	iwl_set_bit(trans, CSR_RESET, CSR_RESET_REG_FLAG_SW_RESET);
-
-	udelay(10);
+	usleep_range(1000, 2000);
 
 	/*
 	 * Set "initialization complete" bit to move adapter from
@@ -408,8 +406,7 @@ static void iwl_pcie_apm_lp_xtal_enable(struct iwl_trans *trans)
 	 * SHRD_HW_RST). Turn MAC off before proceeding.
 	 */
 	iwl_set_bit(trans, CSR_RESET, CSR_RESET_REG_FLAG_SW_RESET);
-
-	udelay(10);
+	usleep_range(1000, 2000);
 
 	/* Enable LP XTAL by indirect access through CSR */
 	apmg_gp1_reg = iwl_trans_pcie_read_shr(trans, SHR_APMG_GP1_REG);
@@ -506,8 +503,7 @@ static void iwl_pcie_apm_stop(struct iwl_trans *trans, bool op_mode_leave)
 
 	/* Reset the entire device */
 	iwl_set_bit(trans, CSR_RESET, CSR_RESET_REG_FLAG_SW_RESET);
-
-	udelay(10);
+	usleep_range(1000, 2000);
 
 	/*
 	 * Clear "initialization complete" bit to move adapter from
@@ -586,7 +582,7 @@ static int iwl_pcie_prepare_card_hw(struct iwl_trans *trans)
 
 	iwl_set_bit(trans, CSR_DBG_LINK_PWR_MGMT_REG,
 		    CSR_RESET_LINK_PWR_MGMT_DISABLED);
-	msleep(1);
+	usleep_range(1000, 2000);
 
 	for (iter = 0; iter < 10; iter++) {
 		/* If HW is not ready, prepare the conditions to check again */
@@ -1074,7 +1070,7 @@ static void _iwl_trans_pcie_stop_device(struct iwl_trans *trans, bool low_power)
 
 	/* stop and reset the on-board processor */
 	iwl_write32(trans, CSR_RESET, CSR_RESET_REG_FLAG_SW_RESET);
-	udelay(20);
+	usleep_range(1000, 2000);
 
 	/*
 	 * Upon stop, the APM issues an interrupt if HW RF kill is set.
@@ -1321,6 +1317,7 @@ static int iwl_trans_pcie_d3_resume(struct iwl_trans *trans,
 	 * after this call.
 	 */
 	iwl_pcie_reset_ict(trans);
+	iwl_enable_interrupts(trans);
 
 	iwl_set_bit(trans, CSR_GP_CNTRL, CSR_GP_CNTRL_REG_FLAG_MAC_ACCESS_REQ);
 	iwl_set_bit(trans, CSR_GP_CNTRL, CSR_GP_CNTRL_REG_FLAG_INIT_DONE);
@@ -1434,7 +1431,7 @@ static void iwl_pcie_set_interrupt_capa(struct pci_dev *pdev,
 	int ret, i;
 
 	if (trans->cfg->mq_rx_supported) {
-		max_vector = min_t(u32, (num_possible_cpus() + 1),
+		max_vector = min_t(u32, (num_possible_cpus() + 2),
 				   IWL_MAX_RX_HW_QUEUES);
 		for (i = 0; i < max_vector; i++)
 			trans_pcie->msix_entries[i].entry = i;
@@ -1465,7 +1462,7 @@ static void iwl_pcie_set_interrupt_capa(struct pci_dev *pdev,
 
 	ret = pci_enable_msi(pdev);
 	if (ret) {
-		dev_err(&pdev->dev, "pci_enable_msi failed(0X%x)\n", ret);
+		dev_err(&pdev->dev, "pci_enable_msi failed - %d\n", ret);
 		/* enable rfkill interrupt: hw bug w/a */
 		pci_read_config_word(pdev, PCI_COMMAND, &pci_cmd);
 		if (pci_cmd & PCI_COMMAND_INTX_DISABLE) {
@@ -1499,8 +1496,8 @@ static int iwl_pcie_init_msix_handler(struct pci_dev *pdev,
 			IWL_ERR(trans_pcie->trans,
 				"Error allocating IRQ %d\n", i);
 			for (j = 0; j < i; j++)
-				free_irq(trans_pcie->msix_entries[i].vector,
-					 &trans_pcie->msix_entries[i]);
+				free_irq(trans_pcie->msix_entries[j].vector,
+					 &trans_pcie->msix_entries[j]);
 			pci_disable_msix(pdev);
 			return ret;
 		}
@@ -1525,8 +1522,7 @@ static int _iwl_trans_pcie_start_hw(struct iwl_trans *trans, bool low_power)
 
 	/* Reset the entire device */
 	iwl_write32(trans, CSR_RESET, CSR_RESET_REG_FLAG_SW_RESET);
-
-	usleep_range(10, 15);
+	usleep_range(1000, 2000);
 
 	iwl_pcie_apm_init(trans);
 
@@ -1694,6 +1690,7 @@ void iwl_trans_pcie_free(struct iwl_trans *trans)
 	}
 
 	free_percpu(trans_pcie->tso_hdr_page);
+	mutex_destroy(&trans_pcie->mutex);
 	iwl_trans_free(trans);
 }
 
@@ -1948,7 +1945,7 @@ static int iwl_trans_pcie_wait_txq_empty(struct iwl_trans *trans, u32 txq_bm)
 				      "WR pointer moved while flushing %d -> %d\n",
 				      wr_ptr, write_ptr))
 				return -ETIMEDOUT;
-			msleep(1);
+			usleep_range(1000, 2000);
 		}
 
 		if (q->read_ptr != q->write_ptr) {
@@ -2011,41 +2008,35 @@ static void iwl_trans_pcie_set_bits_mask(struct iwl_trans *trans, u32 reg,
 	spin_unlock_irqrestore(&trans_pcie->reg_lock, flags);
 }
 
-void iwl_trans_pcie_ref(struct iwl_trans *trans)
+static void iwl_trans_pcie_ref(struct iwl_trans *trans)
 {
 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
-	unsigned long flags;
 
 	if (iwlwifi_mod_params.d0i3_disable)
 		return;
 
-	spin_lock_irqsave(&trans_pcie->ref_lock, flags);
-	IWL_DEBUG_RPM(trans, "ref_counter: %d\n", trans_pcie->ref_count);
-	trans_pcie->ref_count++;
 	pm_runtime_get(&trans_pcie->pci_dev->dev);
-	spin_unlock_irqrestore(&trans_pcie->ref_lock, flags);
+
+#ifdef CONFIG_PM
+	IWL_DEBUG_RPM(trans, "runtime usage count: %d\n",
+		      atomic_read(&trans_pcie->pci_dev->dev.power.usage_count));
+#endif /* CONFIG_PM */
 }
 
-void iwl_trans_pcie_unref(struct iwl_trans *trans)
+static void iwl_trans_pcie_unref(struct iwl_trans *trans)
 {
 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
-	unsigned long flags;
 
 	if (iwlwifi_mod_params.d0i3_disable)
 		return;
 
-	spin_lock_irqsave(&trans_pcie->ref_lock, flags);
-	IWL_DEBUG_RPM(trans, "ref_counter: %d\n", trans_pcie->ref_count);
-	if (WARN_ON_ONCE(trans_pcie->ref_count == 0)) {
-		spin_unlock_irqrestore(&trans_pcie->ref_lock, flags);
-		return;
-	}
-	trans_pcie->ref_count--;
-
 	pm_runtime_mark_last_busy(&trans_pcie->pci_dev->dev);
 	pm_runtime_put_autosuspend(&trans_pcie->pci_dev->dev);
 
-	spin_unlock_irqrestore(&trans_pcie->ref_lock, flags);
+#ifdef CONFIG_PM
+	IWL_DEBUG_RPM(trans, "runtime usage count: %d\n",
+		      atomic_read(&trans_pcie->pci_dev->dev.power.usage_count));
+#endif /* CONFIG_PM */
 }
 
 static const char *get_csr_string(int cmd)
@@ -2793,7 +2784,6 @@ struct iwl_trans *iwl_trans_pcie_alloc(struct pci_dev *pdev,
 	trans_pcie->trans = trans;
 	spin_lock_init(&trans_pcie->irq_lock);
 	spin_lock_init(&trans_pcie->reg_lock);
-	spin_lock_init(&trans_pcie->ref_lock);
 	mutex_init(&trans_pcie->mutex);
 	init_waitqueue_head(&trans_pcie->ucode_write_waitq);
 	trans_pcie->tso_hdr_page = alloc_percpu(struct iwl_tso_hdr_page);
@@ -2912,6 +2902,8 @@ struct iwl_trans *iwl_trans_pcie_alloc(struct pci_dev *pdev,
 		}
 	}
 
+	trans->hw_rf_id = iwl_read32(trans, CSR_HW_RF_ID);
+
 	iwl_pcie_set_interrupt_capa(pdev, trans);
 	trans->hw_id = (pdev->device << 16) + pdev->subsystem_device;
 	snprintf(trans->hw_id_str, sizeof(trans->hw_id_str),
diff --git a/drivers/net/wireless/intel/iwlwifi/pcie/tx.c b/drivers/net/wireless/intel/iwlwifi/pcie/tx.c
index 16ad820ca..d6beac9af 100644
--- a/drivers/net/wireless/intel/iwlwifi/pcie/tx.c
+++ b/drivers/net/wireless/intel/iwlwifi/pcie/tx.c
@@ -32,6 +32,7 @@
 #include <linux/ieee80211.h>
 #include <linux/slab.h>
 #include <linux/sched.h>
+#include <linux/pm_runtime.h>
 #include <net/ip6_checksum.h>
 #include <net/tso.h>
 
@@ -596,6 +597,28 @@ static void iwl_pcie_free_tso_page(struct sk_buff *skb)
 	}
 }
 
+static void iwl_pcie_clear_cmd_in_flight(struct iwl_trans *trans)
+{
+	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
+
+	lockdep_assert_held(&trans_pcie->reg_lock);
+
+	if (trans_pcie->ref_cmd_in_flight) {
+		trans_pcie->ref_cmd_in_flight = false;
+		IWL_DEBUG_RPM(trans, "clear ref_cmd_in_flight - unref\n");
+		iwl_trans_unref(trans);
+	}
+
+	if (!trans->cfg->base_params->apmg_wake_up_wa)
+		return;
+	if (WARN_ON(!trans_pcie->cmd_hold_nic_awake))
+		return;
+
+	trans_pcie->cmd_hold_nic_awake = false;
+	__iwl_trans_pcie_clear_bit(trans, CSR_GP_CNTRL,
+				   CSR_GP_CNTRL_REG_FLAG_MAC_ACCESS_REQ);
+}
+
 /*
  * iwl_pcie_txq_unmap -  Unmap any remaining DMA mappings and free skb's
  */
@@ -620,6 +643,20 @@ static void iwl_pcie_txq_unmap(struct iwl_trans *trans, int txq_id)
 		}
 		iwl_pcie_txq_free_tfd(trans, txq);
 		q->read_ptr = iwl_queue_inc_wrap(q->read_ptr);
+
+		if (q->read_ptr == q->write_ptr) {
+			unsigned long flags;
+
+			spin_lock_irqsave(&trans_pcie->reg_lock, flags);
+			if (txq_id != trans_pcie->cmd_queue) {
+				IWL_DEBUG_RPM(trans, "Q %d - last tx freed\n",
+					      q->id);
+				iwl_trans_unref(trans);
+			} else {
+				iwl_pcie_clear_cmd_in_flight(trans);
+			}
+			spin_unlock_irqrestore(&trans_pcie->reg_lock, flags);
+		}
 	}
 	txq->active = false;
 
@@ -1098,7 +1135,7 @@ void iwl_trans_pcie_reclaim(struct iwl_trans *trans, int txq_id, int ssn,
 
 	if (q->read_ptr == q->write_ptr) {
 		IWL_DEBUG_RPM(trans, "Q %d - last tx reclaimed\n", q->id);
-		iwl_trans_pcie_unref(trans);
+		iwl_trans_unref(trans);
 	}
 
 out:
@@ -1117,7 +1154,7 @@ static int iwl_pcie_set_cmd_in_flight(struct iwl_trans *trans,
 	    !trans_pcie->ref_cmd_in_flight) {
 		trans_pcie->ref_cmd_in_flight = true;
 		IWL_DEBUG_RPM(trans, "set ref_cmd_in_flight - ref\n");
-		iwl_trans_pcie_ref(trans);
+		iwl_trans_ref(trans);
 	}
 
 	/*
@@ -1148,29 +1185,6 @@ static int iwl_pcie_set_cmd_in_flight(struct iwl_trans *trans,
 	return 0;
 }
 
-static int iwl_pcie_clear_cmd_in_flight(struct iwl_trans *trans)
-{
-	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
-
-	lockdep_assert_held(&trans_pcie->reg_lock);
-
-	if (trans_pcie->ref_cmd_in_flight) {
-		trans_pcie->ref_cmd_in_flight = false;
-		IWL_DEBUG_RPM(trans, "clear ref_cmd_in_flight - unref\n");
-		iwl_trans_pcie_unref(trans);
-	}
-
-	if (trans->cfg->base_params->apmg_wake_up_wa) {
-		if (WARN_ON(!trans_pcie->cmd_hold_nic_awake))
-			return 0;
-
-		trans_pcie->cmd_hold_nic_awake = false;
-		__iwl_trans_pcie_clear_bit(trans, CSR_GP_CNTRL,
-					   CSR_GP_CNTRL_REG_FLAG_MAC_ACCESS_REQ);
-	}
-	return 0;
-}
-
 /*
  * iwl_pcie_cmdq_reclaim - Reclaim TX command queue entries already Tx'd
  *
@@ -1786,6 +1800,16 @@ static int iwl_pcie_send_hcmd_sync(struct iwl_trans *trans,
 	IWL_DEBUG_INFO(trans, "Setting HCMD_ACTIVE for command %s\n",
 		       iwl_get_cmd_string(trans, cmd->id));
 
+	if (pm_runtime_suspended(&trans_pcie->pci_dev->dev)) {
+		ret = wait_event_timeout(trans_pcie->d0i3_waitq,
+				 pm_runtime_active(&trans_pcie->pci_dev->dev),
+				 msecs_to_jiffies(IWL_TRANS_IDLE_TIMEOUT));
+		if (!ret) {
+			IWL_ERR(trans, "Timeout exiting D0i3 before hcmd\n");
+			return -ETIMEDOUT;
+		}
+	}
+
 	cmd_idx = iwl_pcie_enqueue_hcmd(trans, cmd);
 	if (cmd_idx < 0) {
 		ret = cmd_idx;
@@ -2197,6 +2221,7 @@ int iwl_trans_pcie_tx(struct iwl_trans *trans, struct sk_buff *skb,
 	__le16 fc;
 	u8 hdr_len;
 	u16 wifi_seq;
+	bool amsdu;
 
 	txq = &trans_pcie->txq[txq_id];
 	q = &txq->q;
@@ -2288,11 +2313,18 @@ int iwl_trans_pcie_tx(struct iwl_trans *trans, struct sk_buff *skb,
 	 */
 	len = sizeof(struct iwl_tx_cmd) + sizeof(struct iwl_cmd_header) +
 	      hdr_len - IWL_HCMD_SCRATCHBUF_SIZE;
-	tb1_len = ALIGN(len, 4);
-
-	/* Tell NIC about any 2-byte padding after MAC header */
-	if (tb1_len != len)
-		tx_cmd->tx_flags |= TX_CMD_FLG_MH_PAD_MSK;
+	/* do not align A-MSDU to dword as the subframe header aligns it */
+	amsdu = ieee80211_is_data_qos(fc) &&
+		(*ieee80211_get_qos_ctl(hdr) &
+		 IEEE80211_QOS_CTL_A_MSDU_PRESENT);
+	if (trans_pcie->sw_csum_tx || !amsdu) {
+		tb1_len = ALIGN(len, 4);
+		/* Tell NIC about any 2-byte padding after MAC header */
+		if (tb1_len != len)
+			tx_cmd->tx_flags |= TX_CMD_FLG_MH_PAD_MSK;
+	} else {
+		tb1_len = len;
+	}
 
 	/* The first TB points to the scratchbuf data - min_copy bytes */
 	memcpy(&txq->scratchbufs[q->write_ptr], &dev_cmd->hdr,
@@ -2310,8 +2342,7 @@ int iwl_trans_pcie_tx(struct iwl_trans *trans, struct sk_buff *skb,
 		goto out_err;
 	iwl_pcie_txq_build_tfd(trans, txq, tb1_phys, tb1_len, false);
 
-	if (ieee80211_is_data_qos(fc) &&
-	    (*ieee80211_get_qos_ctl(hdr) & IEEE80211_QOS_CTL_A_MSDU_PRESENT)) {
+	if (amsdu) {
 		if (unlikely(iwl_fill_data_tbs_amsdu(trans, skb, txq, hdr_len,
 						     out_meta, dev_cmd,
 						     tb1_len)))
@@ -2342,7 +2373,7 @@ int iwl_trans_pcie_tx(struct iwl_trans *trans, struct sk_buff *skb,
 				txq->frozen_expiry_remainder = txq->wd_timeout;
 		}
 		IWL_DEBUG_RPM(trans, "Q: %d first tx - take ref\n", q->id);
-		iwl_trans_pcie_ref(trans);
+		iwl_trans_ref(trans);
 	}
 
 	/* Tell device the write index *just past* this latest filled TFD */