Add support for the latest 1G/10G Chelsio adapter, T3.

This driver is required by the Chelsio T3 RDMA driver posted by
Steve Wise.

Signed-off-by: Divy Le Ray <divy@chelsio.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>

25 files changed, 17328 insertions, 0 deletions

diff --git a/drivers/net/Kconfig b/drivers/net/Kconfig
index b199456a736..555b3ad6609 100644
--- a/drivers/net/Kconfig
+++ b/drivers/net/Kconfig
@@ -2389,6 +2389,24 @@ config CHELSIO_T1_NAPI
 	  NAPI is a driver API designed to reduce CPU and interrupt load
 	  when the driver is receiving lots of packets from the card.
 
+config CHELSIO_T3
+        tristate "Chelsio Communications T3 10Gb Ethernet support"
+        depends on PCI
+        help
+          This driver supports Chelsio T3-based gigabit and 10Gb Ethernet
+          adapters.
+
+          For general information about Chelsio and our products, visit
+          our website at <http://www.chelsio.com>.
+
+          For customer support, please visit our customer support page at
+          <http://www.chelsio.com/support.htm>.
+
+          Please send feedback to <linux-bugs@chelsio.com>.
+
+          To compile this driver as a module, choose M here: the module
+          will be called cxgb3.
+
 config EHEA
 	tristate "eHEA Ethernet support"
 	depends on IBMEBUS
diff --git a/drivers/net/Makefile b/drivers/net/Makefile
index 6c61f3e7396..88b63369d78 100644
--- a/drivers/net/Makefile
+++ b/drivers/net/Makefile
@@ -6,6 +6,7 @@ obj-$(CONFIG_E1000) += e1000/
 obj-$(CONFIG_IBM_EMAC) += ibm_emac/
 obj-$(CONFIG_IXGB) += ixgb/
 obj-$(CONFIG_CHELSIO_T1) += chelsio/
+obj-$(CONFIG_CHELSIO_T3) += cxgb3/
 obj-$(CONFIG_EHEA) += ehea/
 obj-$(CONFIG_BONDING) += bonding/
 obj-$(CONFIG_GIANFAR) += gianfar_driver.o
diff --git a/drivers/net/cxgb3/Makefile b/drivers/net/cxgb3/Makefile
new file mode 100644
index 00000000000..34346798532
--- /dev/null
+++ b/drivers/net/cxgb3/Makefile
@@ -0,0 +1,8 @@
+#
+# Chelsio T3 driver
+#
+
+obj-$(CONFIG_CHELSIO_T3) += cxgb3.o
+
+cxgb3-objs := cxgb3_main.o ael1002.o vsc8211.o t3_hw.o mc5.o \
+	      xgmac.o sge.o l2t.o cxgb3_offload.o
diff --git a/drivers/net/cxgb3/adapter.h b/drivers/net/cxgb3/adapter.h
new file mode 100644
index 00000000000..16643f6d00a
--- /dev/null
+++ b/drivers/net/cxgb3/adapter.h
@@ -0,0 +1,255 @@
+/*
+ * This file is part of the Chelsio T3 Ethernet driver for Linux.
+ *
+ * Copyright (C) 2003-2006 Chelsio Communications.  All rights reserved.
+ *
+ * This program is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the LICENSE file included in this
+ * release for licensing terms and conditions.
+ */
+
+/* This file should not be included directly.  Include common.h instead. */
+
+#ifndef __T3_ADAPTER_H__
+#define __T3_ADAPTER_H__
+
+#include <linux/pci.h>
+#include <linux/spinlock.h>
+#include <linux/interrupt.h>
+#include <linux/timer.h>
+#include <linux/cache.h>
+#include "t3cdev.h"
+#include <asm/semaphore.h>
+#include <asm/bitops.h>
+#include <asm/io.h>
+
+typedef irqreturn_t(*intr_handler_t) (int, void *);
+
+struct vlan_group;
+
+struct port_info {
+	struct vlan_group *vlan_grp;
+	const struct port_type_info *port_type;
+	u8 port_id;
+	u8 rx_csum_offload;
+	u8 nqsets;
+	u8 first_qset;
+	struct cphy phy;
+	struct cmac mac;
+	struct link_config link_config;
+	struct net_device_stats netstats;
+	int activity;
+};
+
+enum {				/* adapter flags */
+	FULL_INIT_DONE = (1 << 0),
+	USING_MSI = (1 << 1),
+	USING_MSIX = (1 << 2),
+};
+
+struct rx_desc;
+struct rx_sw_desc;
+
+struct sge_fl {			/* SGE per free-buffer list state */
+	unsigned int buf_size;	/* size of each Rx buffer */
+	unsigned int credits;	/* # of available Rx buffers */
+	unsigned int size;	/* capacity of free list */
+	unsigned int cidx;	/* consumer index */
+	unsigned int pidx;	/* producer index */
+	unsigned int gen;	/* free list generation */
+	struct rx_desc *desc;	/* address of HW Rx descriptor ring */
+	struct rx_sw_desc *sdesc;	/* address of SW Rx descriptor ring */
+	dma_addr_t phys_addr;	/* physical address of HW ring start */
+	unsigned int cntxt_id;	/* SGE context id for the free list */
+	unsigned long empty;	/* # of times queue ran out of buffers */
+};
+
+/*
+ * Bundle size for grouping offload RX packets for delivery to the stack.
+ * Don't make this too big as we do prefetch on each packet in a bundle.
+ */
+# define RX_BUNDLE_SIZE 8
+
+struct rsp_desc;
+
+struct sge_rspq {		/* state for an SGE response queue */
+	unsigned int credits;	/* # of pending response credits */
+	unsigned int size;	/* capacity of response queue */
+	unsigned int cidx;	/* consumer index */
+	unsigned int gen;	/* current generation bit */
+	unsigned int polling;	/* is the queue serviced through NAPI? */
+	unsigned int holdoff_tmr;	/* interrupt holdoff timer in 100ns */
+	unsigned int next_holdoff;	/* holdoff time for next interrupt */
+	struct rsp_desc *desc;	/* address of HW response ring */
+	dma_addr_t phys_addr;	/* physical address of the ring */
+	unsigned int cntxt_id;	/* SGE context id for the response q */
+	spinlock_t lock;	/* guards response processing */
+	struct sk_buff *rx_head;	/* offload packet receive queue head */
+	struct sk_buff *rx_tail;	/* offload packet receive queue tail */
+
+	unsigned long offload_pkts;
+	unsigned long offload_bundles;
+	unsigned long eth_pkts;	/* # of ethernet packets */
+	unsigned long pure_rsps;	/* # of pure (non-data) responses */
+	unsigned long imm_data;	/* responses with immediate data */
+	unsigned long rx_drops;	/* # of packets dropped due to no mem */
+	unsigned long async_notif; /* # of asynchronous notification events */
+	unsigned long empty;	/* # of times queue ran out of credits */
+	unsigned long nomem;	/* # of responses deferred due to no mem */
+	unsigned long unhandled_irqs;	/* # of spurious intrs */
+};
+
+struct tx_desc;
+struct tx_sw_desc;
+
+struct sge_txq {		/* state for an SGE Tx queue */
+	unsigned long flags;	/* HW DMA fetch status */
+	unsigned int in_use;	/* # of in-use Tx descriptors */
+	unsigned int size;	/* # of descriptors */
+	unsigned int processed;	/* total # of descs HW has processed */
+	unsigned int cleaned;	/* total # of descs SW has reclaimed */
+	unsigned int stop_thres;	/* SW TX queue suspend threshold */
+	unsigned int cidx;	/* consumer index */
+	unsigned int pidx;	/* producer index */
+	unsigned int gen;	/* current value of generation bit */
+	unsigned int unacked;	/* Tx descriptors used since last COMPL */
+	struct tx_desc *desc;	/* address of HW Tx descriptor ring */
+	struct tx_sw_desc *sdesc;	/* address of SW Tx descriptor ring */
+	spinlock_t lock;	/* guards enqueueing of new packets */
+	unsigned int token;	/* WR token */
+	dma_addr_t phys_addr;	/* physical address of the ring */
+	struct sk_buff_head sendq;	/* List of backpressured offload packets */
+	struct tasklet_struct qresume_tsk;	/* restarts the queue */
+	unsigned int cntxt_id;	/* SGE context id for the Tx q */
+	unsigned long stops;	/* # of times q has been stopped */
+	unsigned long restarts;	/* # of queue restarts */
+};
+
+enum {				/* per port SGE statistics */
+	SGE_PSTAT_TSO,		/* # of TSO requests */
+	SGE_PSTAT_RX_CSUM_GOOD,	/* # of successful RX csum offloads */
+	SGE_PSTAT_TX_CSUM,	/* # of TX checksum offloads */
+	SGE_PSTAT_VLANEX,	/* # of VLAN tag extractions */
+	SGE_PSTAT_VLANINS,	/* # of VLAN tag insertions */
+
+	SGE_PSTAT_MAX		/* must be last */
+};
+
+struct sge_qset {		/* an SGE queue set */
+	struct sge_rspq rspq;
+	struct sge_fl fl[SGE_RXQ_PER_SET];
+	struct sge_txq txq[SGE_TXQ_PER_SET];
+	struct net_device *netdev;	/* associated net device */
+	unsigned long txq_stopped;	/* which Tx queues are stopped */
+	struct timer_list tx_reclaim_timer;	/* reclaims TX buffers */
+	unsigned long port_stats[SGE_PSTAT_MAX];
+} ____cacheline_aligned;
+
+struct sge {
+	struct sge_qset qs[SGE_QSETS];
+	spinlock_t reg_lock;	/* guards non-atomic SGE registers (eg context) */
+};
+
+struct adapter {
+	struct t3cdev tdev;
+	struct list_head adapter_list;
+	void __iomem *regs;
+	struct pci_dev *pdev;
+	unsigned long registered_device_map;
+	unsigned long open_device_map;
+	unsigned long flags;
+
+	const char *name;
+	int msg_enable;
+	unsigned int mmio_len;
+
+	struct adapter_params params;
+	unsigned int slow_intr_mask;
+	unsigned long irq_stats[IRQ_NUM_STATS];
+
+	struct {
+		unsigned short vec;
+		char desc[22];
+	} msix_info[SGE_QSETS + 1];
+
+	/* T3 modules */
+	struct sge sge;
+	struct mc7 pmrx;
+	struct mc7 pmtx;
+	struct mc7 cm;
+	struct mc5 mc5;
+
+	struct net_device *port[MAX_NPORTS];
+	unsigned int check_task_cnt;
+	struct delayed_work adap_check_task;
+	struct work_struct ext_intr_handler_task;
+
+	/*
+	 * Dummy netdevices are needed when using multiple receive queues with
+	 * NAPI as each netdevice can service only one queue.
+	 */
+	struct net_device *dummy_netdev[SGE_QSETS - 1];
+
+	struct dentry *debugfs_root;
+
+	struct mutex mdio_lock;
+	spinlock_t stats_lock;
+	spinlock_t work_lock;
+};
+
+static inline u32 t3_read_reg(struct adapter *adapter, u32 reg_addr)
+{
+	u32 val = readl(adapter->regs + reg_addr);
+
+	CH_DBG(adapter, MMIO, "read register 0x%x value 0x%x\n", reg_addr, val);
+	return val;
+}
+
+static inline void t3_write_reg(struct adapter *adapter, u32 reg_addr, u32 val)
+{
+	CH_DBG(adapter, MMIO, "setting register 0x%x to 0x%x\n", reg_addr, val);
+	writel(val, adapter->regs + reg_addr);
+}
+
+static inline struct port_info *adap2pinfo(struct adapter *adap, int idx)
+{
+	return netdev_priv(adap->port[idx]);
+}
+
+/*
+ * We use the spare atalk_ptr to map a net device to its SGE queue set.
+ * This is a macro so it can be used as l-value.
+ */
+#define dev2qset(netdev) ((netdev)->atalk_ptr)
+
+#define OFFLOAD_DEVMAP_BIT 15
+
+#define tdev2adap(d) container_of(d, struct adapter, tdev)
+
+static inline int offload_running(struct adapter *adapter)
+{
+	return test_bit(OFFLOAD_DEVMAP_BIT, &adapter->open_device_map);
+}
+
+int t3_offload_tx(struct t3cdev *tdev, struct sk_buff *skb);
+
+void t3_os_ext_intr_handler(struct adapter *adapter);
+void t3_os_link_changed(struct adapter *adapter, int port_id, int link_status,
+			int speed, int duplex, int fc);
+
+void t3_sge_start(struct adapter *adap);
+void t3_sge_stop(struct adapter *adap);
+void t3_free_sge_resources(struct adapter *adap);
+void t3_sge_err_intr_handler(struct adapter *adapter);
+intr_handler_t t3_intr_handler(struct adapter *adap, int polling);
+int t3_eth_xmit(struct sk_buff *skb, struct net_device *dev);
+void t3_update_qset_coalesce(struct sge_qset *qs, const struct qset_params *p);
+int t3_sge_alloc_qset(struct adapter *adapter, unsigned int id, int nports,
+		      int irq_vec_idx, const struct qset_params *p,
+		      int ntxq, struct net_device *netdev);
+int t3_get_desc(const struct sge_qset *qs, unsigned int qnum, unsigned int idx,
+		unsigned char *data);
+irqreturn_t t3_sge_intr_msix(int irq, void *cookie);
+
+#endif				/* __T3_ADAPTER_H__ */
diff --git a/drivers/net/cxgb3/ael1002.c b/drivers/net/cxgb3/ael1002.c
new file mode 100644
index 00000000000..93a90d825d8
--- /dev/null
+++ b/drivers/net/cxgb3/ael1002.c
@@ -0,0 +1,231 @@
+/*
+ * This file is part of the Chelsio T3 Ethernet driver.
+ *
+ * Copyright (C) 2005-2006 Chelsio Communications.  All rights reserved.
+ *
+ * This program is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the LICENSE file included in this
+ * release for licensing terms and conditions.
+ */
+
+#include "common.h"
+#include "regs.h"
+
+enum {
+	AEL100X_TX_DISABLE = 9,
+	AEL100X_TX_CONFIG1 = 0xc002,
+	AEL1002_PWR_DOWN_HI = 0xc011,
+	AEL1002_PWR_DOWN_LO = 0xc012,
+	AEL1002_XFI_EQL = 0xc015,
+	AEL1002_LB_EN = 0xc017,
+
+	LASI_CTRL = 0x9002,
+	LASI_STAT = 0x9005
+};
+
+static void ael100x_txon(struct cphy *phy)
+{
+	int tx_on_gpio = phy->addr == 0 ? F_GPIO7_OUT_VAL : F_GPIO2_OUT_VAL;
+
+	msleep(100);
+	t3_set_reg_field(phy->adapter, A_T3DBG_GPIO_EN, 0, tx_on_gpio);
+	msleep(30);
+}
+
+static int ael1002_power_down(struct cphy *phy, int enable)
+{
+	int err;
+
+	err = mdio_write(phy, MDIO_DEV_PMA_PMD, AEL100X_TX_DISABLE, !!enable);
+	if (!err)
+		err = t3_mdio_change_bits(phy, MDIO_DEV_PMA_PMD, MII_BMCR,
+					  BMCR_PDOWN, enable ? BMCR_PDOWN : 0);
+	return err;
+}
+
+static int ael1002_reset(struct cphy *phy, int wait)
+{
+	int err;
+
+	if ((err = ael1002_power_down(phy, 0)) ||
+	    (err = mdio_write(phy, MDIO_DEV_PMA_PMD, AEL100X_TX_CONFIG1, 1)) ||
+	    (err = mdio_write(phy, MDIO_DEV_PMA_PMD, AEL1002_PWR_DOWN_HI, 0)) ||
+	    (err = mdio_write(phy, MDIO_DEV_PMA_PMD, AEL1002_PWR_DOWN_LO, 0)) ||
+	    (err = mdio_write(phy, MDIO_DEV_PMA_PMD, AEL1002_XFI_EQL, 0x18)) ||
+	    (err = t3_mdio_change_bits(phy, MDIO_DEV_PMA_PMD, AEL1002_LB_EN,
+				       0, 1 << 5)))
+		return err;
+	return 0;
+}
+
+static int ael1002_intr_noop(struct cphy *phy)
+{
+	return 0;
+}
+
+static int ael100x_get_link_status(struct cphy *phy, int *link_ok,
+				   int *speed, int *duplex, int *fc)
+{
+	if (link_ok) {
+		unsigned int status;
+		int err = mdio_read(phy, MDIO_DEV_PMA_PMD, MII_BMSR, &status);
+
+		/*
+		 * BMSR_LSTATUS is latch-low, so if it is 0 we need to read it
+		 * once more to get the current link state.
+		 */
+		if (!err && !(status & BMSR_LSTATUS))
+			err = mdio_read(phy, MDIO_DEV_PMA_PMD, MII_BMSR,
+					&status);
+		if (err)
+			return err;
+		*link_ok = !!(status & BMSR_LSTATUS);
+	}
+	if (speed)
+		*speed = SPEED_10000;
+	if (duplex)
+		*duplex = DUPLEX_FULL;
+	return 0;
+}
+
+static struct cphy_ops ael1002_ops = {
+	.reset = ael1002_reset,
+	.intr_enable = ael1002_intr_noop,
+	.intr_disable = ael1002_intr_noop,
+	.intr_clear = ael1002_intr_noop,
+	.intr_handler = ael1002_intr_noop,
+	.get_link_status = ael100x_get_link_status,
+	.power_down = ael1002_power_down,
+};
+
+void t3_ael1002_phy_prep(struct cphy *phy, struct adapter *adapter,
+			 int phy_addr, const struct mdio_ops *mdio_ops)
+{
+	cphy_init(phy, adapter, phy_addr, &ael1002_ops, mdio_ops);
+	ael100x_txon(phy);
+}
+
+static int ael1006_reset(struct cphy *phy, int wait)
+{
+	return t3_phy_reset(phy, MDIO_DEV_PMA_PMD, wait);
+}
+
+static int ael1006_intr_enable(struct cphy *phy)
+{
+	return mdio_write(phy, MDIO_DEV_PMA_PMD, LASI_CTRL, 1);
+}
+
+static int ael1006_intr_disable(struct cphy *phy)
+{
+	return mdio_write(phy, MDIO_DEV_PMA_PMD, LASI_CTRL, 0);
+}
+
+static int ael1006_intr_clear(struct cphy *phy)
+{
+	u32 val;
+
+	return mdio_read(phy, MDIO_DEV_PMA_PMD, LASI_STAT, &val);
+}
+
+static int ael1006_intr_handler(struct cphy *phy)
+{
+	unsigned int status;
+	int err = mdio_read(phy, MDIO_DEV_PMA_PMD, LASI_STAT, &status);
+
+	if (err)
+		return err;
+	return (status & 1) ? cphy_cause_link_change : 0;
+}
+
+static int ael1006_power_down(struct cphy *phy, int enable)
+{
+	return t3_mdio_change_bits(phy, MDIO_DEV_PMA_PMD, MII_BMCR,
+				   BMCR_PDOWN, enable ? BMCR_PDOWN : 0);
+}
+
+static struct cphy_ops ael1006_ops = {
+	.reset = ael1006_reset,
+	.intr_enable = ael1006_intr_enable,
+	.intr_disable = ael1006_intr_disable,
+	.intr_clear = ael1006_intr_clear,
+	.intr_handler = ael1006_intr_handler,
+	.get_link_status = ael100x_get_link_status,
+	.power_down = ael1006_power_down,
+};
+
+void t3_ael1006_phy_prep(struct cphy *phy, struct adapter *adapter,
+			 int phy_addr, const struct mdio_ops *mdio_ops)
+{
+	cphy_init(phy, adapter, phy_addr, &ael1006_ops, mdio_ops);
+	ael100x_txon(phy);
+}
+
+static struct cphy_ops qt2045_ops = {
+	.reset = ael1006_reset,
+	.intr_enable = ael1006_intr_enable,
+	.intr_disable = ael1006_intr_disable,
+	.intr_clear = ael1006_intr_clear,
+	.intr_handler = ael1006_intr_handler,
+	.get_link_status = ael100x_get_link_status,
+	.power_down = ael1006_power_down,
+};
+
+void t3_qt2045_phy_prep(struct cphy *phy, struct adapter *adapter,
+			int phy_addr, const struct mdio_ops *mdio_ops)
+{
+	unsigned int stat;
+
+	cphy_init(phy, adapter, phy_addr, &qt2045_ops, mdio_ops);
+
+	/*
+	 * Some cards where the PHY is supposed to be at address 0 actually
+	 * have it at 1.
+	 */
+	if (!phy_addr && !mdio_read(phy, MDIO_DEV_PMA_PMD, MII_BMSR, &stat) &&
+	    stat == 0xffff)
+		phy->addr = 1;
+}
+
+static int xaui_direct_reset(struct cphy *phy, int wait)
+{
+	return 0;
+}
+
+static int xaui_direct_get_link_status(struct cphy *phy, int *link_ok,
+				       int *speed, int *duplex, int *fc)
+{
+	if (link_ok) {
+		unsigned int status;
+
+		status = t3_read_reg(phy->adapter,
+				     XGM_REG(A_XGM_SERDES_STAT0, phy->addr));
+		*link_ok = !(status & F_LOWSIG0);
+	}
+	if (speed)
+		*speed = SPEED_10000;
+	if (duplex)
+		*duplex = DUPLEX_FULL;
+	return 0;
+}
+
+static int xaui_direct_power_down(struct cphy *phy, int enable)
+{
+	return 0;
+}
+
+static struct cphy_ops xaui_direct_ops = {
+	.reset = xaui_direct_reset,
+	.intr_enable = ael1002_intr_noop,
+	.intr_disable = ael1002_intr_noop,
+	.intr_clear = ael1002_intr_noop,
+	.intr_handler = ael1002_intr_noop,
+	.get_link_status = xaui_direct_get_link_status,
+	.power_down = xaui_direct_power_down,
+};
+
+void t3_xaui_direct_phy_prep(struct cphy *phy, struct adapter *adapter,
+			     int phy_addr, const struct mdio_ops *mdio_ops)
+{
+	cphy_init(phy, adapter, 1, &xaui_direct_ops, mdio_ops);
+}
diff --git a/drivers/net/cxgb3/common.h b/drivers/net/cxgb3/common.h
new file mode 100644
index 00000000000..60a979b62a6
--- /dev/null
+++ b/drivers/net/cxgb3/common.h
@@ -0,0 +1,709 @@
+/*
+ * This file is part of the Chelsio T3 Ethernet driver.
+ *
+ * Copyright (C) 2005-2006 Chelsio Communications.  All rights reserved.
+ *
+ * This program is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the LICENSE file included in this
+ * release for licensing terms and conditions.
+ */
+
+#ifndef __CHELSIO_COMMON_H
+#define __CHELSIO_COMMON_H
+
+#include <linux/kernel.h>
+#include <linux/types.h>
+#include <linux/ctype.h>
+#include <linux/delay.h>
+#include <linux/init.h>
+#include <linux/netdevice.h>
+#include <linux/ethtool.h>
+#include <linux/mii.h>
+#include "version.h"
+
+#define CH_ERR(adap, fmt, ...)   dev_err(&adap->pdev->dev, fmt, ## __VA_ARGS__)
+#define CH_WARN(adap, fmt, ...)  dev_warn(&adap->pdev->dev, fmt, ## __VA_ARGS__)
+#define CH_ALERT(adap, fmt, ...) \
+	dev_printk(KERN_ALERT, &adap->pdev->dev, fmt, ## __VA_ARGS__)
+
+/*
+ * More powerful macro that selectively prints messages based on msg_enable.
+ * For info and debugging messages.
+ */
+#define CH_MSG(adapter, level, category, fmt, ...) do { \
+	if ((adapter)->msg_enable & NETIF_MSG_##category) \
+		dev_printk(KERN_##level, &adapter->pdev->dev, fmt, \
+			   ## __VA_ARGS__); \
+} while (0)
+
+#ifdef DEBUG
+# define CH_DBG(adapter, category, fmt, ...) \
+	CH_MSG(adapter, DEBUG, category, fmt, ## __VA_ARGS__)
+#else
+# define CH_DBG(adapter, category, fmt, ...)
+#endif
+
+/* Additional NETIF_MSG_* categories */
+#define NETIF_MSG_MMIO 0x8000000
+
+struct t3_rx_mode {
+	struct net_device *dev;
+	struct dev_mc_list *mclist;
+	unsigned int idx;
+};
+
+static inline void init_rx_mode(struct t3_rx_mode *p, struct net_device *dev,
+				struct dev_mc_list *mclist)
+{
+	p->dev = dev;
+	p->mclist = mclist;
+	p->idx = 0;
+}
+
+static inline u8 *t3_get_next_mcaddr(struct t3_rx_mode *rm)
+{
+	u8 *addr = NULL;
+
+	if (rm->mclist && rm->idx < rm->dev->mc_count) {
+		addr = rm->mclist->dmi_addr;
+		rm->mclist = rm->mclist->next;
+		rm->idx++;
+	}
+	return addr;
+}
+
+enum {
+	MAX_NPORTS = 2,		/* max # of ports */
+	MAX_FRAME_SIZE = 10240,	/* max MAC frame size, including header + FCS */
+	EEPROMSIZE = 8192,	/* Serial EEPROM size */
+	RSS_TABLE_SIZE = 64,	/* size of RSS lookup and mapping tables */
+	TCB_SIZE = 128,		/* TCB size */
+	NMTUS = 16,		/* size of MTU table */
+	NCCTRL_WIN = 32,	/* # of congestion control windows */
+};
+
+#define MAX_RX_COALESCING_LEN 16224U
+
+enum {
+	PAUSE_RX = 1 << 0,
+	PAUSE_TX = 1 << 1,
+	PAUSE_AUTONEG = 1 << 2
+};
+
+enum {
+	SUPPORTED_OFFLOAD = 1 << 24,
+	SUPPORTED_IRQ = 1 << 25
+};
+
+enum {				/* adapter interrupt-maintained statistics */
+	STAT_ULP_CH0_PBL_OOB,
+	STAT_ULP_CH1_PBL_OOB,
+	STAT_PCI_CORR_ECC,
+
+	IRQ_NUM_STATS		/* keep last */
+};
+
+enum {
+	SGE_QSETS = 8,		/* # of SGE Tx/Rx/RspQ sets */
+	SGE_RXQ_PER_SET = 2,	/* # of Rx queues per set */
+	SGE_TXQ_PER_SET = 3	/* # of Tx queues per set */
+};
+
+enum sge_context_type {		/* SGE egress context types */
+	SGE_CNTXT_RDMA = 0,
+	SGE_CNTXT_ETH = 2,
+	SGE_CNTXT_OFLD = 4,
+	SGE_CNTXT_CTRL = 5
+};
+
+enum {
+	AN_PKT_SIZE = 32,	/* async notification packet size */
+	IMMED_PKT_SIZE = 48	/* packet size for immediate data */
+};
+
+struct sg_ent {			/* SGE scatter/gather entry */
+	u32 len[2];
+	u64 addr[2];
+};
+
+#ifndef SGE_NUM_GENBITS
+/* Must be 1 or 2 */
+# define SGE_NUM_GENBITS 2
+#endif
+
+#define TX_DESC_FLITS 16U
+#define WR_FLITS (TX_DESC_FLITS + 1 - SGE_NUM_GENBITS)
+
+struct cphy;
+struct adapter;
+
+struct mdio_ops {
+	int (*read)(struct adapter *adapter, int phy_addr, int mmd_addr,
+		    int reg_addr, unsigned int *val);
+	int (*write)(struct adapter *adapter, int phy_addr, int mmd_addr,
+		     int reg_addr, unsigned int val);
+};
+
+struct adapter_info {
+	unsigned char nports;	/* # of ports */
+	unsigned char phy_base_addr;	/* MDIO PHY base address */
+	unsigned char mdien;
+	unsigned char mdiinv;
+	unsigned int gpio_out;	/* GPIO output settings */
+	unsigned int gpio_intr;	/* GPIO IRQ enable mask */
+	unsigned long caps;	/* adapter capabilities */
+	const struct mdio_ops *mdio_ops;	/* MDIO operations */
+	const char *desc;	/* product description */
+};
+
+struct port_type_info {
+	void (*phy_prep)(struct cphy *phy, struct adapter *adapter,
+			 int phy_addr, const struct mdio_ops *ops);
+	unsigned int caps;
+	const char *desc;
+};
+
+struct mc5_stats {
+	unsigned long parity_err;
+	unsigned long active_rgn_full;
+	unsigned long nfa_srch_err;
+	unsigned long unknown_cmd;
+	unsigned long reqq_parity_err;
+	unsigned long dispq_parity_err;
+	unsigned long del_act_empty;
+};
+
+struct mc7_stats {
+	unsigned long corr_err;
+	unsigned long uncorr_err;
+	unsigned long parity_err;
+	unsigned long addr_err;
+};
+
+struct mac_stats {
+	u64 tx_octets;		/* total # of octets in good frames */
+	u64 tx_octets_bad;	/* total # of octets in error frames */
+	u64 tx_frames;		/* all good frames */
+	u64 tx_mcast_frames;	/* good multicast frames */
+	u64 tx_bcast_frames;	/* good broadcast frames */
+	u64 tx_pause;		/* # of transmitted pause frames */
+	u64 tx_deferred;	/* frames with deferred transmissions */
+	u64 tx_late_collisions;	/* # of late collisions */
+	u64 tx_total_collisions;	/* # of total collisions */
+	u64 tx_excess_collisions;	/* frame errors from excessive collissions */
+	u64 tx_underrun;	/* # of Tx FIFO underruns */
+	u64 tx_len_errs;	/* # of Tx length errors */
+	u64 tx_mac_internal_errs;	/* # of internal MAC errors on Tx */
+	u64 tx_excess_deferral;	/* # of frames with excessive deferral */
+	u64 tx_fcs_errs;	/* # of frames with bad FCS */
+
+	u64 tx_frames_64;	/* # of Tx frames in a particular range */
+	u64 tx_frames_65_127;
+	u64 tx_frames_128_255;
+	u64 tx_frames_256_511;
+	u64 tx_frames_512_1023;
+	u64 tx_frames_1024_1518;
+	u64 tx_frames_1519_max;
+
+	u64 rx_octets;		/* total # of octets in good frames */
+	u64 rx_octets_bad;	/* total # of octets in error frames */
+	u64 rx_frames;		/* all good frames */
+	u64 rx_mcast_frames;	/* good multicast frames */
+	u64 rx_bcast_frames;	/* good broadcast frames */
+	u64 rx_pause;		/* # of received pause frames */
+	u64 rx_fcs_errs;	/* # of received frames with bad FCS */
+	u64 rx_align_errs;	/* alignment errors */
+	u64 rx_symbol_errs;	/* symbol errors */
+	u64 rx_data_errs;	/* data errors */
+	u64 rx_sequence_errs;	/* sequence errors */
+	u64 rx_runt;		/* # of runt frames */
+	u64 rx_jabber;		/* # of jabber frames */
+	u64 rx_short;		/* # of short frames */
+	u64 rx_too_long;	/* # of oversized frames */
+	u64 rx_mac_internal_errs;	/* # of internal MAC errors on Rx */
+
+	u64 rx_frames_64;	/* # of Rx frames in a particular range */
+	u64 rx_frames_65_127;
+	u64 rx_frames_128_255;
+	u64 rx_frames_256_511;
+	u64 rx_frames_512_1023;
+	u64 rx_frames_1024_1518;
+	u64 rx_frames_1519_max;
+
+	u64 rx_cong_drops;	/* # of Rx drops due to SGE congestion */
+
+	unsigned long tx_fifo_parity_err;
+	unsigned long rx_fifo_parity_err;
+	unsigned long tx_fifo_urun;
+	unsigned long rx_fifo_ovfl;
+	unsigned long serdes_signal_loss;
+	unsigned long xaui_pcs_ctc_err;
+	unsigned long xaui_pcs_align_change;
+};
+
+struct tp_mib_stats {
+	u32 ipInReceive_hi;
+	u32 ipInReceive_lo;
+	u32 ipInHdrErrors_hi;
+	u32 ipInHdrErrors_lo;
+	u32 ipInAddrErrors_hi;
+	u32 ipInAddrErrors_lo;
+	u32 ipInUnknownProtos_hi;
+	u32 ipInUnknownProtos_lo;
+	u32 ipInDiscards_hi;
+	u32 ipInDiscards_lo;
+	u32 ipInDelivers_hi;
+	u32 ipInDelivers_lo;
+	u32 ipOutRequests_hi;
+	u32 ipOutRequests_lo;
+	u32 ipOutDiscards_hi;
+	u32 ipOutDiscards_lo;
+	u32 ipOutNoRoutes_hi;
+	u32 ipOutNoRoutes_lo;
+	u32 ipReasmTimeout;
+	u32 ipReasmReqds;
+	u32 ipReasmOKs;
+	u32 ipReasmFails;
+
+	u32 reserved[8];
+
+	u32 tcpActiveOpens;
+	u32 tcpPassiveOpens;
+	u32 tcpAttemptFails;
+	u32 tcpEstabResets;
+	u32 tcpOutRsts;
+	u32 tcpCurrEstab;
+	u32 tcpInSegs_hi;
+	u32 tcpInSegs_lo;
+	u32 tcpOutSegs_hi;
+	u32 tcpOutSegs_lo;
+	u32 tcpRetransSeg_hi;
+	u32 tcpRetransSeg_lo;
+	u32 tcpInErrs_hi;
+	u32 tcpInErrs_lo;
+	u32 tcpRtoMin;
+	u32 tcpRtoMax;
+};
+
+struct tp_params {
+	unsigned int nchan;	/* # of channels */
+	unsigned int pmrx_size;	/* total PMRX capacity */
+	unsigned int pmtx_size;	/* total PMTX capacity */
+	unsigned int cm_size;	/* total CM capacity */
+	unsigned int chan_rx_size;	/* per channel Rx size */
+	unsigned int chan_tx_size;	/* per channel Tx size */
+	unsigned int rx_pg_size;	/* Rx page size */
+	unsigned int tx_pg_size;	/* Tx page size */
+	unsigned int rx_num_pgs;	/* # of Rx pages */
+	unsigned int tx_num_pgs;	/* # of Tx pages */
+	unsigned int ntimer_qs;	/* # of timer queues */
+};
+
+struct qset_params {		/* SGE queue set parameters */
+	unsigned int polling;	/* polling/interrupt service for rspq */
+	unsigned int coalesce_usecs;	/* irq coalescing timer */
+	unsigned int rspq_size;	/* # of entries in response queue */
+	unsigned int fl_size;	/* # of entries in regular free list */
+	unsigned int jumbo_size;	/* # of entries in jumbo free list */
+	unsigned int txq_size[SGE_TXQ_PER_SET];	/* Tx queue sizes */
+	unsigned int cong_thres;	/* FL congestion threshold */
+};
+
+struct sge_params {
+	unsigned int max_pkt_size;	/* max offload pkt size */
+	struct qset_params qset[SGE_QSETS];
+};
+
+struct mc5_params {
+	unsigned int mode;	/* selects MC5 width */
+	unsigned int nservers;	/* size of server region */
+	unsigned int nfilters;	/* size of filter region */
+	unsigned int nroutes;	/* size of routing region */
+};
+
+/* Default MC5 region sizes */
+enum {
+	DEFAULT_NSERVERS = 512,
+	DEFAULT_NFILTERS = 128
+};
+
+/* MC5 modes, these must be non-0 */
+enum {
+	MC5_MODE_144_BIT = 1,
+	MC5_MODE_72_BIT = 2
+};
+
+struct vpd_params {
+	unsigned int cclk;
+	unsigned int mclk;
+	unsigned int uclk;
+	unsigned int mdc;
+	unsigned int mem_timing;
+	u8 eth_base[6];
+	u8 port_type[MAX_NPORTS];
+	unsigned short xauicfg[2];
+};
+
+struct pci_params {
+	unsigned int vpd_cap_addr;
+	unsigned int pcie_cap_addr;
+	unsigned short speed;
+	unsigned char width;
+	unsigned char variant;
+};
+
+enum {
+	PCI_VARIANT_PCI,
+	PCI_VARIANT_PCIX_MODE1_PARITY,
+	PCI_VARIANT_PCIX_MODE1_ECC,
+	PCI_VARIANT_PCIX_266_MODE2,
+	PCI_VARIANT_PCIE
+};
+
+struct adapter_params {
+	struct sge_params sge;
+	struct mc5_params mc5;
+	struct tp_params tp;
+	struct vpd_params vpd;
+	struct pci_params pci;
+
+	const struct adapter_info *info;
+
+	unsigned short mtus[NMTUS];
+	unsigned short a_wnd[NCCTRL_WIN];
+	unsigned short b_wnd[NCCTRL_WIN];
+
+	unsigned int nports;	/* # of ethernet ports */
+	unsigned int stats_update_period;	/* MAC stats accumulation period */
+	unsigned int linkpoll_period;	/* link poll period in 0.1s */
+	unsigned int rev;	/* chip revision */
+};
+
+struct trace_params {
+	u32 sip;
+	u32 sip_mask;
+	u32 dip;
+	u32 dip_mask;
+	u16 sport;
+	u16 sport_mask;
+	u16 dport;
+	u16 dport_mask;
+	u32 vlan:12;
+	u32 vlan_mask:12;
+	u32 intf:4;
+	u32 intf_mask:4;
+	u8 proto;
+	u8 proto_mask;
+};
+
+struct link_config {
+	unsigned int supported;	/* link capabilities */
+	unsigned int advertising;	/* advertised capabilities */
+	unsigned short requested_speed;	/* speed user has requested */
+	unsigned short speed;	/* actual link speed */
+	unsigned char requested_duplex;	/* duplex user has requested */
+	unsigned char duplex;	/* actual link duplex */
+	unsigned char requested_fc;	/* flow control user has requested */
+	unsigned char fc;	/* actual link flow control */
+	unsigned char autoneg;	/* autonegotiating? */
+	unsigned int link_ok;	/* link up? */
+};
+
+#define SPEED_INVALID   0xffff
+#define DUPLEX_INVALID  0xff
+
+struct mc5 {
+	struct adapter *adapter;
+	unsigned int tcam_size;
+	unsigned char part_type;
+	unsigned char parity_enabled;
+	unsigned char mode;
+	struct mc5_stats stats;
+};
+
+static inline unsigned int t3_mc5_size(const struct mc5 *p)
+{
+	return p->tcam_size;
+}
+
+struct mc7 {
+	struct adapter *adapter;	/* backpointer to adapter */
+	unsigned int size;	/* memory size in bytes */
+	unsigned int width;	/* MC7 interface width */
+	unsigned int offset;	/* register address offset for MC7 instance */
+	const char *name;	/* name of MC7 instance */
+	struct mc7_stats stats;	/* MC7 statistics */
+};
+
+static inline unsigned int t3_mc7_size(const struct mc7 *p)
+{
+	return p->size;
+}
+
+struct cmac {
+	struct adapter *adapter;
+	unsigned int offset;
+	unsigned int nucast;	/* # of address filters for unicast MACs */
+	struct mac_stats stats;
+};
+
+enum {
+	MAC_DIRECTION_RX = 1,
+	MAC_DIRECTION_TX = 2,
+	MAC_RXFIFO_SIZE = 32768
+};
+
+/* IEEE 802.3ae specified MDIO devices */
+enum {
+	MDIO_DEV_PMA_PMD = 1,
+	MDIO_DEV_WIS = 2,
+	MDIO_DEV_PCS = 3,
+	MDIO_DEV_XGXS = 4
+};
+
+/* PHY loopback direction */
+enum {
+	PHY_LOOPBACK_TX = 1,
+	PHY_LOOPBACK_RX = 2
+};
+
+/* PHY interrupt types */
+enum {
+	cphy_cause_link_change = 1,
+	cphy_cause_fifo_error = 2
+};
+
+/* PHY operations */
+struct cphy_ops {
+	void (*destroy)(struct cphy *phy);
+	int (*reset)(struct cphy *phy, int wait);
+
+	int (*intr_enable)(struct cphy *phy);
+	int (*intr_disable)(struct cphy *phy);
+	int (*intr_clear)(struct cphy *phy);
+	int (*intr_handler)(struct cphy *phy);
+
+	int (*autoneg_enable)(struct cphy *phy);
+	int (*autoneg_restart)(struct cphy *phy);
+
+	int (*advertise)(struct cphy *phy, unsigned int advertise_map);
+	int (*set_loopback)(struct cphy *phy, int mmd, int dir, int enable);
+	int (*set_speed_duplex)(struct cphy *phy, int speed, int duplex);
+	int (*get_link_status)(struct cphy *phy, int *link_ok, int *speed,
+			       int *duplex, int *fc);
+	int (*power_down)(struct cphy *phy, int enable);
+};
+
+/* A PHY instance */
+struct cphy {
+	int addr;		/* PHY address */
+	struct adapter *adapter;	/* associated adapter */
+	unsigned long fifo_errors;	/* FIFO over/under-flows */
+	const struct cphy_ops *ops;	/* PHY operations */
+	int (*mdio_read)(struct adapter *adapter, int phy_addr, int mmd_addr,
+			 int reg_addr, unsigned int *val);
+	int (*mdio_write)(struct adapter *adapter, int phy_addr, int mmd_addr,
+			  int reg_addr, unsigned int val);
+};
+
+/* Convenience MDIO read/write wrappers */
+static inline int mdio_read(struct cphy *phy, int mmd, int reg,
+			    unsigned int *valp)
+{
+	return phy->mdio_read(phy->adapter, phy->addr, mmd, reg, valp);
+}
+
+static inline int mdio_write(struct cphy *phy, int mmd, int reg,
+			     unsigned int val)
+{
+	return phy->mdio_write(phy->adapter, phy->addr, mmd, reg, val);
+}
+
+/* Convenience initializer */
+static inline void cphy_init(struct cphy *phy, struct adapter *adapter,
+			     int phy_addr, struct cphy_ops *phy_ops,
+			     const struct mdio_ops *mdio_ops)
+{
+	phy->adapter = adapter;
+	phy->addr = phy_addr;
+	phy->ops = phy_ops;
+	if (mdio_ops) {
+		phy->mdio_read = mdio_ops->read;
+		phy->mdio_write = mdio_ops->write;
+	}
+}
+
+/* Accumulate MAC statistics every 180 seconds.  For 1G we multiply by 10. */
+#define MAC_STATS_ACCUM_SECS 180
+
+#define XGM_REG(reg_addr, idx) \
+	((reg_addr) + (idx) * (XGMAC0_1_BASE_ADDR - XGMAC0_0_BASE_ADDR))
+
+struct addr_val_pair {
+	unsigned int reg_addr;
+	unsigned int val;
+};
+
+#include "adapter.h"
+
+#ifndef PCI_VENDOR_ID_CHELSIO
+# define PCI_VENDOR_ID_CHELSIO 0x1425
+#endif
+
+#define for_each_port(adapter, iter) \
+	for (iter = 0; iter < (adapter)->params.nports; ++iter)
+
+#define adapter_info(adap) ((adap)->params.info)
+
+static inline int uses_xaui(const struct adapter *adap)
+{
+	return adapter_info(adap)->caps & SUPPORTED_AUI;
+}
+
+static inline int is_10G(const struct adapter *adap)
+{
+	return adapter_info(adap)->caps & SUPPORTED_10000baseT_Full;
+}
+
+static inline int is_offload(const struct adapter *adap)
+{
+	return adapter_info(adap)->caps & SUPPORTED_OFFLOAD;
+}
+
+static inline unsigned int core_ticks_per_usec(const struct adapter *adap)
+{
+	return adap->params.vpd.cclk / 1000;
+}
+
+static inline unsigned int is_pcie(const struct adapter *adap)
+{
+	return adap->params.pci.variant == PCI_VARIANT_PCIE;
+}
+
+void t3_set_reg_field(struct adapter *adap, unsigned int addr, u32 mask,
+		      u32 val);
+void t3_write_regs(struct adapter *adapter, const struct addr_val_pair *p,
+		   int n, unsigned int offset);
+int t3_wait_op_done_val(struct adapter *adapter, int reg, u32 mask,
+			int polarity, int attempts, int delay, u32 *valp);
+static inline int t3_wait_op_done(struct adapter *adapter, int reg, u32 mask,
+				  int polarity, int attempts, int delay)
+{
+	return t3_wait_op_done_val(adapter, reg, mask, polarity, attempts,
+				   delay, NULL);
+}
+int t3_mdio_change_bits(struct cphy *phy, int mmd, int reg, unsigned int clear,
+			unsigned int set);
+int t3_phy_reset(struct cphy *phy, int mmd, int wait);
+int t3_phy_advertise(struct cphy *phy, unsigned int advert);
+int t3_set_phy_speed_duplex(struct cphy *phy, int speed, int duplex);
+
+void t3_intr_enable(struct adapter *adapter);
+void t3_intr_disable(struct adapter *adapter);
+void t3_intr_clear(struct adapter *adapter);
+void t3_port_intr_enable(struct adapter *adapter, int idx);
+void t3_port_intr_disable(struct adapter *adapter, int idx);
+void t3_port_intr_clear(struct adapter *adapter, int idx);
+int t3_slow_intr_handler(struct adapter *adapter);
+int t3_phy_intr_handler(struct adapter *adapter);
+
+void t3_link_changed(struct adapter *adapter, int port_id);
+int t3_link_start(struct cphy *phy, struct cmac *mac, struct link_config *lc);
+const struct adapter_info *t3_get_adapter_info(unsigned int board_id);
+int t3_seeprom_read(struct adapter *adapter, u32 addr, u32 *data);
+int t3_seeprom_write(struct adapter *adapter, u32 addr, u32 data);
+int t3_seeprom_wp(struct adapter *adapter, int enable);
+int t3_read_flash(struct adapter *adapter, unsigned int addr,
+		  unsigned int nwords, u32 *data, int byte_oriented);
+int t3_load_fw(struct adapter *adapter, const u8 * fw_data, unsigned int size);
+int t3_get_fw_version(struct adapter *adapter, u32 *vers);
+int t3_check_fw_version(struct adapter *adapter);
+int t3_init_hw(struct adapter *adapter, u32 fw_params);
+void mac_prep(struct cmac *mac, struct adapter *adapter, int index);
+void early_hw_init(struct adapter *adapter, const struct adapter_info *ai);
+int t3_prep_adapter(struct adapter *adapter, const struct adapter_info *ai,
+		    int reset);
+void t3_led_ready(struct adapter *adapter);
+void t3_fatal_err(struct adapter *adapter);
+void t3_set_vlan_accel(struct adapter *adapter, unsigned int ports, int on);
+void t3_config_rss(struct adapter *adapter, unsigned int rss_config,
+		   const u8 * cpus, const u16 *rspq);
+int t3_read_rss(struct adapter *adapter, u8 * lkup, u16 *map);
+int t3_mps_set_active_ports(struct adapter *adap, unsigned int port_mask);
+int t3_cim_ctl_blk_read(struct adapter *adap, unsigned int addr,
+			unsigned int n, unsigned int *valp);
+int t3_mc7_bd_read(struct mc7 *mc7, unsigned int start, unsigned int n,
+		   u64 *buf);
+
+int t3_mac_reset(struct cmac *mac);
+void t3b_pcs_reset(struct cmac *mac);
+int t3_mac_enable(struct cmac *mac, int which);
+int t3_mac_disable(struct cmac *mac, int which);
+int t3_mac_set_mtu(struct cmac *mac, unsigned int mtu);
+int t3_mac_set_rx_mode(struct cmac *mac, struct t3_rx_mode *rm);
+int t3_mac_set_address(struct cmac *mac, unsigned int idx, u8 addr[6]);
+int t3_mac_set_num_ucast(struct cmac *mac, int n);
+const struct mac_stats *t3_mac_update_stats(struct cmac *mac);
+int t3_mac_set_speed_duplex_fc(struct cmac *mac, int speed, int duplex, int fc);
+
+void t3_mc5_prep(struct adapter *adapter, struct mc5 *mc5, int mode);
+int t3_mc5_init(struct mc5 *mc5, unsigned int nservers, unsigned int nfilters,
+		unsigned int nroutes);
+void t3_mc5_intr_handler(struct mc5 *mc5);
+int t3_read_mc5_range(const struct mc5 *mc5, unsigned int start, unsigned int n,
+		      u32 *buf);
+
+int t3_tp_set_coalescing_size(struct adapter *adap, unsigned int size, int psh);
+void t3_tp_set_max_rxsize(struct adapter *adap, unsigned int size);
+void t3_tp_set_offload_mode(struct adapter *adap, int enable);
+void t3_tp_get_mib_stats(struct adapter *adap, struct tp_mib_stats *tps);
+void t3_load_mtus(struct adapter *adap, unsigned short mtus[NMTUS],
+		  unsigned short alpha[NCCTRL_WIN],
+		  unsigned short beta[NCCTRL_WIN], unsigned short mtu_cap);
+void t3_read_hw_mtus(struct adapter *adap, unsigned short mtus[NMTUS]);
+void t3_get_cong_cntl_tab(struct adapter *adap,
+			  unsigned short incr[NMTUS][NCCTRL_WIN]);
+void t3_config_trace_filter(struct adapter *adapter,
+			    const struct trace_params *tp, int filter_index,
+			    int invert, int enable);
+int t3_config_sched(struct adapter *adap, unsigned int kbps, int sched);
+
+void t3_sge_prep(struct adapter *adap, struct sge_params *p);
+void t3_sge_init(struct adapter *adap, struct sge_params *p);
+int t3_sge_init_ecntxt(struct adapter *adapter, unsigned int id, int gts_enable,
+		       enum sge_context_type type, int respq, u64 base_addr,
+		       unsigned int size, unsigned int token, int gen,
+		       unsigned int cidx);
+int t3_sge_init_flcntxt(struct adapter *adapter, unsigned int id,
+			int gts_enable, u64 base_addr, unsigned int size,
+			unsigned int esize, unsigned int cong_thres, int gen,
+			unsigned int cidx);
+int t3_sge_init_rspcntxt(struct adapter *adapter, unsigned int id,
+			 int irq_vec_idx, u64 base_addr, unsigned int size,
+			 unsigned int fl_thres, int gen, unsigned int cidx);
+int t3_sge_init_cqcntxt(struct adapter *adapter, unsigned int id, u64 base_addr,
+			unsigned int size, int rspq, int ovfl_mode,
+			unsigned int credits, unsigned int credit_thres);
+int t3_sge_enable_ecntxt(struct adapter *adapter, unsigned int id, int enable);
+int t3_sge_disable_fl(struct adapter *adapter, unsigned int id);
+int t3_sge_disable_rspcntxt(struct adapter *adapter, unsigned int id);
+int t3_sge_disable_cqcntxt(struct adapter *adapter, unsigned int id);
+int t3_sge_read_ecntxt(struct adapter *adapter, unsigned int id, u32 data[4]);
+int t3_sge_read_fl(struct adapter *adapter, unsigned int id, u32 data[4]);
+int t3_sge_read_cq(struct adapter *adapter, unsigned int id, u32 data[4]);
+int t3_sge_read_rspq(struct adapter *adapter, unsigned int id, u32 data[4]);
+int t3_sge_cqcntxt_op(struct adapter *adapter, unsigned int id, unsigned int op,
+		      unsigned int credits);
+
+void t3_vsc8211_phy_prep(struct cphy *phy, struct adapter *adapter,
+			 int phy_addr, const struct mdio_ops *mdio_ops);
+void t3_ael1002_phy_prep(struct cphy *phy, struct adapter *adapter,
+			 int phy_addr, const struct mdio_ops *mdio_ops);
+void t3_ael1006_phy_prep(struct cphy *phy, struct adapter *adapter,
+			 int phy_addr, const struct mdio_ops *mdio_ops);
+void t3_qt2045_phy_prep(struct cphy *phy, struct adapter *adapter, int phy_addr,
+			const struct mdio_ops *mdio_ops);
+void t3_xaui_direct_phy_prep(struct cphy *phy, struct adapter *adapter,
+			     int phy_addr, const struct mdio_ops *mdio_ops);
+#endif				/* __CHELSIO_COMMON_H */
diff --git a/drivers/net/cxgb3/cxgb3_ctl_defs.h b/drivers/net/cxgb3/cxgb3_ctl_defs.h
new file mode 100644
index 00000000000..0fdc36529eb
--- /dev/null
+++ b/drivers/net/cxgb3/cxgb3_ctl_defs.h
@@ -0,0 +1,142 @@
+/*
+ * Copyright (C) 2003-2006 Chelsio Communications.  All rights reserved.
+ *
+ * This program is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the LICENSE file included in this
+ * release for licensing terms and conditions.
+ */
+
+#ifndef _CXGB3_OFFLOAD_CTL_DEFS_H
+#define _CXGB3_OFFLOAD_CTL_DEFS_H
+
+enum {
+	GET_MAX_OUTSTANDING_WR,
+	GET_TX_MAX_CHUNK,
+	GET_TID_RANGE,
+	GET_STID_RANGE,
+	GET_RTBL_RANGE,
+	GET_L2T_CAPACITY,
+	GET_MTUS,
+	GET_WR_LEN,
+	GET_IFF_FROM_MAC,
+	GET_DDP_PARAMS,
+	GET_PORTS,
+
+	ULP_ISCSI_GET_PARAMS,
+	ULP_ISCSI_SET_PARAMS,
+
+	RDMA_GET_PARAMS,
+	RDMA_CQ_OP,
+	RDMA_CQ_SETUP,
+	RDMA_CQ_DISABLE,
+	RDMA_CTRL_QP_SETUP,
+	RDMA_GET_MEM,
+};
+
+/*
+ * Structure used to describe a TID range.  Valid TIDs are [base, base+num).
+ */
+struct tid_range {
+	unsigned int base;	/* first TID */
+	unsigned int num;	/* number of TIDs in range */
+};
+
+/*
+ * Structure used to request the size and contents of the MTU table.
+ */
+struct mtutab {
+	unsigned int size;	/* # of entries in the MTU table */
+	const unsigned short *mtus;	/* the MTU table values */
+};
+
+struct net_device;
+
+/*
+ * Structure used to request the adapter net_device owning a given MAC address.
+ */
+struct iff_mac {
+	struct net_device *dev;	/* the net_device */
+	const unsigned char *mac_addr;	/* MAC address to lookup */
+	u16 vlan_tag;
+};
+
+struct pci_dev;
+
+/*
+ * Structure used to request the TCP DDP parameters.
+ */
+struct ddp_params {
+	unsigned int llimit;	/* TDDP region start address */
+	unsigned int ulimit;	/* TDDP region end address */
+	unsigned int tag_mask;	/* TDDP tag mask */
+	struct pci_dev *pdev;
+};
+
+struct adap_ports {
+	unsigned int nports;	/* number of ports on this adapter */
+	struct net_device *lldevs[2];
+};
+
+/*
+ * Structure used to return information to the iscsi layer.
+ */
+struct ulp_iscsi_info {
+	unsigned int offset;
+	unsigned int llimit;
+	unsigned int ulimit;
+	unsigned int tagmask;
+	unsigned int pgsz3;
+	unsigned int pgsz2;
+	unsigned int pgsz1;
+	unsigned int pgsz0;
+	unsigned int max_rxsz;
+	unsigned int max_txsz;
+	struct pci_dev *pdev;
+};
+
+/*
+ * Structure used to return information to the RDMA layer.
+ */
+struct rdma_info {
+	unsigned int tpt_base;	/* TPT base address */
+	unsigned int tpt_top;	/* TPT last entry address */
+	unsigned int pbl_base;	/* PBL base address */
+	unsigned int pbl_top;	/* PBL last entry address */
+	unsigned int rqt_base;	/* RQT base address */
+	unsigned int rqt_top;	/* RQT last entry address */
+	unsigned int udbell_len;	/* user doorbell region length */
+	unsigned long udbell_physbase;	/* user doorbell physical start addr */
+	void __iomem *kdb_addr;	/* kernel doorbell register address */
+	struct pci_dev *pdev;	/* associated PCI device */
+};
+
+/*
+ * Structure used to request an operation on an RDMA completion queue.
+ */
+struct rdma_cq_op {
+	unsigned int id;
+	unsigned int op;
+	unsigned int credits;
+};
+
+/*
+ * Structure used to setup RDMA completion queues.
+ */
+struct rdma_cq_setup {
+	unsigned int id;
+	unsigned long long base_addr;
+	unsigned int size;
+	unsigned int credits;
+	unsigned int credit_thres;
+	unsigned int ovfl_mode;
+};
+
+/*
+ * Structure used to setup the RDMA control egress context.
+ */
+struct rdma_ctrlqp_setup {
+	unsigned long long base_addr;
+	unsigned int size;
+};
+#endif				/* _CXGB3_OFFLOAD_CTL_DEFS_H */
diff --git a/drivers/net/cxgb3/cxgb3_defs.h b/drivers/net/cxgb3/cxgb3_defs.h
new file mode 100644
index 00000000000..82344c2bbc1
--- /dev/null
+++ b/drivers/net/cxgb3/cxgb3_defs.h
@@ -0,0 +1,99 @@
+/*
+ * Copyright (c) 2006 Chelsio, Inc. All rights reserved.
+ * Copyright (c) 2006 Open Grid Computing, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef _CHELSIO_DEFS_H
+#define _CHELSIO_DEFS_H
+
+#include <linux/skbuff.h>
+#include <net/tcp.h>
+
+#include "t3cdev.h"
+
+#include "cxgb3_offload.h"
+
+#define VALIDATE_TID 1
+
+void *cxgb_alloc_mem(unsigned long size);
+void cxgb_free_mem(void *addr);
+void cxgb_neigh_update(struct neighbour *neigh);
+void cxgb_redirect(struct dst_entry *old, struct dst_entry *new);
+
+/*
+ * Map an ATID or STID to their entries in the corresponding TID tables.
+ */
+static inline union active_open_entry *atid2entry(const struct tid_info *t,
+						  unsigned int atid)
+{
+	return &t->atid_tab[atid - t->atid_base];
+}
+
+static inline union listen_entry *stid2entry(const struct tid_info *t,
+					     unsigned int stid)
+{
+	return &t->stid_tab[stid - t->stid_base];
+}
+
+/*
+ * Find the connection corresponding to a TID.
+ */
+static inline struct t3c_tid_entry *lookup_tid(const struct tid_info *t,
+					       unsigned int tid)
+{
+	return tid < t->ntids ? &(t->tid_tab[tid]) : NULL;
+}
+
+/*
+ * Find the connection corresponding to a server TID.
+ */
+static inline struct t3c_tid_entry *lookup_stid(const struct tid_info *t,
+						unsigned int tid)
+{
+	if (tid < t->stid_base || tid >= t->stid_base + t->nstids)
+		return NULL;
+	return &(stid2entry(t, tid)->t3c_tid);
+}
+
+/*
+ * Find the connection corresponding to an active-open TID.
+ */
+static inline struct t3c_tid_entry *lookup_atid(const struct tid_info *t,
+						unsigned int tid)
+{
+	if (tid < t->atid_base || tid >= t->atid_base + t->natids)
+		return NULL;
+	return &(atid2entry(t, tid)->t3c_tid);
+}
+
+int process_rx(struct t3cdev *dev, struct sk_buff **skbs, int n);
+int attach_t3cdev(struct t3cdev *dev);
+void detach_t3cdev(struct t3cdev *dev);
+#endif
diff --git a/drivers/net/cxgb3/cxgb3_ioctl.h b/drivers/net/cxgb3/cxgb3_ioctl.h
new file mode 100644
index 00000000000..1ee77b28cdf
--- /dev/null
+++ b/drivers/net/cxgb3/cxgb3_ioctl.h
@@ -0,0 +1,165 @@
+/*
+ * This file is part of the Chelsio T3 Ethernet driver for Linux.
+ *
+ * Copyright (C) 2003-2006 Chelsio Communications.  All rights reserved.
+ *
+ * This program is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the LICENSE file included in this
+ * release for licensing terms and conditions.
+ */
+
+#ifndef __CHIOCTL_H__
+#define __CHIOCTL_H__
+
+/*
+ * Ioctl commands specific to this driver.
+ */
+enum {
+	CHELSIO_SETREG = 1024,
+	CHELSIO_GETREG,
+	CHELSIO_SETTPI,
+	CHELSIO_GETTPI,
+	CHELSIO_GETMTUTAB,
+	CHELSIO_SETMTUTAB,
+	CHELSIO_GETMTU,
+	CHELSIO_SET_PM,
+	CHELSIO_GET_PM,
+	CHELSIO_GET_TCAM,
+	CHELSIO_SET_TCAM,
+	CHELSIO_GET_TCB,
+	CHELSIO_GET_MEM,
+	CHELSIO_LOAD_FW,
+	CHELSIO_GET_PROTO,
+	CHELSIO_SET_PROTO,
+	CHELSIO_SET_TRACE_FILTER,
+	CHELSIO_SET_QSET_PARAMS,
+	CHELSIO_GET_QSET_PARAMS,
+	CHELSIO_SET_QSET_NUM,
+	CHELSIO_GET_QSET_NUM,
+	CHELSIO_SET_PKTSCHED,
+};
+
+struct ch_reg {
+	uint32_t cmd;
+	uint32_t addr;
+	uint32_t val;
+};
+
+struct ch_cntxt {
+	uint32_t cmd;
+	uint32_t cntxt_type;
+	uint32_t cntxt_id;
+	uint32_t data[4];
+};
+
+/* context types */
+enum { CNTXT_TYPE_EGRESS, CNTXT_TYPE_FL, CNTXT_TYPE_RSP, CNTXT_TYPE_CQ };
+
+struct ch_desc {
+	uint32_t cmd;
+	uint32_t queue_num;
+	uint32_t idx;
+	uint32_t size;
+	uint8_t data[128];
+};
+
+struct ch_mem_range {
+	uint32_t cmd;
+	uint32_t mem_id;
+	uint32_t addr;
+	uint32_t len;
+	uint32_t version;
+	uint8_t buf[0];
+};
+
+struct ch_qset_params {
+	uint32_t cmd;
+	uint32_t qset_idx;
+	int32_t txq_size[3];
+	int32_t rspq_size;
+	int32_t fl_size[2];
+	int32_t intr_lat;
+	int32_t polling;
+	int32_t cong_thres;
+};
+
+struct ch_pktsched_params {
+	uint32_t cmd;
+	uint8_t sched;
+	uint8_t idx;
+	uint8_t min;
+	uint8_t max;
+	uint8_t binding;
+};
+
+#ifndef TCB_SIZE
+# define TCB_SIZE   128
+#endif
+
+/* TCB size in 32-bit words */
+#define TCB_WORDS (TCB_SIZE / 4)
+
+enum { MEM_CM, MEM_PMRX, MEM_PMTX };	/* ch_mem_range.mem_id values */
+
+struct ch_mtus {
+	uint32_t cmd;
+	uint32_t nmtus;
+	uint16_t mtus[NMTUS];
+};
+
+struct ch_pm {
+	uint32_t cmd;
+	uint32_t tx_pg_sz;
+	uint32_t tx_num_pg;
+	uint32_t rx_pg_sz;
+	uint32_t rx_num_pg;
+	uint32_t pm_total;
+};
+
+struct ch_tcam {
+	uint32_t cmd;
+	uint32_t tcam_size;
+	uint32_t nservers;
+	uint32_t nroutes;
+	uint32_t nfilters;
+};
+
+struct ch_tcb {
+	uint32_t cmd;
+	uint32_t tcb_index;
+	uint32_t tcb_data[TCB_WORDS];
+};
+
+struct ch_tcam_word {
+	uint32_t cmd;
+	uint32_t addr;
+	uint32_t buf[3];
+};
+
+struct ch_trace {
+	uint32_t cmd;
+	uint32_t sip;
+	uint32_t sip_mask;
+	uint32_t dip;
+	uint32_t dip_mask;
+	uint16_t sport;
+	uint16_t sport_mask;
+	uint16_t dport;
+	uint16_t dport_mask;
+	uint32_t vlan:12;
+	uint32_t vlan_mask:12;
+	uint32_t intf:4;
+	uint32_t intf_mask:4;
+	uint8_t proto;
+	uint8_t proto_mask;
+	uint8_t invert_match:1;
+	uint8_t config_tx:1;
+	uint8_t config_rx:1;
+	uint8_t trace_tx:1;
+	uint8_t trace_rx:1;
+};
+
+#define SIOCCHIOCTL SIOCDEVPRIVATE
+
+#endif
diff --git a/drivers/net/cxgb3/cxgb3_main.c b/drivers/net/cxgb3/cxgb3_main.c
new file mode 100644
index 00000000000..54c49acd86b
--- /dev/null
+++ b/drivers/net/cxgb3/cxgb3_main.c
@@ -0,0 +1,2474 @@
+/*
+ * This file is part of the Chelsio T3 Ethernet driver for Linux.
+ *
+ * Copyright (C) 2003-2006 Chelsio Communications.  All rights reserved.
+ *
+ * This program is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the LICENSE file included in this
+ * release for licensing terms and conditions.
+ */
+
+#include <linux/module.h>
+#include <linux/moduleparam.h>
+#include <linux/init.h>
+#include <linux/pci.h>
+#include <linux/dma-mapping.h>
+#include <linux/netdevice.h>
+#include <linux/etherdevice.h>
+#include <linux/if_vlan.h>
+#include <linux/mii.h>
+#include <linux/sockios.h>
+#include <linux/workqueue.h>
+#include <linux/proc_fs.h>
+#include <linux/rtnetlink.h>
+#include <asm/uaccess.h>
+
+#include "common.h"
+#include "cxgb3_ioctl.h"
+#include "regs.h"
+#include "cxgb3_offload.h"
+#include "version.h"
+
+#include "cxgb3_ctl_defs.h"
+#include "t3_cpl.h"
+#include "firmware_exports.h"
+
+enum {
+	MAX_TXQ_ENTRIES = 16384,
+	MAX_CTRL_TXQ_ENTRIES = 1024,
+	MAX_RSPQ_ENTRIES = 16384,
+	MAX_RX_BUFFERS = 16384,
+	MAX_RX_JUMBO_BUFFERS = 16384,
+	MIN_TXQ_ENTRIES = 4,
+	MIN_CTRL_TXQ_ENTRIES = 4,
+	MIN_RSPQ_ENTRIES = 32,
+	MIN_FL_ENTRIES = 32
+};
+
+#define PORT_MASK ((1 << MAX_NPORTS) - 1)
+
+#define DFLT_MSG_ENABLE (NETIF_MSG_DRV | NETIF_MSG_PROBE | NETIF_MSG_LINK | \
+			 NETIF_MSG_TIMER | NETIF_MSG_IFDOWN | NETIF_MSG_IFUP |\
+			 NETIF_MSG_RX_ERR | NETIF_MSG_TX_ERR)
+
+#define EEPROM_MAGIC 0x38E2F10C
+
+#define to_net_dev(class) container_of(class, struct net_device, class_dev)
+
+#define CH_DEVICE(devid, ssid, idx) \
+	{ PCI_VENDOR_ID_CHELSIO, devid, PCI_ANY_ID, ssid, 0, 0, idx }
+
+static const struct pci_device_id cxgb3_pci_tbl[] = {
+	CH_DEVICE(0x20, 1, 0),	/* PE9000 */
+	CH_DEVICE(0x21, 1, 1),	/* T302E */
+	CH_DEVICE(0x22, 1, 2),	/* T310E */
+	CH_DEVICE(0x23, 1, 3),	/* T320X */
+	CH_DEVICE(0x24, 1, 1),	/* T302X */
+	CH_DEVICE(0x25, 1, 3),	/* T320E */
+	CH_DEVICE(0x26, 1, 2),	/* T310X */
+	CH_DEVICE(0x30, 1, 2),	/* T3B10 */
+	CH_DEVICE(0x31, 1, 3),	/* T3B20 */
+	CH_DEVICE(0x32, 1, 1),	/* T3B02 */
+	{0,}
+};
+
+MODULE_DESCRIPTION(DRV_DESC);
+MODULE_AUTHOR("Chelsio Communications");
+MODULE_LICENSE("GPL");
+MODULE_VERSION(DRV_VERSION);
+MODULE_DEVICE_TABLE(pci, cxgb3_pci_tbl);
+
+static int dflt_msg_enable = DFLT_MSG_ENABLE;
+
+module_param(dflt_msg_enable, int, 0644);
+MODULE_PARM_DESC(dflt_msg_enable, "Chelsio T3 default message enable bitmap");
+
+/*
+ * The driver uses the best interrupt scheme available on a platform in the
+ * order MSI-X, MSI, legacy pin interrupts.  This parameter determines which
+ * of these schemes the driver may consider as follows:
+ *
+ * msi = 2: choose from among all three options
+ * msi = 1: only consider MSI and pin interrupts
+ * msi = 0: force pin interrupts
+ */
+static int msi = 2;
+
+module_param(msi, int, 0644);
+MODULE_PARM_DESC(msi, "whether to use MSI or MSI-X");
+
+/*
+ * The driver enables offload as a default.
+ * To disable it, use ofld_disable = 1.
+ */
+
+static int ofld_disable = 0;
+
+module_param(ofld_disable, int, 0644);
+MODULE_PARM_DESC(ofld_disable, "whether to enable offload at init time or not");
+
+/*
+ * We have work elements that we need to cancel when an interface is taken
+ * down.  Normally the work elements would be executed by keventd but that
+ * can deadlock because of linkwatch.  If our close method takes the rtnl
+ * lock and linkwatch is ahead of our work elements in keventd, linkwatch
+ * will block keventd as it needs the rtnl lock, and we'll deadlock waiting
+ * for our work to complete.  Get our own work queue to solve this.
+ */
+static struct workqueue_struct *cxgb3_wq;
+
+/**
+ *	link_report - show link status and link speed/duplex
+ *	@p: the port whose settings are to be reported
+ *
+ *	Shows the link status, speed, and duplex of a port.
+ */
+static void link_report(struct net_device *dev)
+{
+	if (!netif_carrier_ok(dev))
+		printk(KERN_INFO "%s: link down\n", dev->name);
+	else {
+		const char *s = "10Mbps";
+		const struct port_info *p = netdev_priv(dev);
+
+		switch (p->link_config.speed) {
+		case SPEED_10000:
+			s = "10Gbps";
+			break;
+		case SPEED_1000:
+			s = "1000Mbps";
+			break;
+		case SPEED_100:
+			s = "100Mbps";
+			break;
+		}
+
+		printk(KERN_INFO "%s: link up, %s, %s-duplex\n", dev->name, s,
+		       p->link_config.duplex == DUPLEX_FULL ? "full" : "half");
+	}
+}
+
+/**
+ *	t3_os_link_changed - handle link status changes
+ *	@adapter: the adapter associated with the link change
+ *	@port_id: the port index whose limk status has changed
+ *	@link_stat: the new status of the link
+ *	@speed: the new speed setting
+ *	@duplex: the new duplex setting
+ *	@pause: the new flow-control setting
+ *
+ *	This is the OS-dependent handler for link status changes.  The OS
+ *	neutral handler takes care of most of the processing for these events,
+ *	then calls this handler for any OS-specific processing.
+ */
+void t3_os_link_changed(struct adapter *adapter, int port_id, int link_stat,
+			int speed, int duplex, int pause)
+{
+	struct net_device *dev = adapter->port[port_id];
+
+	/* Skip changes from disabled ports. */
+	if (!netif_running(dev))
+		return;
+
+	if (link_stat != netif_carrier_ok(dev)) {
+		if (link_stat)
+			netif_carrier_on(dev);
+		else
+			netif_carrier_off(dev);
+		link_report(dev);
+	}
+}
+
+static void cxgb_set_rxmode(struct net_device *dev)
+{
+	struct t3_rx_mode rm;
+	struct port_info *pi = netdev_priv(dev);
+
+	init_rx_mode(&rm, dev, dev->mc_list);
+	t3_mac_set_rx_mode(&pi->mac, &rm);
+}
+
+/**
+ *	link_start - enable a port
+ *	@dev: the device to enable
+ *
+ *	Performs the MAC and PHY actions needed to enable a port.
+ */
+static void link_start(struct net_device *dev)
+{
+	struct t3_rx_mode rm;
+	struct port_info *pi = netdev_priv(dev);
+	struct cmac *mac = &pi->mac;
+
+	init_rx_mode(&rm, dev, dev->mc_list);
+	t3_mac_reset(mac);
+	t3_mac_set_mtu(mac, dev->mtu);
+	t3_mac_set_address(mac, 0, dev->dev_addr);
+	t3_mac_set_rx_mode(mac, &rm);
+	t3_link_start(&pi->phy, mac, &pi->link_config);
+	t3_mac_enable(mac, MAC_DIRECTION_RX | MAC_DIRECTION_TX);
+}
+
+static inline void cxgb_disable_msi(struct adapter *adapter)
+{
+	if (adapter->flags & USING_MSIX) {
+		pci_disable_msix(adapter->pdev);
+		adapter->flags &= ~USING_MSIX;
+	} else if (adapter->flags & USING_MSI) {
+		pci_disable_msi(adapter->pdev);
+		adapter->flags &= ~USING_MSI;
+	}
+}
+
+/*
+ * Interrupt handler for asynchronous events used with MSI-X.
+ */
+static irqreturn_t t3_async_intr_handler(int irq, void *cookie)
+{
+	t3_slow_intr_handler(cookie);
+	return IRQ_HANDLED;
+}
+
+/*
+ * Name the MSI-X interrupts.
+ */
+static void name_msix_vecs(struct adapter *adap)
+{
+	int i, j, msi_idx = 1, n = sizeof(adap->msix_info[0].desc) - 1;
+
+	snprintf(adap->msix_info[0].desc, n, "%s", adap->name);
+	adap->msix_info[0].desc[n] = 0;
+
+	for_each_port(adap, j) {
+		struct net_device *d = adap->port[j];
+		const struct port_info *pi = netdev_priv(d);
+
+		for (i = 0; i < pi->nqsets; i++, msi_idx++) {
+			snprintf(adap->msix_info[msi_idx].desc, n,
+				 "%s (queue %d)", d->name, i);
+			adap->msix_info[msi_idx].desc[n] = 0;
+		}
+ 	}
+}
+
+static int request_msix_data_irqs(struct adapter *adap)
+{
+	int i, j, err, qidx = 0;
+
+	for_each_port(adap, i) {
+		int nqsets = adap2pinfo(adap, i)->nqsets;
+
+		for (j = 0; j < nqsets; ++j) {
+			err = request_irq(adap->msix_info[qidx + 1].vec,
+					  t3_intr_handler(adap,
+							  adap->sge.qs[qidx].
+							  rspq.polling), 0,
+					  adap->msix_info[qidx + 1].desc,
+					  &adap->sge.qs[qidx]);
+			if (err) {
+				while (--qidx >= 0)
+					free_irq(adap->msix_info[qidx + 1].vec,
+						 &adap->sge.qs[qidx]);
+				return err;
+			}
+			qidx++;
+		}
+	}
+	return 0;
+}
+
+/**
+ *	setup_rss - configure RSS
+ *	@adap: the adapter
+ *
+ *	Sets up RSS to distribute packets to multiple receive queues.  We
+ *	configure the RSS CPU lookup table to distribute to the number of HW
+ *	receive queues, and the response queue lookup table to narrow that
+ *	down to the response queues actually configured for each port.
+ *	We always configure the RSS mapping for two ports since the mapping
+ *	table has plenty of entries.
+ */
+static void setup_rss(struct adapter *adap)
+{
+	int i;
+	unsigned int nq0 = adap2pinfo(adap, 0)->nqsets;
+	unsigned int nq1 = adap->port[1] ? adap2pinfo(adap, 1)->nqsets : 1;
+	u8 cpus[SGE_QSETS + 1];
+	u16 rspq_map[RSS_TABLE_SIZE];
+
+	for (i = 0; i < SGE_QSETS; ++i)
+		cpus[i] = i;
+	cpus[SGE_QSETS] = 0xff;	/* terminator */
+
+	for (i = 0; i < RSS_TABLE_SIZE / 2; ++i) {
+		rspq_map[i] = i % nq0;
+		rspq_map[i + RSS_TABLE_SIZE / 2] = (i % nq1) + nq0;
+	}
+
+	t3_config_rss(adap, F_RQFEEDBACKENABLE | F_TNLLKPEN | F_TNLMAPEN |
+		      F_TNLPRTEN | F_TNL2TUPEN | F_TNL4TUPEN |
+		      V_RRCPLCPUSIZE(6), cpus, rspq_map);
+}
+
+/*
+ * If we have multiple receive queues per port serviced by NAPI we need one
+ * netdevice per queue as NAPI operates on netdevices.  We already have one
+ * netdevice, namely the one associated with the interface, so we use dummy
+ * ones for any additional queues.  Note that these netdevices exist purely
+ * so that NAPI has something to work with, they do not represent network
+ * ports and are not registered.
+ */
+static int init_dummy_netdevs(struct adapter *adap)
+{
+	int i, j, dummy_idx = 0;
+	struct net_device *nd;
+
+	for_each_port(adap, i) {
+		struct net_device *dev = adap->port[i];
+		const struct port_info *pi = netdev_priv(dev);
+
+		for (j = 0; j < pi->nqsets - 1; j++) {
+			if (!adap->dummy_netdev[dummy_idx]) {
+				nd = alloc_netdev(0, "", ether_setup);
+				if (!nd)
+					goto free_all;
+
+				nd->priv = adap;
+				nd->weight = 64;
+				set_bit(__LINK_STATE_START, &nd->state);
+				adap->dummy_netdev[dummy_idx] = nd;
+			}
+			strcpy(adap->dummy_netdev[dummy_idx]->name, dev->name);
+			dummy_idx++;
+		}
+	}
+	return 0;
+
+free_all:
+	while (--dummy_idx >= 0) {
+		free_netdev(adap->dummy_netdev[dummy_idx]);
+		adap->dummy_netdev[dummy_idx] = NULL;
+	}
+	return -ENOMEM;
+}
+
+/*
+ * Wait until all NAPI handlers are descheduled.  This includes the handlers of
+ * both netdevices representing interfaces and the dummy ones for the extra
+ * queues.
+ */
+static void quiesce_rx(struct adapter *adap)
+{
+	int i;
+	struct net_device *dev;
+
+	for_each_port(adap, i) {
+		dev = adap->port[i];
+		while (test_bit(__LINK_STATE_RX_SCHED, &dev->state))
+			msleep(1);
+	}
+
+	for (i = 0; i < ARRAY_SIZE(adap->dummy_netdev); i++) {
+		dev = adap->dummy_netdev[i];
+		if (dev)
+			while (test_bit(__LINK_STATE_RX_SCHED, &dev->state))
+				msleep(1);
+	}
+}
+
+/**
+ *	setup_sge_qsets - configure SGE Tx/Rx/response queues
+ *	@adap: the adapter
+ *
+ *	Determines how many sets of SGE queues to use and initializes them.
+ *	We support multiple queue sets per port if we have MSI-X, otherwise
+ *	just one queue set per port.
+ */
+static int setup_sge_qsets(struct adapter *adap)
+{
+	int i, j, err, irq_idx = 0, qset_idx = 0, dummy_dev_idx = 0;
+	unsigned int ntxq = is_offload(adap) ? SGE_TXQ_PER_SET : 1;
+
+	if (adap->params.rev > 0 && !(adap->flags & USING_MSI))
+		irq_idx = -1;
+
+	for_each_port(adap, i) {
+		struct net_device *dev = adap->port[i];
+		const struct port_info *pi = netdev_priv(dev);
+
+		for (j = 0; j < pi->nqsets; ++j, ++qset_idx) {
+			err = t3_sge_alloc_qset(adap, qset_idx, 1,
+				(adap->flags & USING_MSIX) ? qset_idx + 1 :
+							     irq_idx,
+				&adap->params.sge.qset[qset_idx], ntxq,
+				j == 0 ? dev :
+					 adap-> dummy_netdev[dummy_dev_idx++]);
+			if (err) {
+				t3_free_sge_resources(adap);
+				return err;
+			}
+		}
+	}
+
+	return 0;
+}
+
+static ssize_t attr_show(struct class_device *cd, char *buf,
+			 ssize_t(*format) (struct adapter *, char *))
+{
+	ssize_t len;
+	struct adapter *adap = to_net_dev(cd)->priv;
+
+	/* Synchronize with ioctls that may shut down the device */
+	rtnl_lock();
+	len = (*format) (adap, buf);
+	rtnl_unlock();
+	return len;
+}
+
+static ssize_t attr_store(struct class_device *cd, const char *buf, size_t len,
+			  ssize_t(*set) (struct adapter *, unsigned int),
+			  unsigned int min_val, unsigned int max_val)
+{
+	char *endp;
+	ssize_t ret;
+	unsigned int val;
+	struct adapter *adap = to_net_dev(cd)->priv;
+
+	if (!capable(CAP_NET_ADMIN))
+		return -EPERM;
+
+	val = simple_strtoul(buf, &endp, 0);
+	if (endp == buf || val < min_val || val > max_val)
+		return -EINVAL;
+
+	rtnl_lock();
+	ret = (*set) (adap, val);
+	if (!ret)
+		ret = len;
+	rtnl_unlock();
+	return ret;
+}
+
+#define CXGB3_SHOW(name, val_expr) \
+static ssize_t format_##name(struct adapter *adap, char *buf) \
+{ \
+	return sprintf(buf, "%u\n", val_expr); \
+} \
+static ssize_t show_##name(struct class_device *cd, char *buf) \
+{ \
+	return attr_show(cd, buf, format_##name); \
+}
+
+static ssize_t set_nfilters(struct adapter *adap, unsigned int val)
+{
+	if (adap->flags & FULL_INIT_DONE)
+		return -EBUSY;
+	if (val && adap->params.rev == 0)
+		return -EINVAL;
+	if (val > t3_mc5_size(&adap->mc5) - adap->params.mc5.nservers)
+		return -EINVAL;
+	adap->params.mc5.nfilters = val;
+	return 0;
+}
+
+static ssize_t store_nfilters(struct class_device *cd, const char *buf,
+			      size_t len)
+{
+	return attr_store(cd, buf, len, set_nfilters, 0, ~0);
+}
+
+static ssize_t set_nservers(struct adapter *adap, unsigned int val)
+{
+	if (adap->flags & FULL_INIT_DONE)
+		return -EBUSY;
+	if (val > t3_mc5_size(&adap->mc5) - adap->params.mc5.nfilters)
+		return -EINVAL;
+	adap->params.mc5.nservers = val;
+	return 0;
+}
+
+static ssize_t store_nservers(struct class_device *cd, const char *buf,
+			      size_t len)
+{
+	return attr_store(cd, buf, len, set_nservers, 0, ~0);
+}
+
+#define CXGB3_ATTR_R(name, val_expr) \
+CXGB3_SHOW(name, val_expr) \
+static CLASS_DEVICE_ATTR(name, S_IRUGO, show_##name, NULL)
+
+#define CXGB3_ATTR_RW(name, val_expr, store_method) \
+CXGB3_SHOW(name, val_expr) \
+static CLASS_DEVICE_ATTR(name, S_IRUGO | S_IWUSR, show_##name, store_method)
+
+CXGB3_ATTR_R(cam_size, t3_mc5_size(&adap->mc5));
+CXGB3_ATTR_RW(nfilters, adap->params.mc5.nfilters, store_nfilters);
+CXGB3_ATTR_RW(nservers, adap->params.mc5.nservers, store_nservers);
+
+static struct attribute *cxgb3_attrs[] = {
+	&class_device_attr_cam_size.attr,
+	&class_device_attr_nfilters.attr,
+	&class_device_attr_nservers.attr,
+	NULL
+};
+
+static struct attribute_group cxgb3_attr_group = {.attrs = cxgb3_attrs };
+
+static ssize_t tm_attr_show(struct class_device *cd, char *buf, int sched)
+{
+	ssize_t len;
+	unsigned int v, addr, bpt, cpt;
+	struct adapter *adap = to_net_dev(cd)->priv;
+
+	addr = A_TP_TX_MOD_Q1_Q0_RATE_LIMIT - sched / 2;
+	rtnl_lock();
+	t3_write_reg(adap, A_TP_TM_PIO_ADDR, addr);
+	v = t3_read_reg(adap, A_TP_TM_PIO_DATA);
+	if (sched & 1)
+		v >>= 16;
+	bpt = (v >> 8) & 0xff;
+	cpt = v & 0xff;
+	if (!cpt)
+		len = sprintf(buf, "disabled\n");
+	else {
+		v = (adap->params.vpd.cclk * 1000) / cpt;
+		len = sprintf(buf, "%u Kbps\n", (v * bpt) / 125);
+	}
+	rtnl_unlock();
+	return len;
+}
+
+static ssize_t tm_attr_store(struct class_device *cd, const char *buf,
+			     size_t len, int sched)
+{
+	char *endp;
+	ssize_t ret;
+	unsigned int val;
+	struct adapter *adap = to_net_dev(cd)->priv;
+
+	if (!capable(CAP_NET_ADMIN))
+		return -EPERM;
+
+	val = simple_strtoul(buf, &endp, 0);
+	if (endp == buf || val > 10000000)
+		return -EINVAL;
+
+	rtnl_lock();
+	ret = t3_config_sched(adap, val, sched);
+	if (!ret)
+		ret = len;
+	rtnl_unlock();
+	return ret;
+}
+
+#define TM_ATTR(name, sched) \
+static ssize_t show_##name(struct class_device *cd, char *buf) \
+{ \
+	return tm_attr_show(cd, buf, sched); \
+} \
+static ssize_t store_##name(struct class_device *cd, const char *buf, size_t len) \
+{ \
+	return tm_attr_store(cd, buf, len, sched); \
+} \
+static CLASS_DEVICE_ATTR(name, S_IRUGO | S_IWUSR, show_##name, store_##name)
+
+TM_ATTR(sched0, 0);
+TM_ATTR(sched1, 1);
+TM_ATTR(sched2, 2);
+TM_ATTR(sched3, 3);
+TM_ATTR(sched4, 4);
+TM_ATTR(sched5, 5);
+TM_ATTR(sched6, 6);
+TM_ATTR(sched7, 7);
+
+static struct attribute *offload_attrs[] = {
+	&class_device_attr_sched0.attr,
+	&class_device_attr_sched1.attr,
+	&class_device_attr_sched2.attr,
+	&class_device_attr_sched3.attr,
+	&class_device_attr_sched4.attr,
+	&class_device_attr_sched5.attr,
+	&class_device_attr_sched6.attr,
+	&class_device_attr_sched7.attr,
+	NULL
+};
+
+static struct attribute_group offload_attr_group = {.attrs = offload_attrs };
+
+/*
+ * Sends an sk_buff to an offload queue driver
+ * after dealing with any active network taps.
+ */
+static inline int offload_tx(struct t3cdev *tdev, struct sk_buff *skb)
+{
+	int ret;
+
+	local_bh_disable();
+	ret = t3_offload_tx(tdev, skb);
+	local_bh_enable();
+	return ret;
+}
+
+static int write_smt_entry(struct adapter *adapter, int idx)
+{
+	struct cpl_smt_write_req *req;
+	struct sk_buff *skb = alloc_skb(sizeof(*req), GFP_KERNEL);
+
+	if (!skb)
+		return -ENOMEM;
+
+	req = (struct cpl_smt_write_req *)__skb_put(skb, sizeof(*req));
+	req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
+	OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_SMT_WRITE_REQ, idx));
+	req->mtu_idx = NMTUS - 1;	/* should be 0 but there's a T3 bug */
+	req->iff = idx;
+	memset(req->src_mac1, 0, sizeof(req->src_mac1));
+	memcpy(req->src_mac0, adapter->port[idx]->dev_addr, ETH_ALEN);
+	skb->priority = 1;
+	offload_tx(&adapter->tdev, skb);
+	return 0;
+}
+
+static int init_smt(struct adapter *adapter)
+{
+	int i;
+
+	for_each_port(adapter, i)
+	    write_smt_entry(adapter, i);
+	return 0;
+}
+
+static void init_port_mtus(struct adapter *adapter)
+{
+	unsigned int mtus = adapter->port[0]->mtu;
+
+	if (adapter->port[1])
+		mtus |= adapter->port[1]->mtu << 16;
+	t3_write_reg(adapter, A_TP_MTU_PORT_TABLE, mtus);
+}
+
+/**
+ *	cxgb_up - enable the adapter
+ *	@adapter: adapter being enabled
+ *
+ *	Called when the first port is enabled, this function performs the
+ *	actions necessary to make an adapter operational, such as completing
+ *	the initialization of HW modules, and enabling interrupts.
+ *
+ *	Must be called with the rtnl lock held.
+ */
+static int cxgb_up(struct adapter *adap)
+{
+	int err = 0;
+
+	if (!(adap->flags & FULL_INIT_DONE)) {
+		err = t3_check_fw_version(adap);
+		if (err) {
+			dev_err(&adap->pdev->dev,
+				"adapter FW is not compatible with driver\n");
+			goto out;
+		}
+
+		err = init_dummy_netdevs(adap);
+		if (err)
+			goto out;
+
+		err = t3_init_hw(adap, 0);
+		if (err)
+			goto out;
+
+		err = setup_sge_qsets(adap);
+		if (err)
+			goto out;
+
+		setup_rss(adap);
+		adap->flags |= FULL_INIT_DONE;
+	}
+
+	t3_intr_clear(adap);
+
+	if (adap->flags & USING_MSIX) {
+		name_msix_vecs(adap);
+		err = request_irq(adap->msix_info[0].vec,
+				  t3_async_intr_handler, 0,
+				  adap->msix_info[0].desc, adap);
+		if (err)
+			goto irq_err;
+
+		if (request_msix_data_irqs(adap)) {
+			free_irq(adap->msix_info[0].vec, adap);
+			goto irq_err;
+		}
+	} else if ((err = request_irq(adap->pdev->irq,
+				      t3_intr_handler(adap,
+						      adap->sge.qs[0].rspq.
+						      polling),
+				      (adap->flags & USING_MSI) ? 0 : SA_SHIRQ,
+				      adap->name, adap)))
+		goto irq_err;
+
+	t3_sge_start(adap);
+	t3_intr_enable(adap);
+out:
+	return err;
+irq_err:
+	CH_ERR(adap, "request_irq failed, err %d\n", err);
+	goto out;
+}
+
+/*
+ * Release resources when all the ports and offloading have been stopped.
+ */
+static void cxgb_down(struct adapter *adapter)
+{
+	t3_sge_stop(adapter);
+	spin_lock_irq(&adapter->work_lock);	/* sync with PHY intr task */
+	t3_intr_disable(adapter);
+	spin_unlock_irq(&adapter->work_lock);
+
+	if (adapter->flags & USING_MSIX) {
+		int i, n = 0;
+
+		free_irq(adapter->msix_info[0].vec, adapter);
+		for_each_port(adapter, i)
+		    n += adap2pinfo(adapter, i)->nqsets;
+
+		for (i = 0; i < n; ++i)
+			free_irq(adapter->msix_info[i + 1].vec,
+				 &adapter->sge.qs[i]);
+	} else
+		free_irq(adapter->pdev->irq, adapter);
+
+	flush_workqueue(cxgb3_wq);	/* wait for external IRQ handler */
+	quiesce_rx(adapter);
+}
+
+static void schedule_chk_task(struct adapter *adap)
+{
+	unsigned int timeo;
+
+	timeo = adap->params.linkpoll_period ?
+	    (HZ * adap->params.linkpoll_period) / 10 :
+	    adap->params.stats_update_period * HZ;
+	if (timeo)
+		queue_delayed_work(cxgb3_wq, &adap->adap_check_task, timeo);
+}
+
+static int offload_open(struct net_device *dev)
+{
+	struct adapter *adapter = dev->priv;
+	struct t3cdev *tdev = T3CDEV(dev);
+	int adap_up = adapter->open_device_map & PORT_MASK;
+	int err = 0;
+
+	if (test_and_set_bit(OFFLOAD_DEVMAP_BIT, &adapter->open_device_map))
+		return 0;
+
+	if (!adap_up && (err = cxgb_up(adapter)) < 0)
+		return err;
+
+	t3_tp_set_offload_mode(adapter, 1);
+	tdev->lldev = adapter->port[0];
+	err = cxgb3_offload_activate(adapter);
+	if (err)
+		goto out;
+
+	init_port_mtus(adapter);
+	t3_load_mtus(adapter, adapter->params.mtus, adapter->params.a_wnd,
+		     adapter->params.b_wnd,
+		     adapter->params.rev == 0 ?
+		     adapter->port[0]->mtu : 0xffff);
+	init_smt(adapter);
+
+	/* Never mind if the next step fails */
+	sysfs_create_group(&tdev->lldev->class_dev.kobj, &offload_attr_group);
+
+	/* Call back all registered clients */
+	cxgb3_add_clients(tdev);
+
+out:
+	/* restore them in case the offload module has changed them */
+	if (err) {
+		t3_tp_set_offload_mode(adapter, 0);
+		clear_bit(OFFLOAD_DEVMAP_BIT, &adapter->open_device_map);
+		cxgb3_set_dummy_ops(tdev);
+	}
+	return err;
+}
+
+static int offload_close(struct t3cdev *tdev)
+{
+	struct adapter *adapter = tdev2adap(tdev);
+
+	if (!test_bit(OFFLOAD_DEVMAP_BIT, &adapter->open_device_map))
+		return 0;
+
+	/* Call back all registered clients */
+	cxgb3_remove_clients(tdev);
+
+	sysfs_remove_group(&tdev->lldev->class_dev.kobj, &offload_attr_group);
+
+	tdev->lldev = NULL;
+	cxgb3_set_dummy_ops(tdev);
+	t3_tp_set_offload_mode(adapter, 0);
+	clear_bit(OFFLOAD_DEVMAP_BIT, &adapter->open_device_map);
+
+	if (!adapter->open_device_map)
+		cxgb_down(adapter);
+
+	cxgb3_offload_deactivate(adapter);
+	return 0;
+}
+
+static int cxgb_open(struct net_device *dev)
+{
+	int err;
+	struct adapter *adapter = dev->priv;
+	struct port_info *pi = netdev_priv(dev);
+	int other_ports = adapter->open_device_map & PORT_MASK;
+
+	if (!adapter->open_device_map && (err = cxgb_up(adapter)) < 0)
+		return err;
+
+	set_bit(pi->port_id, &adapter->open_device_map);
+	if (!ofld_disable) {
+		err = offload_open(dev);
+		if (err)
+			printk(KERN_WARNING
+			       "Could not initialize offload capabilities\n");
+	}
+
+	link_start(dev);
+	t3_port_intr_enable(adapter, pi->port_id);
+	netif_start_queue(dev);
+	if (!other_ports)
+		schedule_chk_task(adapter);
+
+	return 0;
+}
+
+static int cxgb_close(struct net_device *dev)
+{
+	struct adapter *adapter = dev->priv;
+	struct port_info *p = netdev_priv(dev);
+
+	t3_port_intr_disable(adapter, p->port_id);
+	netif_stop_queue(dev);
+	p->phy.ops->power_down(&p->phy, 1);
+	netif_carrier_off(dev);
+	t3_mac_disable(&p->mac, MAC_DIRECTION_TX | MAC_DIRECTION_RX);
+
+	spin_lock(&adapter->work_lock);	/* sync with update task */
+	clear_bit(p->port_id, &adapter->open_device_map);
+	spin_unlock(&adapter->work_lock);
+
+	if (!(adapter->open_device_map & PORT_MASK))
+		cancel_rearming_delayed_workqueue(cxgb3_wq,
+						  &adapter->adap_check_task);
+
+	if (!adapter->open_device_map)
+		cxgb_down(adapter);
+
+	return 0;
+}
+
+static struct net_device_stats *cxgb_get_stats(struct net_device *dev)
+{
+	struct adapter *adapter = dev->priv;
+	struct port_info *p = netdev_priv(dev);
+	struct net_device_stats *ns = &p->netstats;
+	const struct mac_stats *pstats;
+
+	spin_lock(&adapter->stats_lock);
+	pstats = t3_mac_update_stats(&p->mac);
+	spin_unlock(&adapter->stats_lock);
+
+	ns->tx_bytes = pstats->tx_octets;
+	ns->tx_packets = pstats->tx_frames;
+	ns->rx_bytes = pstats->rx_octets;
+	ns->rx_packets = pstats->rx_frames;
+	ns->multicast = pstats->rx_mcast_frames;
+
+	ns->tx_errors = pstats->tx_underrun;
+	ns->rx_errors = pstats->rx_symbol_errs + pstats->rx_fcs_errs +
+	    pstats->rx_too_long + pstats->rx_jabber + pstats->rx_short +
+	    pstats->rx_fifo_ovfl;
+
+	/* detailed rx_errors */
+	ns->rx_length_errors = pstats->rx_jabber + pstats->rx_too_long;
+	ns->rx_over_errors = 0;
+	ns->rx_crc_errors = pstats->rx_fcs_errs;
+	ns->rx_frame_errors = pstats->rx_symbol_errs;
+	ns->rx_fifo_errors = pstats->rx_fifo_ovfl;
+	ns->rx_missed_errors = pstats->rx_cong_drops;
+
+	/* detailed tx_errors */
+	ns->tx_aborted_errors = 0;
+	ns->tx_carrier_errors = 0;
+	ns->tx_fifo_errors = pstats->tx_underrun;
+	ns->tx_heartbeat_errors = 0;
+	ns->tx_window_errors = 0;
+	return ns;
+}
+
+static u32 get_msglevel(struct net_device *dev)
+{
+	struct adapter *adapter = dev->priv;
+
+	return adapter->msg_enable;
+}
+
+static void set_msglevel(struct net_device *dev, u32 val)
+{
+	struct adapter *adapter = dev->priv;
+
+	adapter->msg_enable = val;
+}
+
+static char stats_strings[][ETH_GSTRING_LEN] = {
+	"TxOctetsOK         ",
+	"TxFramesOK         ",
+	"TxMulticastFramesOK",
+	"TxBroadcastFramesOK",
+	"TxPauseFrames      ",
+	"TxUnderrun         ",
+	"TxExtUnderrun      ",
+
+	"TxFrames64         ",
+	"TxFrames65To127    ",
+	"TxFrames128To255   ",
+	"TxFrames256To511   ",
+	"TxFrames512To1023  ",
+	"TxFrames1024To1518 ",
+	"TxFrames1519ToMax  ",
+
+	"RxOctetsOK         ",
+	"RxFramesOK         ",
+	"RxMulticastFramesOK",
+	"RxBroadcastFramesOK",
+	"RxPauseFrames      ",
+	"RxFCSErrors        ",
+	"RxSymbolErrors     ",
+	"RxShortErrors      ",
+	"RxJabberErrors     ",
+	"RxLengthErrors     ",
+	"RxFIFOoverflow     ",
+
+	"RxFrames64         ",
+	"RxFrames65To127    ",
+	"RxFrames128To255   ",
+	"RxFrames256To511   ",
+	"RxFrames512To1023  ",
+	"RxFrames1024To1518 ",
+	"RxFrames1519ToMax  ",
+
+	"PhyFIFOErrors      ",
+	"TSO                ",
+	"VLANextractions    ",
+	"VLANinsertions     ",
+	"TxCsumOffload      ",
+	"RxCsumGood         ",
+	"RxDrops            "
+};
+
+static int get_stats_count(struct net_device *dev)
+{
+	return ARRAY_SIZE(stats_strings);
+}
+
+#define T3_REGMAP_SIZE (3 * 1024)
+
+static int get_regs_len(struct net_device *dev)
+{
+	return T3_REGMAP_SIZE;
+}
+
+static int get_eeprom_len(struct net_device *dev)
+{
+	return EEPROMSIZE;
+}
+
+static void get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
+{
+	u32 fw_vers = 0;
+	struct adapter *adapter = dev->priv;
+
+	t3_get_fw_version(adapter, &fw_vers);
+
+	strcpy(info->driver, DRV_NAME);
+	strcpy(info->version, DRV_VERSION);
+	strcpy(info->bus_info, pci_name(adapter->pdev));
+	if (!fw_vers)
+		strcpy(info->fw_version, "N/A");
+	else
+		snprintf(info->fw_version, sizeof(info->fw_version),
+			 "%s %u.%u", (fw_vers >> 24) ? "T" : "N",
+			 (fw_vers >> 12) & 0xfff, fw_vers & 0xfff);
+}
+
+static void get_strings(struct net_device *dev, u32 stringset, u8 * data)
+{
+	if (stringset == ETH_SS_STATS)
+		memcpy(data, stats_strings, sizeof(stats_strings));
+}
+
+static unsigned long collect_sge_port_stats(struct adapter *adapter,
+					    struct port_info *p, int idx)
+{
+	int i;
+	unsigned long tot = 0;
+
+	for (i = 0; i < p->nqsets; ++i)
+		tot += adapter->sge.qs[i + p->first_qset].port_stats[idx];
+	return tot;
+}
+
+static void get_stats(struct net_device *dev, struct ethtool_stats *stats,
+		      u64 *data)
+{
+	struct adapter *adapter = dev->priv;
+	struct port_info *pi = netdev_priv(dev);
+	const struct mac_stats *s;
+
+	spin_lock(&adapter->stats_lock);
+	s = t3_mac_update_stats(&pi->mac);
+	spin_unlock(&adapter->stats_lock);
+
+	*data++ = s->tx_octets;
+	*data++ = s->tx_frames;
+	*data++ = s->tx_mcast_frames;
+	*data++ = s->tx_bcast_frames;
+	*data++ = s->tx_pause;
+	*data++ = s->tx_underrun;
+	*data++ = s->tx_fifo_urun;
+
+	*data++ = s->tx_frames_64;
+	*data++ = s->tx_frames_65_127;
+	*data++ = s->tx_frames_128_255;
+	*data++ = s->tx_frames_256_511;
+	*data++ = s->tx_frames_512_1023;
+	*data++ = s->tx_frames_1024_1518;
+	*data++ = s->tx_frames_1519_max;
+
+	*data++ = s->rx_octets;
+	*data++ = s->rx_frames;
+	*data++ = s->rx_mcast_frames;
+	*data++ = s->rx_bcast_frames;
+	*data++ = s->rx_pause;
+	*data++ = s->rx_fcs_errs;
+	*data++ = s->rx_symbol_errs;
+	*data++ = s->rx_short;
+	*data++ = s->rx_jabber;
+	*data++ = s->rx_too_long;
+	*data++ = s->rx_fifo_ovfl;
+
+	*data++ = s->rx_frames_64;
+	*data++ = s->rx_frames_65_127;
+	*data++ = s->rx_frames_128_255;
+	*data++ = s->rx_frames_256_511;
+	*data++ = s->rx_frames_512_1023;
+	*data++ = s->rx_frames_1024_1518;
+	*data++ = s->rx_frames_1519_max;
+
+	*data++ = pi->phy.fifo_errors;
+
+	*data++ = collect_sge_port_stats(adapter, pi, SGE_PSTAT_TSO);
+	*data++ = collect_sge_port_stats(adapter, pi, SGE_PSTAT_VLANEX);
+	*data++ = collect_sge_port_stats(adapter, pi, SGE_PSTAT_VLANINS);
+	*data++ = collect_sge_port_stats(adapter, pi, SGE_PSTAT_TX_CSUM);
+	*data++ = collect_sge_port_stats(adapter, pi, SGE_PSTAT_RX_CSUM_GOOD);
+	*data++ = s->rx_cong_drops;
+}
+
+static inline void reg_block_dump(struct adapter *ap, void *buf,
+				  unsigned int start, unsigned int end)
+{
+	u32 *p = buf + start;
+
+	for (; start <= end; start += sizeof(u32))
+		*p++ = t3_read_reg(ap, start);
+}
+
+static void get_regs(struct net_device *dev, struct ethtool_regs *regs,
+		     void *buf)
+{
+	struct adapter *ap = dev->priv;
+
+	/*
+	 * Version scheme:
+	 * bits 0..9: chip version
+	 * bits 10..15: chip revision
+	 * bit 31: set for PCIe cards
+	 */
+	regs->version = 3 | (ap->params.rev << 10) | (is_pcie(ap) << 31);
+
+	/*
+	 * We skip the MAC statistics registers because they are clear-on-read.
+	 * Also reading multi-register stats would need to synchronize with the
+	 * periodic mac stats accumulation.  Hard to justify the complexity.
+	 */
+	memset(buf, 0, T3_REGMAP_SIZE);
+	reg_block_dump(ap, buf, 0, A_SG_RSPQ_CREDIT_RETURN);
+	reg_block_dump(ap, buf, A_SG_HI_DRB_HI_THRSH, A_ULPRX_PBL_ULIMIT);
+	reg_block_dump(ap, buf, A_ULPTX_CONFIG, A_MPS_INT_CAUSE);
+	reg_block_dump(ap, buf, A_CPL_SWITCH_CNTRL, A_CPL_MAP_TBL_DATA);
+	reg_block_dump(ap, buf, A_SMB_GLOBAL_TIME_CFG, A_XGM_SERDES_STAT3);
+	reg_block_dump(ap, buf, A_XGM_SERDES_STATUS0,
+		       XGM_REG(A_XGM_SERDES_STAT3, 1));
+	reg_block_dump(ap, buf, XGM_REG(A_XGM_SERDES_STATUS0, 1),
+		       XGM_REG(A_XGM_RX_SPI4_SOP_EOP_CNT, 1));
+}
+
+static int restart_autoneg(struct net_device *dev)
+{
+	struct port_info *p = netdev_priv(dev);
+
+	if (!netif_running(dev))
+		return -EAGAIN;
+	if (p->link_config.autoneg != AUTONEG_ENABLE)
+		return -EINVAL;
+	p->phy.ops->autoneg_restart(&p->phy);
+	return 0;
+}
+
+static int cxgb3_phys_id(struct net_device *dev, u32 data)
+{
+	int i;
+	struct adapter *adapter = dev->priv;
+
+	if (data == 0)
+		data = 2;
+
+	for (i = 0; i < data * 2; i++) {
+		t3_set_reg_field(adapter, A_T3DBG_GPIO_EN, F_GPIO0_OUT_VAL,
+				 (i & 1) ? F_GPIO0_OUT_VAL : 0);
+		if (msleep_interruptible(500))
+			break;
+	}
+	t3_set_reg_field(adapter, A_T3DBG_GPIO_EN, F_GPIO0_OUT_VAL,
+			 F_GPIO0_OUT_VAL);
+	return 0;
+}
+
+static int get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
+{
+	struct port_info *p = netdev_priv(dev);
+
+	cmd->supported = p->link_config.supported;
+	cmd->advertising = p->link_config.advertising;
+
+	if (netif_carrier_ok(dev)) {
+		cmd->speed = p->link_config.speed;
+		cmd->duplex = p->link_config.duplex;
+	} else {
+		cmd->speed = -1;
+		cmd->duplex = -1;
+	}
+
+	cmd->port = (cmd->supported & SUPPORTED_TP) ? PORT_TP : PORT_FIBRE;
+	cmd->phy_address = p->phy.addr;
+	cmd->transceiver = XCVR_EXTERNAL;
+	cmd->autoneg = p->link_config.autoneg;
+	cmd->maxtxpkt = 0;
+	cmd->maxrxpkt = 0;
+	return 0;
+}
+
+static int speed_duplex_to_caps(int speed, int duplex)
+{
+	int cap = 0;
+
+	switch (speed) {
+	case SPEED_10:
+		if (duplex == DUPLEX_FULL)
+			cap = SUPPORTED_10baseT_Full;
+		else
+			cap = SUPPORTED_10baseT_Half;
+		break;
+	case SPEED_100:
+		if (duplex == DUPLEX_FULL)
+			cap = SUPPORTED_100baseT_Full;
+		else
+			cap = SUPPORTED_100baseT_Half;
+		break;
+	case SPEED_1000:
+		if (duplex == DUPLEX_FULL)
+			cap = SUPPORTED_1000baseT_Full;
+		else
+			cap = SUPPORTED_1000baseT_Half;
+		break;
+	case SPEED_10000:
+		if (duplex == DUPLEX_FULL)
+			cap = SUPPORTED_10000baseT_Full;
+	}
+	return cap;
+}
+
+#define ADVERTISED_MASK (ADVERTISED_10baseT_Half | ADVERTISED_10baseT_Full | \
+		      ADVERTISED_100baseT_Half | ADVERTISED_100baseT_Full | \
+		      ADVERTISED_1000baseT_Half | ADVERTISED_1000baseT_Full | \
+		      ADVERTISED_10000baseT_Full)
+
+static int set_settings(struct net_device *dev, struct ethtool_cmd *cmd)
+{
+	struct port_info *p = netdev_priv(dev);
+	struct link_config *lc = &p->link_config;
+
+	if (!(lc->supported & SUPPORTED_Autoneg))
+		return -EOPNOTSUPP;	/* can't change speed/duplex */
+
+	if (cmd->autoneg == AUTONEG_DISABLE) {
+		int cap = speed_duplex_to_caps(cmd->speed, cmd->duplex);
+
+		if (!(lc->supported & cap) || cmd->speed == SPEED_1000)
+			return -EINVAL;
+		lc->requested_speed = cmd->speed;
+		lc->requested_duplex = cmd->duplex;
+		lc->advertising = 0;
+	} else {
+		cmd->advertising &= ADVERTISED_MASK;
+		cmd->advertising &= lc->supported;
+		if (!cmd->advertising)
+			return -EINVAL;
+		lc->requested_speed = SPEED_INVALID;
+		lc->requested_duplex = DUPLEX_INVALID;
+		lc->advertising = cmd->advertising | ADVERTISED_Autoneg;
+	}
+	lc->autoneg = cmd->autoneg;
+	if (netif_running(dev))
+		t3_link_start(&p->phy, &p->mac, lc);
+	return 0;
+}
+
+static void get_pauseparam(struct net_device *dev,
+			   struct ethtool_pauseparam *epause)
+{
+	struct port_info *p = netdev_priv(dev);
+
+	epause->autoneg = (p->link_config.requested_fc & PAUSE_AUTONEG) != 0;
+	epause->rx_pause = (p->link_config.fc & PAUSE_RX) != 0;
+	epause->tx_pause = (p->link_config.fc & PAUSE_TX) != 0;
+}
+
+static int set_pauseparam(struct net_device *dev,
+			  struct ethtool_pauseparam *epause)
+{
+	struct port_info *p = netdev_priv(dev);
+	struct link_config *lc = &p->link_config;
+
+	if (epause->autoneg == AUTONEG_DISABLE)
+		lc->requested_fc = 0;
+	else if (lc->supported & SUPPORTED_Autoneg)
+		lc->requested_fc = PAUSE_AUTONEG;
+	else
+		return -EINVAL;
+
+	if (epause->rx_pause)
+		lc->requested_fc |= PAUSE_RX;
+	if (epause->tx_pause)
+		lc->requested_fc |= PAUSE_TX;
+	if (lc->autoneg == AUTONEG_ENABLE) {
+		if (netif_running(dev))
+			t3_link_start(&p->phy, &p->mac, lc);
+	} else {
+		lc->fc = lc->requested_fc & (PAUSE_RX | PAUSE_TX);
+		if (netif_running(dev))
+			t3_mac_set_speed_duplex_fc(&p->mac, -1, -1, lc->fc);
+	}
+	return 0;
+}
+
+static u32 get_rx_csum(struct net_device *dev)
+{
+	struct port_info *p = netdev_priv(dev);
+
+	return p->rx_csum_offload;
+}
+
+static int set_rx_csum(struct net_device *dev, u32 data)
+{
+	struct port_info *p = netdev_priv(dev);
+
+	p->rx_csum_offload = data;
+	return 0;
+}
+
+static void get_sge_param(struct net_device *dev, struct ethtool_ringparam *e)
+{
+	struct adapter *adapter = dev->priv;
+
+	e->rx_max_pending = MAX_RX_BUFFERS;
+	e->rx_mini_max_pending = 0;
+	e->rx_jumbo_max_pending = MAX_RX_JUMBO_BUFFERS;
+	e->tx_max_pending = MAX_TXQ_ENTRIES;
+
+	e->rx_pending = adapter->params.sge.qset[0].fl_size;
+	e->rx_mini_pending = adapter->params.sge.qset[0].rspq_size;
+	e->rx_jumbo_pending = adapter->params.sge.qset[0].jumbo_size;
+	e->tx_pending = adapter->params.sge.qset[0].txq_size[0];
+}
+
+static int set_sge_param(struct net_device *dev, struct ethtool_ringparam *e)
+{
+	int i;
+	struct adapter *adapter = dev->priv;
+
+	if (e->rx_pending > MAX_RX_BUFFERS ||
+	    e->rx_jumbo_pending > MAX_RX_JUMBO_BUFFERS ||
+	    e->tx_pending > MAX_TXQ_ENTRIES ||
+	    e->rx_mini_pending > MAX_RSPQ_ENTRIES ||
+	    e->rx_mini_pending < MIN_RSPQ_ENTRIES ||
+	    e->rx_pending < MIN_FL_ENTRIES ||
+	    e->rx_jumbo_pending < MIN_FL_ENTRIES ||
+	    e->tx_pending < adapter->params.nports * MIN_TXQ_ENTRIES)
+		return -EINVAL;
+
+	if (adapter->flags & FULL_INIT_DONE)
+		return -EBUSY;
+
+	for (i = 0; i < SGE_QSETS; ++i) {
+		struct qset_params *q = &adapter->params.sge.qset[i];
+
+		q->rspq_size = e->rx_mini_pending;
+		q->fl_size = e->rx_pending;
+		q->jumbo_size = e->rx_jumbo_pending;
+		q->txq_size[0] = e->tx_pending;
+		q->txq_size[1] = e->tx_pending;
+		q->txq_size[2] = e->tx_pending;
+	}
+	return 0;
+}
+
+static int set_coalesce(struct net_device *dev, struct ethtool_coalesce *c)
+{
+	struct adapter *adapter = dev->priv;
+	struct qset_params *qsp = &adapter->params.sge.qset[0];
+	struct sge_qset *qs = &adapter->sge.qs[0];
+
+	if (c->rx_coalesce_usecs * 10 > M_NEWTIMER)
+		return -EINVAL;
+
+	qsp->coalesce_usecs = c->rx_coalesce_usecs;
+	t3_update_qset_coalesce(qs, qsp);
+	return 0;
+}
+
+static int get_coalesce(struct net_device *dev, struct ethtool_coalesce *c)
+{
+	struct adapter *adapter = dev->priv;
+	struct qset_params *q = adapter->params.sge.qset;
+
+	c->rx_coalesce_usecs = q->coalesce_usecs;
+	return 0;
+}
+
+static int get_eeprom(struct net_device *dev, struct ethtool_eeprom *e,
+		      u8 * data)
+{
+	int i, err = 0;
+	struct adapter *adapter = dev->priv;
+
+	u8 *buf = kmalloc(EEPROMSIZE, GFP_KERNEL);
+	if (!buf)
+		return -ENOMEM;
+
+	e->magic = EEPROM_MAGIC;
+	for (i = e->offset & ~3; !err && i < e->offset + e->len; i += 4)
+		err = t3_seeprom_read(adapter, i, (u32 *) & buf[i]);
+
+	if (!err)
+		memcpy(data, buf + e->offset, e->len);
+	kfree(buf);
+	return err;
+}
+
+static int set_eeprom(struct net_device *dev, struct ethtool_eeprom *eeprom,
+		      u8 * data)
+{
+	u8 *buf;
+	int err = 0;
+	u32 aligned_offset, aligned_len, *p;
+	struct adapter *adapter = dev->priv;
+
+	if (eeprom->magic != EEPROM_MAGIC)
+		return -EINVAL;
+
+	aligned_offset = eeprom->offset & ~3;
+	aligned_len = (eeprom->len + (eeprom->offset & 3) + 3) & ~3;
+
+	if (aligned_offset != eeprom->offset || aligned_len != eeprom->len) {
+		buf = kmalloc(aligned_len, GFP_KERNEL);
+		if (!buf)
+			return -ENOMEM;
+		err = t3_seeprom_read(adapter, aligned_offset, (u32 *) buf);
+		if (!err && aligned_len > 4)
+			err = t3_seeprom_read(adapter,
+					      aligned_offset + aligned_len - 4,
+					      (u32 *) & buf[aligned_len - 4]);
+		if (err)
+			goto out;
+		memcpy(buf + (eeprom->offset & 3), data, eeprom->len);
+	} else
+		buf = data;
+
+	err = t3_seeprom_wp(adapter, 0);
+	if (err)
+		goto out;
+
+	for (p = (u32 *) buf; !err && aligned_len; aligned_len -= 4, p++) {
+		err = t3_seeprom_write(adapter, aligned_offset, *p);
+		aligned_offset += 4;
+	}
+
+	if (!err)
+		err = t3_seeprom_wp(adapter, 1);
+out:
+	if (buf != data)
+		kfree(buf);
+	return err;
+}
+
+static void get_wol(struct net_device *dev, struct ethtool_wolinfo *wol)
+{
+	wol->supported = 0;
+	wol->wolopts = 0;
+	memset(&wol->sopass, 0, sizeof(wol->sopass));
+}
+
+static const struct ethtool_ops cxgb_ethtool_ops = {
+	.get_settings = get_settings,
+	.set_settings = set_settings,
+	.get_drvinfo = get_drvinfo,
+	.get_msglevel = get_msglevel,
+	.set_msglevel = set_msglevel,
+	.get_ringparam = get_sge_param,
+	.set_ringparam = set_sge_param,
+	.get_coalesce = get_coalesce,
+	.set_coalesce = set_coalesce,
+	.get_eeprom_len = get_eeprom_len,
+	.get_eeprom = get_eeprom,
+	.set_eeprom = set_eeprom,
+	.get_pauseparam = get_pauseparam,
+	.set_pauseparam = set_pauseparam,
+	.get_rx_csum = get_rx_csum,
+	.set_rx_csum = set_rx_csum,
+	.get_tx_csum = ethtool_op_get_tx_csum,
+	.set_tx_csum = ethtool_op_set_tx_csum,
+	.get_sg = ethtool_op_get_sg,
+	.set_sg = ethtool_op_set_sg,
+	.get_link = ethtool_op_get_link,
+	.get_strings = get_strings,
+	.phys_id = cxgb3_phys_id,
+	.nway_reset = restart_autoneg,
+	.get_stats_count = get_stats_count,
+	.get_ethtool_stats = get_stats,
+	.get_regs_len = get_regs_len,
+	.get_regs = get_regs,
+	.get_wol = get_wol,
+	.get_tso = ethtool_op_get_tso,
+	.set_tso = ethtool_op_set_tso,
+	.get_perm_addr = ethtool_op_get_perm_addr
+};
+
+static int in_range(int val, int lo, int hi)
+{
+	return val < 0 || (val <= hi && val >= lo);
+}
+
+static int cxgb_extension_ioctl(struct net_device *dev, void __user *useraddr)
+{
+	int ret;
+	u32 cmd;
+	struct adapter *adapter = dev->priv;
+
+	if (copy_from_user(&cmd, useraddr, sizeof(cmd)))
+		return -EFAULT;
+
+	switch (cmd) {
+	case CHELSIO_SETREG:{
+		struct ch_reg edata;
+
+		if (!capable(CAP_NET_ADMIN))
+			return -EPERM;
+		if (copy_from_user(&edata, useraddr, sizeof(edata)))
+			return -EFAULT;
+		if ((edata.addr & 3) != 0
+			|| edata.addr >= adapter->mmio_len)
+			return -EINVAL;
+		writel(edata.val, adapter->regs + edata.addr);
+		break;
+	}
+	case CHELSIO_GETREG:{
+		struct ch_reg edata;
+
+		if (copy_from_user(&edata, useraddr, sizeof(edata)))
+			return -EFAULT;
+		if ((edata.addr & 3) != 0
+			|| edata.addr >= adapter->mmio_len)
+			return -EINVAL;
+		edata.val = readl(adapter->regs + edata.addr);
+		if (copy_to_user(useraddr, &edata, sizeof(edata)))
+			return -EFAULT;
+		break;
+	}
+	case CHELSIO_SET_QSET_PARAMS:{
+		int i;
+		struct qset_params *q;
+		struct ch_qset_params t;
+
+		if (!capable(CAP_NET_ADMIN))
+			return -EPERM;
+		if (copy_from_user(&t, useraddr, sizeof(t)))
+			return -EFAULT;
+		if (t.qset_idx >= SGE_QSETS)
+			return -EINVAL;
+		if (!in_range(t.intr_lat, 0, M_NEWTIMER) ||
+			!in_range(t.cong_thres, 0, 255) ||
+			!in_range(t.txq_size[0], MIN_TXQ_ENTRIES,
+				MAX_TXQ_ENTRIES) ||
+			!in_range(t.txq_size[1], MIN_TXQ_ENTRIES,
+				MAX_TXQ_ENTRIES) ||
+			!in_range(t.txq_size[2], MIN_CTRL_TXQ_ENTRIES,
+				MAX_CTRL_TXQ_ENTRIES) ||
+			!in_range(t.fl_size[0], MIN_FL_ENTRIES,
+				MAX_RX_BUFFERS)
+			|| !in_range(t.fl_size[1], MIN_FL_ENTRIES,
+					MAX_RX_JUMBO_BUFFERS)
+			|| !in_range(t.rspq_size, MIN_RSPQ_ENTRIES,
+					MAX_RSPQ_ENTRIES))
+			return -EINVAL;
+		if ((adapter->flags & FULL_INIT_DONE) &&
+			(t.rspq_size >= 0 || t.fl_size[0] >= 0 ||
+			t.fl_size[1] >= 0 || t.txq_size[0] >= 0 ||
+			t.txq_size[1] >= 0 || t.txq_size[2] >= 0 ||
+			t.polling >= 0 || t.cong_thres >= 0))
+			return -EBUSY;
+
+		q = &adapter->params.sge.qset[t.qset_idx];
+
+		if (t.rspq_size >= 0)
+			q->rspq_size = t.rspq_size;
+		if (t.fl_size[0] >= 0)
+			q->fl_size = t.fl_size[0];
+		if (t.fl_size[1] >= 0)
+			q->jumbo_size = t.fl_size[1];
+		if (t.txq_size[0] >= 0)
+			q->txq_size[0] = t.txq_size[0];
+		if (t.txq_size[1] >= 0)
+			q->txq_size[1] = t.txq_size[1];
+		if (t.txq_size[2] >= 0)
+			q->txq_size[2] = t.txq_size[2];
+		if (t.cong_thres >= 0)
+			q->cong_thres = t.cong_thres;
+		if (t.intr_lat >= 0) {
+			struct sge_qset *qs =
+				&adapter->sge.qs[t.qset_idx];
+
+			q->coalesce_usecs = t.intr_lat;
+			t3_update_qset_coalesce(qs, q);
+		}
+		if (t.polling >= 0) {
+			if (adapter->flags & USING_MSIX)
+				q->polling = t.polling;
+			else {
+				/* No polling with INTx for T3A */
+				if (adapter->params.rev == 0 &&
+					!(adapter->flags & USING_MSI))
+					t.polling = 0;
+
+				for (i = 0; i < SGE_QSETS; i++) {
+					q = &adapter->params.sge.
+						qset[i];
+					q->polling = t.polling;
+				}
+			}
+		}
+		break;
+	}
+	case CHELSIO_GET_QSET_PARAMS:{
+		struct qset_params *q;
+		struct ch_qset_params t;
+
+		if (copy_from_user(&t, useraddr, sizeof(t)))
+			return -EFAULT;
+		if (t.qset_idx >= SGE_QSETS)
+			return -EINVAL;
+
+		q = &adapter->params.sge.qset[t.qset_idx];
+		t.rspq_size = q->rspq_size;
+		t.txq_size[0] = q->txq_size[0];
+		t.txq_size[1] = q->txq_size[1];
+		t.txq_size[2] = q->txq_size[2];
+		t.fl_size[0] = q->fl_size;
+		t.fl_size[1] = q->jumbo_size;
+		t.polling = q->polling;
+		t.intr_lat = q->coalesce_usecs;
+		t.cong_thres = q->cong_thres;
+
+		if (copy_to_user(useraddr, &t, sizeof(t)))
+			return -EFAULT;
+		break;
+	}
+	case CHELSIO_SET_QSET_NUM:{
+		struct ch_reg edata;
+		struct port_info *pi = netdev_priv(dev);
+		unsigned int i, first_qset = 0, other_qsets = 0;
+
+		if (!capable(CAP_NET_ADMIN))
+			return -EPERM;
+		if (adapter->flags & FULL_INIT_DONE)
+			return -EBUSY;
+		if (copy_from_user(&edata, useraddr, sizeof(edata)))
+			return -EFAULT;
+		if (edata.val < 1 ||
+			(edata.val > 1 && !(adapter->flags & USING_MSIX)))
+			return -EINVAL;
+
+		for_each_port(adapter, i)
+			if (adapter->port[i] && adapter->port[i] != dev)
+				other_qsets += adap2pinfo(adapter, i)->nqsets;
+
+		if (edata.val + other_qsets > SGE_QSETS)
+			return -EINVAL;
+
+		pi->nqsets = edata.val;
+
+		for_each_port(adapter, i)
+			if (adapter->port[i]) {
+				pi = adap2pinfo(adapter, i);
+				pi->first_qset = first_qset;
+				first_qset += pi->nqsets;
+			}
+		break;
+	}
+	case CHELSIO_GET_QSET_NUM:{
+		struct ch_reg edata;
+		struct port_info *pi = netdev_priv(dev);
+
+		edata.cmd = CHELSIO_GET_QSET_NUM;
+		edata.val = pi->nqsets;
+		if (copy_to_user(useraddr, &edata, sizeof(edata)))
+			return -EFAULT;
+		break;
+	}
+	case CHELSIO_LOAD_FW:{
+		u8 *fw_data;
+		struct ch_mem_range t;
+
+		if (!capable(CAP_NET_ADMIN))
+			return -EPERM;
+		if (copy_from_user(&t, useraddr, sizeof(t)))
+			return -EFAULT;
+
+		fw_data = kmalloc(t.len, GFP_KERNEL);
+		if (!fw_data)
+			return -ENOMEM;
+
+		if (copy_from_user
+			(fw_data, useraddr + sizeof(t), t.len)) {
+			kfree(fw_data);
+			return -EFAULT;
+		}
+
+		ret = t3_load_fw(adapter, fw_data, t.len);
+		kfree(fw_data);
+		if (ret)
+			return ret;
+		break;
+	}
+	case CHELSIO_SETMTUTAB:{
+		struct ch_mtus m;
+		int i;
+
+		if (!is_offload(adapter))
+			return -EOPNOTSUPP;
+		if (!capable(CAP_NET_ADMIN))
+			return -EPERM;
+		if (offload_running(adapter))
+			return -EBUSY;
+		if (copy_from_user(&m, useraddr, sizeof(m)))
+			return -EFAULT;
+		if (m.nmtus != NMTUS)
+			return -EINVAL;
+		if (m.mtus[0] < 81)	/* accommodate SACK */
+			return -EINVAL;
+
+		/* MTUs must be in ascending order */
+		for (i = 1; i < NMTUS; ++i)
+			if (m.mtus[i] < m.mtus[i - 1])
+				return -EINVAL;
+
+		memcpy(adapter->params.mtus, m.mtus,
+			sizeof(adapter->params.mtus));
+		break;
+	}
+	case CHELSIO_GET_PM:{
+		struct tp_params *p = &adapter->params.tp;
+		struct ch_pm m = {.cmd = CHELSIO_GET_PM };
+
+		if (!is_offload(adapter))
+			return -EOPNOTSUPP;
+		m.tx_pg_sz = p->tx_pg_size;
+		m.tx_num_pg = p->tx_num_pgs;
+		m.rx_pg_sz = p->rx_pg_size;
+		m.rx_num_pg = p->rx_num_pgs;
+		m.pm_total = p->pmtx_size + p->chan_rx_size * p->nchan;
+		if (copy_to_user(useraddr, &m, sizeof(m)))
+			return -EFAULT;
+		break;
+	}
+	case CHELSIO_SET_PM:{
+		struct ch_pm m;
+		struct tp_params *p = &adapter->params.tp;
+
+		if (!is_offload(adapter))
+			return -EOPNOTSUPP;
+		if (!capable(CAP_NET_ADMIN))
+			return -EPERM;
+		if (adapter->flags & FULL_INIT_DONE)
+			return -EBUSY;
+		if (copy_from_user(&m, useraddr, sizeof(m)))
+			return -EFAULT;
+		if (!m.rx_pg_sz || (m.rx_pg_sz & (m.rx_pg_sz - 1)) ||
+			!m.tx_pg_sz || (m.tx_pg_sz & (m.tx_pg_sz - 1)))
+			return -EINVAL;	/* not power of 2 */
+		if (!(m.rx_pg_sz & 0x14000))
+			return -EINVAL;	/* not 16KB or 64KB */
+		if (!(m.tx_pg_sz & 0x1554000))
+			return -EINVAL;
+		if (m.tx_num_pg == -1)
+			m.tx_num_pg = p->tx_num_pgs;
+		if (m.rx_num_pg == -1)
+			m.rx_num_pg = p->rx_num_pgs;
+		if (m.tx_num_pg % 24 || m.rx_num_pg % 24)
+			return -EINVAL;
+		if (m.rx_num_pg * m.rx_pg_sz > p->chan_rx_size ||
+			m.tx_num_pg * m.tx_pg_sz > p->chan_tx_size)
+			return -EINVAL;
+		p->rx_pg_size = m.rx_pg_sz;
+		p->tx_pg_size = m.tx_pg_sz;
+		p->rx_num_pgs = m.rx_num_pg;
+		p->tx_num_pgs = m.tx_num_pg;
+		break;
+	}
+	case CHELSIO_GET_MEM:{
+		struct ch_mem_range t;
+		struct mc7 *mem;
+		u64 buf[32];
+
+		if (!is_offload(adapter))
+			return -EOPNOTSUPP;
+		if (!(adapter->flags & FULL_INIT_DONE))
+			return -EIO;	/* need the memory controllers */
+		if (copy_from_user(&t, useraddr, sizeof(t)))
+			return -EFAULT;
+		if ((t.addr & 7) || (t.len & 7))
+			return -EINVAL;
+		if (t.mem_id == MEM_CM)
+			mem = &adapter->cm;
+		else if (t.mem_id == MEM_PMRX)
+			mem = &adapter->pmrx;
+		else if (t.mem_id == MEM_PMTX)
+			mem = &adapter->pmtx;
+		else
+			return -EINVAL;
+
+		/*
+			* Version scheme:
+			* bits 0..9: chip version
+			* bits 10..15: chip revision
+			*/
+		t.version = 3 | (adapter->params.rev << 10);
+		if (copy_to_user(useraddr, &t, sizeof(t)))
+			return -EFAULT;
+
+		/*
+		 * Read 256 bytes at a time as len can be large and we don't
+		 * want to use huge intermediate buffers.
+		 */
+		useraddr += sizeof(t);	/* advance to start of buffer */
+		while (t.len) {
+			unsigned int chunk =
+				min_t(unsigned int, t.len, sizeof(buf));
+
+			ret =
+				t3_mc7_bd_read(mem, t.addr / 8, chunk / 8,
+						buf);
+			if (ret)
+				return ret;
+			if (copy_to_user(useraddr, buf, chunk))
+				return -EFAULT;
+			useraddr += chunk;
+			t.addr += chunk;
+			t.len -= chunk;
+		}
+		break;
+	}
+	case CHELSIO_SET_TRACE_FILTER:{
+		struct ch_trace t;
+		const struct trace_params *tp;
+
+		if (!capable(CAP_NET_ADMIN))
+			return -EPERM;
+		if (!offload_running(adapter))
+			return -EAGAIN;
+		if (copy_from_user(&t, useraddr, sizeof(t)))
+			return -EFAULT;
+
+		tp = (const struct trace_params *)&t.sip;
+		if (t.config_tx)
+			t3_config_trace_filter(adapter, tp, 0,
+						t.invert_match,
+						t.trace_tx);
+		if (t.config_rx)
+			t3_config_trace_filter(adapter, tp, 1,
+						t.invert_match,
+						t.trace_rx);
+		break;
+	}
+	case CHELSIO_SET_PKTSCHED:{
+		struct sk_buff *skb;
+		struct ch_pktsched_params p;
+		struct mngt_pktsched_wr *req;
+
+		if (!(adapter->flags & FULL_INIT_DONE))
+			return -EIO;	/* uP must be up and running */
+		if (copy_from_user(&p, useraddr, sizeof(p)))
+			return -EFAULT;
+		skb = alloc_skb(sizeof(*req), GFP_KERNEL);
+		if (!skb)
+			return -ENOMEM;
+		req =
+			(struct mngt_pktsched_wr *)skb_put(skb,
+							sizeof(*req));
+		req->wr_hi = htonl(V_WR_OP(FW_WROPCODE_MNGT));
+		req->mngt_opcode = FW_MNGTOPCODE_PKTSCHED_SET;
+		req->sched = p.sched;
+		req->idx = p.idx;
+		req->min = p.min;
+		req->max = p.max;
+		req->binding = p.binding;
+		printk(KERN_INFO
+			"pktsched: sched %u idx %u min %u max %u binding %u\n",
+			req->sched, req->idx, req->min, req->max,
+			req->binding);
+		skb->priority = 1;
+		offload_tx(&adapter->tdev, skb);
+		break;
+	}
+	default:
+		return -EOPNOTSUPP;
+	}
+	return 0;
+}
+
+static int cxgb_ioctl(struct net_device *dev, struct ifreq *req, int cmd)
+{
+	int ret, mmd;
+	struct adapter *adapter = dev->priv;
+	struct port_info *pi = netdev_priv(dev);
+	struct mii_ioctl_data *data = if_mii(req);
+
+	switch (cmd) {
+	case SIOCGMIIPHY:
+		data->phy_id = pi->phy.addr;
+		/* FALLTHRU */
+	case SIOCGMIIREG:{
+		u32 val;
+		struct cphy *phy = &pi->phy;
+
+		if (!phy->mdio_read)
+			return -EOPNOTSUPP;
+		if (is_10G(adapter)) {
+			mmd = data->phy_id >> 8;
+			if (!mmd)
+				mmd = MDIO_DEV_PCS;
+			else if (mmd > MDIO_DEV_XGXS)
+				return -EINVAL;
+
+			ret =
+				phy->mdio_read(adapter, data->phy_id & 0x1f,
+						mmd, data->reg_num, &val);
+		} else
+			ret =
+				phy->mdio_read(adapter, data->phy_id & 0x1f,
+						0, data->reg_num & 0x1f,
+						&val);
+		if (!ret)
+			data->val_out = val;
+		break;
+	}
+	case SIOCSMIIREG:{
+		struct cphy *phy = &pi->phy;
+
+		if (!capable(CAP_NET_ADMIN))
+			return -EPERM;
+		if (!phy->mdio_write)
+			return -EOPNOTSUPP;
+		if (is_10G(adapter)) {
+			mmd = data->phy_id >> 8;
+			if (!mmd)
+				mmd = MDIO_DEV_PCS;
+			else if (mmd > MDIO_DEV_XGXS)
+				return -EINVAL;
+
+			ret =
+				phy->mdio_write(adapter,
+						data->phy_id & 0x1f, mmd,
+						data->reg_num,
+						data->val_in);
+		} else
+			ret =
+				phy->mdio_write(adapter,
+						data->phy_id & 0x1f, 0,
+						data->reg_num & 0x1f,
+						data->val_in);
+		break;
+	}
+	case SIOCCHIOCTL:
+		return cxgb_extension_ioctl(dev, req->ifr_data);
+	default:
+		return -EOPNOTSUPP;
+	}
+	return ret;
+}
+
+static int cxgb_change_mtu(struct net_device *dev, int new_mtu)
+{
+	int ret;
+	struct adapter *adapter = dev->priv;
+	struct port_info *pi = netdev_priv(dev);
+
+	if (new_mtu < 81)	/* accommodate SACK */
+		return -EINVAL;
+	if ((ret = t3_mac_set_mtu(&pi->mac, new_mtu)))
+		return ret;
+	dev->mtu = new_mtu;
+	init_port_mtus(adapter);
+	if (adapter->params.rev == 0 && offload_running(adapter))
+		t3_load_mtus(adapter, adapter->params.mtus,
+			     adapter->params.a_wnd, adapter->params.b_wnd,
+			     adapter->port[0]->mtu);
+	return 0;
+}
+
+static int cxgb_set_mac_addr(struct net_device *dev, void *p)
+{
+	struct adapter *adapter = dev->priv;
+	struct port_info *pi = netdev_priv(dev);
+	struct sockaddr *addr = p;
+
+	if (!is_valid_ether_addr(addr->sa_data))
+		return -EINVAL;
+
+	memcpy(dev->dev_addr, addr->sa_data, dev->addr_len);
+	t3_mac_set_address(&pi->mac, 0, dev->dev_addr);
+	if (offload_running(adapter))
+		write_smt_entry(adapter, pi->port_id);
+	return 0;
+}
+
+/**
+ * t3_synchronize_rx - wait for current Rx processing on a port to complete
+ * @adap: the adapter
+ * @p: the port
+ *
+ * Ensures that current Rx processing on any of the queues associated with
+ * the given port completes before returning.  We do this by acquiring and
+ * releasing the locks of the response queues associated with the port.
+ */
+static void t3_synchronize_rx(struct adapter *adap, const struct port_info *p)
+{
+	int i;
+
+	for (i = 0; i < p->nqsets; i++) {
+		struct sge_rspq *q = &adap->sge.qs[i + p->first_qset].rspq;
+
+		spin_lock_irq(&q->lock);
+		spin_unlock_irq(&q->lock);
+	}
+}
+
+static void vlan_rx_register(struct net_device *dev, struct vlan_group *grp)
+{
+	struct adapter *adapter = dev->priv;
+	struct port_info *pi = netdev_priv(dev);
+
+	pi->vlan_grp = grp;
+	if (adapter->params.rev > 0)
+		t3_set_vlan_accel(adapter, 1 << pi->port_id, grp != NULL);
+	else {
+		/* single control for all ports */
+		unsigned int i, have_vlans = 0;
+		for_each_port(adapter, i)
+		    have_vlans |= adap2pinfo(adapter, i)->vlan_grp != NULL;
+
+		t3_set_vlan_accel(adapter, 1, have_vlans);
+	}
+	t3_synchronize_rx(adapter, pi);
+}
+
+static void vlan_rx_kill_vid(struct net_device *dev, unsigned short vid)
+{
+	/* nothing */
+}
+
+#ifdef CONFIG_NET_POLL_CONTROLLER
+static void cxgb_netpoll(struct net_device *dev)
+{
+	struct adapter *adapter = dev->priv;
+	struct sge_qset *qs = dev2qset(dev);
+
+	t3_intr_handler(adapter, qs->rspq.polling) (adapter->pdev->irq,
+						    adapter);
+}
+#endif
+
+/*
+ * Periodic accumulation of MAC statistics.
+ */
+static void mac_stats_update(struct adapter *adapter)
+{
+	int i;
+
+	for_each_port(adapter, i) {
+		struct net_device *dev = adapter->port[i];
+		struct port_info *p = netdev_priv(dev);
+
+		if (netif_running(dev)) {
+			spin_lock(&adapter->stats_lock);
+			t3_mac_update_stats(&p->mac);
+			spin_unlock(&adapter->stats_lock);
+		}
+	}
+}
+
+static void check_link_status(struct adapter *adapter)
+{
+	int i;
+
+	for_each_port(adapter, i) {
+		struct net_device *dev = adapter->port[i];
+		struct port_info *p = netdev_priv(dev);
+
+		if (!(p->port_type->caps & SUPPORTED_IRQ) && netif_running(dev))
+			t3_link_changed(adapter, i);
+	}
+}
+
+static void t3_adap_check_task(struct work_struct *work)
+{
+	struct adapter *adapter = container_of(work, struct adapter,
+					       adap_check_task.work);
+	const struct adapter_params *p = &adapter->params;
+
+	adapter->check_task_cnt++;
+
+	/* Check link status for PHYs without interrupts */
+	if (p->linkpoll_period)
+		check_link_status(adapter);
+
+	/* Accumulate MAC stats if needed */
+	if (!p->linkpoll_period ||
+	    (adapter->check_task_cnt * p->linkpoll_period) / 10 >=
+	    p->stats_update_period) {
+		mac_stats_update(adapter);
+		adapter->check_task_cnt = 0;
+	}
+
+	/* Schedule the next check update if any port is active. */
+	spin_lock(&adapter->work_lock);
+	if (adapter->open_device_map & PORT_MASK)
+		schedule_chk_task(adapter);
+	spin_unlock(&adapter->work_lock);
+}
+
+/*
+ * Processes external (PHY) interrupts in process context.
+ */
+static void ext_intr_task(struct work_struct *work)
+{
+	struct adapter *adapter = container_of(work, struct adapter,
+					       ext_intr_handler_task);
+
+	t3_phy_intr_handler(adapter);
+
+	/* Now reenable external interrupts */
+	spin_lock_irq(&adapter->work_lock);
+	if (adapter->slow_intr_mask) {
+		adapter->slow_intr_mask |= F_T3DBG;
+		t3_write_reg(adapter, A_PL_INT_CAUSE0, F_T3DBG);
+		t3_write_reg(adapter, A_PL_INT_ENABLE0,
+			     adapter->slow_intr_mask);
+	}
+	spin_unlock_irq(&adapter->work_lock);
+}
+
+/*
+ * Interrupt-context handler for external (PHY) interrupts.
+ */
+void t3_os_ext_intr_handler(struct adapter *adapter)
+{
+	/*
+	 * Schedule a task to handle external interrupts as they may be slow
+	 * and we use a mutex to protect MDIO registers.  We disable PHY
+	 * interrupts in the meantime and let the task reenable them when
+	 * it's done.
+	 */
+	spin_lock(&adapter->work_lock);
+	if (adapter->slow_intr_mask) {
+		adapter->slow_intr_mask &= ~F_T3DBG;
+		t3_write_reg(adapter, A_PL_INT_ENABLE0,
+			     adapter->slow_intr_mask);
+		queue_work(cxgb3_wq, &adapter->ext_intr_handler_task);
+	}
+	spin_unlock(&adapter->work_lock);
+}
+
+void t3_fatal_err(struct adapter *adapter)
+{
+	unsigned int fw_status[4];
+
+	if (adapter->flags & FULL_INIT_DONE) {
+		t3_sge_stop(adapter);
+		t3_intr_disable(adapter);
+	}
+	CH_ALERT(adapter, "encountered fatal error, operation suspended\n");
+	if (!t3_cim_ctl_blk_read(adapter, 0xa0, 4, fw_status))
+		CH_ALERT(adapter, "FW status: 0x%x, 0x%x, 0x%x, 0x%x\n",
+			 fw_status[0], fw_status[1],
+			 fw_status[2], fw_status[3]);
+
+}
+
+static int __devinit cxgb_enable_msix(struct adapter *adap)
+{
+	struct msix_entry entries[SGE_QSETS + 1];
+	int i, err;
+
+	for (i = 0; i < ARRAY_SIZE(entries); ++i)
+		entries[i].entry = i;
+
+	err = pci_enable_msix(adap->pdev, entries, ARRAY_SIZE(entries));
+	if (!err) {
+		for (i = 0; i < ARRAY_SIZE(entries); ++i)
+			adap->msix_info[i].vec = entries[i].vector;
+	} else if (err > 0)
+		dev_info(&adap->pdev->dev,
+		       "only %d MSI-X vectors left, not using MSI-X\n", err);
+	return err;
+}
+
+static void __devinit print_port_info(struct adapter *adap,
+				      const struct adapter_info *ai)
+{
+	static const char *pci_variant[] = {
+		"PCI", "PCI-X", "PCI-X ECC", "PCI-X 266", "PCI Express"
+	};
+
+	int i;
+	char buf[80];
+
+	if (is_pcie(adap))
+		snprintf(buf, sizeof(buf), "%s x%d",
+			 pci_variant[adap->params.pci.variant],
+			 adap->params.pci.width);
+	else
+		snprintf(buf, sizeof(buf), "%s %dMHz/%d-bit",
+			 pci_variant[adap->params.pci.variant],
+			 adap->params.pci.speed, adap->params.pci.width);
+
+	for_each_port(adap, i) {
+		struct net_device *dev = adap->port[i];
+		const struct port_info *pi = netdev_priv(dev);
+
+		if (!test_bit(i, &adap->registered_device_map))
+			continue;
+		printk(KERN_INFO "%s: %s %s RNIC (rev %d) %s%s\n",
+		       dev->name, ai->desc, pi->port_type->desc,
+		       adap->params.rev, buf,
+		       (adap->flags & USING_MSIX) ? " MSI-X" :
+		       (adap->flags & USING_MSI) ? " MSI" : "");
+		if (adap->name == dev->name && adap->params.vpd.mclk)
+			printk(KERN_INFO "%s: %uMB CM, %uMB PMTX, %uMB PMRX\n",
+			       adap->name, t3_mc7_size(&adap->cm) >> 20,
+			       t3_mc7_size(&adap->pmtx) >> 20,
+			       t3_mc7_size(&adap->pmrx) >> 20);
+	}
+}
+
+static int __devinit init_one(struct pci_dev *pdev,
+			      const struct pci_device_id *ent)
+{
+	static int version_printed;
+
+	int i, err, pci_using_dac = 0;
+	unsigned long mmio_start, mmio_len;
+	const struct adapter_info *ai;
+	struct adapter *adapter = NULL;
+	struct port_info *pi;
+
+	if (!version_printed) {
+		printk(KERN_INFO "%s - version %s\n", DRV_DESC, DRV_VERSION);
+		++version_printed;
+	}
+
+	if (!cxgb3_wq) {
+		cxgb3_wq = create_singlethread_workqueue(DRV_NAME);
+		if (!cxgb3_wq) {
+			printk(KERN_ERR DRV_NAME
+			       ": cannot initialize work queue\n");
+			return -ENOMEM;
+		}
+	}
+
+	err = pci_request_regions(pdev, DRV_NAME);
+	if (err) {
+		/* Just info, some other driver may have claimed the device. */
+		dev_info(&pdev->dev, "cannot obtain PCI resources\n");
+		return err;
+	}
+
+	err = pci_enable_device(pdev);
+	if (err) {
+		dev_err(&pdev->dev, "cannot enable PCI device\n");
+		goto out_release_regions;
+	}
+
+	if (!pci_set_dma_mask(pdev, DMA_64BIT_MASK)) {
+		pci_using_dac = 1;
+		err = pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK);
+		if (err) {
+			dev_err(&pdev->dev, "unable to obtain 64-bit DMA for "
+			       "coherent allocations\n");
+			goto out_disable_device;
+		}
+	} else if ((err = pci_set_dma_mask(pdev, DMA_32BIT_MASK)) != 0) {
+		dev_err(&pdev->dev, "no usable DMA configuration\n");
+		goto out_disable_device;
+	}
+
+	pci_set_master(pdev);
+
+	mmio_start = pci_resource_start(pdev, 0);
+	mmio_len = pci_resource_len(pdev, 0);
+	ai = t3_get_adapter_info(ent->driver_data);
+
+	adapter = kzalloc(sizeof(*adapter), GFP_KERNEL);
+	if (!adapter) {
+		err = -ENOMEM;
+		goto out_disable_device;
+	}
+
+	adapter->regs = ioremap_nocache(mmio_start, mmio_len);
+	if (!adapter->regs) {
+		dev_err(&pdev->dev, "cannot map device registers\n");
+		err = -ENOMEM;
+		goto out_free_adapter;
+	}
+
+	adapter->pdev = pdev;
+	adapter->name = pci_name(pdev);
+	adapter->msg_enable = dflt_msg_enable;
+	adapter->mmio_len = mmio_len;
+
+	mutex_init(&adapter->mdio_lock);
+	spin_lock_init(&adapter->work_lock);
+	spin_lock_init(&adapter->stats_lock);
+
+	INIT_LIST_HEAD(&adapter->adapter_list);
+	INIT_WORK(&adapter->ext_intr_handler_task, ext_intr_task);
+	INIT_DELAYED_WORK(&adapter->adap_check_task, t3_adap_check_task);
+
+	for (i = 0; i < ai->nports; ++i) {
+		struct net_device *netdev;
+
+		netdev = alloc_etherdev(sizeof(struct port_info));
+		if (!netdev) {
+			err = -ENOMEM;
+			goto out_free_dev;
+		}
+
+		SET_MODULE_OWNER(netdev);
+		SET_NETDEV_DEV(netdev, &pdev->dev);
+
+		adapter->port[i] = netdev;
+		pi = netdev_priv(netdev);
+		pi->rx_csum_offload = 1;
+		pi->nqsets = 1;
+		pi->first_qset = i;
+		pi->activity = 0;
+		pi->port_id = i;
+		netif_carrier_off(netdev);
+		netdev->irq = pdev->irq;
+		netdev->mem_start = mmio_start;
+		netdev->mem_end = mmio_start + mmio_len - 1;
+		netdev->priv = adapter;
+		netdev->features |= NETIF_F_SG | NETIF_F_IP_CSUM | NETIF_F_TSO;
+		netdev->features |= NETIF_F_LLTX;
+		if (pci_using_dac)
+			netdev->features |= NETIF_F_HIGHDMA;
+
+		netdev->features |= NETIF_F_HW_VLAN_TX | NETIF_F_HW_VLAN_RX;
+		netdev->vlan_rx_register = vlan_rx_register;
+		netdev->vlan_rx_kill_vid = vlan_rx_kill_vid;
+
+		netdev->open = cxgb_open;
+		netdev->stop = cxgb_close;
+		netdev->hard_start_xmit = t3_eth_xmit;
+		netdev->get_stats = cxgb_get_stats;
+		netdev->set_multicast_list = cxgb_set_rxmode;
+		netdev->do_ioctl = cxgb_ioctl;
+		netdev->change_mtu = cxgb_change_mtu;
+		netdev->set_mac_address = cxgb_set_mac_addr;
+#ifdef CONFIG_NET_POLL_CONTROLLER
+		netdev->poll_controller = cxgb_netpoll;
+#endif
+		netdev->weight = 64;
+
+		SET_ETHTOOL_OPS(netdev, &cxgb_ethtool_ops);
+	}
+
+	pci_set_drvdata(pdev, adapter->port[0]);
+	if (t3_prep_adapter(adapter, ai, 1) < 0) {
+		err = -ENODEV;
+		goto out_free_dev;
+	}
+
+	/*
+	 * The card is now ready to go.  If any errors occur during device
+	 * registration we do not fail the whole card but rather proceed only
+	 * with the ports we manage to register successfully.  However we must
+	 * register at least one net device.
+	 */
+	for_each_port(adapter, i) {
+		err = register_netdev(adapter->port[i]);
+		if (err)
+			dev_warn(&pdev->dev,
+				 "cannot register net device %s, skipping\n",
+				 adapter->port[i]->name);
+		else {
+			/*
+			 * Change the name we use for messages to the name of
+			 * the first successfully registered interface.
+			 */
+			if (!adapter->registered_device_map)
+				adapter->name = adapter->port[i]->name;
+
+			__set_bit(i, &adapter->registered_device_map);
+		}
+	}
+	if (!adapter->registered_device_map) {
+		dev_err(&pdev->dev, "could not register any net devices\n");
+		goto out_free_dev;
+	}
+
+	/* Driver's ready. Reflect it on LEDs */
+	t3_led_ready(adapter);
+
+	if (is_offload(adapter)) {
+		__set_bit(OFFLOAD_DEVMAP_BIT, &adapter->registered_device_map);
+		cxgb3_adapter_ofld(adapter);
+	}
+
+	/* See what interrupts we'll be using */
+	if (msi > 1 && cxgb_enable_msix(adapter) == 0)
+		adapter->flags |= USING_MSIX;
+	else if (msi > 0 && pci_enable_msi(pdev) == 0)
+		adapter->flags |= USING_MSI;
+
+	err = sysfs_create_group(&adapter->port[0]->class_dev.kobj,
+				 &cxgb3_attr_group);
+
+	print_port_info(adapter, ai);
+	return 0;
+
+out_free_dev:
+	iounmap(adapter->regs);
+	for (i = ai->nports - 1; i >= 0; --i)
+		if (adapter->port[i])
+			free_netdev(adapter->port[i]);
+
+out_free_adapter:
+	kfree(adapter);
+
+out_disable_device:
+	pci_disable_device(pdev);
+out_release_regions:
+	pci_release_regions(pdev);
+	pci_set_drvdata(pdev, NULL);
+	return err;
+}
+
+static void __devexit remove_one(struct pci_dev *pdev)
+{
+	struct net_device *dev = pci_get_drvdata(pdev);
+
+	if (dev) {
+		int i;
+		struct adapter *adapter = dev->priv;
+
+		t3_sge_stop(adapter);
+		sysfs_remove_group(&adapter->port[0]->class_dev.kobj,
+				   &cxgb3_attr_group);
+
+		for_each_port(adapter, i)
+		    if (test_bit(i, &adapter->registered_device_map))
+			unregister_netdev(adapter->port[i]);
+
+		if (is_offload(adapter)) {
+			cxgb3_adapter_unofld(adapter);
+			if (test_bit(OFFLOAD_DEVMAP_BIT,
+				     &adapter->open_device_map))
+				offload_close(&adapter->tdev);
+		}
+
+		t3_free_sge_resources(adapter);
+		cxgb_disable_msi(adapter);
+
+		for (i = 0; i < ARRAY_SIZE(adapter->dummy_netdev); i++)
+			if (adapter->dummy_netdev[i]) {
+				free_netdev(adapter->dummy_netdev[i]);
+				adapter->dummy_netdev[i] = NULL;
+			}
+
+		for_each_port(adapter, i)
+			if (adapter->port[i])
+				free_netdev(adapter->port[i]);
+
+		iounmap(adapter->regs);
+		kfree(adapter);
+		pci_release_regions(pdev);
+		pci_disable_device(pdev);
+		pci_set_drvdata(pdev, NULL);
+	}
+}
+
+static struct pci_driver driver = {
+	.name = DRV_NAME,
+	.id_table = cxgb3_pci_tbl,
+	.probe = init_one,
+	.remove = __devexit_p(remove_one),
+};
+
+static int __init cxgb3_init_module(void)
+{
+	int ret;
+
+	cxgb3_offload_init();
+
+	ret = pci_register_driver(&driver);
+	return ret;
+}
+
+static void __exit cxgb3_cleanup_module(void)
+{
+	pci_unregister_driver(&driver);
+	if (cxgb3_wq)
+		destroy_workqueue(cxgb3_wq);
+}
+
+module_init(cxgb3_init_module);
+module_exit(cxgb3_cleanup_module);
diff --git a/drivers/net/cxgb3/cxgb3_offload.c b/drivers/net/cxgb3/cxgb3_offload.c
new file mode 100644
index 00000000000..3abd4d25c3b
--- /dev/null
+++ b/drivers/net/cxgb3/cxgb3_offload.c
@@ -0,0 +1,1222 @@
+/*
+ * Copyright (c) 2006 Chelsio, Inc. All rights reserved.
+ * Copyright (c) 2006 Open Grid Computing, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/list.h>
+#include <net/neighbour.h>
+#include <linux/notifier.h>
+#include <asm/atomic.h>
+#include <linux/proc_fs.h>
+#include <linux/if_vlan.h>
+#include <net/netevent.h>
+#include <linux/highmem.h>
+#include <linux/vmalloc.h>
+
+#include "common.h"
+#include "regs.h"
+#include "cxgb3_ioctl.h"
+#include "cxgb3_ctl_defs.h"
+#include "cxgb3_defs.h"
+#include "l2t.h"
+#include "firmware_exports.h"
+#include "cxgb3_offload.h"
+
+static LIST_HEAD(client_list);
+static LIST_HEAD(ofld_dev_list);
+static DEFINE_MUTEX(cxgb3_db_lock);
+
+static DEFINE_RWLOCK(adapter_list_lock);
+static LIST_HEAD(adapter_list);
+
+static const unsigned int MAX_ATIDS = 64 * 1024;
+static const unsigned int ATID_BASE = 0x100000;
+
+static inline int offload_activated(struct t3cdev *tdev)
+{
+	const struct adapter *adapter = tdev2adap(tdev);
+
+	return (test_bit(OFFLOAD_DEVMAP_BIT, &adapter->open_device_map));
+}
+
+/**
+ *	cxgb3_register_client - register an offload client
+ *	@client: the client
+ *
+ *	Add the client to the client list,
+ *	and call backs the client for each activated offload device
+ */
+void cxgb3_register_client(struct cxgb3_client *client)
+{
+	struct t3cdev *tdev;
+
+	mutex_lock(&cxgb3_db_lock);
+	list_add_tail(&client->client_list, &client_list);
+
+	if (client->add) {
+		list_for_each_entry(tdev, &ofld_dev_list, ofld_dev_list) {
+			if (offload_activated(tdev))
+				client->add(tdev);
+		}
+	}
+	mutex_unlock(&cxgb3_db_lock);
+}
+
+EXPORT_SYMBOL(cxgb3_register_client);
+
+/**
+ *	cxgb3_unregister_client - unregister an offload client
+ *	@client: the client
+ *
+ *	Remove the client to the client list,
+ *	and call backs the client for each activated offload device.
+ */
+void cxgb3_unregister_client(struct cxgb3_client *client)
+{
+	struct t3cdev *tdev;
+
+	mutex_lock(&cxgb3_db_lock);
+	list_del(&client->client_list);
+
+	if (client->remove) {
+		list_for_each_entry(tdev, &ofld_dev_list, ofld_dev_list) {
+			if (offload_activated(tdev))
+				client->remove(tdev);
+		}
+	}
+	mutex_unlock(&cxgb3_db_lock);
+}
+
+EXPORT_SYMBOL(cxgb3_unregister_client);
+
+/**
+ *	cxgb3_add_clients - activate registered clients for an offload device
+ *	@tdev: the offload device
+ *
+ *	Call backs all registered clients once a offload device is activated
+ */
+void cxgb3_add_clients(struct t3cdev *tdev)
+{
+	struct cxgb3_client *client;
+
+	mutex_lock(&cxgb3_db_lock);
+	list_for_each_entry(client, &client_list, client_list) {
+		if (client->add)
+			client->add(tdev);
+	}
+	mutex_unlock(&cxgb3_db_lock);
+}
+
+/**
+ *	cxgb3_remove_clients - deactivates registered clients
+ *			       for an offload device
+ *	@tdev: the offload device
+ *
+ *	Call backs all registered clients once a offload device is deactivated
+ */
+void cxgb3_remove_clients(struct t3cdev *tdev)
+{
+	struct cxgb3_client *client;
+
+	mutex_lock(&cxgb3_db_lock);
+	list_for_each_entry(client, &client_list, client_list) {
+		if (client->remove)
+			client->remove(tdev);
+	}
+	mutex_unlock(&cxgb3_db_lock);
+}
+
+static struct net_device *get_iff_from_mac(struct adapter *adapter,
+					   const unsigned char *mac,
+					   unsigned int vlan)
+{
+	int i;
+
+	for_each_port(adapter, i) {
+		const struct vlan_group *grp;
+		struct net_device *dev = adapter->port[i];
+		const struct port_info *p = netdev_priv(dev);
+
+		if (!memcmp(dev->dev_addr, mac, ETH_ALEN)) {
+			if (vlan && vlan != VLAN_VID_MASK) {
+				grp = p->vlan_grp;
+				dev = grp ? grp->vlan_devices[vlan] : NULL;
+			} else
+				while (dev->master)
+					dev = dev->master;
+			return dev;
+		}
+	}
+	return NULL;
+}
+
+static int cxgb_ulp_iscsi_ctl(struct adapter *adapter, unsigned int req,
+			      void *data)
+{
+	int ret = 0;
+	struct ulp_iscsi_info *uiip = data;
+
+	switch (req) {
+	case ULP_ISCSI_GET_PARAMS:
+		uiip->pdev = adapter->pdev;
+		uiip->llimit = t3_read_reg(adapter, A_ULPRX_ISCSI_LLIMIT);
+		uiip->ulimit = t3_read_reg(adapter, A_ULPRX_ISCSI_ULIMIT);
+		uiip->tagmask = t3_read_reg(adapter, A_ULPRX_ISCSI_TAGMASK);
+		/*
+		 * On tx, the iscsi pdu has to be <= tx page size and has to
+		 * fit into the Tx PM FIFO.
+		 */
+		uiip->max_txsz = min(adapter->params.tp.tx_pg_size,
+				     t3_read_reg(adapter, A_PM1_TX_CFG) >> 17);
+		/* on rx, the iscsi pdu has to be < rx page size and the
+		   whole pdu + cpl headers has to fit into one sge buffer */
+		uiip->max_rxsz = min_t(unsigned int,
+				       adapter->params.tp.rx_pg_size,
+				       (adapter->sge.qs[0].fl[1].buf_size -
+					sizeof(struct cpl_rx_data) * 2 -
+					sizeof(struct cpl_rx_data_ddp)));
+		break;
+	case ULP_ISCSI_SET_PARAMS:
+		t3_write_reg(adapter, A_ULPRX_ISCSI_TAGMASK, uiip->tagmask);
+		break;
+	default:
+		ret = -EOPNOTSUPP;
+	}
+	return ret;
+}
+
+/* Response queue used for RDMA events. */
+#define ASYNC_NOTIF_RSPQ 0
+
+static int cxgb_rdma_ctl(struct adapter *adapter, unsigned int req, void *data)
+{
+	int ret = 0;
+
+	switch (req) {
+	case RDMA_GET_PARAMS:{
+		struct rdma_info *req = data;
+		struct pci_dev *pdev = adapter->pdev;
+
+		req->udbell_physbase = pci_resource_start(pdev, 2);
+		req->udbell_len = pci_resource_len(pdev, 2);
+		req->tpt_base =
+			t3_read_reg(adapter, A_ULPTX_TPT_LLIMIT);
+		req->tpt_top = t3_read_reg(adapter, A_ULPTX_TPT_ULIMIT);
+		req->pbl_base =
+			t3_read_reg(adapter, A_ULPTX_PBL_LLIMIT);
+		req->pbl_top = t3_read_reg(adapter, A_ULPTX_PBL_ULIMIT);
+		req->rqt_base = t3_read_reg(adapter, A_ULPRX_RQ_LLIMIT);
+		req->rqt_top = t3_read_reg(adapter, A_ULPRX_RQ_ULIMIT);
+		req->kdb_addr = adapter->regs + A_SG_KDOORBELL;
+		req->pdev = pdev;
+		break;
+	}
+	case RDMA_CQ_OP:{
+		unsigned long flags;
+		struct rdma_cq_op *req = data;
+
+		/* may be called in any context */
+		spin_lock_irqsave(&adapter->sge.reg_lock, flags);
+		ret = t3_sge_cqcntxt_op(adapter, req->id, req->op,
+					req->credits);
+		spin_unlock_irqrestore(&adapter->sge.reg_lock, flags);
+		break;
+	}
+	case RDMA_GET_MEM:{
+		struct ch_mem_range *t = data;
+		struct mc7 *mem;
+
+		if ((t->addr & 7) || (t->len & 7))
+			return -EINVAL;
+		if (t->mem_id == MEM_CM)
+			mem = &adapter->cm;
+		else if (t->mem_id == MEM_PMRX)
+			mem = &adapter->pmrx;
+		else if (t->mem_id == MEM_PMTX)
+			mem = &adapter->pmtx;
+		else
+			return -EINVAL;
+
+		ret =
+			t3_mc7_bd_read(mem, t->addr / 8, t->len / 8,
+					(u64 *) t->buf);
+		if (ret)
+			return ret;
+		break;
+	}
+	case RDMA_CQ_SETUP:{
+		struct rdma_cq_setup *req = data;
+
+		spin_lock_irq(&adapter->sge.reg_lock);
+		ret =
+			t3_sge_init_cqcntxt(adapter, req->id,
+					req->base_addr, req->size,
+					ASYNC_NOTIF_RSPQ,
+					req->ovfl_mode, req->credits,
+					req->credit_thres);
+		spin_unlock_irq(&adapter->sge.reg_lock);
+		break;
+	}
+	case RDMA_CQ_DISABLE:
+		spin_lock_irq(&adapter->sge.reg_lock);
+		ret = t3_sge_disable_cqcntxt(adapter, *(unsigned int *)data);
+		spin_unlock_irq(&adapter->sge.reg_lock);
+		break;
+	case RDMA_CTRL_QP_SETUP:{
+		struct rdma_ctrlqp_setup *req = data;
+
+		spin_lock_irq(&adapter->sge.reg_lock);
+		ret = t3_sge_init_ecntxt(adapter, FW_RI_SGEEC_START, 0,
+						SGE_CNTXT_RDMA,
+						ASYNC_NOTIF_RSPQ,
+						req->base_addr, req->size,
+						FW_RI_TID_START, 1, 0);
+		spin_unlock_irq(&adapter->sge.reg_lock);
+		break;
+	}
+	default:
+		ret = -EOPNOTSUPP;
+	}
+	return ret;
+}
+
+static int cxgb_offload_ctl(struct t3cdev *tdev, unsigned int req, void *data)
+{
+	struct adapter *adapter = tdev2adap(tdev);
+	struct tid_range *tid;
+	struct mtutab *mtup;
+	struct iff_mac *iffmacp;
+	struct ddp_params *ddpp;
+	struct adap_ports *ports;
+	int i;
+
+	switch (req) {
+	case GET_MAX_OUTSTANDING_WR:
+		*(unsigned int *)data = FW_WR_NUM;
+		break;
+	case GET_WR_LEN:
+		*(unsigned int *)data = WR_FLITS;
+		break;
+	case GET_TX_MAX_CHUNK:
+		*(unsigned int *)data = 1 << 20;	/* 1MB */
+		break;
+	case GET_TID_RANGE:
+		tid = data;
+		tid->num = t3_mc5_size(&adapter->mc5) -
+		    adapter->params.mc5.nroutes -
+		    adapter->params.mc5.nfilters - adapter->params.mc5.nservers;
+		tid->base = 0;
+		break;
+	case GET_STID_RANGE:
+		tid = data;
+		tid->num = adapter->params.mc5.nservers;
+		tid->base = t3_mc5_size(&adapter->mc5) - tid->num -
+		    adapter->params.mc5.nfilters - adapter->params.mc5.nroutes;
+		break;
+	case GET_L2T_CAPACITY:
+		*(unsigned int *)data = 2048;
+		break;
+	case GET_MTUS:
+		mtup = data;
+		mtup->size = NMTUS;
+		mtup->mtus = adapter->params.mtus;
+		break;
+	case GET_IFF_FROM_MAC:
+		iffmacp = data;
+		iffmacp->dev = get_iff_from_mac(adapter, iffmacp->mac_addr,
+						iffmacp->vlan_tag &
+						VLAN_VID_MASK);
+		break;
+	case GET_DDP_PARAMS:
+		ddpp = data;
+		ddpp->llimit = t3_read_reg(adapter, A_ULPRX_TDDP_LLIMIT);
+		ddpp->ulimit = t3_read_reg(adapter, A_ULPRX_TDDP_ULIMIT);
+		ddpp->tag_mask = t3_read_reg(adapter, A_ULPRX_TDDP_TAGMASK);
+		break;
+	case GET_PORTS:
+		ports = data;
+		ports->nports = adapter->params.nports;
+		for_each_port(adapter, i)
+			ports->lldevs[i] = adapter->port[i];
+		break;
+	case ULP_ISCSI_GET_PARAMS:
+	case ULP_ISCSI_SET_PARAMS:
+		if (!offload_running(adapter))
+			return -EAGAIN;
+		return cxgb_ulp_iscsi_ctl(adapter, req, data);
+	case RDMA_GET_PARAMS:
+	case RDMA_CQ_OP:
+	case RDMA_CQ_SETUP:
+	case RDMA_CQ_DISABLE:
+	case RDMA_CTRL_QP_SETUP:
+	case RDMA_GET_MEM:
+		if (!offload_running(adapter))
+			return -EAGAIN;
+		return cxgb_rdma_ctl(adapter, req, data);
+	default:
+		return -EOPNOTSUPP;
+	}
+	return 0;
+}
+
+/*
+ * Dummy handler for Rx offload packets in case we get an offload packet before
+ * proper processing is setup.  This complains and drops the packet as it isn't
+ * normal to get offload packets at this stage.
+ */
+static int rx_offload_blackhole(struct t3cdev *dev, struct sk_buff **skbs,
+				int n)
+{
+	CH_ERR(tdev2adap(dev), "%d unexpected offload packets, first data %u\n",
+	       n, ntohl(*(u32 *)skbs[0]->data));
+	while (n--)
+		dev_kfree_skb_any(skbs[n]);
+	return 0;
+}
+
+static void dummy_neigh_update(struct t3cdev *dev, struct neighbour *neigh)
+{
+}
+
+void cxgb3_set_dummy_ops(struct t3cdev *dev)
+{
+	dev->recv = rx_offload_blackhole;
+	dev->neigh_update = dummy_neigh_update;
+}
+
+/*
+ * Free an active-open TID.
+ */
+void *cxgb3_free_atid(struct t3cdev *tdev, int atid)
+{
+	struct tid_info *t = &(T3C_DATA(tdev))->tid_maps;
+	union active_open_entry *p = atid2entry(t, atid);
+	void *ctx = p->t3c_tid.ctx;
+
+	spin_lock_bh(&t->atid_lock);
+	p->next = t->afree;
+	t->afree = p;
+	t->atids_in_use--;
+	spin_unlock_bh(&t->atid_lock);
+
+	return ctx;
+}
+
+EXPORT_SYMBOL(cxgb3_free_atid);
+
+/*
+ * Free a server TID and return it to the free pool.
+ */
+void cxgb3_free_stid(struct t3cdev *tdev, int stid)
+{
+	struct tid_info *t = &(T3C_DATA(tdev))->tid_maps;
+	union listen_entry *p = stid2entry(t, stid);
+
+	spin_lock_bh(&t->stid_lock);
+	p->next = t->sfree;
+	t->sfree = p;
+	t->stids_in_use--;
+	spin_unlock_bh(&t->stid_lock);
+}
+
+EXPORT_SYMBOL(cxgb3_free_stid);
+
+void cxgb3_insert_tid(struct t3cdev *tdev, struct cxgb3_client *client,
+		      void *ctx, unsigned int tid)
+{
+	struct tid_info *t = &(T3C_DATA(tdev))->tid_maps;
+
+	t->tid_tab[tid].client = client;
+	t->tid_tab[tid].ctx = ctx;
+	atomic_inc(&t->tids_in_use);
+}
+
+EXPORT_SYMBOL(cxgb3_insert_tid);
+
+/*
+ * Populate a TID_RELEASE WR.  The skb must be already propely sized.
+ */
+static inline void mk_tid_release(struct sk_buff *skb, unsigned int tid)
+{
+	struct cpl_tid_release *req;
+
+	skb->priority = CPL_PRIORITY_SETUP;
+	req = (struct cpl_tid_release *)__skb_put(skb, sizeof(*req));
+	req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
+	OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_TID_RELEASE, tid));
+}
+
+static void t3_process_tid_release_list(struct work_struct *work)
+{
+	struct t3c_data *td = container_of(work, struct t3c_data,
+					   tid_release_task);
+	struct sk_buff *skb;
+	struct t3cdev *tdev = td->dev;
+	
+
+	spin_lock_bh(&td->tid_release_lock);
+	while (td->tid_release_list) {
+		struct t3c_tid_entry *p = td->tid_release_list;
+
+		td->tid_release_list = (struct t3c_tid_entry *)p->ctx;
+		spin_unlock_bh(&td->tid_release_lock);
+
+		skb = alloc_skb(sizeof(struct cpl_tid_release),
+				GFP_KERNEL | __GFP_NOFAIL);
+		mk_tid_release(skb, p - td->tid_maps.tid_tab);
+		cxgb3_ofld_send(tdev, skb);
+		p->ctx = NULL;
+		spin_lock_bh(&td->tid_release_lock);
+	}
+	spin_unlock_bh(&td->tid_release_lock);
+}
+
+/* use ctx as a next pointer in the tid release list */
+void cxgb3_queue_tid_release(struct t3cdev *tdev, unsigned int tid)
+{
+	struct t3c_data *td = T3C_DATA(tdev);
+	struct t3c_tid_entry *p = &td->tid_maps.tid_tab[tid];
+
+	spin_lock_bh(&td->tid_release_lock);
+	p->ctx = (void *)td->tid_release_list;
+	td->tid_release_list = p;
+	if (!p->ctx)
+		schedule_work(&td->tid_release_task);
+	spin_unlock_bh(&td->tid_release_lock);
+}
+
+EXPORT_SYMBOL(cxgb3_queue_tid_release);
+
+/*
+ * Remove a tid from the TID table.  A client may defer processing its last
+ * CPL message if it is locked at the time it arrives, and while the message
+ * sits in the client's backlog the TID may be reused for another connection.
+ * To handle this we atomically switch the TID association if it still points
+ * to the original client context.
+ */
+void cxgb3_remove_tid(struct t3cdev *tdev, void *ctx, unsigned int tid)
+{
+	struct tid_info *t = &(T3C_DATA(tdev))->tid_maps;
+
+	BUG_ON(tid >= t->ntids);
+	if (tdev->type == T3A)
+		(void)cmpxchg(&t->tid_tab[tid].ctx, ctx, NULL);
+	else {
+		struct sk_buff *skb;
+
+		skb = alloc_skb(sizeof(struct cpl_tid_release), GFP_ATOMIC);
+		if (likely(skb)) {
+			mk_tid_release(skb, tid);
+			cxgb3_ofld_send(tdev, skb);
+			t->tid_tab[tid].ctx = NULL;
+		} else
+			cxgb3_queue_tid_release(tdev, tid);
+	}
+	atomic_dec(&t->tids_in_use);
+}
+
+EXPORT_SYMBOL(cxgb3_remove_tid);
+
+int cxgb3_alloc_atid(struct t3cdev *tdev, struct cxgb3_client *client,
+		     void *ctx)
+{
+	int atid = -1;
+	struct tid_info *t = &(T3C_DATA(tdev))->tid_maps;
+
+	spin_lock_bh(&t->atid_lock);
+	if (t->afree) {
+		union active_open_entry *p = t->afree;
+
+		atid = (p - t->atid_tab) + t->atid_base;
+		t->afree = p->next;
+		p->t3c_tid.ctx = ctx;
+		p->t3c_tid.client = client;
+		t->atids_in_use++;
+	}
+	spin_unlock_bh(&t->atid_lock);
+	return atid;
+}
+
+EXPORT_SYMBOL(cxgb3_alloc_atid);
+
+int cxgb3_alloc_stid(struct t3cdev *tdev, struct cxgb3_client *client,
+		     void *ctx)
+{
+	int stid = -1;
+	struct tid_info *t = &(T3C_DATA(tdev))->tid_maps;
+
+	spin_lock_bh(&t->stid_lock);
+	if (t->sfree) {
+		union listen_entry *p = t->sfree;
+
+		stid = (p - t->stid_tab) + t->stid_base;
+		t->sfree = p->next;
+		p->t3c_tid.ctx = ctx;
+		p->t3c_tid.client = client;
+		t->stids_in_use++;
+	}
+	spin_unlock_bh(&t->stid_lock);
+	return stid;
+}
+
+EXPORT_SYMBOL(cxgb3_alloc_stid);
+
+static int do_smt_write_rpl(struct t3cdev *dev, struct sk_buff *skb)
+{
+	struct cpl_smt_write_rpl *rpl = cplhdr(skb);
+
+	if (rpl->status != CPL_ERR_NONE)
+		printk(KERN_ERR
+		       "Unexpected SMT_WRITE_RPL status %u for entry %u\n",
+		       rpl->status, GET_TID(rpl));
+
+	return CPL_RET_BUF_DONE;
+}
+
+static int do_l2t_write_rpl(struct t3cdev *dev, struct sk_buff *skb)
+{
+	struct cpl_l2t_write_rpl *rpl = cplhdr(skb);
+
+	if (rpl->status != CPL_ERR_NONE)
+		printk(KERN_ERR
+		       "Unexpected L2T_WRITE_RPL status %u for entry %u\n",
+		       rpl->status, GET_TID(rpl));
+
+	return CPL_RET_BUF_DONE;
+}
+
+static int do_act_open_rpl(struct t3cdev *dev, struct sk_buff *skb)
+{
+	struct cpl_act_open_rpl *rpl = cplhdr(skb);
+	unsigned int atid = G_TID(ntohl(rpl->atid));
+	struct t3c_tid_entry *t3c_tid;
+
+	t3c_tid = lookup_atid(&(T3C_DATA(dev))->tid_maps, atid);
+	if (t3c_tid->ctx && t3c_tid->client && t3c_tid->client->handlers &&
+	    t3c_tid->client->handlers[CPL_ACT_OPEN_RPL]) {
+		return t3c_tid->client->handlers[CPL_ACT_OPEN_RPL] (dev, skb,
+								    t3c_tid->
+								    ctx);
+	} else {
+		printk(KERN_ERR "%s: received clientless CPL command 0x%x\n",
+		       dev->name, CPL_ACT_OPEN_RPL);
+		return CPL_RET_BUF_DONE | CPL_RET_BAD_MSG;
+	}
+}
+
+static int do_stid_rpl(struct t3cdev *dev, struct sk_buff *skb)
+{
+	union opcode_tid *p = cplhdr(skb);
+	unsigned int stid = G_TID(ntohl(p->opcode_tid));
+	struct t3c_tid_entry *t3c_tid;
+
+	t3c_tid = lookup_stid(&(T3C_DATA(dev))->tid_maps, stid);
+	if (t3c_tid->ctx && t3c_tid->client->handlers &&
+	    t3c_tid->client->handlers[p->opcode]) {
+		return t3c_tid->client->handlers[p->opcode] (dev, skb,
+							     t3c_tid->ctx);
+	} else {
+		printk(KERN_ERR "%s: received clientless CPL command 0x%x\n",
+		       dev->name, p->opcode);
+		return CPL_RET_BUF_DONE | CPL_RET_BAD_MSG;
+	}
+}
+
+static int do_hwtid_rpl(struct t3cdev *dev, struct sk_buff *skb)
+{
+	union opcode_tid *p = cplhdr(skb);
+	unsigned int hwtid = G_TID(ntohl(p->opcode_tid));
+	struct t3c_tid_entry *t3c_tid;
+
+	t3c_tid = lookup_tid(&(T3C_DATA(dev))->tid_maps, hwtid);
+	if (t3c_tid->ctx && t3c_tid->client->handlers &&
+	    t3c_tid->client->handlers[p->opcode]) {
+		return t3c_tid->client->handlers[p->opcode]
+		    (dev, skb, t3c_tid->ctx);
+	} else {
+		printk(KERN_ERR "%s: received clientless CPL command 0x%x\n",
+		       dev->name, p->opcode);
+		return CPL_RET_BUF_DONE | CPL_RET_BAD_MSG;
+	}
+}
+
+static int do_cr(struct t3cdev *dev, struct sk_buff *skb)
+{
+	struct cpl_pass_accept_req *req = cplhdr(skb);
+	unsigned int stid = G_PASS_OPEN_TID(ntohl(req->tos_tid));
+	struct t3c_tid_entry *t3c_tid;
+
+	t3c_tid = lookup_stid(&(T3C_DATA(dev))->tid_maps, stid);
+	if (t3c_tid->ctx && t3c_tid->client->handlers &&
+	    t3c_tid->client->handlers[CPL_PASS_ACCEPT_REQ]) {
+		return t3c_tid->client->handlers[CPL_PASS_ACCEPT_REQ]
+		    (dev, skb, t3c_tid->ctx);
+	} else {
+		printk(KERN_ERR "%s: received clientless CPL command 0x%x\n",
+		       dev->name, CPL_PASS_ACCEPT_REQ);
+		return CPL_RET_BUF_DONE | CPL_RET_BAD_MSG;
+	}
+}
+
+static int do_abort_req_rss(struct t3cdev *dev, struct sk_buff *skb)
+{
+	union opcode_tid *p = cplhdr(skb);
+	unsigned int hwtid = G_TID(ntohl(p->opcode_tid));
+	struct t3c_tid_entry *t3c_tid;
+
+	t3c_tid = lookup_tid(&(T3C_DATA(dev))->tid_maps, hwtid);
+	if (t3c_tid->ctx && t3c_tid->client->handlers &&
+	    t3c_tid->client->handlers[p->opcode]) {
+		return t3c_tid->client->handlers[p->opcode]
+		    (dev, skb, t3c_tid->ctx);
+	} else {
+		struct cpl_abort_req_rss *req = cplhdr(skb);
+		struct cpl_abort_rpl *rpl;
+
+		struct sk_buff *skb =
+		    alloc_skb(sizeof(struct cpl_abort_rpl), GFP_ATOMIC);
+		if (!skb) {
+			printk("do_abort_req_rss: couldn't get skb!\n");
+			goto out;
+		}
+		skb->priority = CPL_PRIORITY_DATA;
+		__skb_put(skb, sizeof(struct cpl_abort_rpl));
+		rpl = cplhdr(skb);
+		rpl->wr.wr_hi =
+		    htonl(V_WR_OP(FW_WROPCODE_OFLD_HOST_ABORT_CON_RPL));
+		rpl->wr.wr_lo = htonl(V_WR_TID(GET_TID(req)));
+		OPCODE_TID(rpl) =
+		    htonl(MK_OPCODE_TID(CPL_ABORT_RPL, GET_TID(req)));
+		rpl->cmd = req->status;
+		cxgb3_ofld_send(dev, skb);
+out:
+		return CPL_RET_BUF_DONE;
+	}
+}
+
+static int do_act_establish(struct t3cdev *dev, struct sk_buff *skb)
+{
+	struct cpl_act_establish *req = cplhdr(skb);
+	unsigned int atid = G_PASS_OPEN_TID(ntohl(req->tos_tid));
+	struct t3c_tid_entry *t3c_tid;
+
+	t3c_tid = lookup_atid(&(T3C_DATA(dev))->tid_maps, atid);
+	if (t3c_tid->ctx && t3c_tid->client->handlers &&
+	    t3c_tid->client->handlers[CPL_ACT_ESTABLISH]) {
+		return t3c_tid->client->handlers[CPL_ACT_ESTABLISH]
+		    (dev, skb, t3c_tid->ctx);
+	} else {
+		printk(KERN_ERR "%s: received clientless CPL command 0x%x\n",
+		       dev->name, CPL_PASS_ACCEPT_REQ);
+		return CPL_RET_BUF_DONE | CPL_RET_BAD_MSG;
+	}
+}
+
+static int do_set_tcb_rpl(struct t3cdev *dev, struct sk_buff *skb)
+{
+	struct cpl_set_tcb_rpl *rpl = cplhdr(skb);
+
+	if (rpl->status != CPL_ERR_NONE)
+		printk(KERN_ERR
+		       "Unexpected SET_TCB_RPL status %u for tid %u\n",
+		       rpl->status, GET_TID(rpl));
+	return CPL_RET_BUF_DONE;
+}
+
+static int do_trace(struct t3cdev *dev, struct sk_buff *skb)
+{
+	struct cpl_trace_pkt *p = cplhdr(skb);
+
+	skb->protocol = 0xffff;
+	skb->dev = dev->lldev;
+	skb_pull(skb, sizeof(*p));
+	skb->mac.raw = skb->data;
+	netif_receive_skb(skb);
+	return 0;
+}
+
+static int do_term(struct t3cdev *dev, struct sk_buff *skb)
+{
+	unsigned int hwtid = ntohl(skb->priority) >> 8 & 0xfffff;
+	unsigned int opcode = G_OPCODE(ntohl(skb->csum));
+	struct t3c_tid_entry *t3c_tid;
+
+	t3c_tid = lookup_tid(&(T3C_DATA(dev))->tid_maps, hwtid);
+	if (t3c_tid->ctx && t3c_tid->client->handlers &&
+	    t3c_tid->client->handlers[opcode]) {
+		return t3c_tid->client->handlers[opcode] (dev, skb,
+							  t3c_tid->ctx);
+	} else {
+		printk(KERN_ERR "%s: received clientless CPL command 0x%x\n",
+		       dev->name, opcode);
+		return CPL_RET_BUF_DONE | CPL_RET_BAD_MSG;
+	}
+}
+
+static int nb_callback(struct notifier_block *self, unsigned long event,
+		       void *ctx)
+{
+	switch (event) {
+	case (NETEVENT_NEIGH_UPDATE):{
+		cxgb_neigh_update((struct neighbour *)ctx);
+		break;
+	}
+	case (NETEVENT_PMTU_UPDATE):
+		break;
+	case (NETEVENT_REDIRECT):{
+		struct netevent_redirect *nr = ctx;
+		cxgb_redirect(nr->old, nr->new);
+		cxgb_neigh_update(nr->new->neighbour);
+		break;
+	}
+	default:
+		break;
+	}
+	return 0;
+}
+
+static struct notifier_block nb = {
+	.notifier_call = nb_callback
+};
+
+/*
+ * Process a received packet with an unknown/unexpected CPL opcode.
+ */
+static int do_bad_cpl(struct t3cdev *dev, struct sk_buff *skb)
+{
+	printk(KERN_ERR "%s: received bad CPL command 0x%x\n", dev->name,
+	       *skb->data);
+	return CPL_RET_BUF_DONE | CPL_RET_BAD_MSG;
+}
+
+/*
+ * Handlers for each CPL opcode
+ */
+static cpl_handler_func cpl_handlers[NUM_CPL_CMDS];
+
+/*
+ * Add a new handler to the CPL dispatch table.  A NULL handler may be supplied
+ * to unregister an existing handler.
+ */
+void t3_register_cpl_handler(unsigned int opcode, cpl_handler_func h)
+{
+	if (opcode < NUM_CPL_CMDS)
+		cpl_handlers[opcode] = h ? h : do_bad_cpl;
+	else
+		printk(KERN_ERR "T3C: handler registration for "
+		       "opcode %x failed\n", opcode);
+}
+
+EXPORT_SYMBOL(t3_register_cpl_handler);
+
+/*
+ * T3CDEV's receive method.
+ */
+int process_rx(struct t3cdev *dev, struct sk_buff **skbs, int n)
+{
+	while (n--) {
+		struct sk_buff *skb = *skbs++;
+		unsigned int opcode = G_OPCODE(ntohl(skb->csum));
+		int ret = cpl_handlers[opcode] (dev, skb);
+
+#if VALIDATE_TID
+		if (ret & CPL_RET_UNKNOWN_TID) {
+			union opcode_tid *p = cplhdr(skb);
+
+			printk(KERN_ERR "%s: CPL message (opcode %u) had "
+			       "unknown TID %u\n", dev->name, opcode,
+			       G_TID(ntohl(p->opcode_tid)));
+		}
+#endif
+		if (ret & CPL_RET_BUF_DONE)
+			kfree_skb(skb);
+	}
+	return 0;
+}
+
+/*
+ * Sends an sk_buff to a T3C driver after dealing with any active network taps.
+ */
+int cxgb3_ofld_send(struct t3cdev *dev, struct sk_buff *skb)
+{
+	int r;
+
+	local_bh_disable();
+	r = dev->send(dev, skb);
+	local_bh_enable();
+	return r;
+}
+
+EXPORT_SYMBOL(cxgb3_ofld_send);
+
+static int is_offloading(struct net_device *dev)
+{
+	struct adapter *adapter;
+	int i;
+
+	read_lock_bh(&adapter_list_lock);
+	list_for_each_entry(adapter, &adapter_list, adapter_list) {
+		for_each_port(adapter, i) {
+			if (dev == adapter->port[i]) {
+				read_unlock_bh(&adapter_list_lock);
+				return 1;
+			}
+		}
+	}
+	read_unlock_bh(&adapter_list_lock);
+	return 0;
+}
+
+void cxgb_neigh_update(struct neighbour *neigh)
+{
+	struct net_device *dev = neigh->dev;
+
+	if (dev && (is_offloading(dev))) {
+		struct t3cdev *tdev = T3CDEV(dev);
+
+		BUG_ON(!tdev);
+		t3_l2t_update(tdev, neigh);
+	}
+}
+
+static void set_l2t_ix(struct t3cdev *tdev, u32 tid, struct l2t_entry *e)
+{
+	struct sk_buff *skb;
+	struct cpl_set_tcb_field *req;
+
+	skb = alloc_skb(sizeof(*req), GFP_ATOMIC);
+	if (!skb) {
+		printk(KERN_ERR "%s: cannot allocate skb!\n", __FUNCTION__);
+		return;
+	}
+	skb->priority = CPL_PRIORITY_CONTROL;
+	req = (struct cpl_set_tcb_field *)skb_put(skb, sizeof(*req));
+	req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
+	OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_SET_TCB_FIELD, tid));
+	req->reply = 0;
+	req->cpu_idx = 0;
+	req->word = htons(W_TCB_L2T_IX);
+	req->mask = cpu_to_be64(V_TCB_L2T_IX(M_TCB_L2T_IX));
+	req->val = cpu_to_be64(V_TCB_L2T_IX(e->idx));
+	tdev->send(tdev, skb);
+}
+
+void cxgb_redirect(struct dst_entry *old, struct dst_entry *new)
+{
+	struct net_device *olddev, *newdev;
+	struct tid_info *ti;
+	struct t3cdev *tdev;
+	u32 tid;
+	int update_tcb;
+	struct l2t_entry *e;
+	struct t3c_tid_entry *te;
+
+	olddev = old->neighbour->dev;
+	newdev = new->neighbour->dev;
+	if (!is_offloading(olddev))
+		return;
+	if (!is_offloading(newdev)) {
+		printk(KERN_WARNING "%s: Redirect to non-offload"
+		       "device ignored.\n", __FUNCTION__);
+		return;
+	}
+	tdev = T3CDEV(olddev);
+	BUG_ON(!tdev);
+	if (tdev != T3CDEV(newdev)) {
+		printk(KERN_WARNING "%s: Redirect to different "
+		       "offload device ignored.\n", __FUNCTION__);
+		return;
+	}
+
+	/* Add new L2T entry */
+	e = t3_l2t_get(tdev, new->neighbour, newdev);
+	if (!e) {
+		printk(KERN_ERR "%s: couldn't allocate new l2t entry!\n",
+		       __FUNCTION__);
+		return;
+	}
+
+	/* Walk tid table and notify clients of dst change. */
+	ti = &(T3C_DATA(tdev))->tid_maps;
+	for (tid = 0; tid < ti->ntids; tid++) {
+		te = lookup_tid(ti, tid);
+		BUG_ON(!te);
+		if (te->ctx && te->client && te->client->redirect) {
+			update_tcb = te->client->redirect(te->ctx, old, new, e);
+			if (update_tcb) {
+				l2t_hold(L2DATA(tdev), e);
+				set_l2t_ix(tdev, tid, e);
+			}
+		}
+	}
+	l2t_release(L2DATA(tdev), e);
+}
+
+/*
+ * Allocate a chunk of memory using kmalloc or, if that fails, vmalloc.
+ * The allocated memory is cleared.
+ */
+void *cxgb_alloc_mem(unsigned long size)
+{
+	void *p = kmalloc(size, GFP_KERNEL);
+
+	if (!p)
+		p = vmalloc(size);
+	if (p)
+		memset(p, 0, size);
+	return p;
+}
+
+/*
+ * Free memory allocated through t3_alloc_mem().
+ */
+void cxgb_free_mem(void *addr)
+{
+	unsigned long p = (unsigned long)addr;
+
+	if (p >= VMALLOC_START && p < VMALLOC_END)
+		vfree(addr);
+	else
+		kfree(addr);
+}
+
+/*
+ * Allocate and initialize the TID tables.  Returns 0 on success.
+ */
+static int init_tid_tabs(struct tid_info *t, unsigned int ntids,
+			 unsigned int natids, unsigned int nstids,
+			 unsigned int atid_base, unsigned int stid_base)
+{
+	unsigned long size = ntids * sizeof(*t->tid_tab) +
+	    natids * sizeof(*t->atid_tab) + nstids * sizeof(*t->stid_tab);
+
+	t->tid_tab = cxgb_alloc_mem(size);
+	if (!t->tid_tab)
+		return -ENOMEM;
+
+	t->stid_tab = (union listen_entry *)&t->tid_tab[ntids];
+	t->atid_tab = (union active_open_entry *)&t->stid_tab[nstids];
+	t->ntids = ntids;
+	t->nstids = nstids;
+	t->stid_base = stid_base;
+	t->sfree = NULL;
+	t->natids = natids;
+	t->atid_base = atid_base;
+	t->afree = NULL;
+	t->stids_in_use = t->atids_in_use = 0;
+	atomic_set(&t->tids_in_use, 0);
+	spin_lock_init(&t->stid_lock);
+	spin_lock_init(&t->atid_lock);
+
+	/*
+	 * Setup the free lists for stid_tab and atid_tab.
+	 */
+	if (nstids) {
+		while (--nstids)
+			t->stid_tab[nstids - 1].next = &t->stid_tab[nstids];
+		t->sfree = t->stid_tab;
+	}
+	if (natids) {
+		while (--natids)
+			t->atid_tab[natids - 1].next = &t->atid_tab[natids];
+		t->afree = t->atid_tab;
+	}
+	return 0;
+}
+
+static void free_tid_maps(struct tid_info *t)
+{
+	cxgb_free_mem(t->tid_tab);
+}
+
+static inline void add_adapter(struct adapter *adap)
+{
+	write_lock_bh(&adapter_list_lock);
+	list_add_tail(&adap->adapter_list, &adapter_list);
+	write_unlock_bh(&adapter_list_lock);
+}
+
+static inline void remove_adapter(struct adapter *adap)
+{
+	write_lock_bh(&adapter_list_lock);
+	list_del(&adap->adapter_list);
+	write_unlock_bh(&adapter_list_lock);
+}
+
+int cxgb3_offload_activate(struct adapter *adapter)
+{
+	struct t3cdev *dev = &adapter->tdev;
+	int natids, err;
+	struct t3c_data *t;
+	struct tid_range stid_range, tid_range;
+	struct mtutab mtutab;
+	unsigned int l2t_capacity;
+
+	t = kcalloc(1, sizeof(*t), GFP_KERNEL);
+	if (!t)
+		return -ENOMEM;
+
+	err = -EOPNOTSUPP;
+	if (dev->ctl(dev, GET_TX_MAX_CHUNK, &t->tx_max_chunk) < 0 ||
+	    dev->ctl(dev, GET_MAX_OUTSTANDING_WR, &t->max_wrs) < 0 ||
+	    dev->ctl(dev, GET_L2T_CAPACITY, &l2t_capacity) < 0 ||
+	    dev->ctl(dev, GET_MTUS, &mtutab) < 0 ||
+	    dev->ctl(dev, GET_TID_RANGE, &tid_range) < 0 ||
+	    dev->ctl(dev, GET_STID_RANGE, &stid_range) < 0)
+		goto out_free;
+
+	err = -ENOMEM;
+	L2DATA(dev) = t3_init_l2t(l2t_capacity);
+	if (!L2DATA(dev))
+		goto out_free;
+
+	natids = min(tid_range.num / 2, MAX_ATIDS);
+	err = init_tid_tabs(&t->tid_maps, tid_range.num, natids,
+			    stid_range.num, ATID_BASE, stid_range.base);
+	if (err)
+		goto out_free_l2t;
+
+	t->mtus = mtutab.mtus;
+	t->nmtus = mtutab.size;
+
+	INIT_WORK(&t->tid_release_task, t3_process_tid_release_list);
+	spin_lock_init(&t->tid_release_lock);
+	INIT_LIST_HEAD(&t->list_node);
+	t->dev = dev;
+
+	T3C_DATA(dev) = t;
+	dev->recv = process_rx;
+	dev->neigh_update = t3_l2t_update;
+
+	/* Register netevent handler once */
+	if (list_empty(&adapter_list))
+		register_netevent_notifier(&nb);
+
+	add_adapter(adapter);
+	return 0;
+
+out_free_l2t:
+	t3_free_l2t(L2DATA(dev));
+	L2DATA(dev) = NULL;
+out_free:
+	kfree(t);
+	return err;
+}
+
+void cxgb3_offload_deactivate(struct adapter *adapter)
+{
+	struct t3cdev *tdev = &adapter->tdev;
+	struct t3c_data *t = T3C_DATA(tdev);
+
+	remove_adapter(adapter);
+	if (list_empty(&adapter_list))
+		unregister_netevent_notifier(&nb);
+
+	free_tid_maps(&t->tid_maps);
+	T3C_DATA(tdev) = NULL;
+	t3_free_l2t(L2DATA(tdev));
+	L2DATA(tdev) = NULL;
+	kfree(t);
+}
+
+static inline void register_tdev(struct t3cdev *tdev)
+{
+	static int unit;
+
+	mutex_lock(&cxgb3_db_lock);
+	snprintf(tdev->name, sizeof(tdev->name), "ofld_dev%d", unit++);
+	list_add_tail(&tdev->ofld_dev_list, &ofld_dev_list);
+	mutex_unlock(&cxgb3_db_lock);
+}
+
+static inline void unregister_tdev(struct t3cdev *tdev)
+{
+	mutex_lock(&cxgb3_db_lock);
+	list_del(&tdev->ofld_dev_list);
+	mutex_unlock(&cxgb3_db_lock);
+}
+
+void __devinit cxgb3_adapter_ofld(struct adapter *adapter)
+{
+	struct t3cdev *tdev = &adapter->tdev;
+
+	INIT_LIST_HEAD(&tdev->ofld_dev_list);
+
+	cxgb3_set_dummy_ops(tdev);
+	tdev->send = t3_offload_tx;
+	tdev->ctl = cxgb_offload_ctl;
+	tdev->type = adapter->params.rev == 0 ? T3A : T3B;
+
+	register_tdev(tdev);
+}
+
+void __devexit cxgb3_adapter_unofld(struct adapter *adapter)
+{
+	struct t3cdev *tdev = &adapter->tdev;
+
+	tdev->recv = NULL;
+	tdev->neigh_update = NULL;
+
+	unregister_tdev(tdev);
+}
+
+void __init cxgb3_offload_init(void)
+{
+	int i;
+
+	for (i = 0; i < NUM_CPL_CMDS; ++i)
+		cpl_handlers[i] = do_bad_cpl;
+
+	t3_register_cpl_handler(CPL_SMT_WRITE_RPL, do_smt_write_rpl);
+	t3_register_cpl_handler(CPL_L2T_WRITE_RPL, do_l2t_write_rpl);
+	t3_register_cpl_handler(CPL_PASS_OPEN_RPL, do_stid_rpl);
+	t3_register_cpl_handler(CPL_CLOSE_LISTSRV_RPL, do_stid_rpl);
+	t3_register_cpl_handler(CPL_PASS_ACCEPT_REQ, do_cr);
+	t3_register_cpl_handler(CPL_PASS_ESTABLISH, do_hwtid_rpl);
+	t3_register_cpl_handler(CPL_ABORT_RPL_RSS, do_hwtid_rpl);
+	t3_register_cpl_handler(CPL_ABORT_RPL, do_hwtid_rpl);
+	t3_register_cpl_handler(CPL_RX_URG_NOTIFY, do_hwtid_rpl);
+	t3_register_cpl_handler(CPL_RX_DATA, do_hwtid_rpl);
+	t3_register_cpl_handler(CPL_TX_DATA_ACK, do_hwtid_rpl);
+	t3_register_cpl_handler(CPL_TX_DMA_ACK, do_hwtid_rpl);
+	t3_register_cpl_handler(CPL_ACT_OPEN_RPL, do_act_open_rpl);
+	t3_register_cpl_handler(CPL_PEER_CLOSE, do_hwtid_rpl);
+	t3_register_cpl_handler(CPL_CLOSE_CON_RPL, do_hwtid_rpl);
+	t3_register_cpl_handler(CPL_ABORT_REQ_RSS, do_abort_req_rss);
+	t3_register_cpl_handler(CPL_ACT_ESTABLISH, do_act_establish);
+	t3_register_cpl_handler(CPL_SET_TCB_RPL, do_set_tcb_rpl);
+	t3_register_cpl_handler(CPL_RDMA_TERMINATE, do_term);
+	t3_register_cpl_handler(CPL_RDMA_EC_STATUS, do_hwtid_rpl);
+	t3_register_cpl_handler(CPL_TRACE_PKT, do_trace);
+	t3_register_cpl_handler(CPL_RX_DATA_DDP, do_hwtid_rpl);
+	t3_register_cpl_handler(CPL_RX_DDP_COMPLETE, do_hwtid_rpl);
+	t3_register_cpl_handler(CPL_ISCSI_HDR, do_hwtid_rpl);
+}
diff --git a/drivers/net/cxgb3/cxgb3_offload.h b/drivers/net/cxgb3/cxgb3_offload.h
new file mode 100644
index 00000000000..7b7798384ae
--- /dev/null
+++ b/drivers/net/cxgb3/cxgb3_offload.h
@@ -0,0 +1,193 @@
+/*
+ * Copyright (c) 2006 Chelsio, Inc. All rights reserved.
+ * Copyright (c) 2006 Open Grid Computing, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef _CXGB3_OFFLOAD_H
+#define _CXGB3_OFFLOAD_H
+
+#include <linux/list.h>
+#include <linux/skbuff.h>
+
+#include "l2t.h"
+
+#include "t3cdev.h"
+#include "t3_cpl.h"
+
+struct adapter;
+
+void cxgb3_offload_init(void);
+
+void cxgb3_adapter_ofld(struct adapter *adapter);
+void cxgb3_adapter_unofld(struct adapter *adapter);
+int cxgb3_offload_activate(struct adapter *adapter);
+void cxgb3_offload_deactivate(struct adapter *adapter);
+
+void cxgb3_set_dummy_ops(struct t3cdev *dev);
+
+/*
+ * Client registration.  Users of T3 driver must register themselves.
+ * The T3 driver will call the add function of every client for each T3
+ * adapter activated, passing up the t3cdev ptr.  Each client fills out an
+ * array of callback functions to process CPL messages.
+ */
+
+void cxgb3_register_client(struct cxgb3_client *client);
+void cxgb3_unregister_client(struct cxgb3_client *client);
+void cxgb3_add_clients(struct t3cdev *tdev);
+void cxgb3_remove_clients(struct t3cdev *tdev);
+
+typedef int (*cxgb3_cpl_handler_func)(struct t3cdev *dev,
+				      struct sk_buff *skb, void *ctx);
+
+struct cxgb3_client {
+	char *name;
+	void (*add) (struct t3cdev *);
+	void (*remove) (struct t3cdev *);
+	cxgb3_cpl_handler_func *handlers;
+	int (*redirect)(void *ctx, struct dst_entry *old,
+			struct dst_entry *new, struct l2t_entry *l2t);
+	struct list_head client_list;
+};
+
+/*
+ * TID allocation services.
+ */
+int cxgb3_alloc_atid(struct t3cdev *dev, struct cxgb3_client *client,
+		     void *ctx);
+int cxgb3_alloc_stid(struct t3cdev *dev, struct cxgb3_client *client,
+		     void *ctx);
+void *cxgb3_free_atid(struct t3cdev *dev, int atid);
+void cxgb3_free_stid(struct t3cdev *dev, int stid);
+void cxgb3_insert_tid(struct t3cdev *dev, struct cxgb3_client *client,
+		      void *ctx, unsigned int tid);
+void cxgb3_queue_tid_release(struct t3cdev *dev, unsigned int tid);
+void cxgb3_remove_tid(struct t3cdev *dev, void *ctx, unsigned int tid);
+
+struct t3c_tid_entry {
+	struct cxgb3_client *client;
+	void *ctx;
+};
+
+/* CPL message priority levels */
+enum {
+	CPL_PRIORITY_DATA = 0,	/* data messages */
+	CPL_PRIORITY_SETUP = 1,	/* connection setup messages */
+	CPL_PRIORITY_TEARDOWN = 0,	/* connection teardown messages */
+	CPL_PRIORITY_LISTEN = 1,	/* listen start/stop messages */
+	CPL_PRIORITY_ACK = 1,	/* RX ACK messages */
+	CPL_PRIORITY_CONTROL = 1	/* offload control messages */
+};
+
+/* Flags for return value of CPL message handlers */
+enum {
+	CPL_RET_BUF_DONE = 1, /* buffer processing done, buffer may be freed */
+	CPL_RET_BAD_MSG = 2,  /* bad CPL message (e.g., unknown opcode) */
+	CPL_RET_UNKNOWN_TID = 4	/* unexpected unknown TID */
+};
+
+typedef int (*cpl_handler_func)(struct t3cdev *dev, struct sk_buff *skb);
+
+/*
+ * Returns a pointer to the first byte of the CPL header in an sk_buff that
+ * contains a CPL message.
+ */
+static inline void *cplhdr(struct sk_buff *skb)
+{
+	return skb->data;
+}
+
+void t3_register_cpl_handler(unsigned int opcode, cpl_handler_func h);
+
+union listen_entry {
+	struct t3c_tid_entry t3c_tid;
+	union listen_entry *next;
+};
+
+union active_open_entry {
+	struct t3c_tid_entry t3c_tid;
+	union active_open_entry *next;
+};
+
+/*
+ * Holds the size, base address, free list start, etc of the TID, server TID,
+ * and active-open TID tables for a offload device.
+ * The tables themselves are allocated dynamically.
+ */
+struct tid_info {
+	struct t3c_tid_entry *tid_tab;
+	unsigned int ntids;
+	atomic_t tids_in_use;
+
+	union listen_entry *stid_tab;
+	unsigned int nstids;
+	unsigned int stid_base;
+
+	union active_open_entry *atid_tab;
+	unsigned int natids;
+	unsigned int atid_base;
+
+	/*
+	 * The following members are accessed R/W so we put them in their own
+	 * cache lines.
+	 *
+	 * XXX We could combine the atid fields above with the lock here since
+	 * atids are use once (unlike other tids).  OTOH the above fields are
+	 * usually in cache due to tid_tab.
+	 */
+	spinlock_t atid_lock ____cacheline_aligned_in_smp;
+	union active_open_entry *afree;
+	unsigned int atids_in_use;
+
+	spinlock_t stid_lock ____cacheline_aligned;
+	union listen_entry *sfree;
+	unsigned int stids_in_use;
+};
+
+struct t3c_data {
+	struct list_head list_node;
+	struct t3cdev *dev;
+	unsigned int tx_max_chunk;	/* max payload for TX_DATA */
+	unsigned int max_wrs;	/* max in-flight WRs per connection */
+	unsigned int nmtus;
+	const unsigned short *mtus;
+	struct tid_info tid_maps;
+
+	struct t3c_tid_entry *tid_release_list;
+	spinlock_t tid_release_lock;
+	struct work_struct tid_release_task;
+};
+
+/*
+ * t3cdev -> t3c_data accessor
+ */
+#define T3C_DATA(dev) (*(struct t3c_data **)&(dev)->l4opt)
+
+#endif
diff --git a/drivers/net/cxgb3/firmware_exports.h b/drivers/net/cxgb3/firmware_exports.h
new file mode 100644
index 00000000000..3565f481801
--- /dev/null
+++ b/drivers/net/cxgb3/firmware_exports.h
@@ -0,0 +1,144 @@
+/* 
+ * ----------------------------------------------------------------------------
+ * >>>>>>>>>>>>>>>>>>>>>>>>>>>>> COPYRIGHT NOTICE <<<<<<<<<<<<<<<<<<<<<<<<<<<<<
+ * ----------------------------------------------------------------------------
+ * Copyright 2004 (C) Chelsio Communications, Inc. (Chelsio)
+ *
+ * Chelsio Communications, Inc. owns the sole copyright to this software.
+ * You may not make a copy, you may not derive works herefrom, and you may
+ * not distribute this work to others. Other restrictions of rights may apply
+ * as well. This is unpublished, confidential information. All rights reserved.
+ * This software contains confidential information and trade secrets of Chelsio
+ * Communications, Inc. Use, disclosure, or reproduction is prohibited without
+ * the prior express written permission of Chelsio Communications, Inc.
+ * ----------------------------------------------------------------------------
+ * >>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>> Warranty <<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<
+ * ----------------------------------------------------------------------------
+ * CHELSIO MAKES NO WARRANTY OF ANY KIND WITH REGARD TO THE USE OF THIS
+ * SOFTWARE, EITHER EXPRESSED OR IMPLIED, INCLUDING BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.
+ * ----------------------------------------------------------------------------
+ *
+ * This is the firmware_exports.h header file, firmware interface defines.
+ *
+ * Written January 2005 by felix marti (felix@chelsio.com)
+ */
+#ifndef _FIRMWARE_EXPORTS_H_
+#define _FIRMWARE_EXPORTS_H_
+
+/* WR OPCODES supported by the firmware.
+ */
+#define	FW_WROPCODE_FORWARD			0x01
+#define FW_WROPCODE_BYPASS			0x05
+
+#define FW_WROPCODE_TUNNEL_TX_PKT		0x03
+
+#define FW_WROPOCDE_ULPTX_DATA_SGL		0x00
+#define FW_WROPCODE_ULPTX_MEM_READ		0x02
+#define FW_WROPCODE_ULPTX_PKT			0x04
+#define FW_WROPCODE_ULPTX_INVALIDATE		0x06
+
+#define FW_WROPCODE_TUNNEL_RX_PKT		0x07
+
+#define FW_WROPCODE_OFLD_GETTCB_RPL		0x08
+#define FW_WROPCODE_OFLD_CLOSE_CON		0x09
+#define FW_WROPCODE_OFLD_TP_ABORT_CON_REQ	0x0A
+#define FW_WROPCODE_OFLD_HOST_ABORT_CON_RPL	0x0F
+#define FW_WROPCODE_OFLD_HOST_ABORT_CON_REQ	0x0B
+#define FW_WROPCODE_OFLD_TP_ABORT_CON_RPL	0x0C
+#define FW_WROPCODE_OFLD_TX_DATA		0x0D
+#define FW_WROPCODE_OFLD_TX_DATA_ACK		0x0E
+
+#define FW_WROPCODE_RI_RDMA_INIT		0x10
+#define FW_WROPCODE_RI_RDMA_WRITE		0x11
+#define FW_WROPCODE_RI_RDMA_READ_REQ		0x12
+#define FW_WROPCODE_RI_RDMA_READ_RESP		0x13
+#define FW_WROPCODE_RI_SEND			0x14
+#define FW_WROPCODE_RI_TERMINATE		0x15
+#define FW_WROPCODE_RI_RDMA_READ		0x16
+#define FW_WROPCODE_RI_RECEIVE			0x17
+#define FW_WROPCODE_RI_BIND_MW			0x18
+#define FW_WROPCODE_RI_FASTREGISTER_MR		0x19
+#define FW_WROPCODE_RI_LOCAL_INV		0x1A
+#define FW_WROPCODE_RI_MODIFY_QP		0x1B
+#define FW_WROPCODE_RI_BYPASS			0x1C
+
+#define FW_WROPOCDE_RSVD			0x1E
+
+#define FW_WROPCODE_SGE_EGRESSCONTEXT_RR	0x1F
+
+#define FW_WROPCODE_MNGT			0x1D
+#define FW_MNGTOPCODE_PKTSCHED_SET		0x00
+
+/* Maximum size of a WR sent from the host, limited by the SGE. 
+ *
+ * Note: WR coming from ULP or TP are only limited by CIM. 
+ */
+#define FW_WR_SIZE			128
+
+/* Maximum number of outstanding WRs sent from the host. Value must be
+ * programmed in the CTRL/TUNNEL/QP SGE Egress Context and used by 
+ * offload modules to limit the number of WRs per connection.
+ */
+#define FW_T3_WR_NUM			16
+#define FW_N3_WR_NUM			7
+
+#ifndef N3
+# define FW_WR_NUM			FW_T3_WR_NUM
+#else
+# define FW_WR_NUM			FW_N3_WR_NUM
+#endif
+
+/* FW_TUNNEL_NUM corresponds to the number of supported TUNNEL Queues. These
+ * queues must start at SGE Egress Context FW_TUNNEL_SGEEC_START and must
+ * start at 'TID' (or 'uP Token') FW_TUNNEL_TID_START.
+ *
+ * Ingress Traffic (e.g. DMA completion credit)  for TUNNEL Queue[i] is sent 
+ * to RESP Queue[i].
+ */
+#define FW_TUNNEL_NUM			8
+#define FW_TUNNEL_SGEEC_START		8
+#define FW_TUNNEL_TID_START		65544
+
+/* FW_CTRL_NUM corresponds to the number of supported CTRL Queues. These queues
+ * must start at SGE Egress Context FW_CTRL_SGEEC_START and must start at 'TID'
+ * (or 'uP Token') FW_CTRL_TID_START.
+ *
+ * Ingress Traffic for CTRL Queue[i] is sent to RESP Queue[i].
+ */
+#define FW_CTRL_NUM			8
+#define FW_CTRL_SGEEC_START		65528
+#define FW_CTRL_TID_START		65536
+
+/* FW_OFLD_NUM corresponds to the number of supported OFFLOAD Queues. These 
+ * queues must start at SGE Egress Context FW_OFLD_SGEEC_START. 
+ * 
+ * Note: the 'uP Token' in the SGE Egress Context fields is irrelevant for 
+ * OFFLOAD Queues, as the host is responsible for providing the correct TID in
+ * every WR.
+ *
+ * Ingress Trafffic for OFFLOAD Queue[i] is sent to RESP Queue[i].
+ */
+#define FW_OFLD_NUM			8
+#define FW_OFLD_SGEEC_START		0
+
+/*
+ * 
+ */
+#define FW_RI_NUM			1
+#define FW_RI_SGEEC_START		65527
+#define FW_RI_TID_START			65552
+
+/*
+ * The RX_PKT_TID 
+ */
+#define FW_RX_PKT_NUM			1
+#define FW_RX_PKT_TID_START		65553
+
+/* FW_WRC_NUM corresponds to the number of Work Request Context that supported
+ * by the firmware.
+ */
+#define FW_WRC_NUM			\
+    (65536 + FW_TUNNEL_NUM + FW_CTRL_NUM + FW_RI_NUM + FW_RX_PKT_NUM)
+
+#endif				/* _FIRMWARE_EXPORTS_H_ */
diff --git a/drivers/net/cxgb3/l2t.c b/drivers/net/cxgb3/l2t.c
new file mode 100644
index 00000000000..9997138a4fd
--- /dev/null
+++ b/drivers/net/cxgb3/l2t.c
@@ -0,0 +1,450 @@
+/*
+ * Copyright (c) 2006 Chelsio, Inc. All rights reserved.
+ * Copyright (c) 2006 Open Grid Computing, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include <linux/skbuff.h>
+#include <linux/netdevice.h>
+#include <linux/if.h>
+#include <linux/if_vlan.h>
+#include <linux/jhash.h>
+#include <net/neighbour.h>
+#include "common.h"
+#include "t3cdev.h"
+#include "cxgb3_defs.h"
+#include "l2t.h"
+#include "t3_cpl.h"
+#include "firmware_exports.h"
+
+#define VLAN_NONE 0xfff
+
+/*
+ * Module locking notes:  There is a RW lock protecting the L2 table as a
+ * whole plus a spinlock per L2T entry.  Entry lookups and allocations happen
+ * under the protection of the table lock, individual entry changes happen
+ * while holding that entry's spinlock.  The table lock nests outside the
+ * entry locks.  Allocations of new entries take the table lock as writers so
+ * no other lookups can happen while allocating new entries.  Entry updates
+ * take the table lock as readers so multiple entries can be updated in
+ * parallel.  An L2T entry can be dropped by decrementing its reference count
+ * and therefore can happen in parallel with entry allocation but no entry
+ * can change state or increment its ref count during allocation as both of
+ * these perform lookups.
+ */
+
+static inline unsigned int vlan_prio(const struct l2t_entry *e)
+{
+	return e->vlan >> 13;
+}
+
+static inline unsigned int arp_hash(u32 key, int ifindex,
+				    const struct l2t_data *d)
+{
+	return jhash_2words(key, ifindex, 0) & (d->nentries - 1);
+}
+
+static inline void neigh_replace(struct l2t_entry *e, struct neighbour *n)
+{
+	neigh_hold(n);
+	if (e->neigh)
+		neigh_release(e->neigh);
+	e->neigh = n;
+}
+
+/*
+ * Set up an L2T entry and send any packets waiting in the arp queue.  The
+ * supplied skb is used for the CPL_L2T_WRITE_REQ.  Must be called with the
+ * entry locked.
+ */
+static int setup_l2e_send_pending(struct t3cdev *dev, struct sk_buff *skb,
+				  struct l2t_entry *e)
+{
+	struct cpl_l2t_write_req *req;
+
+	if (!skb) {
+		skb = alloc_skb(sizeof(*req), GFP_ATOMIC);
+		if (!skb)
+			return -ENOMEM;
+	}
+
+	req = (struct cpl_l2t_write_req *)__skb_put(skb, sizeof(*req));
+	req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
+	OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_L2T_WRITE_REQ, e->idx));
+	req->params = htonl(V_L2T_W_IDX(e->idx) | V_L2T_W_IFF(e->smt_idx) |
+			    V_L2T_W_VLAN(e->vlan & VLAN_VID_MASK) |
+			    V_L2T_W_PRIO(vlan_prio(e)));
+	memcpy(e->dmac, e->neigh->ha, sizeof(e->dmac));
+	memcpy(req->dst_mac, e->dmac, sizeof(req->dst_mac));
+	skb->priority = CPL_PRIORITY_CONTROL;
+	cxgb3_ofld_send(dev, skb);
+	while (e->arpq_head) {
+		skb = e->arpq_head;
+		e->arpq_head = skb->next;
+		skb->next = NULL;
+		cxgb3_ofld_send(dev, skb);
+	}
+	e->arpq_tail = NULL;
+	e->state = L2T_STATE_VALID;
+
+	return 0;
+}
+
+/*
+ * Add a packet to the an L2T entry's queue of packets awaiting resolution.
+ * Must be called with the entry's lock held.
+ */
+static inline void arpq_enqueue(struct l2t_entry *e, struct sk_buff *skb)
+{
+	skb->next = NULL;
+	if (e->arpq_head)
+		e->arpq_tail->next = skb;
+	else
+		e->arpq_head = skb;
+	e->arpq_tail = skb;
+}
+
+int t3_l2t_send_slow(struct t3cdev *dev, struct sk_buff *skb,
+		     struct l2t_entry *e)
+{
+again:
+	switch (e->state) {
+	case L2T_STATE_STALE:	/* entry is stale, kick off revalidation */
+		neigh_event_send(e->neigh, NULL);
+		spin_lock_bh(&e->lock);
+		if (e->state == L2T_STATE_STALE)
+			e->state = L2T_STATE_VALID;
+		spin_unlock_bh(&e->lock);
+	case L2T_STATE_VALID:	/* fast-path, send the packet on */
+		return cxgb3_ofld_send(dev, skb);
+	case L2T_STATE_RESOLVING:
+		spin_lock_bh(&e->lock);
+		if (e->state != L2T_STATE_RESOLVING) {
+			/* ARP already completed */
+			spin_unlock_bh(&e->lock);
+			goto again;
+		}
+		arpq_enqueue(e, skb);
+		spin_unlock_bh(&e->lock);
+
+		/*
+		 * Only the first packet added to the arpq should kick off
+		 * resolution.  However, because the alloc_skb below can fail,
+		 * we allow each packet added to the arpq to retry resolution
+		 * as a way of recovering from transient memory exhaustion.
+		 * A better way would be to use a work request to retry L2T
+		 * entries when there's no memory.
+		 */
+		if (!neigh_event_send(e->neigh, NULL)) {
+			skb = alloc_skb(sizeof(struct cpl_l2t_write_req),
+					GFP_ATOMIC);
+			if (!skb)
+				break;
+
+			spin_lock_bh(&e->lock);
+			if (e->arpq_head)
+				setup_l2e_send_pending(dev, skb, e);
+			else	/* we lost the race */
+				__kfree_skb(skb);
+			spin_unlock_bh(&e->lock);
+		}
+	}
+	return 0;
+}
+
+EXPORT_SYMBOL(t3_l2t_send_slow);
+
+void t3_l2t_send_event(struct t3cdev *dev, struct l2t_entry *e)
+{
+again:
+	switch (e->state) {
+	case L2T_STATE_STALE:	/* entry is stale, kick off revalidation */
+		neigh_event_send(e->neigh, NULL);
+		spin_lock_bh(&e->lock);
+		if (e->state == L2T_STATE_STALE) {
+			e->state = L2T_STATE_VALID;
+		}
+		spin_unlock_bh(&e->lock);
+		return;
+	case L2T_STATE_VALID:	/* fast-path, send the packet on */
+		return;
+	case L2T_STATE_RESOLVING:
+		spin_lock_bh(&e->lock);
+		if (e->state != L2T_STATE_RESOLVING) {
+			/* ARP already completed */
+			spin_unlock_bh(&e->lock);
+			goto again;
+		}
+		spin_unlock_bh(&e->lock);
+
+		/*
+		 * Only the first packet added to the arpq should kick off
+		 * resolution.  However, because the alloc_skb below can fail,
+		 * we allow each packet added to the arpq to retry resolution
+		 * as a way of recovering from transient memory exhaustion.
+		 * A better way would be to use a work request to retry L2T
+		 * entries when there's no memory.
+		 */
+		neigh_event_send(e->neigh, NULL);
+	}
+	return;
+}
+
+EXPORT_SYMBOL(t3_l2t_send_event);
+
+/*
+ * Allocate a free L2T entry.  Must be called with l2t_data.lock held.
+ */
+static struct l2t_entry *alloc_l2e(struct l2t_data *d)
+{
+	struct l2t_entry *end, *e, **p;
+
+	if (!atomic_read(&d->nfree))
+		return NULL;
+
+	/* there's definitely a free entry */
+	for (e = d->rover, end = &d->l2tab[d->nentries]; e != end; ++e)
+		if (atomic_read(&e->refcnt) == 0)
+			goto found;
+
+	for (e = &d->l2tab[1]; atomic_read(&e->refcnt); ++e) ;
+found:
+	d->rover = e + 1;
+	atomic_dec(&d->nfree);
+
+	/*
+	 * The entry we found may be an inactive entry that is
+	 * presently in the hash table.  We need to remove it.
+	 */
+	if (e->state != L2T_STATE_UNUSED) {
+		int hash = arp_hash(e->addr, e->ifindex, d);
+
+		for (p = &d->l2tab[hash].first; *p; p = &(*p)->next)
+			if (*p == e) {
+				*p = e->next;
+				break;
+			}
+		e->state = L2T_STATE_UNUSED;
+	}
+	return e;
+}
+
+/*
+ * Called when an L2T entry has no more users.  The entry is left in the hash
+ * table since it is likely to be reused but we also bump nfree to indicate
+ * that the entry can be reallocated for a different neighbor.  We also drop
+ * the existing neighbor reference in case the neighbor is going away and is
+ * waiting on our reference.
+ *
+ * Because entries can be reallocated to other neighbors once their ref count
+ * drops to 0 we need to take the entry's lock to avoid races with a new
+ * incarnation.
+ */
+void t3_l2e_free(struct l2t_data *d, struct l2t_entry *e)
+{
+	spin_lock_bh(&e->lock);
+	if (atomic_read(&e->refcnt) == 0) {	/* hasn't been recycled */
+		if (e->neigh) {
+			neigh_release(e->neigh);
+			e->neigh = NULL;
+		}
+	}
+	spin_unlock_bh(&e->lock);
+	atomic_inc(&d->nfree);
+}
+
+EXPORT_SYMBOL(t3_l2e_free);
+
+/*
+ * Update an L2T entry that was previously used for the same next hop as neigh.
+ * Must be called with softirqs disabled.
+ */
+static inline void reuse_entry(struct l2t_entry *e, struct neighbour *neigh)
+{
+	unsigned int nud_state;
+
+	spin_lock(&e->lock);	/* avoid race with t3_l2t_free */
+
+	if (neigh != e->neigh)
+		neigh_replace(e, neigh);
+	nud_state = neigh->nud_state;
+	if (memcmp(e->dmac, neigh->ha, sizeof(e->dmac)) ||
+	    !(nud_state & NUD_VALID))
+		e->state = L2T_STATE_RESOLVING;
+	else if (nud_state & NUD_CONNECTED)
+		e->state = L2T_STATE_VALID;
+	else
+		e->state = L2T_STATE_STALE;
+	spin_unlock(&e->lock);
+}
+
+struct l2t_entry *t3_l2t_get(struct t3cdev *cdev, struct neighbour *neigh,
+			     struct net_device *dev)
+{
+	struct l2t_entry *e;
+	struct l2t_data *d = L2DATA(cdev);
+	u32 addr = *(u32 *) neigh->primary_key;
+	int ifidx = neigh->dev->ifindex;
+	int hash = arp_hash(addr, ifidx, d);
+	struct port_info *p = netdev_priv(dev);
+	int smt_idx = p->port_id;
+
+	write_lock_bh(&d->lock);
+	for (e = d->l2tab[hash].first; e; e = e->next)
+		if (e->addr == addr && e->ifindex == ifidx &&
+		    e->smt_idx == smt_idx) {
+			l2t_hold(d, e);
+			if (atomic_read(&e->refcnt) == 1)
+				reuse_entry(e, neigh);
+			goto done;
+		}
+
+	/* Need to allocate a new entry */
+	e = alloc_l2e(d);
+	if (e) {
+		spin_lock(&e->lock);	/* avoid race with t3_l2t_free */
+		e->next = d->l2tab[hash].first;
+		d->l2tab[hash].first = e;
+		e->state = L2T_STATE_RESOLVING;
+		e->addr = addr;
+		e->ifindex = ifidx;
+		e->smt_idx = smt_idx;
+		atomic_set(&e->refcnt, 1);
+		neigh_replace(e, neigh);
+		if (neigh->dev->priv_flags & IFF_802_1Q_VLAN)
+			e->vlan = VLAN_DEV_INFO(neigh->dev)->vlan_id;
+		else
+			e->vlan = VLAN_NONE;
+		spin_unlock(&e->lock);
+	}
+done:
+	write_unlock_bh(&d->lock);
+	return e;
+}
+
+EXPORT_SYMBOL(t3_l2t_get);
+
+/*
+ * Called when address resolution fails for an L2T entry to handle packets
+ * on the arpq head.  If a packet specifies a failure handler it is invoked,
+ * otherwise the packets is sent to the offload device.
+ *
+ * XXX: maybe we should abandon the latter behavior and just require a failure
+ * handler.
+ */
+static void handle_failed_resolution(struct t3cdev *dev, struct sk_buff *arpq)
+{
+	while (arpq) {
+		struct sk_buff *skb = arpq;
+		struct l2t_skb_cb *cb = L2T_SKB_CB(skb);
+
+		arpq = skb->next;
+		skb->next = NULL;
+		if (cb->arp_failure_handler)
+			cb->arp_failure_handler(dev, skb);
+		else
+			cxgb3_ofld_send(dev, skb);
+	}
+}
+
+/*
+ * Called when the host's ARP layer makes a change to some entry that is
+ * loaded into the HW L2 table.
+ */
+void t3_l2t_update(struct t3cdev *dev, struct neighbour *neigh)
+{
+	struct l2t_entry *e;
+	struct sk_buff *arpq = NULL;
+	struct l2t_data *d = L2DATA(dev);
+	u32 addr = *(u32 *) neigh->primary_key;
+	int ifidx = neigh->dev->ifindex;
+	int hash = arp_hash(addr, ifidx, d);
+
+	read_lock_bh(&d->lock);
+	for (e = d->l2tab[hash].first; e; e = e->next)
+		if (e->addr == addr && e->ifindex == ifidx) {
+			spin_lock(&e->lock);
+			goto found;
+		}
+	read_unlock_bh(&d->lock);
+	return;
+
+found:
+	read_unlock(&d->lock);
+	if (atomic_read(&e->refcnt)) {
+		if (neigh != e->neigh)
+			neigh_replace(e, neigh);
+
+		if (e->state == L2T_STATE_RESOLVING) {
+			if (neigh->nud_state & NUD_FAILED) {
+				arpq = e->arpq_head;
+				e->arpq_head = e->arpq_tail = NULL;
+			} else if (neigh_is_connected(neigh))
+				setup_l2e_send_pending(dev, NULL, e);
+		} else {
+			e->state = neigh_is_connected(neigh) ?
+			    L2T_STATE_VALID : L2T_STATE_STALE;
+			if (memcmp(e->dmac, neigh->ha, 6))
+				setup_l2e_send_pending(dev, NULL, e);
+		}
+	}
+	spin_unlock_bh(&e->lock);
+
+	if (arpq)
+		handle_failed_resolution(dev, arpq);
+}
+
+struct l2t_data *t3_init_l2t(unsigned int l2t_capacity)
+{
+	struct l2t_data *d;
+	int i, size = sizeof(*d) + l2t_capacity * sizeof(struct l2t_entry);
+
+	d = cxgb_alloc_mem(size);
+	if (!d)
+		return NULL;
+
+	d->nentries = l2t_capacity;
+	d->rover = &d->l2tab[1];	/* entry 0 is not used */
+	atomic_set(&d->nfree, l2t_capacity - 1);
+	rwlock_init(&d->lock);
+
+	for (i = 0; i < l2t_capacity; ++i) {
+		d->l2tab[i].idx = i;
+		d->l2tab[i].state = L2T_STATE_UNUSED;
+		spin_lock_init(&d->l2tab[i].lock);
+		atomic_set(&d->l2tab[i].refcnt, 0);
+	}
+	return d;
+}
+
+void t3_free_l2t(struct l2t_data *d)
+{
+	cxgb_free_mem(d);
+}
+
diff --git a/drivers/net/cxgb3/l2t.h b/drivers/net/cxgb3/l2t.h
new file mode 100644
index 00000000000..51a9c1f2c58
--- /dev/null
+++ b/drivers/net/cxgb3/l2t.h
@@ -0,0 +1,143 @@
+/*
+ * Copyright (c) 2006 Chelsio, Inc. All rights reserved.
+ * Copyright (c) 2006 Open Grid Computing, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef _CHELSIO_L2T_H
+#define _CHELSIO_L2T_H
+
+#include <linux/spinlock.h>
+#include "t3cdev.h"
+#include <asm/atomic.h>
+
+enum {
+	L2T_STATE_VALID,	/* entry is up to date */
+	L2T_STATE_STALE,	/* entry may be used but needs revalidation */
+	L2T_STATE_RESOLVING,	/* entry needs address resolution */
+	L2T_STATE_UNUSED	/* entry not in use */
+};
+
+struct neighbour;
+struct sk_buff;
+
+/*
+ * Each L2T entry plays multiple roles.  First of all, it keeps state for the
+ * corresponding entry of the HW L2 table and maintains a queue of offload
+ * packets awaiting address resolution.  Second, it is a node of a hash table
+ * chain, where the nodes of the chain are linked together through their next
+ * pointer.  Finally, each node is a bucket of a hash table, pointing to the
+ * first element in its chain through its first pointer.
+ */
+struct l2t_entry {
+	u16 state;		/* entry state */
+	u16 idx;		/* entry index */
+	u32 addr;		/* dest IP address */
+	int ifindex;		/* neighbor's net_device's ifindex */
+	u16 smt_idx;		/* SMT index */
+	u16 vlan;		/* VLAN TCI (id: bits 0-11, prio: 13-15 */
+	struct neighbour *neigh;	/* associated neighbour */
+	struct l2t_entry *first;	/* start of hash chain */
+	struct l2t_entry *next;	/* next l2t_entry on chain */
+	struct sk_buff *arpq_head;	/* queue of packets awaiting resolution */
+	struct sk_buff *arpq_tail;
+	spinlock_t lock;
+	atomic_t refcnt;	/* entry reference count */
+	u8 dmac[6];		/* neighbour's MAC address */
+};
+
+struct l2t_data {
+	unsigned int nentries;	/* number of entries */
+	struct l2t_entry *rover;	/* starting point for next allocation */
+	atomic_t nfree;		/* number of free entries */
+	rwlock_t lock;
+	struct l2t_entry l2tab[0];
+};
+
+typedef void (*arp_failure_handler_func)(struct t3cdev * dev,
+					 struct sk_buff * skb);
+
+/*
+ * Callback stored in an skb to handle address resolution failure.
+ */
+struct l2t_skb_cb {
+	arp_failure_handler_func arp_failure_handler;
+};
+
+#define L2T_SKB_CB(skb) ((struct l2t_skb_cb *)(skb)->cb)
+
+static inline void set_arp_failure_handler(struct sk_buff *skb,
+					   arp_failure_handler_func hnd)
+{
+	L2T_SKB_CB(skb)->arp_failure_handler = hnd;
+}
+
+/*
+ * Getting to the L2 data from an offload device.
+ */
+#define L2DATA(dev) ((dev)->l2opt)
+
+#define W_TCB_L2T_IX    0
+#define S_TCB_L2T_IX    7
+#define M_TCB_L2T_IX    0x7ffULL
+#define V_TCB_L2T_IX(x) ((x) << S_TCB_L2T_IX)
+
+void t3_l2e_free(struct l2t_data *d, struct l2t_entry *e);
+void t3_l2t_update(struct t3cdev *dev, struct neighbour *neigh);
+struct l2t_entry *t3_l2t_get(struct t3cdev *cdev, struct neighbour *neigh,
+			     struct net_device *dev);
+int t3_l2t_send_slow(struct t3cdev *dev, struct sk_buff *skb,
+		     struct l2t_entry *e);
+void t3_l2t_send_event(struct t3cdev *dev, struct l2t_entry *e);
+struct l2t_data *t3_init_l2t(unsigned int l2t_capacity);
+void t3_free_l2t(struct l2t_data *d);
+
+int cxgb3_ofld_send(struct t3cdev *dev, struct sk_buff *skb);
+
+static inline int l2t_send(struct t3cdev *dev, struct sk_buff *skb,
+			   struct l2t_entry *e)
+{
+	if (likely(e->state == L2T_STATE_VALID))
+		return cxgb3_ofld_send(dev, skb);
+	return t3_l2t_send_slow(dev, skb, e);
+}
+
+static inline void l2t_release(struct l2t_data *d, struct l2t_entry *e)
+{
+	if (atomic_dec_and_test(&e->refcnt))
+		t3_l2e_free(d, e);
+}
+
+static inline void l2t_hold(struct l2t_data *d, struct l2t_entry *e)
+{
+	if (atomic_add_return(1, &e->refcnt) == 1)	/* 0 -> 1 transition */
+		atomic_dec(&d->nfree);
+}
+
+#endif
diff --git a/drivers/net/cxgb3/mc5.c b/drivers/net/cxgb3/mc5.c
new file mode 100644
index 00000000000..44fa9eaedcc
--- /dev/null
+++ b/drivers/net/cxgb3/mc5.c
@@ -0,0 +1,453 @@
+/*
+ * This file is part of the Chelsio T3 Ethernet driver.
+ *
+ * Copyright (C) 2003-2006 Chelsio Communications.  All rights reserved.
+ *
+ * This program is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the LICENSE file included in this
+ * release for licensing terms and conditions.
+ */
+
+#include "common.h"
+#include "regs.h"
+
+enum {
+	IDT75P52100 = 4,
+	IDT75N43102 = 5
+};
+
+/* DBGI command mode */
+enum {
+	DBGI_MODE_MBUS = 0,
+	DBGI_MODE_IDT52100 = 5
+};
+
+/* IDT 75P52100 commands */
+#define IDT_CMD_READ   0
+#define IDT_CMD_WRITE  1
+#define IDT_CMD_SEARCH 2
+#define IDT_CMD_LEARN  3
+
+/* IDT LAR register address and value for 144-bit mode (low 32 bits) */
+#define IDT_LAR_ADR0   	0x180006
+#define IDT_LAR_MODE144	0xffff0000
+
+/* IDT SCR and SSR addresses (low 32 bits) */
+#define IDT_SCR_ADR0  0x180000
+#define IDT_SSR0_ADR0 0x180002
+#define IDT_SSR1_ADR0 0x180004
+
+/* IDT GMR base address (low 32 bits) */
+#define IDT_GMR_BASE_ADR0 0x180020
+
+/* IDT data and mask array base addresses (low 32 bits) */
+#define IDT_DATARY_BASE_ADR0 0
+#define IDT_MSKARY_BASE_ADR0 0x80000
+
+/* IDT 75N43102 commands */
+#define IDT4_CMD_SEARCH144 3
+#define IDT4_CMD_WRITE     4
+#define IDT4_CMD_READ      5
+
+/* IDT 75N43102 SCR address (low 32 bits) */
+#define IDT4_SCR_ADR0  0x3
+
+/* IDT 75N43102 GMR base addresses (low 32 bits) */
+#define IDT4_GMR_BASE0 0x10
+#define IDT4_GMR_BASE1 0x20
+#define IDT4_GMR_BASE2 0x30
+
+/* IDT 75N43102 data and mask array base addresses (low 32 bits) */
+#define IDT4_DATARY_BASE_ADR0 0x1000000
+#define IDT4_MSKARY_BASE_ADR0 0x2000000
+
+#define MAX_WRITE_ATTEMPTS 5
+
+#define MAX_ROUTES 2048
+
+/*
+ * Issue a command to the TCAM and wait for its completion.  The address and
+ * any data required by the command must have been setup by the caller.
+ */
+static int mc5_cmd_write(struct adapter *adapter, u32 cmd)
+{
+	t3_write_reg(adapter, A_MC5_DB_DBGI_REQ_CMD, cmd);
+	return t3_wait_op_done(adapter, A_MC5_DB_DBGI_RSP_STATUS,
+			       F_DBGIRSPVALID, 1, MAX_WRITE_ATTEMPTS, 1);
+}
+
+static inline void dbgi_wr_addr3(struct adapter *adapter, u32 v1, u32 v2,
+				 u32 v3)
+{
+	t3_write_reg(adapter, A_MC5_DB_DBGI_REQ_ADDR0, v1);
+	t3_write_reg(adapter, A_MC5_DB_DBGI_REQ_ADDR1, v2);
+	t3_write_reg(adapter, A_MC5_DB_DBGI_REQ_ADDR2, v3);
+}
+
+static inline void dbgi_wr_data3(struct adapter *adapter, u32 v1, u32 v2,
+				 u32 v3)
+{
+	t3_write_reg(adapter, A_MC5_DB_DBGI_REQ_DATA0, v1);
+	t3_write_reg(adapter, A_MC5_DB_DBGI_REQ_DATA1, v2);
+	t3_write_reg(adapter, A_MC5_DB_DBGI_REQ_DATA2, v3);
+}
+
+static inline void dbgi_rd_rsp3(struct adapter *adapter, u32 *v1, u32 *v2,
+				u32 *v3)
+{
+	*v1 = t3_read_reg(adapter, A_MC5_DB_DBGI_RSP_DATA0);
+	*v2 = t3_read_reg(adapter, A_MC5_DB_DBGI_RSP_DATA1);
+	*v3 = t3_read_reg(adapter, A_MC5_DB_DBGI_RSP_DATA2);
+}
+
+/*
+ * Write data to the TCAM register at address (0, 0, addr_lo) using the TCAM
+ * command cmd.  The data to be written must have been set up by the caller.
+ * Returns -1 on failure, 0 on success.
+ */
+static int mc5_write(struct adapter *adapter, u32 addr_lo, u32 cmd)
+{
+	t3_write_reg(adapter, A_MC5_DB_DBGI_REQ_ADDR0, addr_lo);
+	if (mc5_cmd_write(adapter, cmd) == 0)
+		return 0;
+	CH_ERR(adapter, "MC5 timeout writing to TCAM address 0x%x\n",
+	       addr_lo);
+	return -1;
+}
+
+static int init_mask_data_array(struct mc5 *mc5, u32 mask_array_base,
+				u32 data_array_base, u32 write_cmd,
+				int addr_shift)
+{
+	unsigned int i;
+	struct adapter *adap = mc5->adapter;
+
+	/*
+	 * We need the size of the TCAM data and mask arrays in terms of
+	 * 72-bit entries.
+	 */
+	unsigned int size72 = mc5->tcam_size;
+	unsigned int server_base = t3_read_reg(adap, A_MC5_DB_SERVER_INDEX);
+
+	if (mc5->mode == MC5_MODE_144_BIT) {
+		size72 *= 2;	/* 1 144-bit entry is 2 72-bit entries */
+		server_base *= 2;
+	}
+
+	/* Clear the data array */
+	dbgi_wr_data3(adap, 0, 0, 0);
+	for (i = 0; i < size72; i++)
+		if (mc5_write(adap, data_array_base + (i << addr_shift),
+			      write_cmd))
+			return -1;
+
+	/* Initialize the mask array. */
+	dbgi_wr_data3(adap, 0xffffffff, 0xffffffff, 0xff);
+	for (i = 0; i < size72; i++) {
+		if (i == server_base)	/* entering server or routing region */
+			t3_write_reg(adap, A_MC5_DB_DBGI_REQ_DATA0,
+				     mc5->mode == MC5_MODE_144_BIT ?
+				     0xfffffff9 : 0xfffffffd);
+		if (mc5_write(adap, mask_array_base + (i << addr_shift),
+			      write_cmd))
+			return -1;
+	}
+	return 0;
+}
+
+static int init_idt52100(struct mc5 *mc5)
+{
+	int i;
+	struct adapter *adap = mc5->adapter;
+
+	t3_write_reg(adap, A_MC5_DB_RSP_LATENCY,
+		     V_RDLAT(0x15) | V_LRNLAT(0x15) | V_SRCHLAT(0x15));
+	t3_write_reg(adap, A_MC5_DB_PART_ID_INDEX, 2);
+
+	/*
+	 * Use GMRs 14-15 for ELOOKUP, GMRs 12-13 for SYN lookups, and
+	 * GMRs 8-9 for ACK- and AOPEN searches.
+	 */
+	t3_write_reg(adap, A_MC5_DB_POPEN_DATA_WR_CMD, IDT_CMD_WRITE);
+	t3_write_reg(adap, A_MC5_DB_POPEN_MASK_WR_CMD, IDT_CMD_WRITE);
+	t3_write_reg(adap, A_MC5_DB_AOPEN_SRCH_CMD, IDT_CMD_SEARCH);
+	t3_write_reg(adap, A_MC5_DB_AOPEN_LRN_CMD, IDT_CMD_LEARN);
+	t3_write_reg(adap, A_MC5_DB_SYN_SRCH_CMD, IDT_CMD_SEARCH | 0x6000);
+	t3_write_reg(adap, A_MC5_DB_SYN_LRN_CMD, IDT_CMD_LEARN);
+	t3_write_reg(adap, A_MC5_DB_ACK_SRCH_CMD, IDT_CMD_SEARCH);
+	t3_write_reg(adap, A_MC5_DB_ACK_LRN_CMD, IDT_CMD_LEARN);
+	t3_write_reg(adap, A_MC5_DB_ILOOKUP_CMD, IDT_CMD_SEARCH);
+	t3_write_reg(adap, A_MC5_DB_ELOOKUP_CMD, IDT_CMD_SEARCH | 0x7000);
+	t3_write_reg(adap, A_MC5_DB_DATA_WRITE_CMD, IDT_CMD_WRITE);
+	t3_write_reg(adap, A_MC5_DB_DATA_READ_CMD, IDT_CMD_READ);
+
+	/* Set DBGI command mode for IDT TCAM. */
+	t3_write_reg(adap, A_MC5_DB_DBGI_CONFIG, DBGI_MODE_IDT52100);
+
+	/* Set up LAR */
+	dbgi_wr_data3(adap, IDT_LAR_MODE144, 0, 0);
+	if (mc5_write(adap, IDT_LAR_ADR0, IDT_CMD_WRITE))
+		goto err;
+
+	/* Set up SSRs */
+	dbgi_wr_data3(adap, 0xffffffff, 0xffffffff, 0);
+	if (mc5_write(adap, IDT_SSR0_ADR0, IDT_CMD_WRITE) ||
+	    mc5_write(adap, IDT_SSR1_ADR0, IDT_CMD_WRITE))
+		goto err;
+
+	/* Set up GMRs */
+	for (i = 0; i < 32; ++i) {
+		if (i >= 12 && i < 15)
+			dbgi_wr_data3(adap, 0xfffffff9, 0xffffffff, 0xff);
+		else if (i == 15)
+			dbgi_wr_data3(adap, 0xfffffff9, 0xffff8007, 0xff);
+		else
+			dbgi_wr_data3(adap, 0xffffffff, 0xffffffff, 0xff);
+
+		if (mc5_write(adap, IDT_GMR_BASE_ADR0 + i, IDT_CMD_WRITE))
+			goto err;
+	}
+
+	/* Set up SCR */
+	dbgi_wr_data3(adap, 1, 0, 0);
+	if (mc5_write(adap, IDT_SCR_ADR0, IDT_CMD_WRITE))
+		goto err;
+
+	return init_mask_data_array(mc5, IDT_MSKARY_BASE_ADR0,
+				    IDT_DATARY_BASE_ADR0, IDT_CMD_WRITE, 0);
+err:
+	return -EIO;
+}
+
+static int init_idt43102(struct mc5 *mc5)
+{
+	int i;
+	struct adapter *adap = mc5->adapter;
+
+	t3_write_reg(adap, A_MC5_DB_RSP_LATENCY,
+		     adap->params.rev == 0 ? V_RDLAT(0xd) | V_SRCHLAT(0x11) :
+		     V_RDLAT(0xd) | V_SRCHLAT(0x12));
+
+	/*
+	 * Use GMRs 24-25 for ELOOKUP, GMRs 20-21 for SYN lookups, and no mask
+	 * for ACK- and AOPEN searches.
+	 */
+	t3_write_reg(adap, A_MC5_DB_POPEN_DATA_WR_CMD, IDT4_CMD_WRITE);
+	t3_write_reg(adap, A_MC5_DB_POPEN_MASK_WR_CMD, IDT4_CMD_WRITE);
+	t3_write_reg(adap, A_MC5_DB_AOPEN_SRCH_CMD,
+		     IDT4_CMD_SEARCH144 | 0x3800);
+	t3_write_reg(adap, A_MC5_DB_SYN_SRCH_CMD, IDT4_CMD_SEARCH144);
+	t3_write_reg(adap, A_MC5_DB_ACK_SRCH_CMD, IDT4_CMD_SEARCH144 | 0x3800);
+	t3_write_reg(adap, A_MC5_DB_ILOOKUP_CMD, IDT4_CMD_SEARCH144 | 0x3800);
+	t3_write_reg(adap, A_MC5_DB_ELOOKUP_CMD, IDT4_CMD_SEARCH144 | 0x800);
+	t3_write_reg(adap, A_MC5_DB_DATA_WRITE_CMD, IDT4_CMD_WRITE);
+	t3_write_reg(adap, A_MC5_DB_DATA_READ_CMD, IDT4_CMD_READ);
+
+	t3_write_reg(adap, A_MC5_DB_PART_ID_INDEX, 3);
+
+	/* Set DBGI command mode for IDT TCAM. */
+	t3_write_reg(adap, A_MC5_DB_DBGI_CONFIG, DBGI_MODE_IDT52100);
+
+	/* Set up GMRs */
+	dbgi_wr_data3(adap, 0xffffffff, 0xffffffff, 0xff);
+	for (i = 0; i < 7; ++i)
+		if (mc5_write(adap, IDT4_GMR_BASE0 + i, IDT4_CMD_WRITE))
+			goto err;
+
+	for (i = 0; i < 4; ++i)
+		if (mc5_write(adap, IDT4_GMR_BASE2 + i, IDT4_CMD_WRITE))
+			goto err;
+
+	dbgi_wr_data3(adap, 0xfffffff9, 0xffffffff, 0xff);
+	if (mc5_write(adap, IDT4_GMR_BASE1, IDT4_CMD_WRITE) ||
+	    mc5_write(adap, IDT4_GMR_BASE1 + 1, IDT4_CMD_WRITE) ||
+	    mc5_write(adap, IDT4_GMR_BASE1 + 4, IDT4_CMD_WRITE))
+		goto err;
+
+	dbgi_wr_data3(adap, 0xfffffff9, 0xffff8007, 0xff);
+	if (mc5_write(adap, IDT4_GMR_BASE1 + 5, IDT4_CMD_WRITE))
+		goto err;
+
+	/* Set up SCR */
+	dbgi_wr_data3(adap, 0xf0000000, 0, 0);
+	if (mc5_write(adap, IDT4_SCR_ADR0, IDT4_CMD_WRITE))
+		goto err;
+
+	return init_mask_data_array(mc5, IDT4_MSKARY_BASE_ADR0,
+				    IDT4_DATARY_BASE_ADR0, IDT4_CMD_WRITE, 1);
+err:
+	return -EIO;
+}
+
+/* Put MC5 in DBGI mode. */
+static inline void mc5_dbgi_mode_enable(const struct mc5 *mc5)
+{
+	t3_write_reg(mc5->adapter, A_MC5_DB_CONFIG,
+		     V_TMMODE(mc5->mode == MC5_MODE_72_BIT) | F_DBGIEN);
+}
+
+/* Put MC5 in M-Bus mode. */
+static void mc5_dbgi_mode_disable(const struct mc5 *mc5)
+{
+	t3_write_reg(mc5->adapter, A_MC5_DB_CONFIG,
+		     V_TMMODE(mc5->mode == MC5_MODE_72_BIT) |
+		     V_COMPEN(mc5->mode == MC5_MODE_72_BIT) |
+		     V_PRTYEN(mc5->parity_enabled) | F_MBUSEN);
+}
+
+/*
+ * Initialization that requires the OS and protocol layers to already
+ * be intialized goes here.
+ */
+int t3_mc5_init(struct mc5 *mc5, unsigned int nservers, unsigned int nfilters,
+		unsigned int nroutes)
+{
+	u32 cfg;
+	int err;
+	unsigned int tcam_size = mc5->tcam_size;
+	struct adapter *adap = mc5->adapter;
+
+	if (nroutes > MAX_ROUTES || nroutes + nservers + nfilters > tcam_size)
+		return -EINVAL;
+
+	/* Reset the TCAM */
+	cfg = t3_read_reg(adap, A_MC5_DB_CONFIG) & ~F_TMMODE;
+	cfg |= V_TMMODE(mc5->mode == MC5_MODE_72_BIT) | F_TMRST;
+	t3_write_reg(adap, A_MC5_DB_CONFIG, cfg);
+	if (t3_wait_op_done(adap, A_MC5_DB_CONFIG, F_TMRDY, 1, 500, 0)) {
+		CH_ERR(adap, "TCAM reset timed out\n");
+		return -1;
+	}
+
+	t3_write_reg(adap, A_MC5_DB_ROUTING_TABLE_INDEX, tcam_size - nroutes);
+	t3_write_reg(adap, A_MC5_DB_FILTER_TABLE,
+		     tcam_size - nroutes - nfilters);
+	t3_write_reg(adap, A_MC5_DB_SERVER_INDEX,
+		     tcam_size - nroutes - nfilters - nservers);
+
+	mc5->parity_enabled = 1;
+
+	/* All the TCAM addresses we access have only the low 32 bits non 0 */
+	t3_write_reg(adap, A_MC5_DB_DBGI_REQ_ADDR1, 0);
+	t3_write_reg(adap, A_MC5_DB_DBGI_REQ_ADDR2, 0);
+
+	mc5_dbgi_mode_enable(mc5);
+
+	switch (mc5->part_type) {
+	case IDT75P52100:
+		err = init_idt52100(mc5);
+		break;
+	case IDT75N43102:
+		err = init_idt43102(mc5);
+		break;
+	default:
+		CH_ERR(adap, "Unsupported TCAM type %d\n", mc5->part_type);
+		err = -EINVAL;
+		break;
+	}
+
+	mc5_dbgi_mode_disable(mc5);
+	return err;
+}
+
+/*
+ *	read_mc5_range - dump a part of the memory managed by MC5
+ *	@mc5: the MC5 handle
+ *	@start: the start address for the dump
+ *	@n: number of 72-bit words to read
+ *	@buf: result buffer
+ *
+ *	Read n 72-bit words from MC5 memory from the given start location.
+ */
+int t3_read_mc5_range(const struct mc5 *mc5, unsigned int start,
+		      unsigned int n, u32 *buf)
+{
+	u32 read_cmd;
+	int err = 0;
+	struct adapter *adap = mc5->adapter;
+
+	if (mc5->part_type == IDT75P52100)
+		read_cmd = IDT_CMD_READ;
+	else if (mc5->part_type == IDT75N43102)
+		read_cmd = IDT4_CMD_READ;
+	else
+		return -EINVAL;
+
+	mc5_dbgi_mode_enable(mc5);
+
+	while (n--) {
+		t3_write_reg(adap, A_MC5_DB_DBGI_REQ_ADDR0, start++);
+		if (mc5_cmd_write(adap, read_cmd)) {
+			err = -EIO;
+			break;
+		}
+		dbgi_rd_rsp3(adap, buf + 2, buf + 1, buf);
+		buf += 3;
+	}
+
+	mc5_dbgi_mode_disable(mc5);
+	return 0;
+}
+
+#define MC5_INT_FATAL (F_PARITYERR | F_REQQPARERR | F_DISPQPARERR)
+
+/*
+ * MC5 interrupt handler
+ */
+void t3_mc5_intr_handler(struct mc5 *mc5)
+{
+	struct adapter *adap = mc5->adapter;
+	u32 cause = t3_read_reg(adap, A_MC5_DB_INT_CAUSE);
+
+	if ((cause & F_PARITYERR) && mc5->parity_enabled) {
+		CH_ALERT(adap, "MC5 parity error\n");
+		mc5->stats.parity_err++;
+	}
+
+	if (cause & F_REQQPARERR) {
+		CH_ALERT(adap, "MC5 request queue parity error\n");
+		mc5->stats.reqq_parity_err++;
+	}
+
+	if (cause & F_DISPQPARERR) {
+		CH_ALERT(adap, "MC5 dispatch queue parity error\n");
+		mc5->stats.dispq_parity_err++;
+	}
+
+	if (cause & F_ACTRGNFULL)
+		mc5->stats.active_rgn_full++;
+	if (cause & F_NFASRCHFAIL)
+		mc5->stats.nfa_srch_err++;
+	if (cause & F_UNKNOWNCMD)
+		mc5->stats.unknown_cmd++;
+	if (cause & F_DELACTEMPTY)
+		mc5->stats.del_act_empty++;
+	if (cause & MC5_INT_FATAL)
+		t3_fatal_err(adap);
+
+	t3_write_reg(adap, A_MC5_DB_INT_CAUSE, cause);
+}
+
+void __devinit t3_mc5_prep(struct adapter *adapter, struct mc5 *mc5, int mode)
+{
+#define K * 1024
+
+	static unsigned int tcam_part_size[] = {	/* in K 72-bit entries */
+		64 K, 128 K, 256 K, 32 K
+	};
+
+#undef K
+
+	u32 cfg = t3_read_reg(adapter, A_MC5_DB_CONFIG);
+
+	mc5->adapter = adapter;
+	mc5->mode = (unsigned char)mode;
+	mc5->part_type = (unsigned char)G_TMTYPE(cfg);
+	if (cfg & F_TMTYPEHI)
+		mc5->part_type |= 4;
+
+	mc5->tcam_size = tcam_part_size[G_TMPARTSIZE(cfg)];
+	if (mode == MC5_MODE_144_BIT)
+		mc5->tcam_size /= 2;
+}
diff --git a/drivers/net/cxgb3/regs.h b/drivers/net/cxgb3/regs.h
new file mode 100644
index 00000000000..b56c5f52bcd
--- /dev/null
+++ b/drivers/net/cxgb3/regs.h
@@ -0,0 +1,2195 @@
+#define A_SG_CONTROL 0x0
+
+#define S_DROPPKT    20
+#define V_DROPPKT(x) ((x) << S_DROPPKT)
+#define F_DROPPKT    V_DROPPKT(1U)
+
+#define S_EGRGENCTRL    19
+#define V_EGRGENCTRL(x) ((x) << S_EGRGENCTRL)
+#define F_EGRGENCTRL    V_EGRGENCTRL(1U)
+
+#define S_USERSPACESIZE    14
+#define M_USERSPACESIZE    0x1f
+#define V_USERSPACESIZE(x) ((x) << S_USERSPACESIZE)
+
+#define S_HOSTPAGESIZE    11
+#define M_HOSTPAGESIZE    0x7
+#define V_HOSTPAGESIZE(x) ((x) << S_HOSTPAGESIZE)
+
+#define S_FLMODE    9
+#define V_FLMODE(x) ((x) << S_FLMODE)
+#define F_FLMODE    V_FLMODE(1U)
+
+#define S_PKTSHIFT    6
+#define M_PKTSHIFT    0x7
+#define V_PKTSHIFT(x) ((x) << S_PKTSHIFT)
+
+#define S_ONEINTMULTQ    5
+#define V_ONEINTMULTQ(x) ((x) << S_ONEINTMULTQ)
+#define F_ONEINTMULTQ    V_ONEINTMULTQ(1U)
+
+#define S_BIGENDIANINGRESS    2
+#define V_BIGENDIANINGRESS(x) ((x) << S_BIGENDIANINGRESS)
+#define F_BIGENDIANINGRESS    V_BIGENDIANINGRESS(1U)
+
+#define S_ISCSICOALESCING    1
+#define V_ISCSICOALESCING(x) ((x) << S_ISCSICOALESCING)
+#define F_ISCSICOALESCING    V_ISCSICOALESCING(1U)
+
+#define S_GLOBALENABLE    0
+#define V_GLOBALENABLE(x) ((x) << S_GLOBALENABLE)
+#define F_GLOBALENABLE    V_GLOBALENABLE(1U)
+
+#define S_AVOIDCQOVFL    24
+#define V_AVOIDCQOVFL(x) ((x) << S_AVOIDCQOVFL)
+#define F_AVOIDCQOVFL    V_AVOIDCQOVFL(1U)
+
+#define S_OPTONEINTMULTQ    23
+#define V_OPTONEINTMULTQ(x) ((x) << S_OPTONEINTMULTQ)
+#define F_OPTONEINTMULTQ    V_OPTONEINTMULTQ(1U)
+
+#define S_CQCRDTCTRL    22
+#define V_CQCRDTCTRL(x) ((x) << S_CQCRDTCTRL)
+#define F_CQCRDTCTRL    V_CQCRDTCTRL(1U)
+
+#define A_SG_KDOORBELL 0x4
+
+#define S_SELEGRCNTX    31
+#define V_SELEGRCNTX(x) ((x) << S_SELEGRCNTX)
+#define F_SELEGRCNTX    V_SELEGRCNTX(1U)
+
+#define S_EGRCNTX    0
+#define M_EGRCNTX    0xffff
+#define V_EGRCNTX(x) ((x) << S_EGRCNTX)
+
+#define A_SG_GTS 0x8
+
+#define S_RSPQ    29
+#define M_RSPQ    0x7
+#define V_RSPQ(x) ((x) << S_RSPQ)
+#define G_RSPQ(x) (((x) >> S_RSPQ) & M_RSPQ)
+
+#define S_NEWTIMER    16
+#define M_NEWTIMER    0x1fff
+#define V_NEWTIMER(x) ((x) << S_NEWTIMER)
+
+#define S_NEWINDEX    0
+#define M_NEWINDEX    0xffff
+#define V_NEWINDEX(x) ((x) << S_NEWINDEX)
+
+#define A_SG_CONTEXT_CMD 0xc
+
+#define S_CONTEXT_CMD_OPCODE    28
+#define M_CONTEXT_CMD_OPCODE    0xf
+#define V_CONTEXT_CMD_OPCODE(x) ((x) << S_CONTEXT_CMD_OPCODE)
+
+#define S_CONTEXT_CMD_BUSY    27
+#define V_CONTEXT_CMD_BUSY(x) ((x) << S_CONTEXT_CMD_BUSY)
+#define F_CONTEXT_CMD_BUSY    V_CONTEXT_CMD_BUSY(1U)
+
+#define S_CQ_CREDIT    20
+
+#define M_CQ_CREDIT    0x7f
+
+#define V_CQ_CREDIT(x) ((x) << S_CQ_CREDIT)
+
+#define G_CQ_CREDIT(x) (((x) >> S_CQ_CREDIT) & M_CQ_CREDIT)
+
+#define S_CQ    19
+
+#define V_CQ(x) ((x) << S_CQ)
+#define F_CQ    V_CQ(1U)
+
+#define S_RESPONSEQ    18
+#define V_RESPONSEQ(x) ((x) << S_RESPONSEQ)
+#define F_RESPONSEQ    V_RESPONSEQ(1U)
+
+#define S_EGRESS    17
+#define V_EGRESS(x) ((x) << S_EGRESS)
+#define F_EGRESS    V_EGRESS(1U)
+
+#define S_FREELIST    16
+#define V_FREELIST(x) ((x) << S_FREELIST)
+#define F_FREELIST    V_FREELIST(1U)
+
+#define S_CONTEXT    0
+#define M_CONTEXT    0xffff
+#define V_CONTEXT(x) ((x) << S_CONTEXT)
+
+#define G_CONTEXT(x) (((x) >> S_CONTEXT) & M_CONTEXT)
+
+#define A_SG_CONTEXT_DATA0 0x10
+
+#define A_SG_CONTEXT_DATA1 0x14
+
+#define A_SG_CONTEXT_DATA2 0x18
+
+#define A_SG_CONTEXT_DATA3 0x1c
+
+#define A_SG_CONTEXT_MASK0 0x20
+
+#define A_SG_CONTEXT_MASK1 0x24
+
+#define A_SG_CONTEXT_MASK2 0x28
+
+#define A_SG_CONTEXT_MASK3 0x2c
+
+#define A_SG_RSPQ_CREDIT_RETURN 0x30
+
+#define S_CREDITS    0
+#define M_CREDITS    0xffff
+#define V_CREDITS(x) ((x) << S_CREDITS)
+
+#define A_SG_DATA_INTR 0x34
+
+#define S_ERRINTR    31
+#define V_ERRINTR(x) ((x) << S_ERRINTR)
+#define F_ERRINTR    V_ERRINTR(1U)
+
+#define A_SG_HI_DRB_HI_THRSH 0x38
+
+#define A_SG_HI_DRB_LO_THRSH 0x3c
+
+#define A_SG_LO_DRB_HI_THRSH 0x40
+
+#define A_SG_LO_DRB_LO_THRSH 0x44
+
+#define A_SG_RSPQ_FL_STATUS 0x4c
+
+#define S_RSPQ0DISABLED    8
+
+#define A_SG_EGR_RCQ_DRB_THRSH 0x54
+
+#define S_HIRCQDRBTHRSH    16
+#define M_HIRCQDRBTHRSH    0x7ff
+#define V_HIRCQDRBTHRSH(x) ((x) << S_HIRCQDRBTHRSH)
+
+#define S_LORCQDRBTHRSH    0
+#define M_LORCQDRBTHRSH    0x7ff
+#define V_LORCQDRBTHRSH(x) ((x) << S_LORCQDRBTHRSH)
+
+#define A_SG_EGR_CNTX_BADDR 0x58
+
+#define A_SG_INT_CAUSE 0x5c
+
+#define S_RSPQDISABLED    3
+#define V_RSPQDISABLED(x) ((x) << S_RSPQDISABLED)
+#define F_RSPQDISABLED    V_RSPQDISABLED(1U)
+
+#define S_RSPQCREDITOVERFOW    2
+#define V_RSPQCREDITOVERFOW(x) ((x) << S_RSPQCREDITOVERFOW)
+#define F_RSPQCREDITOVERFOW    V_RSPQCREDITOVERFOW(1U)
+
+#define A_SG_INT_ENABLE 0x60
+
+#define A_SG_CMDQ_CREDIT_TH 0x64
+
+#define S_TIMEOUT    8
+#define M_TIMEOUT    0xffffff
+#define V_TIMEOUT(x) ((x) << S_TIMEOUT)
+
+#define S_THRESHOLD    0
+#define M_THRESHOLD    0xff
+#define V_THRESHOLD(x) ((x) << S_THRESHOLD)
+
+#define A_SG_TIMER_TICK 0x68
+
+#define A_SG_CQ_CONTEXT_BADDR 0x6c
+
+#define A_SG_OCO_BASE 0x70
+
+#define S_BASE1    16
+#define M_BASE1    0xffff
+#define V_BASE1(x) ((x) << S_BASE1)
+
+#define A_SG_DRB_PRI_THRESH 0x74
+
+#define A_PCIX_INT_ENABLE 0x80
+
+#define S_MSIXPARERR    22
+#define M_MSIXPARERR    0x7
+
+#define V_MSIXPARERR(x) ((x) << S_MSIXPARERR)
+
+#define S_CFPARERR    18
+#define M_CFPARERR    0xf
+
+#define V_CFPARERR(x) ((x) << S_CFPARERR)
+
+#define S_RFPARERR    14
+#define M_RFPARERR    0xf
+
+#define V_RFPARERR(x) ((x) << S_RFPARERR)
+
+#define S_WFPARERR    12
+#define M_WFPARERR    0x3
+
+#define V_WFPARERR(x) ((x) << S_WFPARERR)
+
+#define S_PIOPARERR    11
+#define V_PIOPARERR(x) ((x) << S_PIOPARERR)
+#define F_PIOPARERR    V_PIOPARERR(1U)
+
+#define S_DETUNCECCERR    10
+#define V_DETUNCECCERR(x) ((x) << S_DETUNCECCERR)
+#define F_DETUNCECCERR    V_DETUNCECCERR(1U)
+
+#define S_DETCORECCERR    9
+#define V_DETCORECCERR(x) ((x) << S_DETCORECCERR)
+#define F_DETCORECCERR    V_DETCORECCERR(1U)
+
+#define S_RCVSPLCMPERR    8
+#define V_RCVSPLCMPERR(x) ((x) << S_RCVSPLCMPERR)
+#define F_RCVSPLCMPERR    V_RCVSPLCMPERR(1U)
+
+#define S_UNXSPLCMP    7
+#define V_UNXSPLCMP(x) ((x) << S_UNXSPLCMP)
+#define F_UNXSPLCMP    V_UNXSPLCMP(1U)
+
+#define S_SPLCMPDIS    6
+#define V_SPLCMPDIS(x) ((x) << S_SPLCMPDIS)
+#define F_SPLCMPDIS    V_SPLCMPDIS(1U)
+
+#define S_DETPARERR    5
+#define V_DETPARERR(x) ((x) << S_DETPARERR)
+#define F_DETPARERR    V_DETPARERR(1U)
+
+#define S_SIGSYSERR    4
+#define V_SIGSYSERR(x) ((x) << S_SIGSYSERR)
+#define F_SIGSYSERR    V_SIGSYSERR(1U)
+
+#define S_RCVMSTABT    3
+#define V_RCVMSTABT(x) ((x) << S_RCVMSTABT)
+#define F_RCVMSTABT    V_RCVMSTABT(1U)
+
+#define S_RCVTARABT    2
+#define V_RCVTARABT(x) ((x) << S_RCVTARABT)
+#define F_RCVTARABT    V_RCVTARABT(1U)
+
+#define S_SIGTARABT    1
+#define V_SIGTARABT(x) ((x) << S_SIGTARABT)
+#define F_SIGTARABT    V_SIGTARABT(1U)
+
+#define S_MSTDETPARERR    0
+#define V_MSTDETPARERR(x) ((x) << S_MSTDETPARERR)
+#define F_MSTDETPARERR    V_MSTDETPARERR(1U)
+
+#define A_PCIX_INT_CAUSE 0x84
+
+#define A_PCIX_CFG 0x88
+
+#define S_CLIDECEN    18
+#define V_CLIDECEN(x) ((x) << S_CLIDECEN)
+#define F_CLIDECEN    V_CLIDECEN(1U)
+
+#define A_PCIX_MODE 0x8c
+
+#define S_PCLKRANGE    6
+#define M_PCLKRANGE    0x3
+#define V_PCLKRANGE(x) ((x) << S_PCLKRANGE)
+#define G_PCLKRANGE(x) (((x) >> S_PCLKRANGE) & M_PCLKRANGE)
+
+#define S_PCIXINITPAT    2
+#define M_PCIXINITPAT    0xf
+#define V_PCIXINITPAT(x) ((x) << S_PCIXINITPAT)
+#define G_PCIXINITPAT(x) (((x) >> S_PCIXINITPAT) & M_PCIXINITPAT)
+
+#define S_64BIT    0
+#define V_64BIT(x) ((x) << S_64BIT)
+#define F_64BIT    V_64BIT(1U)
+
+#define A_PCIE_INT_ENABLE 0x80
+
+#define S_BISTERR    15
+#define M_BISTERR    0xff
+
+#define V_BISTERR(x) ((x) << S_BISTERR)
+
+#define S_PCIE_MSIXPARERR    12
+#define M_PCIE_MSIXPARERR    0x7
+
+#define V_PCIE_MSIXPARERR(x) ((x) << S_PCIE_MSIXPARERR)
+
+#define S_PCIE_CFPARERR    11
+#define V_PCIE_CFPARERR(x) ((x) << S_PCIE_CFPARERR)
+#define F_PCIE_CFPARERR    V_PCIE_CFPARERR(1U)
+
+#define S_PCIE_RFPARERR    10
+#define V_PCIE_RFPARERR(x) ((x) << S_PCIE_RFPARERR)
+#define F_PCIE_RFPARERR    V_PCIE_RFPARERR(1U)
+
+#define S_PCIE_WFPARERR    9
+#define V_PCIE_WFPARERR(x) ((x) << S_PCIE_WFPARERR)
+#define F_PCIE_WFPARERR    V_PCIE_WFPARERR(1U)
+
+#define S_PCIE_PIOPARERR    8
+#define V_PCIE_PIOPARERR(x) ((x) << S_PCIE_PIOPARERR)
+#define F_PCIE_PIOPARERR    V_PCIE_PIOPARERR(1U)
+
+#define S_UNXSPLCPLERRC    7
+#define V_UNXSPLCPLERRC(x) ((x) << S_UNXSPLCPLERRC)
+#define F_UNXSPLCPLERRC    V_UNXSPLCPLERRC(1U)
+
+#define S_UNXSPLCPLERRR    6
+#define V_UNXSPLCPLERRR(x) ((x) << S_UNXSPLCPLERRR)
+#define F_UNXSPLCPLERRR    V_UNXSPLCPLERRR(1U)
+
+#define S_PEXERR    0
+#define V_PEXERR(x) ((x) << S_PEXERR)
+#define F_PEXERR    V_PEXERR(1U)
+
+#define A_PCIE_INT_CAUSE 0x84
+
+#define A_PCIE_CFG 0x88
+
+#define S_PCIE_CLIDECEN    16
+#define V_PCIE_CLIDECEN(x) ((x) << S_PCIE_CLIDECEN)
+#define F_PCIE_CLIDECEN    V_PCIE_CLIDECEN(1U)
+
+#define S_CRSTWRMMODE    0
+#define V_CRSTWRMMODE(x) ((x) << S_CRSTWRMMODE)
+#define F_CRSTWRMMODE    V_CRSTWRMMODE(1U)
+
+#define A_PCIE_MODE 0x8c
+
+#define S_NUMFSTTRNSEQRX    10
+#define M_NUMFSTTRNSEQRX    0xff
+#define V_NUMFSTTRNSEQRX(x) ((x) << S_NUMFSTTRNSEQRX)
+#define G_NUMFSTTRNSEQRX(x) (((x) >> S_NUMFSTTRNSEQRX) & M_NUMFSTTRNSEQRX)
+
+#define A_PCIE_PEX_CTRL0 0x98
+
+#define S_NUMFSTTRNSEQ    22
+#define M_NUMFSTTRNSEQ    0xff
+#define V_NUMFSTTRNSEQ(x) ((x) << S_NUMFSTTRNSEQ)
+#define G_NUMFSTTRNSEQ(x) (((x) >> S_NUMFSTTRNSEQ) & M_NUMFSTTRNSEQ)
+
+#define S_REPLAYLMT    2
+#define M_REPLAYLMT    0xfffff
+
+#define V_REPLAYLMT(x) ((x) << S_REPLAYLMT)
+
+#define A_PCIE_PEX_CTRL1 0x9c
+
+#define S_T3A_ACKLAT    0
+#define M_T3A_ACKLAT    0x7ff
+
+#define V_T3A_ACKLAT(x) ((x) << S_T3A_ACKLAT)
+
+#define S_ACKLAT    0
+#define M_ACKLAT    0x1fff
+
+#define V_ACKLAT(x) ((x) << S_ACKLAT)
+
+#define A_PCIE_PEX_ERR 0xa4
+
+#define A_T3DBG_GPIO_EN 0xd0
+
+#define S_GPIO11_OEN    27
+#define V_GPIO11_OEN(x) ((x) << S_GPIO11_OEN)
+#define F_GPIO11_OEN    V_GPIO11_OEN(1U)
+
+#define S_GPIO10_OEN    26
+#define V_GPIO10_OEN(x) ((x) << S_GPIO10_OEN)
+#define F_GPIO10_OEN    V_GPIO10_OEN(1U)
+
+#define S_GPIO7_OEN    23
+#define V_GPIO7_OEN(x) ((x) << S_GPIO7_OEN)
+#define F_GPIO7_OEN    V_GPIO7_OEN(1U)
+
+#define S_GPIO6_OEN    22
+#define V_GPIO6_OEN(x) ((x) << S_GPIO6_OEN)
+#define F_GPIO6_OEN    V_GPIO6_OEN(1U)
+
+#define S_GPIO5_OEN    21
+#define V_GPIO5_OEN(x) ((x) << S_GPIO5_OEN)
+#define F_GPIO5_OEN    V_GPIO5_OEN(1U)
+
+#define S_GPIO4_OEN    20
+#define V_GPIO4_OEN(x) ((x) << S_GPIO4_OEN)
+#define F_GPIO4_OEN    V_GPIO4_OEN(1U)
+
+#define S_GPIO2_OEN    18
+#define V_GPIO2_OEN(x) ((x) << S_GPIO2_OEN)
+#define F_GPIO2_OEN    V_GPIO2_OEN(1U)
+
+#define S_GPIO1_OEN    17
+#define V_GPIO1_OEN(x) ((x) << S_GPIO1_OEN)
+#define F_GPIO1_OEN    V_GPIO1_OEN(1U)
+
+#define S_GPIO0_OEN    16
+#define V_GPIO0_OEN(x) ((x) << S_GPIO0_OEN)
+#define F_GPIO0_OEN    V_GPIO0_OEN(1U)
+
+#define S_GPIO10_OUT_VAL    10
+#define V_GPIO10_OUT_VAL(x) ((x) << S_GPIO10_OUT_VAL)
+#define F_GPIO10_OUT_VAL    V_GPIO10_OUT_VAL(1U)
+
+#define S_GPIO7_OUT_VAL    7
+#define V_GPIO7_OUT_VAL(x) ((x) << S_GPIO7_OUT_VAL)
+#define F_GPIO7_OUT_VAL    V_GPIO7_OUT_VAL(1U)
+
+#define S_GPIO6_OUT_VAL    6
+#define V_GPIO6_OUT_VAL(x) ((x) << S_GPIO6_OUT_VAL)
+#define F_GPIO6_OUT_VAL    V_GPIO6_OUT_VAL(1U)
+
+#define S_GPIO5_OUT_VAL    5
+#define V_GPIO5_OUT_VAL(x) ((x) << S_GPIO5_OUT_VAL)
+#define F_GPIO5_OUT_VAL    V_GPIO5_OUT_VAL(1U)
+
+#define S_GPIO4_OUT_VAL    4
+#define V_GPIO4_OUT_VAL(x) ((x) << S_GPIO4_OUT_VAL)
+#define F_GPIO4_OUT_VAL    V_GPIO4_OUT_VAL(1U)
+
+#define S_GPIO2_OUT_VAL    2
+#define V_GPIO2_OUT_VAL(x) ((x) << S_GPIO2_OUT_VAL)
+#define F_GPIO2_OUT_VAL    V_GPIO2_OUT_VAL(1U)
+
+#define S_GPIO1_OUT_VAL    1
+#define V_GPIO1_OUT_VAL(x) ((x) << S_GPIO1_OUT_VAL)
+#define F_GPIO1_OUT_VAL    V_GPIO1_OUT_VAL(1U)
+
+#define S_GPIO0_OUT_VAL    0
+#define V_GPIO0_OUT_VAL(x) ((x) << S_GPIO0_OUT_VAL)
+#define F_GPIO0_OUT_VAL    V_GPIO0_OUT_VAL(1U)
+
+#define A_T3DBG_INT_ENABLE 0xd8
+
+#define S_GPIO11    11
+#define V_GPIO11(x) ((x) << S_GPIO11)
+#define F_GPIO11    V_GPIO11(1U)
+
+#define S_GPIO10    10
+#define V_GPIO10(x) ((x) << S_GPIO10)
+#define F_GPIO10    V_GPIO10(1U)
+
+#define S_GPIO7    7
+#define V_GPIO7(x) ((x) << S_GPIO7)
+#define F_GPIO7    V_GPIO7(1U)
+
+#define S_GPIO6    6
+#define V_GPIO6(x) ((x) << S_GPIO6)
+#define F_GPIO6    V_GPIO6(1U)
+
+#define S_GPIO5    5
+#define V_GPIO5(x) ((x) << S_GPIO5)
+#define F_GPIO5    V_GPIO5(1U)
+
+#define S_GPIO4    4
+#define V_GPIO4(x) ((x) << S_GPIO4)
+#define F_GPIO4    V_GPIO4(1U)
+
+#define S_GPIO3    3
+#define V_GPIO3(x) ((x) << S_GPIO3)
+#define F_GPIO3    V_GPIO3(1U)
+
+#define S_GPIO2    2
+#define V_GPIO2(x) ((x) << S_GPIO2)
+#define F_GPIO2    V_GPIO2(1U)
+
+#define S_GPIO1    1
+#define V_GPIO1(x) ((x) << S_GPIO1)
+#define F_GPIO1    V_GPIO1(1U)
+
+#define S_GPIO0    0
+#define V_GPIO0(x) ((x) << S_GPIO0)
+#define F_GPIO0    V_GPIO0(1U)
+
+#define A_T3DBG_INT_CAUSE 0xdc
+
+#define A_T3DBG_GPIO_ACT_LOW 0xf0
+
+#define MC7_PMRX_BASE_ADDR 0x100
+
+#define A_MC7_CFG 0x100
+
+#define S_IFEN    13
+#define V_IFEN(x) ((x) << S_IFEN)
+#define F_IFEN    V_IFEN(1U)
+
+#define S_TERM150    11
+#define V_TERM150(x) ((x) << S_TERM150)
+#define F_TERM150    V_TERM150(1U)
+
+#define S_SLOW    10
+#define V_SLOW(x) ((x) << S_SLOW)
+#define F_SLOW    V_SLOW(1U)
+
+#define S_WIDTH    8
+#define M_WIDTH    0x3
+#define V_WIDTH(x) ((x) << S_WIDTH)
+#define G_WIDTH(x) (((x) >> S_WIDTH) & M_WIDTH)
+
+#define S_BKS    6
+#define V_BKS(x) ((x) << S_BKS)
+#define F_BKS    V_BKS(1U)
+
+#define S_ORG    5
+#define V_ORG(x) ((x) << S_ORG)
+#define F_ORG    V_ORG(1U)
+
+#define S_DEN    2
+#define M_DEN    0x7
+#define V_DEN(x) ((x) << S_DEN)
+#define G_DEN(x) (((x) >> S_DEN) & M_DEN)
+
+#define S_RDY    1
+#define V_RDY(x) ((x) << S_RDY)
+#define F_RDY    V_RDY(1U)
+
+#define S_CLKEN    0
+#define V_CLKEN(x) ((x) << S_CLKEN)
+#define F_CLKEN    V_CLKEN(1U)
+
+#define A_MC7_MODE 0x104
+
+#define S_BUSY    31
+#define V_BUSY(x) ((x) << S_BUSY)
+#define F_BUSY    V_BUSY(1U)
+
+#define S_BUSY    31
+#define V_BUSY(x) ((x) << S_BUSY)
+#define F_BUSY    V_BUSY(1U)
+
+#define A_MC7_EXT_MODE1 0x108
+
+#define A_MC7_EXT_MODE2 0x10c
+
+#define A_MC7_EXT_MODE3 0x110
+
+#define A_MC7_PRE 0x114
+
+#define A_MC7_REF 0x118
+
+#define S_PREREFDIV    1
+#define M_PREREFDIV    0x3fff
+#define V_PREREFDIV(x) ((x) << S_PREREFDIV)
+
+#define S_PERREFEN    0
+#define V_PERREFEN(x) ((x) << S_PERREFEN)
+#define F_PERREFEN    V_PERREFEN(1U)
+
+#define A_MC7_DLL 0x11c
+
+#define S_DLLENB    1
+#define V_DLLENB(x) ((x) << S_DLLENB)
+#define F_DLLENB    V_DLLENB(1U)
+
+#define S_DLLRST    0
+#define V_DLLRST(x) ((x) << S_DLLRST)
+#define F_DLLRST    V_DLLRST(1U)
+
+#define A_MC7_PARM 0x120
+
+#define S_ACTTOPREDLY    26
+#define M_ACTTOPREDLY    0xf
+#define V_ACTTOPREDLY(x) ((x) << S_ACTTOPREDLY)
+
+#define S_ACTTORDWRDLY    23
+#define M_ACTTORDWRDLY    0x7
+#define V_ACTTORDWRDLY(x) ((x) << S_ACTTORDWRDLY)
+
+#define S_PRECYC    20
+#define M_PRECYC    0x7
+#define V_PRECYC(x) ((x) << S_PRECYC)
+
+#define S_REFCYC    13
+#define M_REFCYC    0x7f
+#define V_REFCYC(x) ((x) << S_REFCYC)
+
+#define S_BKCYC    8
+#define M_BKCYC    0x1f
+#define V_BKCYC(x) ((x) << S_BKCYC)
+
+#define S_WRTORDDLY    4
+#define M_WRTORDDLY    0xf
+#define V_WRTORDDLY(x) ((x) << S_WRTORDDLY)
+
+#define S_RDTOWRDLY    0
+#define M_RDTOWRDLY    0xf
+#define V_RDTOWRDLY(x) ((x) << S_RDTOWRDLY)
+
+#define A_MC7_CAL 0x128
+
+#define S_BUSY    31
+#define V_BUSY(x) ((x) << S_BUSY)
+#define F_BUSY    V_BUSY(1U)
+
+#define S_BUSY    31
+#define V_BUSY(x) ((x) << S_BUSY)
+#define F_BUSY    V_BUSY(1U)
+
+#define S_CAL_FAULT    30
+#define V_CAL_FAULT(x) ((x) << S_CAL_FAULT)
+#define F_CAL_FAULT    V_CAL_FAULT(1U)
+
+#define S_SGL_CAL_EN    20
+#define V_SGL_CAL_EN(x) ((x) << S_SGL_CAL_EN)
+#define F_SGL_CAL_EN    V_SGL_CAL_EN(1U)
+
+#define A_MC7_ERR_ADDR 0x12c
+
+#define A_MC7_ECC 0x130
+
+#define S_ECCCHKEN    1
+#define V_ECCCHKEN(x) ((x) << S_ECCCHKEN)
+#define F_ECCCHKEN    V_ECCCHKEN(1U)
+
+#define S_ECCGENEN    0
+#define V_ECCGENEN(x) ((x) << S_ECCGENEN)
+#define F_ECCGENEN    V_ECCGENEN(1U)
+
+#define A_MC7_CE_ADDR 0x134
+
+#define A_MC7_CE_DATA0 0x138
+
+#define A_MC7_CE_DATA1 0x13c
+
+#define A_MC7_CE_DATA2 0x140
+
+#define S_DATA    0
+#define M_DATA    0xff
+
+#define G_DATA(x) (((x) >> S_DATA) & M_DATA)
+
+#define A_MC7_UE_ADDR 0x144
+
+#define A_MC7_UE_DATA0 0x148
+
+#define A_MC7_UE_DATA1 0x14c
+
+#define A_MC7_UE_DATA2 0x150
+
+#define A_MC7_BD_ADDR 0x154
+
+#define S_ADDR    3
+
+#define M_ADDR    0x1fffffff
+
+#define A_MC7_BD_DATA0 0x158
+
+#define A_MC7_BD_DATA1 0x15c
+
+#define A_MC7_BD_OP 0x164
+
+#define S_OP    0
+
+#define V_OP(x) ((x) << S_OP)
+#define F_OP    V_OP(1U)
+
+#define F_OP    V_OP(1U)
+#define A_SF_OP 0x6dc
+
+#define A_MC7_BIST_ADDR_BEG 0x168
+
+#define A_MC7_BIST_ADDR_END 0x16c
+
+#define A_MC7_BIST_DATA 0x170
+
+#define A_MC7_BIST_OP 0x174
+
+#define S_CONT    3
+#define V_CONT(x) ((x) << S_CONT)
+#define F_CONT    V_CONT(1U)
+
+#define F_CONT    V_CONT(1U)
+
+#define A_MC7_INT_ENABLE 0x178
+
+#define S_AE    17
+#define V_AE(x) ((x) << S_AE)
+#define F_AE    V_AE(1U)
+
+#define S_PE    2
+#define M_PE    0x7fff
+
+#define V_PE(x) ((x) << S_PE)
+
+#define G_PE(x) (((x) >> S_PE) & M_PE)
+
+#define S_UE    1
+#define V_UE(x) ((x) << S_UE)
+#define F_UE    V_UE(1U)
+
+#define S_CE    0
+#define V_CE(x) ((x) << S_CE)
+#define F_CE    V_CE(1U)
+
+#define A_MC7_INT_CAUSE 0x17c
+
+#define MC7_PMTX_BASE_ADDR 0x180
+
+#define MC7_CM_BASE_ADDR 0x200
+
+#define A_CIM_BOOT_CFG 0x280
+
+#define S_BOOTADDR    2
+#define M_BOOTADDR    0x3fffffff
+#define V_BOOTADDR(x) ((x) << S_BOOTADDR)
+
+#define A_CIM_SDRAM_BASE_ADDR 0x28c
+
+#define A_CIM_SDRAM_ADDR_SIZE 0x290
+
+#define A_CIM_HOST_INT_ENABLE 0x298
+
+#define A_CIM_HOST_INT_CAUSE 0x29c
+
+#define S_BLKWRPLINT    12
+#define V_BLKWRPLINT(x) ((x) << S_BLKWRPLINT)
+#define F_BLKWRPLINT    V_BLKWRPLINT(1U)
+
+#define S_BLKRDPLINT    11
+#define V_BLKRDPLINT(x) ((x) << S_BLKRDPLINT)
+#define F_BLKRDPLINT    V_BLKRDPLINT(1U)
+
+#define S_BLKWRCTLINT    10
+#define V_BLKWRCTLINT(x) ((x) << S_BLKWRCTLINT)
+#define F_BLKWRCTLINT    V_BLKWRCTLINT(1U)
+
+#define S_BLKRDCTLINT    9
+#define V_BLKRDCTLINT(x) ((x) << S_BLKRDCTLINT)
+#define F_BLKRDCTLINT    V_BLKRDCTLINT(1U)
+
+#define S_BLKWRFLASHINT    8
+#define V_BLKWRFLASHINT(x) ((x) << S_BLKWRFLASHINT)
+#define F_BLKWRFLASHINT    V_BLKWRFLASHINT(1U)
+
+#define S_BLKRDFLASHINT    7
+#define V_BLKRDFLASHINT(x) ((x) << S_BLKRDFLASHINT)
+#define F_BLKRDFLASHINT    V_BLKRDFLASHINT(1U)
+
+#define S_SGLWRFLASHINT    6
+#define V_SGLWRFLASHINT(x) ((x) << S_SGLWRFLASHINT)
+#define F_SGLWRFLASHINT    V_SGLWRFLASHINT(1U)
+
+#define S_WRBLKFLASHINT    5
+#define V_WRBLKFLASHINT(x) ((x) << S_WRBLKFLASHINT)
+#define F_WRBLKFLASHINT    V_WRBLKFLASHINT(1U)
+
+#define S_BLKWRBOOTINT    4
+#define V_BLKWRBOOTINT(x) ((x) << S_BLKWRBOOTINT)
+#define F_BLKWRBOOTINT    V_BLKWRBOOTINT(1U)
+
+#define S_FLASHRANGEINT    2
+#define V_FLASHRANGEINT(x) ((x) << S_FLASHRANGEINT)
+#define F_FLASHRANGEINT    V_FLASHRANGEINT(1U)
+
+#define S_SDRAMRANGEINT    1
+#define V_SDRAMRANGEINT(x) ((x) << S_SDRAMRANGEINT)
+#define F_SDRAMRANGEINT    V_SDRAMRANGEINT(1U)
+
+#define S_RSVDSPACEINT    0
+#define V_RSVDSPACEINT(x) ((x) << S_RSVDSPACEINT)
+#define F_RSVDSPACEINT    V_RSVDSPACEINT(1U)
+
+#define A_CIM_HOST_ACC_CTRL 0x2b0
+
+#define S_HOSTBUSY    17
+#define V_HOSTBUSY(x) ((x) << S_HOSTBUSY)
+#define F_HOSTBUSY    V_HOSTBUSY(1U)
+
+#define A_CIM_HOST_ACC_DATA 0x2b4
+
+#define A_TP_IN_CONFIG 0x300
+
+#define S_NICMODE    14
+#define V_NICMODE(x) ((x) << S_NICMODE)
+#define F_NICMODE    V_NICMODE(1U)
+
+#define F_NICMODE    V_NICMODE(1U)
+
+#define S_IPV6ENABLE    15
+#define V_IPV6ENABLE(x) ((x) << S_IPV6ENABLE)
+#define F_IPV6ENABLE    V_IPV6ENABLE(1U)
+
+#define A_TP_OUT_CONFIG 0x304
+
+#define S_VLANEXTRACTIONENABLE    12
+
+#define A_TP_GLOBAL_CONFIG 0x308
+
+#define S_TXPACINGENABLE    24
+#define V_TXPACINGENABLE(x) ((x) << S_TXPACINGENABLE)
+#define F_TXPACINGENABLE    V_TXPACINGENABLE(1U)
+
+#define S_PATHMTU    15
+#define V_PATHMTU(x) ((x) << S_PATHMTU)
+#define F_PATHMTU    V_PATHMTU(1U)
+
+#define S_IPCHECKSUMOFFLOAD    13
+#define V_IPCHECKSUMOFFLOAD(x) ((x) << S_IPCHECKSUMOFFLOAD)
+#define F_IPCHECKSUMOFFLOAD    V_IPCHECKSUMOFFLOAD(1U)
+
+#define S_UDPCHECKSUMOFFLOAD    12
+#define V_UDPCHECKSUMOFFLOAD(x) ((x) << S_UDPCHECKSUMOFFLOAD)
+#define F_UDPCHECKSUMOFFLOAD    V_UDPCHECKSUMOFFLOAD(1U)
+
+#define S_TCPCHECKSUMOFFLOAD    11
+#define V_TCPCHECKSUMOFFLOAD(x) ((x) << S_TCPCHECKSUMOFFLOAD)
+#define F_TCPCHECKSUMOFFLOAD    V_TCPCHECKSUMOFFLOAD(1U)
+
+#define S_IPTTL    0
+#define M_IPTTL    0xff
+#define V_IPTTL(x) ((x) << S_IPTTL)
+
+#define A_TP_CMM_MM_BASE 0x314
+
+#define A_TP_CMM_TIMER_BASE 0x318
+
+#define S_CMTIMERMAXNUM    28
+#define M_CMTIMERMAXNUM    0x3
+#define V_CMTIMERMAXNUM(x) ((x) << S_CMTIMERMAXNUM)
+
+#define A_TP_PMM_SIZE 0x31c
+
+#define A_TP_PMM_TX_BASE 0x320
+
+#define A_TP_PMM_RX_BASE 0x328
+
+#define A_TP_PMM_RX_PAGE_SIZE 0x32c
+
+#define A_TP_PMM_RX_MAX_PAGE 0x330
+
+#define A_TP_PMM_TX_PAGE_SIZE 0x334
+
+#define A_TP_PMM_TX_MAX_PAGE 0x338
+
+#define A_TP_TCP_OPTIONS 0x340
+
+#define S_MTUDEFAULT    16
+#define M_MTUDEFAULT    0xffff
+#define V_MTUDEFAULT(x) ((x) << S_MTUDEFAULT)
+
+#define S_MTUENABLE    10
+#define V_MTUENABLE(x) ((x) << S_MTUENABLE)
+#define F_MTUENABLE    V_MTUENABLE(1U)
+
+#define S_SACKRX    8
+#define V_SACKRX(x) ((x) << S_SACKRX)
+#define F_SACKRX    V_SACKRX(1U)
+
+#define S_SACKMODE    4
+
+#define M_SACKMODE    0x3
+
+#define V_SACKMODE(x) ((x) << S_SACKMODE)
+
+#define S_WINDOWSCALEMODE    2
+#define M_WINDOWSCALEMODE    0x3
+#define V_WINDOWSCALEMODE(x) ((x) << S_WINDOWSCALEMODE)
+
+#define S_TIMESTAMPSMODE    0
+
+#define M_TIMESTAMPSMODE    0x3
+
+#define V_TIMESTAMPSMODE(x) ((x) << S_TIMESTAMPSMODE)
+
+#define A_TP_DACK_CONFIG 0x344
+
+#define S_AUTOSTATE3    30
+#define M_AUTOSTATE3    0x3
+#define V_AUTOSTATE3(x) ((x) << S_AUTOSTATE3)
+
+#define S_AUTOSTATE2    28
+#define M_AUTOSTATE2    0x3
+#define V_AUTOSTATE2(x) ((x) << S_AUTOSTATE2)
+
+#define S_AUTOSTATE1    26
+#define M_AUTOSTATE1    0x3
+#define V_AUTOSTATE1(x) ((x) << S_AUTOSTATE1)
+
+#define S_BYTETHRESHOLD    5
+#define M_BYTETHRESHOLD    0xfffff
+#define V_BYTETHRESHOLD(x) ((x) << S_BYTETHRESHOLD)
+
+#define S_MSSTHRESHOLD    3
+#define M_MSSTHRESHOLD    0x3
+#define V_MSSTHRESHOLD(x) ((x) << S_MSSTHRESHOLD)
+
+#define S_AUTOCAREFUL    2
+#define V_AUTOCAREFUL(x) ((x) << S_AUTOCAREFUL)
+#define F_AUTOCAREFUL    V_AUTOCAREFUL(1U)
+
+#define S_AUTOENABLE    1
+#define V_AUTOENABLE(x) ((x) << S_AUTOENABLE)
+#define F_AUTOENABLE    V_AUTOENABLE(1U)
+
+#define S_DACK_MODE    0
+#define V_DACK_MODE(x) ((x) << S_DACK_MODE)
+#define F_DACK_MODE    V_DACK_MODE(1U)
+
+#define A_TP_PC_CONFIG 0x348
+
+#define S_TXTOSQUEUEMAPMODE    26
+#define V_TXTOSQUEUEMAPMODE(x) ((x) << S_TXTOSQUEUEMAPMODE)
+#define F_TXTOSQUEUEMAPMODE    V_TXTOSQUEUEMAPMODE(1U)
+
+#define S_ENABLEEPCMDAFULL    23
+#define V_ENABLEEPCMDAFULL(x) ((x) << S_ENABLEEPCMDAFULL)
+#define F_ENABLEEPCMDAFULL    V_ENABLEEPCMDAFULL(1U)
+
+#define S_MODULATEUNIONMODE    22
+#define V_MODULATEUNIONMODE(x) ((x) << S_MODULATEUNIONMODE)
+#define F_MODULATEUNIONMODE    V_MODULATEUNIONMODE(1U)
+
+#define S_TXDEFERENABLE    20
+#define V_TXDEFERENABLE(x) ((x) << S_TXDEFERENABLE)
+#define F_TXDEFERENABLE    V_TXDEFERENABLE(1U)
+
+#define S_RXCONGESTIONMODE    19
+#define V_RXCONGESTIONMODE(x) ((x) << S_RXCONGESTIONMODE)
+#define F_RXCONGESTIONMODE    V_RXCONGESTIONMODE(1U)
+
+#define S_HEARBEATDACK    16
+#define V_HEARBEATDACK(x) ((x) << S_HEARBEATDACK)
+#define F_HEARBEATDACK    V_HEARBEATDACK(1U)
+
+#define S_TXCONGESTIONMODE    15
+#define V_TXCONGESTIONMODE(x) ((x) << S_TXCONGESTIONMODE)
+#define F_TXCONGESTIONMODE    V_TXCONGESTIONMODE(1U)
+
+#define S_ENABLEOCSPIFULL    30
+#define V_ENABLEOCSPIFULL(x) ((x) << S_ENABLEOCSPIFULL)
+#define F_ENABLEOCSPIFULL    V_ENABLEOCSPIFULL(1U)
+
+#define S_LOCKTID    28
+#define V_LOCKTID(x) ((x) << S_LOCKTID)
+#define F_LOCKTID    V_LOCKTID(1U)
+
+#define A_TP_PC_CONFIG2 0x34c
+
+#define S_CHDRAFULL    4
+#define V_CHDRAFULL(x) ((x) << S_CHDRAFULL)
+#define F_CHDRAFULL    V_CHDRAFULL(1U)
+
+#define A_TP_TCP_BACKOFF_REG0 0x350
+
+#define A_TP_TCP_BACKOFF_REG1 0x354
+
+#define A_TP_TCP_BACKOFF_REG2 0x358
+
+#define A_TP_TCP_BACKOFF_REG3 0x35c
+
+#define A_TP_PARA_REG2 0x368
+
+#define S_MAXRXDATA    16
+#define M_MAXRXDATA    0xffff
+#define V_MAXRXDATA(x) ((x) << S_MAXRXDATA)
+
+#define S_RXCOALESCESIZE    0
+#define M_RXCOALESCESIZE    0xffff
+#define V_RXCOALESCESIZE(x) ((x) << S_RXCOALESCESIZE)
+
+#define A_TP_PARA_REG3 0x36c
+
+#define S_TXDATAACKIDX    16
+#define M_TXDATAACKIDX    0xf
+
+#define V_TXDATAACKIDX(x) ((x) << S_TXDATAACKIDX)
+
+#define S_TXPACEAUTOSTRICT    10
+#define V_TXPACEAUTOSTRICT(x) ((x) << S_TXPACEAUTOSTRICT)
+#define F_TXPACEAUTOSTRICT    V_TXPACEAUTOSTRICT(1U)
+
+#define S_TXPACEFIXED    9
+#define V_TXPACEFIXED(x) ((x) << S_TXPACEFIXED)
+#define F_TXPACEFIXED    V_TXPACEFIXED(1U)
+
+#define S_TXPACEAUTO    8
+#define V_TXPACEAUTO(x) ((x) << S_TXPACEAUTO)
+#define F_TXPACEAUTO    V_TXPACEAUTO(1U)
+
+#define S_RXCOALESCEENABLE    1
+#define V_RXCOALESCEENABLE(x) ((x) << S_RXCOALESCEENABLE)
+#define F_RXCOALESCEENABLE    V_RXCOALESCEENABLE(1U)
+
+#define S_RXCOALESCEPSHEN    0
+#define V_RXCOALESCEPSHEN(x) ((x) << S_RXCOALESCEPSHEN)
+#define F_RXCOALESCEPSHEN    V_RXCOALESCEPSHEN(1U)
+
+#define A_TP_PARA_REG4 0x370
+
+#define A_TP_PARA_REG6 0x378
+
+#define S_T3A_ENABLEESND    13
+#define V_T3A_ENABLEESND(x) ((x) << S_T3A_ENABLEESND)
+#define F_T3A_ENABLEESND    V_T3A_ENABLEESND(1U)
+
+#define S_ENABLEESND    11
+#define V_ENABLEESND(x) ((x) << S_ENABLEESND)
+#define F_ENABLEESND    V_ENABLEESND(1U)
+
+#define A_TP_PARA_REG7 0x37c
+
+#define S_PMMAXXFERLEN1    16
+#define M_PMMAXXFERLEN1    0xffff
+#define V_PMMAXXFERLEN1(x) ((x) << S_PMMAXXFERLEN1)
+
+#define S_PMMAXXFERLEN0    0
+#define M_PMMAXXFERLEN0    0xffff
+#define V_PMMAXXFERLEN0(x) ((x) << S_PMMAXXFERLEN0)
+
+#define A_TP_TIMER_RESOLUTION 0x390
+
+#define S_TIMERRESOLUTION    16
+#define M_TIMERRESOLUTION    0xff
+#define V_TIMERRESOLUTION(x) ((x) << S_TIMERRESOLUTION)
+
+#define S_TIMESTAMPRESOLUTION    8
+#define M_TIMESTAMPRESOLUTION    0xff
+#define V_TIMESTAMPRESOLUTION(x) ((x) << S_TIMESTAMPRESOLUTION)
+
+#define S_DELAYEDACKRESOLUTION    0
+#define M_DELAYEDACKRESOLUTION    0xff
+#define V_DELAYEDACKRESOLUTION(x) ((x) << S_DELAYEDACKRESOLUTION)
+
+#define A_TP_MSL 0x394
+
+#define A_TP_RXT_MIN 0x398
+
+#define A_TP_RXT_MAX 0x39c
+
+#define A_TP_PERS_MIN 0x3a0
+
+#define A_TP_PERS_MAX 0x3a4
+
+#define A_TP_KEEP_IDLE 0x3a8
+
+#define A_TP_KEEP_INTVL 0x3ac
+
+#define A_TP_INIT_SRTT 0x3b0
+
+#define A_TP_DACK_TIMER 0x3b4
+
+#define A_TP_FINWAIT2_TIMER 0x3b8
+
+#define A_TP_SHIFT_CNT 0x3c0
+
+#define S_SYNSHIFTMAX    24
+
+#define M_SYNSHIFTMAX    0xff
+
+#define V_SYNSHIFTMAX(x) ((x) << S_SYNSHIFTMAX)
+
+#define S_RXTSHIFTMAXR1    20
+
+#define M_RXTSHIFTMAXR1    0xf
+
+#define V_RXTSHIFTMAXR1(x) ((x) << S_RXTSHIFTMAXR1)
+
+#define S_RXTSHIFTMAXR2    16
+
+#define M_RXTSHIFTMAXR2    0xf
+
+#define V_RXTSHIFTMAXR2(x) ((x) << S_RXTSHIFTMAXR2)
+
+#define S_PERSHIFTBACKOFFMAX    12
+#define M_PERSHIFTBACKOFFMAX    0xf
+#define V_PERSHIFTBACKOFFMAX(x) ((x) << S_PERSHIFTBACKOFFMAX)
+
+#define S_PERSHIFTMAX    8
+#define M_PERSHIFTMAX    0xf
+#define V_PERSHIFTMAX(x) ((x) << S_PERSHIFTMAX)
+
+#define S_KEEPALIVEMAX    0
+
+#define M_KEEPALIVEMAX    0xff
+
+#define V_KEEPALIVEMAX(x) ((x) << S_KEEPALIVEMAX)
+
+#define A_TP_MTU_PORT_TABLE 0x3d0
+
+#define A_TP_CCTRL_TABLE 0x3dc
+
+#define A_TP_MTU_TABLE 0x3e4
+
+#define A_TP_RSS_MAP_TABLE 0x3e8
+
+#define A_TP_RSS_LKP_TABLE 0x3ec
+
+#define A_TP_RSS_CONFIG 0x3f0
+
+#define S_TNL4TUPEN    29
+#define V_TNL4TUPEN(x) ((x) << S_TNL4TUPEN)
+#define F_TNL4TUPEN    V_TNL4TUPEN(1U)
+
+#define S_TNL2TUPEN    28
+#define V_TNL2TUPEN(x) ((x) << S_TNL2TUPEN)
+#define F_TNL2TUPEN    V_TNL2TUPEN(1U)
+
+#define S_TNLPRTEN    26
+#define V_TNLPRTEN(x) ((x) << S_TNLPRTEN)
+#define F_TNLPRTEN    V_TNLPRTEN(1U)
+
+#define S_TNLMAPEN    25
+#define V_TNLMAPEN(x) ((x) << S_TNLMAPEN)
+#define F_TNLMAPEN    V_TNLMAPEN(1U)
+
+#define S_TNLLKPEN    24
+#define V_TNLLKPEN(x) ((x) << S_TNLLKPEN)
+#define F_TNLLKPEN    V_TNLLKPEN(1U)
+
+#define S_RRCPLCPUSIZE    4
+#define M_RRCPLCPUSIZE    0x7
+#define V_RRCPLCPUSIZE(x) ((x) << S_RRCPLCPUSIZE)
+
+#define S_RQFEEDBACKENABLE    3
+#define V_RQFEEDBACKENABLE(x) ((x) << S_RQFEEDBACKENABLE)
+#define F_RQFEEDBACKENABLE    V_RQFEEDBACKENABLE(1U)
+
+#define S_DISABLE    0
+
+#define A_TP_TM_PIO_ADDR 0x418
+
+#define A_TP_TM_PIO_DATA 0x41c
+
+#define A_TP_TX_MOD_QUE_TABLE 0x420
+
+#define A_TP_TX_RESOURCE_LIMIT 0x424
+
+#define A_TP_TX_MOD_QUEUE_REQ_MAP 0x428
+
+#define S_TX_MOD_QUEUE_REQ_MAP    0
+#define M_TX_MOD_QUEUE_REQ_MAP    0xff
+#define V_TX_MOD_QUEUE_REQ_MAP(x) ((x) << S_TX_MOD_QUEUE_REQ_MAP)
+
+#define A_TP_TX_MOD_QUEUE_WEIGHT1 0x42c
+
+#define A_TP_TX_MOD_QUEUE_WEIGHT0 0x430
+
+#define A_TP_MOD_CHANNEL_WEIGHT 0x434
+
+#define A_TP_PIO_ADDR 0x440
+
+#define A_TP_PIO_DATA 0x444
+
+#define A_TP_RESET 0x44c
+
+#define S_FLSTINITENABLE    1
+#define V_FLSTINITENABLE(x) ((x) << S_FLSTINITENABLE)
+#define F_FLSTINITENABLE    V_FLSTINITENABLE(1U)
+
+#define S_TPRESET    0
+#define V_TPRESET(x) ((x) << S_TPRESET)
+#define F_TPRESET    V_TPRESET(1U)
+
+#define A_TP_CMM_MM_RX_FLST_BASE 0x460
+
+#define A_TP_CMM_MM_TX_FLST_BASE 0x464
+
+#define A_TP_CMM_MM_PS_FLST_BASE 0x468
+
+#define A_TP_MIB_INDEX 0x450
+
+#define A_TP_MIB_RDATA 0x454
+
+#define A_TP_CMM_MM_MAX_PSTRUCT 0x46c
+
+#define A_TP_INT_ENABLE 0x470
+
+#define A_TP_INT_CAUSE 0x474
+
+#define A_TP_TX_MOD_Q1_Q0_RATE_LIMIT 0x8
+
+#define A_TP_TX_DROP_CFG_CH0 0x12b
+
+#define A_TP_TX_DROP_MODE 0x12f
+
+#define A_TP_EGRESS_CONFIG 0x145
+
+#define S_REWRITEFORCETOSIZE    0
+#define V_REWRITEFORCETOSIZE(x) ((x) << S_REWRITEFORCETOSIZE)
+#define F_REWRITEFORCETOSIZE    V_REWRITEFORCETOSIZE(1U)
+
+#define A_TP_TX_TRC_KEY0 0x20
+
+#define A_TP_RX_TRC_KEY0 0x120
+
+#define A_ULPRX_CTL 0x500
+
+#define S_ROUND_ROBIN    4
+#define V_ROUND_ROBIN(x) ((x) << S_ROUND_ROBIN)
+#define F_ROUND_ROBIN    V_ROUND_ROBIN(1U)
+
+#define A_ULPRX_INT_ENABLE 0x504
+
+#define S_PARERR    0
+#define V_PARERR(x) ((x) << S_PARERR)
+#define F_PARERR    V_PARERR(1U)
+
+#define A_ULPRX_INT_CAUSE 0x508
+
+#define A_ULPRX_ISCSI_LLIMIT 0x50c
+
+#define A_ULPRX_ISCSI_ULIMIT 0x510
+
+#define A_ULPRX_ISCSI_TAGMASK 0x514
+
+#define A_ULPRX_TDDP_LLIMIT 0x51c
+
+#define A_ULPRX_TDDP_ULIMIT 0x520
+
+#define A_ULPRX_STAG_LLIMIT 0x52c
+
+#define A_ULPRX_STAG_ULIMIT 0x530
+
+#define A_ULPRX_RQ_LLIMIT 0x534
+#define A_ULPRX_RQ_LLIMIT 0x534
+
+#define A_ULPRX_RQ_ULIMIT 0x538
+#define A_ULPRX_RQ_ULIMIT 0x538
+
+#define A_ULPRX_PBL_LLIMIT 0x53c
+
+#define A_ULPRX_PBL_ULIMIT 0x540
+#define A_ULPRX_PBL_ULIMIT 0x540
+
+#define A_ULPRX_TDDP_TAGMASK 0x524
+
+#define A_ULPRX_RQ_LLIMIT 0x534
+#define A_ULPRX_RQ_LLIMIT 0x534
+
+#define A_ULPRX_RQ_ULIMIT 0x538
+#define A_ULPRX_RQ_ULIMIT 0x538
+
+#define A_ULPRX_PBL_ULIMIT 0x540
+#define A_ULPRX_PBL_ULIMIT 0x540
+
+#define A_ULPTX_CONFIG 0x580
+
+#define S_CFG_RR_ARB    0
+#define V_CFG_RR_ARB(x) ((x) << S_CFG_RR_ARB)
+#define F_CFG_RR_ARB    V_CFG_RR_ARB(1U)
+
+#define A_ULPTX_INT_ENABLE 0x584
+
+#define S_PBL_BOUND_ERR_CH1    1
+#define V_PBL_BOUND_ERR_CH1(x) ((x) << S_PBL_BOUND_ERR_CH1)
+#define F_PBL_BOUND_ERR_CH1    V_PBL_BOUND_ERR_CH1(1U)
+
+#define S_PBL_BOUND_ERR_CH0    0
+#define V_PBL_BOUND_ERR_CH0(x) ((x) << S_PBL_BOUND_ERR_CH0)
+#define F_PBL_BOUND_ERR_CH0    V_PBL_BOUND_ERR_CH0(1U)
+
+#define A_ULPTX_INT_CAUSE 0x588
+
+#define A_ULPTX_TPT_LLIMIT 0x58c
+
+#define A_ULPTX_TPT_ULIMIT 0x590
+
+#define A_ULPTX_PBL_LLIMIT 0x594
+
+#define A_ULPTX_PBL_ULIMIT 0x598
+
+#define A_ULPTX_DMA_WEIGHT 0x5ac
+
+#define S_D1_WEIGHT    16
+#define M_D1_WEIGHT    0xffff
+#define V_D1_WEIGHT(x) ((x) << S_D1_WEIGHT)
+
+#define S_D0_WEIGHT    0
+#define M_D0_WEIGHT    0xffff
+#define V_D0_WEIGHT(x) ((x) << S_D0_WEIGHT)
+
+#define A_PM1_RX_CFG 0x5c0
+
+#define A_PM1_RX_INT_ENABLE 0x5d8
+
+#define S_ZERO_E_CMD_ERROR    18
+#define V_ZERO_E_CMD_ERROR(x) ((x) << S_ZERO_E_CMD_ERROR)
+#define F_ZERO_E_CMD_ERROR    V_ZERO_E_CMD_ERROR(1U)
+
+#define S_IESPI0_FIFO2X_RX_FRAMING_ERROR    17
+#define V_IESPI0_FIFO2X_RX_FRAMING_ERROR(x) ((x) << S_IESPI0_FIFO2X_RX_FRAMING_ERROR)
+#define F_IESPI0_FIFO2X_RX_FRAMING_ERROR    V_IESPI0_FIFO2X_RX_FRAMING_ERROR(1U)
+
+#define S_IESPI1_FIFO2X_RX_FRAMING_ERROR    16
+#define V_IESPI1_FIFO2X_RX_FRAMING_ERROR(x) ((x) << S_IESPI1_FIFO2X_RX_FRAMING_ERROR)
+#define F_IESPI1_FIFO2X_RX_FRAMING_ERROR    V_IESPI1_FIFO2X_RX_FRAMING_ERROR(1U)
+
+#define S_IESPI0_RX_FRAMING_ERROR    15
+#define V_IESPI0_RX_FRAMING_ERROR(x) ((x) << S_IESPI0_RX_FRAMING_ERROR)
+#define F_IESPI0_RX_FRAMING_ERROR    V_IESPI0_RX_FRAMING_ERROR(1U)
+
+#define S_IESPI1_RX_FRAMING_ERROR    14
+#define V_IESPI1_RX_FRAMING_ERROR(x) ((x) << S_IESPI1_RX_FRAMING_ERROR)
+#define F_IESPI1_RX_FRAMING_ERROR    V_IESPI1_RX_FRAMING_ERROR(1U)
+
+#define S_IESPI0_TX_FRAMING_ERROR    13
+#define V_IESPI0_TX_FRAMING_ERROR(x) ((x) << S_IESPI0_TX_FRAMING_ERROR)
+#define F_IESPI0_TX_FRAMING_ERROR    V_IESPI0_TX_FRAMING_ERROR(1U)
+
+#define S_IESPI1_TX_FRAMING_ERROR    12
+#define V_IESPI1_TX_FRAMING_ERROR(x) ((x) << S_IESPI1_TX_FRAMING_ERROR)
+#define F_IESPI1_TX_FRAMING_ERROR    V_IESPI1_TX_FRAMING_ERROR(1U)
+
+#define S_OCSPI0_RX_FRAMING_ERROR    11
+#define V_OCSPI0_RX_FRAMING_ERROR(x) ((x) << S_OCSPI0_RX_FRAMING_ERROR)
+#define F_OCSPI0_RX_FRAMING_ERROR    V_OCSPI0_RX_FRAMING_ERROR(1U)
+
+#define S_OCSPI1_RX_FRAMING_ERROR    10
+#define V_OCSPI1_RX_FRAMING_ERROR(x) ((x) << S_OCSPI1_RX_FRAMING_ERROR)
+#define F_OCSPI1_RX_FRAMING_ERROR    V_OCSPI1_RX_FRAMING_ERROR(1U)
+
+#define S_OCSPI0_TX_FRAMING_ERROR    9
+#define V_OCSPI0_TX_FRAMING_ERROR(x) ((x) << S_OCSPI0_TX_FRAMING_ERROR)
+#define F_OCSPI0_TX_FRAMING_ERROR    V_OCSPI0_TX_FRAMING_ERROR(1U)
+
+#define S_OCSPI1_TX_FRAMING_ERROR    8
+#define V_OCSPI1_TX_FRAMING_ERROR(x) ((x) << S_OCSPI1_TX_FRAMING_ERROR)
+#define F_OCSPI1_TX_FRAMING_ERROR    V_OCSPI1_TX_FRAMING_ERROR(1U)
+
+#define S_OCSPI0_OFIFO2X_TX_FRAMING_ERROR    7
+#define V_OCSPI0_OFIFO2X_TX_FRAMING_ERROR(x) ((x) << S_OCSPI0_OFIFO2X_TX_FRAMING_ERROR)
+#define F_OCSPI0_OFIFO2X_TX_FRAMING_ERROR    V_OCSPI0_OFIFO2X_TX_FRAMING_ERROR(1U)
+
+#define S_OCSPI1_OFIFO2X_TX_FRAMING_ERROR    6
+#define V_OCSPI1_OFIFO2X_TX_FRAMING_ERROR(x) ((x) << S_OCSPI1_OFIFO2X_TX_FRAMING_ERROR)
+#define F_OCSPI1_OFIFO2X_TX_FRAMING_ERROR    V_OCSPI1_OFIFO2X_TX_FRAMING_ERROR(1U)
+
+#define S_IESPI_PAR_ERROR    3
+#define M_IESPI_PAR_ERROR    0x7
+
+#define V_IESPI_PAR_ERROR(x) ((x) << S_IESPI_PAR_ERROR)
+
+#define S_OCSPI_PAR_ERROR    0
+#define M_OCSPI_PAR_ERROR    0x7
+
+#define V_OCSPI_PAR_ERROR(x) ((x) << S_OCSPI_PAR_ERROR)
+
+#define A_PM1_RX_INT_CAUSE 0x5dc
+
+#define A_PM1_TX_CFG 0x5e0
+
+#define A_PM1_TX_INT_ENABLE 0x5f8
+
+#define S_ZERO_C_CMD_ERROR    18
+#define V_ZERO_C_CMD_ERROR(x) ((x) << S_ZERO_C_CMD_ERROR)
+#define F_ZERO_C_CMD_ERROR    V_ZERO_C_CMD_ERROR(1U)
+
+#define S_ICSPI0_FIFO2X_RX_FRAMING_ERROR    17
+#define V_ICSPI0_FIFO2X_RX_FRAMING_ERROR(x) ((x) << S_ICSPI0_FIFO2X_RX_FRAMING_ERROR)
+#define F_ICSPI0_FIFO2X_RX_FRAMING_ERROR    V_ICSPI0_FIFO2X_RX_FRAMING_ERROR(1U)
+
+#define S_ICSPI1_FIFO2X_RX_FRAMING_ERROR    16
+#define V_ICSPI1_FIFO2X_RX_FRAMING_ERROR(x) ((x) << S_ICSPI1_FIFO2X_RX_FRAMING_ERROR)
+#define F_ICSPI1_FIFO2X_RX_FRAMING_ERROR    V_ICSPI1_FIFO2X_RX_FRAMING_ERROR(1U)
+
+#define S_ICSPI0_RX_FRAMING_ERROR    15
+#define V_ICSPI0_RX_FRAMING_ERROR(x) ((x) << S_ICSPI0_RX_FRAMING_ERROR)
+#define F_ICSPI0_RX_FRAMING_ERROR    V_ICSPI0_RX_FRAMING_ERROR(1U)
+
+#define S_ICSPI1_RX_FRAMING_ERROR    14
+#define V_ICSPI1_RX_FRAMING_ERROR(x) ((x) << S_ICSPI1_RX_FRAMING_ERROR)
+#define F_ICSPI1_RX_FRAMING_ERROR    V_ICSPI1_RX_FRAMING_ERROR(1U)
+
+#define S_ICSPI0_TX_FRAMING_ERROR    13
+#define V_ICSPI0_TX_FRAMING_ERROR(x) ((x) << S_ICSPI0_TX_FRAMING_ERROR)
+#define F_ICSPI0_TX_FRAMING_ERROR    V_ICSPI0_TX_FRAMING_ERROR(1U)
+
+#define S_ICSPI1_TX_FRAMING_ERROR    12
+#define V_ICSPI1_TX_FRAMING_ERROR(x) ((x) << S_ICSPI1_TX_FRAMING_ERROR)
+#define F_ICSPI1_TX_FRAMING_ERROR    V_ICSPI1_TX_FRAMING_ERROR(1U)
+
+#define S_OESPI0_RX_FRAMING_ERROR    11
+#define V_OESPI0_RX_FRAMING_ERROR(x) ((x) << S_OESPI0_RX_FRAMING_ERROR)
+#define F_OESPI0_RX_FRAMING_ERROR    V_OESPI0_RX_FRAMING_ERROR(1U)
+
+#define S_OESPI1_RX_FRAMING_ERROR    10
+#define V_OESPI1_RX_FRAMING_ERROR(x) ((x) << S_OESPI1_RX_FRAMING_ERROR)
+#define F_OESPI1_RX_FRAMING_ERROR    V_OESPI1_RX_FRAMING_ERROR(1U)
+
+#define S_OESPI0_TX_FRAMING_ERROR    9
+#define V_OESPI0_TX_FRAMING_ERROR(x) ((x) << S_OESPI0_TX_FRAMING_ERROR)
+#define F_OESPI0_TX_FRAMING_ERROR    V_OESPI0_TX_FRAMING_ERROR(1U)
+
+#define S_OESPI1_TX_FRAMING_ERROR    8
+#define V_OESPI1_TX_FRAMING_ERROR(x) ((x) << S_OESPI1_TX_FRAMING_ERROR)
+#define F_OESPI1_TX_FRAMING_ERROR    V_OESPI1_TX_FRAMING_ERROR(1U)
+
+#define S_OESPI0_OFIFO2X_TX_FRAMING_ERROR    7
+#define V_OESPI0_OFIFO2X_TX_FRAMING_ERROR(x) ((x) << S_OESPI0_OFIFO2X_TX_FRAMING_ERROR)
+#define F_OESPI0_OFIFO2X_TX_FRAMING_ERROR    V_OESPI0_OFIFO2X_TX_FRAMING_ERROR(1U)
+
+#define S_OESPI1_OFIFO2X_TX_FRAMING_ERROR    6
+#define V_OESPI1_OFIFO2X_TX_FRAMING_ERROR(x) ((x) << S_OESPI1_OFIFO2X_TX_FRAMING_ERROR)
+#define F_OESPI1_OFIFO2X_TX_FRAMING_ERROR    V_OESPI1_OFIFO2X_TX_FRAMING_ERROR(1U)
+
+#define S_ICSPI_PAR_ERROR    3
+#define M_ICSPI_PAR_ERROR    0x7
+
+#define V_ICSPI_PAR_ERROR(x) ((x) << S_ICSPI_PAR_ERROR)
+
+#define S_OESPI_PAR_ERROR    0
+#define M_OESPI_PAR_ERROR    0x7
+
+#define V_OESPI_PAR_ERROR(x) ((x) << S_OESPI_PAR_ERROR)
+
+#define A_PM1_TX_INT_CAUSE 0x5fc
+
+#define A_MPS_CFG 0x600
+
+#define S_TPRXPORTEN    4
+#define V_TPRXPORTEN(x) ((x) << S_TPRXPORTEN)
+#define F_TPRXPORTEN    V_TPRXPORTEN(1U)
+
+#define S_TPTXPORT1EN    3
+#define V_TPTXPORT1EN(x) ((x) << S_TPTXPORT1EN)
+#define F_TPTXPORT1EN    V_TPTXPORT1EN(1U)
+
+#define S_TPTXPORT0EN    2
+#define V_TPTXPORT0EN(x) ((x) << S_TPTXPORT0EN)
+#define F_TPTXPORT0EN    V_TPTXPORT0EN(1U)
+
+#define S_PORT1ACTIVE    1
+#define V_PORT1ACTIVE(x) ((x) << S_PORT1ACTIVE)
+#define F_PORT1ACTIVE    V_PORT1ACTIVE(1U)
+
+#define S_PORT0ACTIVE    0
+#define V_PORT0ACTIVE(x) ((x) << S_PORT0ACTIVE)
+#define F_PORT0ACTIVE    V_PORT0ACTIVE(1U)
+
+#define S_ENFORCEPKT    11
+#define V_ENFORCEPKT(x) ((x) << S_ENFORCEPKT)
+#define F_ENFORCEPKT    V_ENFORCEPKT(1U)
+
+#define A_MPS_INT_ENABLE 0x61c
+
+#define S_MCAPARERRENB    6
+#define M_MCAPARERRENB    0x7
+
+#define V_MCAPARERRENB(x) ((x) << S_MCAPARERRENB)
+
+#define S_RXTPPARERRENB    4
+#define M_RXTPPARERRENB    0x3
+
+#define V_RXTPPARERRENB(x) ((x) << S_RXTPPARERRENB)
+
+#define S_TX1TPPARERRENB    2
+#define M_TX1TPPARERRENB    0x3
+
+#define V_TX1TPPARERRENB(x) ((x) << S_TX1TPPARERRENB)
+
+#define S_TX0TPPARERRENB    0
+#define M_TX0TPPARERRENB    0x3
+
+#define V_TX0TPPARERRENB(x) ((x) << S_TX0TPPARERRENB)
+
+#define A_MPS_INT_CAUSE 0x620
+
+#define S_MCAPARERR    6
+#define M_MCAPARERR    0x7
+
+#define V_MCAPARERR(x) ((x) << S_MCAPARERR)
+
+#define S_RXTPPARERR    4
+#define M_RXTPPARERR    0x3
+
+#define V_RXTPPARERR(x) ((x) << S_RXTPPARERR)
+
+#define S_TX1TPPARERR    2
+#define M_TX1TPPARERR    0x3
+
+#define V_TX1TPPARERR(x) ((x) << S_TX1TPPARERR)
+
+#define S_TX0TPPARERR    0
+#define M_TX0TPPARERR    0x3
+
+#define V_TX0TPPARERR(x) ((x) << S_TX0TPPARERR)
+
+#define A_CPL_SWITCH_CNTRL 0x640
+
+#define A_CPL_INTR_ENABLE 0x650
+
+#define S_CIM_OVFL_ERROR    4
+#define V_CIM_OVFL_ERROR(x) ((x) << S_CIM_OVFL_ERROR)
+#define F_CIM_OVFL_ERROR    V_CIM_OVFL_ERROR(1U)
+
+#define S_TP_FRAMING_ERROR    3
+#define V_TP_FRAMING_ERROR(x) ((x) << S_TP_FRAMING_ERROR)
+#define F_TP_FRAMING_ERROR    V_TP_FRAMING_ERROR(1U)
+
+#define S_SGE_FRAMING_ERROR    2
+#define V_SGE_FRAMING_ERROR(x) ((x) << S_SGE_FRAMING_ERROR)
+#define F_SGE_FRAMING_ERROR    V_SGE_FRAMING_ERROR(1U)
+
+#define S_CIM_FRAMING_ERROR    1
+#define V_CIM_FRAMING_ERROR(x) ((x) << S_CIM_FRAMING_ERROR)
+#define F_CIM_FRAMING_ERROR    V_CIM_FRAMING_ERROR(1U)
+
+#define S_ZERO_SWITCH_ERROR    0
+#define V_ZERO_SWITCH_ERROR(x) ((x) << S_ZERO_SWITCH_ERROR)
+#define F_ZERO_SWITCH_ERROR    V_ZERO_SWITCH_ERROR(1U)
+
+#define A_CPL_INTR_CAUSE 0x654
+
+#define A_CPL_MAP_TBL_DATA 0x65c
+
+#define A_SMB_GLOBAL_TIME_CFG 0x660
+
+#define A_I2C_CFG 0x6a0
+
+#define S_I2C_CLKDIV    0
+#define M_I2C_CLKDIV    0xfff
+#define V_I2C_CLKDIV(x) ((x) << S_I2C_CLKDIV)
+
+#define A_MI1_CFG 0x6b0
+
+#define S_CLKDIV    5
+#define M_CLKDIV    0xff
+#define V_CLKDIV(x) ((x) << S_CLKDIV)
+
+#define S_ST    3
+
+#define M_ST    0x3
+
+#define V_ST(x) ((x) << S_ST)
+
+#define G_ST(x) (((x) >> S_ST) & M_ST)
+
+#define S_PREEN    2
+#define V_PREEN(x) ((x) << S_PREEN)
+#define F_PREEN    V_PREEN(1U)
+
+#define S_MDIINV    1
+#define V_MDIINV(x) ((x) << S_MDIINV)
+#define F_MDIINV    V_MDIINV(1U)
+
+#define S_MDIEN    0
+#define V_MDIEN(x) ((x) << S_MDIEN)
+#define F_MDIEN    V_MDIEN(1U)
+
+#define A_MI1_ADDR 0x6b4
+
+#define S_PHYADDR    5
+#define M_PHYADDR    0x1f
+#define V_PHYADDR(x) ((x) << S_PHYADDR)
+
+#define S_REGADDR    0
+#define M_REGADDR    0x1f
+#define V_REGADDR(x) ((x) << S_REGADDR)
+
+#define A_MI1_DATA 0x6b8
+
+#define A_MI1_OP 0x6bc
+
+#define S_MDI_OP    0
+#define M_MDI_OP    0x3
+#define V_MDI_OP(x) ((x) << S_MDI_OP)
+
+#define A_SF_DATA 0x6d8
+
+#define A_SF_OP 0x6dc
+
+#define S_BYTECNT    1
+#define M_BYTECNT    0x3
+#define V_BYTECNT(x) ((x) << S_BYTECNT)
+
+#define A_PL_INT_ENABLE0 0x6e0
+
+#define S_T3DBG    23
+#define V_T3DBG(x) ((x) << S_T3DBG)
+#define F_T3DBG    V_T3DBG(1U)
+
+#define S_XGMAC0_1    20
+#define V_XGMAC0_1(x) ((x) << S_XGMAC0_1)
+#define F_XGMAC0_1    V_XGMAC0_1(1U)
+
+#define S_XGMAC0_0    19
+#define V_XGMAC0_0(x) ((x) << S_XGMAC0_0)
+#define F_XGMAC0_0    V_XGMAC0_0(1U)
+
+#define S_MC5A    18
+#define V_MC5A(x) ((x) << S_MC5A)
+#define F_MC5A    V_MC5A(1U)
+
+#define S_CPL_SWITCH    12
+#define V_CPL_SWITCH(x) ((x) << S_CPL_SWITCH)
+#define F_CPL_SWITCH    V_CPL_SWITCH(1U)
+
+#define S_MPS0    11
+#define V_MPS0(x) ((x) << S_MPS0)
+#define F_MPS0    V_MPS0(1U)
+
+#define S_PM1_TX    10
+#define V_PM1_TX(x) ((x) << S_PM1_TX)
+#define F_PM1_TX    V_PM1_TX(1U)
+
+#define S_PM1_RX    9
+#define V_PM1_RX(x) ((x) << S_PM1_RX)
+#define F_PM1_RX    V_PM1_RX(1U)
+
+#define S_ULP2_TX    8
+#define V_ULP2_TX(x) ((x) << S_ULP2_TX)
+#define F_ULP2_TX    V_ULP2_TX(1U)
+
+#define S_ULP2_RX    7
+#define V_ULP2_RX(x) ((x) << S_ULP2_RX)
+#define F_ULP2_RX    V_ULP2_RX(1U)
+
+#define S_TP1    6
+#define V_TP1(x) ((x) << S_TP1)
+#define F_TP1    V_TP1(1U)
+
+#define S_CIM    5
+#define V_CIM(x) ((x) << S_CIM)
+#define F_CIM    V_CIM(1U)
+
+#define S_MC7_CM    4
+#define V_MC7_CM(x) ((x) << S_MC7_CM)
+#define F_MC7_CM    V_MC7_CM(1U)
+
+#define S_MC7_PMTX    3
+#define V_MC7_PMTX(x) ((x) << S_MC7_PMTX)
+#define F_MC7_PMTX    V_MC7_PMTX(1U)
+
+#define S_MC7_PMRX    2
+#define V_MC7_PMRX(x) ((x) << S_MC7_PMRX)
+#define F_MC7_PMRX    V_MC7_PMRX(1U)
+
+#define S_PCIM0    1
+#define V_PCIM0(x) ((x) << S_PCIM0)
+#define F_PCIM0    V_PCIM0(1U)
+
+#define S_SGE3    0
+#define V_SGE3(x) ((x) << S_SGE3)
+#define F_SGE3    V_SGE3(1U)
+
+#define A_PL_INT_CAUSE0 0x6e4
+
+#define A_PL_RST 0x6f0
+
+#define S_CRSTWRM    1
+#define V_CRSTWRM(x) ((x) << S_CRSTWRM)
+#define F_CRSTWRM    V_CRSTWRM(1U)
+
+#define A_PL_REV 0x6f4
+
+#define A_PL_CLI 0x6f8
+
+#define A_MC5_DB_CONFIG 0x704
+
+#define S_TMTYPEHI    30
+#define V_TMTYPEHI(x) ((x) << S_TMTYPEHI)
+#define F_TMTYPEHI    V_TMTYPEHI(1U)
+
+#define S_TMPARTSIZE    28
+#define M_TMPARTSIZE    0x3
+#define V_TMPARTSIZE(x) ((x) << S_TMPARTSIZE)
+#define G_TMPARTSIZE(x) (((x) >> S_TMPARTSIZE) & M_TMPARTSIZE)
+
+#define S_TMTYPE    26
+#define M_TMTYPE    0x3
+#define V_TMTYPE(x) ((x) << S_TMTYPE)
+#define G_TMTYPE(x) (((x) >> S_TMTYPE) & M_TMTYPE)
+
+#define S_COMPEN    17
+#define V_COMPEN(x) ((x) << S_COMPEN)
+#define F_COMPEN    V_COMPEN(1U)
+
+#define S_PRTYEN    6
+#define V_PRTYEN(x) ((x) << S_PRTYEN)
+#define F_PRTYEN    V_PRTYEN(1U)
+
+#define S_MBUSEN    5
+#define V_MBUSEN(x) ((x) << S_MBUSEN)
+#define F_MBUSEN    V_MBUSEN(1U)
+
+#define S_DBGIEN    4
+#define V_DBGIEN(x) ((x) << S_DBGIEN)
+#define F_DBGIEN    V_DBGIEN(1U)
+
+#define S_TMRDY    2
+#define V_TMRDY(x) ((x) << S_TMRDY)
+#define F_TMRDY    V_TMRDY(1U)
+
+#define S_TMRST    1
+#define V_TMRST(x) ((x) << S_TMRST)
+#define F_TMRST    V_TMRST(1U)
+
+#define S_TMMODE    0
+#define V_TMMODE(x) ((x) << S_TMMODE)
+#define F_TMMODE    V_TMMODE(1U)
+
+#define F_TMMODE    V_TMMODE(1U)
+
+#define A_MC5_DB_ROUTING_TABLE_INDEX 0x70c
+
+#define A_MC5_DB_FILTER_TABLE 0x710
+
+#define A_MC5_DB_SERVER_INDEX 0x714
+
+#define A_MC5_DB_RSP_LATENCY 0x720
+
+#define S_RDLAT    16
+#define M_RDLAT    0x1f
+#define V_RDLAT(x) ((x) << S_RDLAT)
+
+#define S_LRNLAT    8
+#define M_LRNLAT    0x1f
+#define V_LRNLAT(x) ((x) << S_LRNLAT)
+
+#define S_SRCHLAT    0
+#define M_SRCHLAT    0x1f
+#define V_SRCHLAT(x) ((x) << S_SRCHLAT)
+
+#define A_MC5_DB_PART_ID_INDEX 0x72c
+
+#define A_MC5_DB_INT_ENABLE 0x740
+
+#define S_DELACTEMPTY    18
+#define V_DELACTEMPTY(x) ((x) << S_DELACTEMPTY)
+#define F_DELACTEMPTY    V_DELACTEMPTY(1U)
+
+#define S_DISPQPARERR    17
+#define V_DISPQPARERR(x) ((x) << S_DISPQPARERR)
+#define F_DISPQPARERR    V_DISPQPARERR(1U)
+
+#define S_REQQPARERR    16
+#define V_REQQPARERR(x) ((x) << S_REQQPARERR)
+#define F_REQQPARERR    V_REQQPARERR(1U)
+
+#define S_UNKNOWNCMD    15
+#define V_UNKNOWNCMD(x) ((x) << S_UNKNOWNCMD)
+#define F_UNKNOWNCMD    V_UNKNOWNCMD(1U)
+
+#define S_NFASRCHFAIL    8
+#define V_NFASRCHFAIL(x) ((x) << S_NFASRCHFAIL)
+#define F_NFASRCHFAIL    V_NFASRCHFAIL(1U)
+
+#define S_ACTRGNFULL    7
+#define V_ACTRGNFULL(x) ((x) << S_ACTRGNFULL)
+#define F_ACTRGNFULL    V_ACTRGNFULL(1U)
+
+#define S_PARITYERR    6
+#define V_PARITYERR(x) ((x) << S_PARITYERR)
+#define F_PARITYERR    V_PARITYERR(1U)
+
+#define A_MC5_DB_INT_CAUSE 0x744
+
+#define A_MC5_DB_DBGI_CONFIG 0x774
+
+#define A_MC5_DB_DBGI_REQ_CMD 0x778
+
+#define A_MC5_DB_DBGI_REQ_ADDR0 0x77c
+
+#define A_MC5_DB_DBGI_REQ_ADDR1 0x780
+
+#define A_MC5_DB_DBGI_REQ_ADDR2 0x784
+
+#define A_MC5_DB_DBGI_REQ_DATA0 0x788
+
+#define A_MC5_DB_DBGI_REQ_DATA1 0x78c
+
+#define A_MC5_DB_DBGI_REQ_DATA2 0x790
+
+#define A_MC5_DB_DBGI_RSP_STATUS 0x7b0
+
+#define S_DBGIRSPVALID    0
+#define V_DBGIRSPVALID(x) ((x) << S_DBGIRSPVALID)
+#define F_DBGIRSPVALID    V_DBGIRSPVALID(1U)
+
+#define A_MC5_DB_DBGI_RSP_DATA0 0x7b4
+
+#define A_MC5_DB_DBGI_RSP_DATA1 0x7b8
+
+#define A_MC5_DB_DBGI_RSP_DATA2 0x7bc
+
+#define A_MC5_DB_POPEN_DATA_WR_CMD 0x7cc
+
+#define A_MC5_DB_POPEN_MASK_WR_CMD 0x7d0
+
+#define A_MC5_DB_AOPEN_SRCH_CMD 0x7d4
+
+#define A_MC5_DB_AOPEN_LRN_CMD 0x7d8
+
+#define A_MC5_DB_SYN_SRCH_CMD 0x7dc
+
+#define A_MC5_DB_SYN_LRN_CMD 0x7e0
+
+#define A_MC5_DB_ACK_SRCH_CMD 0x7e4
+
+#define A_MC5_DB_ACK_LRN_CMD 0x7e8
+
+#define A_MC5_DB_ILOOKUP_CMD 0x7ec
+
+#define A_MC5_DB_ELOOKUP_CMD 0x7f0
+
+#define A_MC5_DB_DATA_WRITE_CMD 0x7f4
+
+#define A_MC5_DB_DATA_READ_CMD 0x7f8
+
+#define XGMAC0_0_BASE_ADDR 0x800
+
+#define A_XGM_TX_CTRL 0x800
+
+#define S_TXEN    0
+#define V_TXEN(x) ((x) << S_TXEN)
+#define F_TXEN    V_TXEN(1U)
+
+#define A_XGM_TX_CFG 0x804
+
+#define S_TXPAUSEEN    0
+#define V_TXPAUSEEN(x) ((x) << S_TXPAUSEEN)
+#define F_TXPAUSEEN    V_TXPAUSEEN(1U)
+
+#define A_XGM_RX_CTRL 0x80c
+
+#define S_RXEN    0
+#define V_RXEN(x) ((x) << S_RXEN)
+#define F_RXEN    V_RXEN(1U)
+
+#define A_XGM_RX_CFG 0x810
+
+#define S_DISPAUSEFRAMES    9
+#define V_DISPAUSEFRAMES(x) ((x) << S_DISPAUSEFRAMES)
+#define F_DISPAUSEFRAMES    V_DISPAUSEFRAMES(1U)
+
+#define S_EN1536BFRAMES    8
+#define V_EN1536BFRAMES(x) ((x) << S_EN1536BFRAMES)
+#define F_EN1536BFRAMES    V_EN1536BFRAMES(1U)
+
+#define S_ENJUMBO    7
+#define V_ENJUMBO(x) ((x) << S_ENJUMBO)
+#define F_ENJUMBO    V_ENJUMBO(1U)
+
+#define S_RMFCS    6
+#define V_RMFCS(x) ((x) << S_RMFCS)
+#define F_RMFCS    V_RMFCS(1U)
+
+#define S_ENHASHMCAST    2
+#define V_ENHASHMCAST(x) ((x) << S_ENHASHMCAST)
+#define F_ENHASHMCAST    V_ENHASHMCAST(1U)
+
+#define S_COPYALLFRAMES    0
+#define V_COPYALLFRAMES(x) ((x) << S_COPYALLFRAMES)
+#define F_COPYALLFRAMES    V_COPYALLFRAMES(1U)
+
+#define A_XGM_RX_HASH_LOW 0x814
+
+#define A_XGM_RX_HASH_HIGH 0x818
+
+#define A_XGM_RX_EXACT_MATCH_LOW_1 0x81c
+
+#define A_XGM_RX_EXACT_MATCH_HIGH_1 0x820
+
+#define A_XGM_RX_EXACT_MATCH_LOW_2 0x824
+
+#define A_XGM_RX_EXACT_MATCH_LOW_3 0x82c
+
+#define A_XGM_RX_EXACT_MATCH_LOW_4 0x834
+
+#define A_XGM_RX_EXACT_MATCH_LOW_5 0x83c
+
+#define A_XGM_RX_EXACT_MATCH_LOW_6 0x844
+
+#define A_XGM_RX_EXACT_MATCH_LOW_7 0x84c
+
+#define A_XGM_RX_EXACT_MATCH_LOW_8 0x854
+
+#define A_XGM_STAT_CTRL 0x880
+
+#define S_CLRSTATS    2
+#define V_CLRSTATS(x) ((x) << S_CLRSTATS)
+#define F_CLRSTATS    V_CLRSTATS(1U)
+
+#define A_XGM_RXFIFO_CFG 0x884
+
+#define S_RXFIFOPAUSEHWM    17
+#define M_RXFIFOPAUSEHWM    0xfff
+
+#define V_RXFIFOPAUSEHWM(x) ((x) << S_RXFIFOPAUSEHWM)
+
+#define G_RXFIFOPAUSEHWM(x) (((x) >> S_RXFIFOPAUSEHWM) & M_RXFIFOPAUSEHWM)
+
+#define S_RXFIFOPAUSELWM    5
+#define M_RXFIFOPAUSELWM    0xfff
+
+#define V_RXFIFOPAUSELWM(x) ((x) << S_RXFIFOPAUSELWM)
+
+#define G_RXFIFOPAUSELWM(x) (((x) >> S_RXFIFOPAUSELWM) & M_RXFIFOPAUSELWM)
+
+#define S_RXSTRFRWRD    1
+#define V_RXSTRFRWRD(x) ((x) << S_RXSTRFRWRD)
+#define F_RXSTRFRWRD    V_RXSTRFRWRD(1U)
+
+#define S_DISERRFRAMES    0
+#define V_DISERRFRAMES(x) ((x) << S_DISERRFRAMES)
+#define F_DISERRFRAMES    V_DISERRFRAMES(1U)
+
+#define A_XGM_TXFIFO_CFG 0x888
+
+#define S_TXFIFOTHRESH    4
+#define M_TXFIFOTHRESH    0x1ff
+
+#define V_TXFIFOTHRESH(x) ((x) << S_TXFIFOTHRESH)
+
+#define A_XGM_SERDES_CTRL 0x890
+#define A_XGM_SERDES_CTRL0 0x8e0
+
+#define S_SERDESRESET_    24
+#define V_SERDESRESET_(x) ((x) << S_SERDESRESET_)
+#define F_SERDESRESET_    V_SERDESRESET_(1U)
+
+#define S_RXENABLE    4
+#define V_RXENABLE(x) ((x) << S_RXENABLE)
+#define F_RXENABLE    V_RXENABLE(1U)
+
+#define S_TXENABLE    3
+#define V_TXENABLE(x) ((x) << S_TXENABLE)
+#define F_TXENABLE    V_TXENABLE(1U)
+
+#define A_XGM_PAUSE_TIMER 0x890
+
+#define A_XGM_RGMII_IMP 0x89c
+
+#define S_XGM_IMPSETUPDATE    6
+#define V_XGM_IMPSETUPDATE(x) ((x) << S_XGM_IMPSETUPDATE)
+#define F_XGM_IMPSETUPDATE    V_XGM_IMPSETUPDATE(1U)
+
+#define S_RGMIIIMPPD    3
+#define M_RGMIIIMPPD    0x7
+#define V_RGMIIIMPPD(x) ((x) << S_RGMIIIMPPD)
+
+#define S_RGMIIIMPPU    0
+#define M_RGMIIIMPPU    0x7
+#define V_RGMIIIMPPU(x) ((x) << S_RGMIIIMPPU)
+
+#define S_CALRESET    8
+#define V_CALRESET(x) ((x) << S_CALRESET)
+#define F_CALRESET    V_CALRESET(1U)
+
+#define S_CALUPDATE    7
+#define V_CALUPDATE(x) ((x) << S_CALUPDATE)
+#define F_CALUPDATE    V_CALUPDATE(1U)
+
+#define A_XGM_XAUI_IMP 0x8a0
+
+#define S_CALBUSY    31
+#define V_CALBUSY(x) ((x) << S_CALBUSY)
+#define F_CALBUSY    V_CALBUSY(1U)
+
+#define S_XGM_CALFAULT    29
+#define V_XGM_CALFAULT(x) ((x) << S_XGM_CALFAULT)
+#define F_XGM_CALFAULT    V_XGM_CALFAULT(1U)
+
+#define S_CALIMP    24
+#define M_CALIMP    0x1f
+#define V_CALIMP(x) ((x) << S_CALIMP)
+#define G_CALIMP(x) (((x) >> S_CALIMP) & M_CALIMP)
+
+#define S_XAUIIMP    0
+#define M_XAUIIMP    0x7
+#define V_XAUIIMP(x) ((x) << S_XAUIIMP)
+
+#define A_XGM_RX_MAX_PKT_SIZE 0x8a8
+#define A_XGM_RX_MAX_PKT_SIZE_ERR_CNT 0x9a4
+
+#define A_XGM_RESET_CTRL 0x8ac
+
+#define S_XG2G_RESET_    3
+#define V_XG2G_RESET_(x) ((x) << S_XG2G_RESET_)
+#define F_XG2G_RESET_    V_XG2G_RESET_(1U)
+
+#define S_RGMII_RESET_    2
+#define V_RGMII_RESET_(x) ((x) << S_RGMII_RESET_)
+#define F_RGMII_RESET_    V_RGMII_RESET_(1U)
+
+#define S_PCS_RESET_    1
+#define V_PCS_RESET_(x) ((x) << S_PCS_RESET_)
+#define F_PCS_RESET_    V_PCS_RESET_(1U)
+
+#define S_MAC_RESET_    0
+#define V_MAC_RESET_(x) ((x) << S_MAC_RESET_)
+#define F_MAC_RESET_    V_MAC_RESET_(1U)
+
+#define A_XGM_PORT_CFG 0x8b8
+
+#define S_CLKDIVRESET_    3
+#define V_CLKDIVRESET_(x) ((x) << S_CLKDIVRESET_)
+#define F_CLKDIVRESET_    V_CLKDIVRESET_(1U)
+
+#define S_PORTSPEED    1
+#define M_PORTSPEED    0x3
+
+#define V_PORTSPEED(x) ((x) << S_PORTSPEED)
+
+#define S_ENRGMII    0
+#define V_ENRGMII(x) ((x) << S_ENRGMII)
+#define F_ENRGMII    V_ENRGMII(1U)
+
+#define A_XGM_INT_ENABLE 0x8d4
+
+#define S_TXFIFO_PRTY_ERR    17
+#define M_TXFIFO_PRTY_ERR    0x7
+
+#define V_TXFIFO_PRTY_ERR(x) ((x) << S_TXFIFO_PRTY_ERR)
+
+#define S_RXFIFO_PRTY_ERR    14
+#define M_RXFIFO_PRTY_ERR    0x7
+
+#define V_RXFIFO_PRTY_ERR(x) ((x) << S_RXFIFO_PRTY_ERR)
+
+#define S_TXFIFO_UNDERRUN    13
+#define V_TXFIFO_UNDERRUN(x) ((x) << S_TXFIFO_UNDERRUN)
+#define F_TXFIFO_UNDERRUN    V_TXFIFO_UNDERRUN(1U)
+
+#define S_RXFIFO_OVERFLOW    12
+#define V_RXFIFO_OVERFLOW(x) ((x) << S_RXFIFO_OVERFLOW)
+#define F_RXFIFO_OVERFLOW    V_RXFIFO_OVERFLOW(1U)
+
+#define S_SERDES_LOS    4
+#define M_SERDES_LOS    0xf
+
+#define V_SERDES_LOS(x) ((x) << S_SERDES_LOS)
+
+#define S_XAUIPCSCTCERR    3
+#define V_XAUIPCSCTCERR(x) ((x) << S_XAUIPCSCTCERR)
+#define F_XAUIPCSCTCERR    V_XAUIPCSCTCERR(1U)
+
+#define S_XAUIPCSALIGNCHANGE    2
+#define V_XAUIPCSALIGNCHANGE(x) ((x) << S_XAUIPCSALIGNCHANGE)
+#define F_XAUIPCSALIGNCHANGE    V_XAUIPCSALIGNCHANGE(1U)
+
+#define A_XGM_INT_CAUSE 0x8d8
+
+#define A_XGM_XAUI_ACT_CTRL 0x8dc
+
+#define S_TXACTENABLE    1
+#define V_TXACTENABLE(x) ((x) << S_TXACTENABLE)
+#define F_TXACTENABLE    V_TXACTENABLE(1U)
+
+#define A_XGM_SERDES_CTRL0 0x8e0
+
+#define S_RESET3    23
+#define V_RESET3(x) ((x) << S_RESET3)
+#define F_RESET3    V_RESET3(1U)
+
+#define S_RESET2    22
+#define V_RESET2(x) ((x) << S_RESET2)
+#define F_RESET2    V_RESET2(1U)
+
+#define S_RESET1    21
+#define V_RESET1(x) ((x) << S_RESET1)
+#define F_RESET1    V_RESET1(1U)
+
+#define S_RESET0    20
+#define V_RESET0(x) ((x) << S_RESET0)
+#define F_RESET0    V_RESET0(1U)
+
+#define S_PWRDN3    19
+#define V_PWRDN3(x) ((x) << S_PWRDN3)
+#define F_PWRDN3    V_PWRDN3(1U)
+
+#define S_PWRDN2    18
+#define V_PWRDN2(x) ((x) << S_PWRDN2)
+#define F_PWRDN2    V_PWRDN2(1U)
+
+#define S_PWRDN1    17
+#define V_PWRDN1(x) ((x) << S_PWRDN1)
+#define F_PWRDN1    V_PWRDN1(1U)
+
+#define S_PWRDN0    16
+#define V_PWRDN0(x) ((x) << S_PWRDN0)
+#define F_PWRDN0    V_PWRDN0(1U)
+
+#define S_RESETPLL23    15
+#define V_RESETPLL23(x) ((x) << S_RESETPLL23)
+#define F_RESETPLL23    V_RESETPLL23(1U)
+
+#define S_RESETPLL01    14
+#define V_RESETPLL01(x) ((x) << S_RESETPLL01)
+#define F_RESETPLL01    V_RESETPLL01(1U)
+
+#define A_XGM_SERDES_STAT0 0x8f0
+
+#define S_LOWSIG0    0
+#define V_LOWSIG0(x) ((x) << S_LOWSIG0)
+#define F_LOWSIG0    V_LOWSIG0(1U)
+
+#define A_XGM_SERDES_STAT3 0x8fc
+
+#define A_XGM_STAT_TX_BYTE_LOW 0x900
+
+#define A_XGM_STAT_TX_BYTE_HIGH 0x904
+
+#define A_XGM_STAT_TX_FRAME_LOW 0x908
+
+#define A_XGM_STAT_TX_FRAME_HIGH 0x90c
+
+#define A_XGM_STAT_TX_BCAST 0x910
+
+#define A_XGM_STAT_TX_MCAST 0x914
+
+#define A_XGM_STAT_TX_PAUSE 0x918
+
+#define A_XGM_STAT_TX_64B_FRAMES 0x91c
+
+#define A_XGM_STAT_TX_65_127B_FRAMES 0x920
+
+#define A_XGM_STAT_TX_128_255B_FRAMES 0x924
+
+#define A_XGM_STAT_TX_256_511B_FRAMES 0x928
+
+#define A_XGM_STAT_TX_512_1023B_FRAMES 0x92c
+
+#define A_XGM_STAT_TX_1024_1518B_FRAMES 0x930
+
+#define A_XGM_STAT_TX_1519_MAXB_FRAMES 0x934
+
+#define A_XGM_STAT_TX_ERR_FRAMES 0x938
+
+#define A_XGM_STAT_RX_BYTES_LOW 0x93c
+
+#define A_XGM_STAT_RX_BYTES_HIGH 0x940
+
+#define A_XGM_STAT_RX_FRAMES_LOW 0x944
+
+#define A_XGM_STAT_RX_FRAMES_HIGH 0x948
+
+#define A_XGM_STAT_RX_BCAST_FRAMES 0x94c
+
+#define A_XGM_STAT_RX_MCAST_FRAMES 0x950
+
+#define A_XGM_STAT_RX_PAUSE_FRAMES 0x954
+
+#define A_XGM_STAT_RX_64B_FRAMES 0x958
+
+#define A_XGM_STAT_RX_65_127B_FRAMES 0x95c
+
+#define A_XGM_STAT_RX_128_255B_FRAMES 0x960
+
+#define A_XGM_STAT_RX_256_511B_FRAMES 0x964
+
+#define A_XGM_STAT_RX_512_1023B_FRAMES 0x968
+
+#define A_XGM_STAT_RX_1024_1518B_FRAMES 0x96c
+
+#define A_XGM_STAT_RX_1519_MAXB_FRAMES 0x970
+
+#define A_XGM_STAT_RX_SHORT_FRAMES 0x974
+
+#define A_XGM_STAT_RX_OVERSIZE_FRAMES 0x978
+
+#define A_XGM_STAT_RX_JABBER_FRAMES 0x97c
+
+#define A_XGM_STAT_RX_CRC_ERR_FRAMES 0x980
+
+#define A_XGM_STAT_RX_LENGTH_ERR_FRAMES 0x984
+
+#define A_XGM_STAT_RX_SYM_CODE_ERR_FRAMES 0x988
+
+#define A_XGM_SERDES_STATUS0 0x98c
+
+#define A_XGM_SERDES_STATUS1 0x990
+
+#define S_CMULOCK    31
+#define V_CMULOCK(x) ((x) << S_CMULOCK)
+#define F_CMULOCK    V_CMULOCK(1U)
+
+#define A_XGM_RX_MAX_PKT_SIZE_ERR_CNT 0x9a4
+
+#define A_XGM_RX_SPI4_SOP_EOP_CNT 0x9ac
+
+#define XGMAC0_1_BASE_ADDR 0xa00
diff --git a/drivers/net/cxgb3/sge.c b/drivers/net/cxgb3/sge.c
new file mode 100644
index 00000000000..6c77f4bab62
--- /dev/null
+++ b/drivers/net/cxgb3/sge.c
@@ -0,0 +1,2702 @@
+/*
+ * This file is part of the Chelsio T3 Ethernet driver.
+ *
+ * Copyright (C) 2005-2006 Chelsio Communications.  All rights reserved.
+ *
+ * This program is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the LICENSE file included in this
+ * release for licensing terms and conditions.
+ */
+
+#include <linux/skbuff.h>
+#include <linux/netdevice.h>
+#include <linux/etherdevice.h>
+#include <linux/if_vlan.h>
+#include <linux/ip.h>
+#include <linux/tcp.h>
+#include <linux/dma-mapping.h>
+#include "common.h"
+#include "regs.h"
+#include "sge_defs.h"
+#include "t3_cpl.h"
+#include "firmware_exports.h"
+
+#define USE_GTS 0
+
+#define SGE_RX_SM_BUF_SIZE 1536
+#define SGE_RX_COPY_THRES  256
+
+# define SGE_RX_DROP_THRES 16
+
+/*
+ * Period of the Tx buffer reclaim timer.  This timer does not need to run
+ * frequently as Tx buffers are usually reclaimed by new Tx packets.
+ */
+#define TX_RECLAIM_PERIOD (HZ / 4)
+
+/* WR size in bytes */
+#define WR_LEN (WR_FLITS * 8)
+
+/*
+ * Types of Tx queues in each queue set.  Order here matters, do not change.
+ */
+enum { TXQ_ETH, TXQ_OFLD, TXQ_CTRL };
+
+/* Values for sge_txq.flags */
+enum {
+	TXQ_RUNNING = 1 << 0,	/* fetch engine is running */
+	TXQ_LAST_PKT_DB = 1 << 1,	/* last packet rang the doorbell */
+};
+
+struct tx_desc {
+	u64 flit[TX_DESC_FLITS];
+};
+
+struct rx_desc {
+	__be32 addr_lo;
+	__be32 len_gen;
+	__be32 gen2;
+	__be32 addr_hi;
+};
+
+struct tx_sw_desc {		/* SW state per Tx descriptor */
+	struct sk_buff *skb;
+};
+
+struct rx_sw_desc {		/* SW state per Rx descriptor */
+	struct sk_buff *skb;
+	 DECLARE_PCI_UNMAP_ADDR(dma_addr);
+};
+
+struct rsp_desc {		/* response queue descriptor */
+	struct rss_header rss_hdr;
+	__be32 flags;
+	__be32 len_cq;
+	u8 imm_data[47];
+	u8 intr_gen;
+};
+
+struct unmap_info {		/* packet unmapping info, overlays skb->cb */
+	int sflit;		/* start flit of first SGL entry in Tx descriptor */
+	u16 fragidx;		/* first page fragment in current Tx descriptor */
+	u16 addr_idx;		/* buffer index of first SGL entry in descriptor */
+	u32 len;		/* mapped length of skb main body */
+};
+
+/*
+ * Maps a number of flits to the number of Tx descriptors that can hold them.
+ * The formula is
+ *
+ * desc = 1 + (flits - 2) / (WR_FLITS - 1).
+ *
+ * HW allows up to 4 descriptors to be combined into a WR.
+ */
+static u8 flit_desc_map[] = {
+	0,
+#if SGE_NUM_GENBITS == 1
+	1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
+	2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
+	3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3,
+	4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4
+#elif SGE_NUM_GENBITS == 2
+	1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
+	2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
+	3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3,
+	4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4,
+#else
+# error "SGE_NUM_GENBITS must be 1 or 2"
+#endif
+};
+
+static inline struct sge_qset *fl_to_qset(const struct sge_fl *q, int qidx)
+{
+	return container_of(q, struct sge_qset, fl[qidx]);
+}
+
+static inline struct sge_qset *rspq_to_qset(const struct sge_rspq *q)
+{
+	return container_of(q, struct sge_qset, rspq);
+}
+
+static inline struct sge_qset *txq_to_qset(const struct sge_txq *q, int qidx)
+{
+	return container_of(q, struct sge_qset, txq[qidx]);
+}
+
+/**
+ *	refill_rspq - replenish an SGE response queue
+ *	@adapter: the adapter
+ *	@q: the response queue to replenish
+ *	@credits: how many new responses to make available
+ *
+ *	Replenishes a response queue by making the supplied number of responses
+ *	available to HW.
+ */
+static inline void refill_rspq(struct adapter *adapter,
+			       const struct sge_rspq *q, unsigned int credits)
+{
+	t3_write_reg(adapter, A_SG_RSPQ_CREDIT_RETURN,
+		     V_RSPQ(q->cntxt_id) | V_CREDITS(credits));
+}
+
+/**
+ *	need_skb_unmap - does the platform need unmapping of sk_buffs?
+ *
+ *	Returns true if the platfrom needs sk_buff unmapping.  The compiler
+ *	optimizes away unecessary code if this returns true.
+ */
+static inline int need_skb_unmap(void)
+{
+	/*
+	 * This structure is used to tell if the platfrom needs buffer
+	 * unmapping by checking if DECLARE_PCI_UNMAP_ADDR defines anything.
+	 */
+	struct dummy {
+		DECLARE_PCI_UNMAP_ADDR(addr);
+	};
+
+	return sizeof(struct dummy) != 0;
+}
+
+/**
+ *	unmap_skb - unmap a packet main body and its page fragments
+ *	@skb: the packet
+ *	@q: the Tx queue containing Tx descriptors for the packet
+ *	@cidx: index of Tx descriptor
+ *	@pdev: the PCI device
+ *
+ *	Unmap the main body of an sk_buff and its page fragments, if any.
+ *	Because of the fairly complicated structure of our SGLs and the desire
+ *	to conserve space for metadata, we keep the information necessary to
+ *	unmap an sk_buff partly in the sk_buff itself (in its cb), and partly
+ *	in the Tx descriptors (the physical addresses of the various data
+ *	buffers).  The send functions initialize the state in skb->cb so we
+ *	can unmap the buffers held in the first Tx descriptor here, and we
+ *	have enough information at this point to update the state for the next
+ *	Tx descriptor.
+ */
+static inline void unmap_skb(struct sk_buff *skb, struct sge_txq *q,
+			     unsigned int cidx, struct pci_dev *pdev)
+{
+	const struct sg_ent *sgp;
+	struct unmap_info *ui = (struct unmap_info *)skb->cb;
+	int nfrags, frag_idx, curflit, j = ui->addr_idx;
+
+	sgp = (struct sg_ent *)&q->desc[cidx].flit[ui->sflit];
+
+	if (ui->len) {
+		pci_unmap_single(pdev, be64_to_cpu(sgp->addr[0]), ui->len,
+				 PCI_DMA_TODEVICE);
+		ui->len = 0;	/* so we know for next descriptor for this skb */
+		j = 1;
+	}
+
+	frag_idx = ui->fragidx;
+	curflit = ui->sflit + 1 + j;
+	nfrags = skb_shinfo(skb)->nr_frags;
+
+	while (frag_idx < nfrags && curflit < WR_FLITS) {
+		pci_unmap_page(pdev, be64_to_cpu(sgp->addr[j]),
+			       skb_shinfo(skb)->frags[frag_idx].size,
+			       PCI_DMA_TODEVICE);
+		j ^= 1;
+		if (j == 0) {
+			sgp++;
+			curflit++;
+		}
+		curflit++;
+		frag_idx++;
+	}
+
+	if (frag_idx < nfrags) {	/* SGL continues into next Tx descriptor */
+		ui->fragidx = frag_idx;
+		ui->addr_idx = j;
+		ui->sflit = curflit - WR_FLITS - j;	/* sflit can be -1 */
+	}
+}
+
+/**
+ *	free_tx_desc - reclaims Tx descriptors and their buffers
+ *	@adapter: the adapter
+ *	@q: the Tx queue to reclaim descriptors from
+ *	@n: the number of descriptors to reclaim
+ *
+ *	Reclaims Tx descriptors from an SGE Tx queue and frees the associated
+ *	Tx buffers.  Called with the Tx queue lock held.
+ */
+static void free_tx_desc(struct adapter *adapter, struct sge_txq *q,
+			 unsigned int n)
+{
+	struct tx_sw_desc *d;
+	struct pci_dev *pdev = adapter->pdev;
+	unsigned int cidx = q->cidx;
+
+	d = &q->sdesc[cidx];
+	while (n--) {
+		if (d->skb) {	/* an SGL is present */
+			if (need_skb_unmap())
+				unmap_skb(d->skb, q, cidx, pdev);
+			if (d->skb->priority == cidx)
+				kfree_skb(d->skb);
+		}
+		++d;
+		if (++cidx == q->size) {
+			cidx = 0;
+			d = q->sdesc;
+		}
+	}
+	q->cidx = cidx;
+}
+
+/**
+ *	reclaim_completed_tx - reclaims completed Tx descriptors
+ *	@adapter: the adapter
+ *	@q: the Tx queue to reclaim completed descriptors from
+ *
+ *	Reclaims Tx descriptors that the SGE has indicated it has processed,
+ *	and frees the associated buffers if possible.  Called with the Tx
+ *	queue's lock held.
+ */
+static inline void reclaim_completed_tx(struct adapter *adapter,
+					struct sge_txq *q)
+{
+	unsigned int reclaim = q->processed - q->cleaned;
+
+	if (reclaim) {
+		free_tx_desc(adapter, q, reclaim);
+		q->cleaned += reclaim;
+		q->in_use -= reclaim;
+	}
+}
+
+/**
+ *	should_restart_tx - are there enough resources to restart a Tx queue?
+ *	@q: the Tx queue
+ *
+ *	Checks if there are enough descriptors to restart a suspended Tx queue.
+ */
+static inline int should_restart_tx(const struct sge_txq *q)
+{
+	unsigned int r = q->processed - q->cleaned;
+
+	return q->in_use - r < (q->size >> 1);
+}
+
+/**
+ *	free_rx_bufs - free the Rx buffers on an SGE free list
+ *	@pdev: the PCI device associated with the adapter
+ *	@rxq: the SGE free list to clean up
+ *
+ *	Release the buffers on an SGE free-buffer Rx queue.  HW fetching from
+ *	this queue should be stopped before calling this function.
+ */
+static void free_rx_bufs(struct pci_dev *pdev, struct sge_fl *q)
+{
+	unsigned int cidx = q->cidx;
+
+	while (q->credits--) {
+		struct rx_sw_desc *d = &q->sdesc[cidx];
+
+		pci_unmap_single(pdev, pci_unmap_addr(d, dma_addr),
+				 q->buf_size, PCI_DMA_FROMDEVICE);
+		kfree_skb(d->skb);
+		d->skb = NULL;
+		if (++cidx == q->size)
+			cidx = 0;
+	}
+}
+
+/**
+ *	add_one_rx_buf - add a packet buffer to a free-buffer list
+ *	@skb: the buffer to add
+ *	@len: the buffer length
+ *	@d: the HW Rx descriptor to write
+ *	@sd: the SW Rx descriptor to write
+ *	@gen: the generation bit value
+ *	@pdev: the PCI device associated with the adapter
+ *
+ *	Add a buffer of the given length to the supplied HW and SW Rx
+ *	descriptors.
+ */
+static inline void add_one_rx_buf(struct sk_buff *skb, unsigned int len,
+				  struct rx_desc *d, struct rx_sw_desc *sd,
+				  unsigned int gen, struct pci_dev *pdev)
+{
+	dma_addr_t mapping;
+
+	sd->skb = skb;
+	mapping = pci_map_single(pdev, skb->data, len, PCI_DMA_FROMDEVICE);
+	pci_unmap_addr_set(sd, dma_addr, mapping);
+
+	d->addr_lo = cpu_to_be32(mapping);
+	d->addr_hi = cpu_to_be32((u64) mapping >> 32);
+	wmb();
+	d->len_gen = cpu_to_be32(V_FLD_GEN1(gen));
+	d->gen2 = cpu_to_be32(V_FLD_GEN2(gen));
+}
+
+/**
+ *	refill_fl - refill an SGE free-buffer list
+ *	@adapter: the adapter
+ *	@q: the free-list to refill
+ *	@n: the number of new buffers to allocate
+ *	@gfp: the gfp flags for allocating new buffers
+ *
+ *	(Re)populate an SGE free-buffer list with up to @n new packet buffers,
+ *	allocated with the supplied gfp flags.  The caller must assure that
+ *	@n does not exceed the queue's capacity.
+ */
+static void refill_fl(struct adapter *adap, struct sge_fl *q, int n, gfp_t gfp)
+{
+	struct rx_sw_desc *sd = &q->sdesc[q->pidx];
+	struct rx_desc *d = &q->desc[q->pidx];
+
+	while (n--) {
+		struct sk_buff *skb = alloc_skb(q->buf_size, gfp);
+
+		if (!skb)
+			break;
+
+		add_one_rx_buf(skb, q->buf_size, d, sd, q->gen, adap->pdev);
+		d++;
+		sd++;
+		if (++q->pidx == q->size) {
+			q->pidx = 0;
+			q->gen ^= 1;
+			sd = q->sdesc;
+			d = q->desc;
+		}
+		q->credits++;
+	}
+
+	t3_write_reg(adap, A_SG_KDOORBELL, V_EGRCNTX(q->cntxt_id));
+}
+
+static inline void __refill_fl(struct adapter *adap, struct sge_fl *fl)
+{
+	refill_fl(adap, fl, min(16U, fl->size - fl->credits), GFP_ATOMIC);
+}
+
+/**
+ *	recycle_rx_buf - recycle a receive buffer
+ *	@adapter: the adapter
+ *	@q: the SGE free list
+ *	@idx: index of buffer to recycle
+ *
+ *	Recycles the specified buffer on the given free list by adding it at
+ *	the next available slot on the list.
+ */
+static void recycle_rx_buf(struct adapter *adap, struct sge_fl *q,
+			   unsigned int idx)
+{
+	struct rx_desc *from = &q->desc[idx];
+	struct rx_desc *to = &q->desc[q->pidx];
+
+	q->sdesc[q->pidx] = q->sdesc[idx];
+	to->addr_lo = from->addr_lo;	/* already big endian */
+	to->addr_hi = from->addr_hi;	/* likewise */
+	wmb();
+	to->len_gen = cpu_to_be32(V_FLD_GEN1(q->gen));
+	to->gen2 = cpu_to_be32(V_FLD_GEN2(q->gen));
+	q->credits++;
+
+	if (++q->pidx == q->size) {
+		q->pidx = 0;
+		q->gen ^= 1;
+	}
+	t3_write_reg(adap, A_SG_KDOORBELL, V_EGRCNTX(q->cntxt_id));
+}
+
+/**
+ *	alloc_ring - allocate resources for an SGE descriptor ring
+ *	@pdev: the PCI device
+ *	@nelem: the number of descriptors
+ *	@elem_size: the size of each descriptor
+ *	@sw_size: the size of the SW state associated with each ring element
+ *	@phys: the physical address of the allocated ring
+ *	@metadata: address of the array holding the SW state for the ring
+ *
+ *	Allocates resources for an SGE descriptor ring, such as Tx queues,
+ *	free buffer lists, or response queues.  Each SGE ring requires
+ *	space for its HW descriptors plus, optionally, space for the SW state
+ *	associated with each HW entry (the metadata).  The function returns
+ *	three values: the virtual address for the HW ring (the return value
+ *	of the function), the physical address of the HW ring, and the address
+ *	of the SW ring.
+ */
+static void *alloc_ring(struct pci_dev *pdev, size_t nelem, size_t elem_size,
+			size_t sw_size, dma_addr_t *phys, void *metadata)
+{
+	size_t len = nelem * elem_size;
+	void *s = NULL;
+	void *p = dma_alloc_coherent(&pdev->dev, len, phys, GFP_KERNEL);
+
+	if (!p)
+		return NULL;
+	if (sw_size) {
+		s = kcalloc(nelem, sw_size, GFP_KERNEL);
+
+		if (!s) {
+			dma_free_coherent(&pdev->dev, len, p, *phys);
+			return NULL;
+		}
+	}
+	if (metadata)
+		*(void **)metadata = s;
+	memset(p, 0, len);
+	return p;
+}
+
+/**
+ *	free_qset - free the resources of an SGE queue set
+ *	@adapter: the adapter owning the queue set
+ *	@q: the queue set
+ *
+ *	Release the HW and SW resources associated with an SGE queue set, such
+ *	as HW contexts, packet buffers, and descriptor rings.  Traffic to the
+ *	queue set must be quiesced prior to calling this.
+ */
+void t3_free_qset(struct adapter *adapter, struct sge_qset *q)
+{
+	int i;
+	struct pci_dev *pdev = adapter->pdev;
+
+	if (q->tx_reclaim_timer.function)
+		del_timer_sync(&q->tx_reclaim_timer);
+
+	for (i = 0; i < SGE_RXQ_PER_SET; ++i)
+		if (q->fl[i].desc) {
+			spin_lock(&adapter->sge.reg_lock);
+			t3_sge_disable_fl(adapter, q->fl[i].cntxt_id);
+			spin_unlock(&adapter->sge.reg_lock);
+			free_rx_bufs(pdev, &q->fl[i]);
+			kfree(q->fl[i].sdesc);
+			dma_free_coherent(&pdev->dev,
+					  q->fl[i].size *
+					  sizeof(struct rx_desc), q->fl[i].desc,
+					  q->fl[i].phys_addr);
+		}
+
+	for (i = 0; i < SGE_TXQ_PER_SET; ++i)
+		if (q->txq[i].desc) {
+			spin_lock(&adapter->sge.reg_lock);
+			t3_sge_enable_ecntxt(adapter, q->txq[i].cntxt_id, 0);
+			spin_unlock(&adapter->sge.reg_lock);
+			if (q->txq[i].sdesc) {
+				free_tx_desc(adapter, &q->txq[i],
+					     q->txq[i].in_use);
+				kfree(q->txq[i].sdesc);
+			}
+			dma_free_coherent(&pdev->dev,
+					  q->txq[i].size *
+					  sizeof(struct tx_desc),
+					  q->txq[i].desc, q->txq[i].phys_addr);
+			__skb_queue_purge(&q->txq[i].sendq);
+		}
+
+	if (q->rspq.desc) {
+		spin_lock(&adapter->sge.reg_lock);
+		t3_sge_disable_rspcntxt(adapter, q->rspq.cntxt_id);
+		spin_unlock(&adapter->sge.reg_lock);
+		dma_free_coherent(&pdev->dev,
+				  q->rspq.size * sizeof(struct rsp_desc),
+				  q->rspq.desc, q->rspq.phys_addr);
+	}
+
+	if (q->netdev)
+		q->netdev->atalk_ptr = NULL;
+
+	memset(q, 0, sizeof(*q));
+}
+
+/**
+ *	init_qset_cntxt - initialize an SGE queue set context info
+ *	@qs: the queue set
+ *	@id: the queue set id
+ *
+ *	Initializes the TIDs and context ids for the queues of a queue set.
+ */
+static void init_qset_cntxt(struct sge_qset *qs, unsigned int id)
+{
+	qs->rspq.cntxt_id = id;
+	qs->fl[0].cntxt_id = 2 * id;
+	qs->fl[1].cntxt_id = 2 * id + 1;
+	qs->txq[TXQ_ETH].cntxt_id = FW_TUNNEL_SGEEC_START + id;
+	qs->txq[TXQ_ETH].token = FW_TUNNEL_TID_START + id;
+	qs->txq[TXQ_OFLD].cntxt_id = FW_OFLD_SGEEC_START + id;
+	qs->txq[TXQ_CTRL].cntxt_id = FW_CTRL_SGEEC_START + id;
+	qs->txq[TXQ_CTRL].token = FW_CTRL_TID_START + id;
+}
+
+/**
+ *	sgl_len - calculates the size of an SGL of the given capacity
+ *	@n: the number of SGL entries
+ *
+ *	Calculates the number of flits needed for a scatter/gather list that
+ *	can hold the given number of entries.
+ */
+static inline unsigned int sgl_len(unsigned int n)
+{
+	/* alternatively: 3 * (n / 2) + 2 * (n & 1) */
+	return (3 * n) / 2 + (n & 1);
+}
+
+/**
+ *	flits_to_desc - returns the num of Tx descriptors for the given flits
+ *	@n: the number of flits
+ *
+ *	Calculates the number of Tx descriptors needed for the supplied number
+ *	of flits.
+ */
+static inline unsigned int flits_to_desc(unsigned int n)
+{
+	BUG_ON(n >= ARRAY_SIZE(flit_desc_map));
+	return flit_desc_map[n];
+}
+
+/**
+ *	get_packet - return the next ingress packet buffer from a free list
+ *	@adap: the adapter that received the packet
+ *	@fl: the SGE free list holding the packet
+ *	@len: the packet length including any SGE padding
+ *	@drop_thres: # of remaining buffers before we start dropping packets
+ *
+ *	Get the next packet from a free list and complete setup of the
+ *	sk_buff.  If the packet is small we make a copy and recycle the
+ *	original buffer, otherwise we use the original buffer itself.  If a
+ *	positive drop threshold is supplied packets are dropped and their
+ *	buffers recycled if (a) the number of remaining buffers is under the
+ *	threshold and the packet is too big to copy, or (b) the packet should
+ *	be copied but there is no memory for the copy.
+ */
+static struct sk_buff *get_packet(struct adapter *adap, struct sge_fl *fl,
+				  unsigned int len, unsigned int drop_thres)
+{
+	struct sk_buff *skb = NULL;
+	struct rx_sw_desc *sd = &fl->sdesc[fl->cidx];
+
+	prefetch(sd->skb->data);
+
+	if (len <= SGE_RX_COPY_THRES) {
+		skb = alloc_skb(len, GFP_ATOMIC);
+		if (likely(skb != NULL)) {
+			__skb_put(skb, len);
+			pci_dma_sync_single_for_cpu(adap->pdev,
+						    pci_unmap_addr(sd,
+								   dma_addr),
+						    len, PCI_DMA_FROMDEVICE);
+			memcpy(skb->data, sd->skb->data, len);
+			pci_dma_sync_single_for_device(adap->pdev,
+						       pci_unmap_addr(sd,
+								      dma_addr),
+						       len, PCI_DMA_FROMDEVICE);
+		} else if (!drop_thres)
+			goto use_orig_buf;
+	      recycle:
+		recycle_rx_buf(adap, fl, fl->cidx);
+		return skb;
+	}
+
+	if (unlikely(fl->credits < drop_thres))
+		goto recycle;
+
+      use_orig_buf:
+	pci_unmap_single(adap->pdev, pci_unmap_addr(sd, dma_addr),
+			 fl->buf_size, PCI_DMA_FROMDEVICE);
+	skb = sd->skb;
+	skb_put(skb, len);
+	__refill_fl(adap, fl);
+	return skb;
+}
+
+/**
+ *	get_imm_packet - return the next ingress packet buffer from a response
+ *	@resp: the response descriptor containing the packet data
+ *
+ *	Return a packet containing the immediate data of the given response.
+ */
+static inline struct sk_buff *get_imm_packet(const struct rsp_desc *resp)
+{
+	struct sk_buff *skb = alloc_skb(IMMED_PKT_SIZE, GFP_ATOMIC);
+
+	if (skb) {
+		__skb_put(skb, IMMED_PKT_SIZE);
+		memcpy(skb->data, resp->imm_data, IMMED_PKT_SIZE);
+	}
+	return skb;
+}
+
+/**
+ *	calc_tx_descs - calculate the number of Tx descriptors for a packet
+ *	@skb: the packet
+ *
+ * 	Returns the number of Tx descriptors needed for the given Ethernet
+ * 	packet.  Ethernet packets require addition of WR and CPL headers.
+ */
+static inline unsigned int calc_tx_descs(const struct sk_buff *skb)
+{
+	unsigned int flits;
+
+	if (skb->len <= WR_LEN - sizeof(struct cpl_tx_pkt))
+		return 1;
+
+	flits = sgl_len(skb_shinfo(skb)->nr_frags + 1) + 2;
+	if (skb_shinfo(skb)->gso_size)
+		flits++;
+	return flits_to_desc(flits);
+}
+
+/**
+ *	make_sgl - populate a scatter/gather list for a packet
+ *	@skb: the packet
+ *	@sgp: the SGL to populate
+ *	@start: start address of skb main body data to include in the SGL
+ *	@len: length of skb main body data to include in the SGL
+ *	@pdev: the PCI device
+ *
+ *	Generates a scatter/gather list for the buffers that make up a packet
+ *	and returns the SGL size in 8-byte words.  The caller must size the SGL
+ *	appropriately.
+ */
+static inline unsigned int make_sgl(const struct sk_buff *skb,
+				    struct sg_ent *sgp, unsigned char *start,
+				    unsigned int len, struct pci_dev *pdev)
+{
+	dma_addr_t mapping;
+	unsigned int i, j = 0, nfrags;
+
+	if (len) {
+		mapping = pci_map_single(pdev, start, len, PCI_DMA_TODEVICE);
+		sgp->len[0] = cpu_to_be32(len);
+		sgp->addr[0] = cpu_to_be64(mapping);
+		j = 1;
+	}
+
+	nfrags = skb_shinfo(skb)->nr_frags;
+	for (i = 0; i < nfrags; i++) {
+		skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
+
+		mapping = pci_map_page(pdev, frag->page, frag->page_offset,
+				       frag->size, PCI_DMA_TODEVICE);
+		sgp->len[j] = cpu_to_be32(frag->size);
+		sgp->addr[j] = cpu_to_be64(mapping);
+		j ^= 1;
+		if (j == 0)
+			++sgp;
+	}
+	if (j)
+		sgp->len[j] = 0;
+	return ((nfrags + (len != 0)) * 3) / 2 + j;
+}
+
+/**
+ *	check_ring_tx_db - check and potentially ring a Tx queue's doorbell
+ *	@adap: the adapter
+ *	@q: the Tx queue
+ *
+ *	Ring the doorbel if a Tx queue is asleep.  There is a natural race,
+ *	where the HW is going to sleep just after we checked, however,
+ *	then the interrupt handler will detect the outstanding TX packet
+ *	and ring the doorbell for us.
+ *
+ *	When GTS is disabled we unconditionally ring the doorbell.
+ */
+static inline void check_ring_tx_db(struct adapter *adap, struct sge_txq *q)
+{
+#if USE_GTS
+	clear_bit(TXQ_LAST_PKT_DB, &q->flags);
+	if (test_and_set_bit(TXQ_RUNNING, &q->flags) == 0) {
+		set_bit(TXQ_LAST_PKT_DB, &q->flags);
+		t3_write_reg(adap, A_SG_KDOORBELL,
+			     F_SELEGRCNTX | V_EGRCNTX(q->cntxt_id));
+	}
+#else
+	wmb();			/* write descriptors before telling HW */
+	t3_write_reg(adap, A_SG_KDOORBELL,
+		     F_SELEGRCNTX | V_EGRCNTX(q->cntxt_id));
+#endif
+}
+
+static inline void wr_gen2(struct tx_desc *d, unsigned int gen)
+{
+#if SGE_NUM_GENBITS == 2
+	d->flit[TX_DESC_FLITS - 1] = cpu_to_be64(gen);
+#endif
+}
+
+/**
+ *	write_wr_hdr_sgl - write a WR header and, optionally, SGL
+ *	@ndesc: number of Tx descriptors spanned by the SGL
+ *	@skb: the packet corresponding to the WR
+ *	@d: first Tx descriptor to be written
+ *	@pidx: index of above descriptors
+ *	@q: the SGE Tx queue
+ *	@sgl: the SGL
+ *	@flits: number of flits to the start of the SGL in the first descriptor
+ *	@sgl_flits: the SGL size in flits
+ *	@gen: the Tx descriptor generation
+ *	@wr_hi: top 32 bits of WR header based on WR type (big endian)
+ *	@wr_lo: low 32 bits of WR header based on WR type (big endian)
+ *
+ *	Write a work request header and an associated SGL.  If the SGL is
+ *	small enough to fit into one Tx descriptor it has already been written
+ *	and we just need to write the WR header.  Otherwise we distribute the
+ *	SGL across the number of descriptors it spans.
+ */
+static void write_wr_hdr_sgl(unsigned int ndesc, struct sk_buff *skb,
+			     struct tx_desc *d, unsigned int pidx,
+			     const struct sge_txq *q,
+			     const struct sg_ent *sgl,
+			     unsigned int flits, unsigned int sgl_flits,
+			     unsigned int gen, unsigned int wr_hi,
+			     unsigned int wr_lo)
+{
+	struct work_request_hdr *wrp = (struct work_request_hdr *)d;
+	struct tx_sw_desc *sd = &q->sdesc[pidx];
+
+	sd->skb = skb;
+	if (need_skb_unmap()) {
+		struct unmap_info *ui = (struct unmap_info *)skb->cb;
+
+		ui->fragidx = 0;
+		ui->addr_idx = 0;
+		ui->sflit = flits;
+	}
+
+	if (likely(ndesc == 1)) {
+		skb->priority = pidx;
+		wrp->wr_hi = htonl(F_WR_SOP | F_WR_EOP | V_WR_DATATYPE(1) |
+				   V_WR_SGLSFLT(flits)) | wr_hi;
+		wmb();
+		wrp->wr_lo = htonl(V_WR_LEN(flits + sgl_flits) |
+				   V_WR_GEN(gen)) | wr_lo;
+		wr_gen2(d, gen);
+	} else {
+		unsigned int ogen = gen;
+		const u64 *fp = (const u64 *)sgl;
+		struct work_request_hdr *wp = wrp;
+
+		wrp->wr_hi = htonl(F_WR_SOP | V_WR_DATATYPE(1) |
+				   V_WR_SGLSFLT(flits)) | wr_hi;
+
+		while (sgl_flits) {
+			unsigned int avail = WR_FLITS - flits;
+
+			if (avail > sgl_flits)
+				avail = sgl_flits;
+			memcpy(&d->flit[flits], fp, avail * sizeof(*fp));
+			sgl_flits -= avail;
+			ndesc--;
+			if (!sgl_flits)
+				break;
+
+			fp += avail;
+			d++;
+			sd++;
+			if (++pidx == q->size) {
+				pidx = 0;
+				gen ^= 1;
+				d = q->desc;
+				sd = q->sdesc;
+			}
+
+			sd->skb = skb;
+			wrp = (struct work_request_hdr *)d;
+			wrp->wr_hi = htonl(V_WR_DATATYPE(1) |
+					   V_WR_SGLSFLT(1)) | wr_hi;
+			wrp->wr_lo = htonl(V_WR_LEN(min(WR_FLITS,
+							sgl_flits + 1)) |
+					   V_WR_GEN(gen)) | wr_lo;
+			wr_gen2(d, gen);
+			flits = 1;
+		}
+		skb->priority = pidx;
+		wrp->wr_hi |= htonl(F_WR_EOP);
+		wmb();
+		wp->wr_lo = htonl(V_WR_LEN(WR_FLITS) | V_WR_GEN(ogen)) | wr_lo;
+		wr_gen2((struct tx_desc *)wp, ogen);
+		WARN_ON(ndesc != 0);
+	}
+}
+
+/**
+ *	write_tx_pkt_wr - write a TX_PKT work request
+ *	@adap: the adapter
+ *	@skb: the packet to send
+ *	@pi: the egress interface
+ *	@pidx: index of the first Tx descriptor to write
+ *	@gen: the generation value to use
+ *	@q: the Tx queue
+ *	@ndesc: number of descriptors the packet will occupy
+ *	@compl: the value of the COMPL bit to use
+ *
+ *	Generate a TX_PKT work request to send the supplied packet.
+ */
+static void write_tx_pkt_wr(struct adapter *adap, struct sk_buff *skb,
+			    const struct port_info *pi,
+			    unsigned int pidx, unsigned int gen,
+			    struct sge_txq *q, unsigned int ndesc,
+			    unsigned int compl)
+{
+	unsigned int flits, sgl_flits, cntrl, tso_info;
+	struct sg_ent *sgp, sgl[MAX_SKB_FRAGS / 2 + 1];
+	struct tx_desc *d = &q->desc[pidx];
+	struct cpl_tx_pkt *cpl = (struct cpl_tx_pkt *)d;
+
+	cpl->len = htonl(skb->len | 0x80000000);
+	cntrl = V_TXPKT_INTF(pi->port_id);
+
+	if (vlan_tx_tag_present(skb) && pi->vlan_grp)
+		cntrl |= F_TXPKT_VLAN_VLD | V_TXPKT_VLAN(vlan_tx_tag_get(skb));
+
+	tso_info = V_LSO_MSS(skb_shinfo(skb)->gso_size);
+	if (tso_info) {
+		int eth_type;
+		struct cpl_tx_pkt_lso *hdr = (struct cpl_tx_pkt_lso *)cpl;
+
+		d->flit[2] = 0;
+		cntrl |= V_TXPKT_OPCODE(CPL_TX_PKT_LSO);
+		hdr->cntrl = htonl(cntrl);
+		eth_type = skb->nh.raw - skb->data == ETH_HLEN ?
+		    CPL_ETH_II : CPL_ETH_II_VLAN;
+		tso_info |= V_LSO_ETH_TYPE(eth_type) |
+		    V_LSO_IPHDR_WORDS(skb->nh.iph->ihl) |
+		    V_LSO_TCPHDR_WORDS(skb->h.th->doff);
+		hdr->lso_info = htonl(tso_info);
+		flits = 3;
+	} else {
+		cntrl |= V_TXPKT_OPCODE(CPL_TX_PKT);
+		cntrl |= F_TXPKT_IPCSUM_DIS;	/* SW calculates IP csum */
+		cntrl |= V_TXPKT_L4CSUM_DIS(skb->ip_summed != CHECKSUM_PARTIAL);
+		cpl->cntrl = htonl(cntrl);
+
+		if (skb->len <= WR_LEN - sizeof(*cpl)) {
+			q->sdesc[pidx].skb = NULL;
+			if (!skb->data_len)
+				memcpy(&d->flit[2], skb->data, skb->len);
+			else
+				skb_copy_bits(skb, 0, &d->flit[2], skb->len);
+
+			flits = (skb->len + 7) / 8 + 2;
+			cpl->wr.wr_hi = htonl(V_WR_BCNTLFLT(skb->len & 7) |
+					      V_WR_OP(FW_WROPCODE_TUNNEL_TX_PKT)
+					      | F_WR_SOP | F_WR_EOP | compl);
+			wmb();
+			cpl->wr.wr_lo = htonl(V_WR_LEN(flits) | V_WR_GEN(gen) |
+					      V_WR_TID(q->token));
+			wr_gen2(d, gen);
+			kfree_skb(skb);
+			return;
+		}
+
+		flits = 2;
+	}
+
+	sgp = ndesc == 1 ? (struct sg_ent *)&d->flit[flits] : sgl;
+	sgl_flits = make_sgl(skb, sgp, skb->data, skb_headlen(skb), adap->pdev);
+	if (need_skb_unmap())
+		((struct unmap_info *)skb->cb)->len = skb_headlen(skb);
+
+	write_wr_hdr_sgl(ndesc, skb, d, pidx, q, sgl, flits, sgl_flits, gen,
+			 htonl(V_WR_OP(FW_WROPCODE_TUNNEL_TX_PKT) | compl),
+			 htonl(V_WR_TID(q->token)));
+}
+
+/**
+ *	eth_xmit - add a packet to the Ethernet Tx queue
+ *	@skb: the packet
+ *	@dev: the egress net device
+ *
+ *	Add a packet to an SGE Tx queue.  Runs with softirqs disabled.
+ */
+int t3_eth_xmit(struct sk_buff *skb, struct net_device *dev)
+{
+	unsigned int ndesc, pidx, credits, gen, compl;
+	const struct port_info *pi = netdev_priv(dev);
+	struct adapter *adap = dev->priv;
+	struct sge_qset *qs = dev2qset(dev);
+	struct sge_txq *q = &qs->txq[TXQ_ETH];
+
+	/*
+	 * The chip min packet length is 9 octets but play safe and reject
+	 * anything shorter than an Ethernet header.
+	 */
+	if (unlikely(skb->len < ETH_HLEN)) {
+		dev_kfree_skb(skb);
+		return NETDEV_TX_OK;
+	}
+
+	spin_lock(&q->lock);
+	reclaim_completed_tx(adap, q);
+
+	credits = q->size - q->in_use;
+	ndesc = calc_tx_descs(skb);
+
+	if (unlikely(credits < ndesc)) {
+		if (!netif_queue_stopped(dev)) {
+			netif_stop_queue(dev);
+			set_bit(TXQ_ETH, &qs->txq_stopped);
+			q->stops++;
+			dev_err(&adap->pdev->dev,
+				"%s: Tx ring %u full while queue awake!\n",
+				dev->name, q->cntxt_id & 7);
+		}
+		spin_unlock(&q->lock);
+		return NETDEV_TX_BUSY;
+	}
+
+	q->in_use += ndesc;
+	if (unlikely(credits - ndesc < q->stop_thres)) {
+		q->stops++;
+		netif_stop_queue(dev);
+		set_bit(TXQ_ETH, &qs->txq_stopped);
+#if !USE_GTS
+		if (should_restart_tx(q) &&
+		    test_and_clear_bit(TXQ_ETH, &qs->txq_stopped)) {
+			q->restarts++;
+			netif_wake_queue(dev);
+		}
+#endif
+	}
+
+	gen = q->gen;
+	q->unacked += ndesc;
+	compl = (q->unacked & 8) << (S_WR_COMPL - 3);
+	q->unacked &= 7;
+	pidx = q->pidx;
+	q->pidx += ndesc;
+	if (q->pidx >= q->size) {
+		q->pidx -= q->size;
+		q->gen ^= 1;
+	}
+
+	/* update port statistics */
+	if (skb->ip_summed == CHECKSUM_COMPLETE)
+		qs->port_stats[SGE_PSTAT_TX_CSUM]++;
+	if (skb_shinfo(skb)->gso_size)
+		qs->port_stats[SGE_PSTAT_TSO]++;
+	if (vlan_tx_tag_present(skb) && pi->vlan_grp)
+		qs->port_stats[SGE_PSTAT_VLANINS]++;
+
+	dev->trans_start = jiffies;
+	spin_unlock(&q->lock);
+
+	/*
+	 * We do not use Tx completion interrupts to free DMAd Tx packets.
+	 * This is good for performamce but means that we rely on new Tx
+	 * packets arriving to run the destructors of completed packets,
+	 * which open up space in their sockets' send queues.  Sometimes
+	 * we do not get such new packets causing Tx to stall.  A single
+	 * UDP transmitter is a good example of this situation.  We have
+	 * a clean up timer that periodically reclaims completed packets
+	 * but it doesn't run often enough (nor do we want it to) to prevent
+	 * lengthy stalls.  A solution to this problem is to run the
+	 * destructor early, after the packet is queued but before it's DMAd.
+	 * A cons is that we lie to socket memory accounting, but the amount
+	 * of extra memory is reasonable (limited by the number of Tx
+	 * descriptors), the packets do actually get freed quickly by new
+	 * packets almost always, and for protocols like TCP that wait for
+	 * acks to really free up the data the extra memory is even less.
+	 * On the positive side we run the destructors on the sending CPU
+	 * rather than on a potentially different completing CPU, usually a
+	 * good thing.  We also run them without holding our Tx queue lock,
+	 * unlike what reclaim_completed_tx() would otherwise do.
+	 *
+	 * Run the destructor before telling the DMA engine about the packet
+	 * to make sure it doesn't complete and get freed prematurely.
+	 */
+	if (likely(!skb_shared(skb)))
+		skb_orphan(skb);
+
+	write_tx_pkt_wr(adap, skb, pi, pidx, gen, q, ndesc, compl);
+	check_ring_tx_db(adap, q);
+	return NETDEV_TX_OK;
+}
+
+/**
+ *	write_imm - write a packet into a Tx descriptor as immediate data
+ *	@d: the Tx descriptor to write
+ *	@skb: the packet
+ *	@len: the length of packet data to write as immediate data
+ *	@gen: the generation bit value to write
+ *
+ *	Writes a packet as immediate data into a Tx descriptor.  The packet
+ *	contains a work request at its beginning.  We must write the packet
+ *	carefully so the SGE doesn't read accidentally before it's written in
+ *	its entirety.
+ */
+static inline void write_imm(struct tx_desc *d, struct sk_buff *skb,
+			     unsigned int len, unsigned int gen)
+{
+	struct work_request_hdr *from = (struct work_request_hdr *)skb->data;
+	struct work_request_hdr *to = (struct work_request_hdr *)d;
+
+	memcpy(&to[1], &from[1], len - sizeof(*from));
+	to->wr_hi = from->wr_hi | htonl(F_WR_SOP | F_WR_EOP |
+					V_WR_BCNTLFLT(len & 7));
+	wmb();
+	to->wr_lo = from->wr_lo | htonl(V_WR_GEN(gen) |
+					V_WR_LEN((len + 7) / 8));
+	wr_gen2(d, gen);
+	kfree_skb(skb);
+}
+
+/**
+ *	check_desc_avail - check descriptor availability on a send queue
+ *	@adap: the adapter
+ *	@q: the send queue
+ *	@skb: the packet needing the descriptors
+ *	@ndesc: the number of Tx descriptors needed
+ *	@qid: the Tx queue number in its queue set (TXQ_OFLD or TXQ_CTRL)
+ *
+ *	Checks if the requested number of Tx descriptors is available on an
+ *	SGE send queue.  If the queue is already suspended or not enough
+ *	descriptors are available the packet is queued for later transmission.
+ *	Must be called with the Tx queue locked.
+ *
+ *	Returns 0 if enough descriptors are available, 1 if there aren't
+ *	enough descriptors and the packet has been queued, and 2 if the caller
+ *	needs to retry because there weren't enough descriptors at the
+ *	beginning of the call but some freed up in the mean time.
+ */
+static inline int check_desc_avail(struct adapter *adap, struct sge_txq *q,
+				   struct sk_buff *skb, unsigned int ndesc,
+				   unsigned int qid)
+{
+	if (unlikely(!skb_queue_empty(&q->sendq))) {
+	      addq_exit:__skb_queue_tail(&q->sendq, skb);
+		return 1;
+	}
+	if (unlikely(q->size - q->in_use < ndesc)) {
+		struct sge_qset *qs = txq_to_qset(q, qid);
+
+		set_bit(qid, &qs->txq_stopped);
+		smp_mb__after_clear_bit();
+
+		if (should_restart_tx(q) &&
+		    test_and_clear_bit(qid, &qs->txq_stopped))
+			return 2;
+
+		q->stops++;
+		goto addq_exit;
+	}
+	return 0;
+}
+
+/**
+ *	reclaim_completed_tx_imm - reclaim completed control-queue Tx descs
+ *	@q: the SGE control Tx queue
+ *
+ *	This is a variant of reclaim_completed_tx() that is used for Tx queues
+ *	that send only immediate data (presently just the control queues) and
+ *	thus do not have any sk_buffs to release.
+ */
+static inline void reclaim_completed_tx_imm(struct sge_txq *q)
+{
+	unsigned int reclaim = q->processed - q->cleaned;
+
+	q->in_use -= reclaim;
+	q->cleaned += reclaim;
+}
+
+static inline int immediate(const struct sk_buff *skb)
+{
+	return skb->len <= WR_LEN && !skb->data_len;
+}
+
+/**
+ *	ctrl_xmit - send a packet through an SGE control Tx queue
+ *	@adap: the adapter
+ *	@q: the control queue
+ *	@skb: the packet
+ *
+ *	Send a packet through an SGE control Tx queue.  Packets sent through
+ *	a control queue must fit entirely as immediate data in a single Tx
+ *	descriptor and have no page fragments.
+ */
+static int ctrl_xmit(struct adapter *adap, struct sge_txq *q,
+		     struct sk_buff *skb)
+{
+	int ret;
+	struct work_request_hdr *wrp = (struct work_request_hdr *)skb->data;
+
+	if (unlikely(!immediate(skb))) {
+		WARN_ON(1);
+		dev_kfree_skb(skb);
+		return NET_XMIT_SUCCESS;
+	}
+
+	wrp->wr_hi |= htonl(F_WR_SOP | F_WR_EOP);
+	wrp->wr_lo = htonl(V_WR_TID(q->token));
+
+	spin_lock(&q->lock);
+      again:reclaim_completed_tx_imm(q);
+
+	ret = check_desc_avail(adap, q, skb, 1, TXQ_CTRL);
+	if (unlikely(ret)) {
+		if (ret == 1) {
+			spin_unlock(&q->lock);
+			return NET_XMIT_CN;
+		}
+		goto again;
+	}
+
+	write_imm(&q->desc[q->pidx], skb, skb->len, q->gen);
+
+	q->in_use++;
+	if (++q->pidx >= q->size) {
+		q->pidx = 0;
+		q->gen ^= 1;
+	}
+	spin_unlock(&q->lock);
+	wmb();
+	t3_write_reg(adap, A_SG_KDOORBELL,
+		     F_SELEGRCNTX | V_EGRCNTX(q->cntxt_id));
+	return NET_XMIT_SUCCESS;
+}
+
+/**
+ *	restart_ctrlq - restart a suspended control queue
+ *	@qs: the queue set cotaining the control queue
+ *
+ *	Resumes transmission on a suspended Tx control queue.
+ */
+static void restart_ctrlq(unsigned long data)
+{
+	struct sk_buff *skb;
+	struct sge_qset *qs = (struct sge_qset *)data;
+	struct sge_txq *q = &qs->txq[TXQ_CTRL];
+	struct adapter *adap = qs->netdev->priv;
+
+	spin_lock(&q->lock);
+      again:reclaim_completed_tx_imm(q);
+
+	while (q->in_use < q->size && (skb = __skb_dequeue(&q->sendq)) != NULL) {
+
+		write_imm(&q->desc[q->pidx], skb, skb->len, q->gen);
+
+		if (++q->pidx >= q->size) {
+			q->pidx = 0;
+			q->gen ^= 1;
+		}
+		q->in_use++;
+	}
+
+	if (!skb_queue_empty(&q->sendq)) {
+		set_bit(TXQ_CTRL, &qs->txq_stopped);
+		smp_mb__after_clear_bit();
+
+		if (should_restart_tx(q) &&
+		    test_and_clear_bit(TXQ_CTRL, &qs->txq_stopped))
+			goto again;
+		q->stops++;
+	}
+
+	spin_unlock(&q->lock);
+	t3_write_reg(adap, A_SG_KDOORBELL,
+		     F_SELEGRCNTX | V_EGRCNTX(q->cntxt_id));
+}
+
+/**
+ *	write_ofld_wr - write an offload work request
+ *	@adap: the adapter
+ *	@skb: the packet to send
+ *	@q: the Tx queue
+ *	@pidx: index of the first Tx descriptor to write
+ *	@gen: the generation value to use
+ *	@ndesc: number of descriptors the packet will occupy
+ *
+ *	Write an offload work request to send the supplied packet.  The packet
+ *	data already carry the work request with most fields populated.
+ */
+static void write_ofld_wr(struct adapter *adap, struct sk_buff *skb,
+			  struct sge_txq *q, unsigned int pidx,
+			  unsigned int gen, unsigned int ndesc)
+{
+	unsigned int sgl_flits, flits;
+	struct work_request_hdr *from;
+	struct sg_ent *sgp, sgl[MAX_SKB_FRAGS / 2 + 1];
+	struct tx_desc *d = &q->desc[pidx];
+
+	if (immediate(skb)) {
+		q->sdesc[pidx].skb = NULL;
+		write_imm(d, skb, skb->len, gen);
+		return;
+	}
+
+	/* Only TX_DATA builds SGLs */
+
+	from = (struct work_request_hdr *)skb->data;
+	memcpy(&d->flit[1], &from[1], skb->h.raw - skb->data - sizeof(*from));
+
+	flits = (skb->h.raw - skb->data) / 8;
+	sgp = ndesc == 1 ? (struct sg_ent *)&d->flit[flits] : sgl;
+	sgl_flits = make_sgl(skb, sgp, skb->h.raw, skb->tail - skb->h.raw,
+			     adap->pdev);
+	if (need_skb_unmap())
+		((struct unmap_info *)skb->cb)->len = skb->tail - skb->h.raw;
+
+	write_wr_hdr_sgl(ndesc, skb, d, pidx, q, sgl, flits, sgl_flits,
+			 gen, from->wr_hi, from->wr_lo);
+}
+
+/**
+ *	calc_tx_descs_ofld - calculate # of Tx descriptors for an offload packet
+ *	@skb: the packet
+ *
+ * 	Returns the number of Tx descriptors needed for the given offload
+ * 	packet.  These packets are already fully constructed.
+ */
+static inline unsigned int calc_tx_descs_ofld(const struct sk_buff *skb)
+{
+	unsigned int flits, cnt = skb_shinfo(skb)->nr_frags;
+
+	if (skb->len <= WR_LEN && cnt == 0)
+		return 1;	/* packet fits as immediate data */
+
+	flits = (skb->h.raw - skb->data) / 8;	/* headers */
+	if (skb->tail != skb->h.raw)
+		cnt++;
+	return flits_to_desc(flits + sgl_len(cnt));
+}
+
+/**
+ *	ofld_xmit - send a packet through an offload queue
+ *	@adap: the adapter
+ *	@q: the Tx offload queue
+ *	@skb: the packet
+ *
+ *	Send an offload packet through an SGE offload queue.
+ */
+static int ofld_xmit(struct adapter *adap, struct sge_txq *q,
+		     struct sk_buff *skb)
+{
+	int ret;
+	unsigned int ndesc = calc_tx_descs_ofld(skb), pidx, gen;
+
+	spin_lock(&q->lock);
+      again:reclaim_completed_tx(adap, q);
+
+	ret = check_desc_avail(adap, q, skb, ndesc, TXQ_OFLD);
+	if (unlikely(ret)) {
+		if (ret == 1) {
+			skb->priority = ndesc;	/* save for restart */
+			spin_unlock(&q->lock);
+			return NET_XMIT_CN;
+		}
+		goto again;
+	}
+
+	gen = q->gen;
+	q->in_use += ndesc;
+	pidx = q->pidx;
+	q->pidx += ndesc;
+	if (q->pidx >= q->size) {
+		q->pidx -= q->size;
+		q->gen ^= 1;
+	}
+	spin_unlock(&q->lock);
+
+	write_ofld_wr(adap, skb, q, pidx, gen, ndesc);
+	check_ring_tx_db(adap, q);
+	return NET_XMIT_SUCCESS;
+}
+
+/**
+ *	restart_offloadq - restart a suspended offload queue
+ *	@qs: the queue set cotaining the offload queue
+ *
+ *	Resumes transmission on a suspended Tx offload queue.
+ */
+static void restart_offloadq(unsigned long data)
+{
+	struct sk_buff *skb;
+	struct sge_qset *qs = (struct sge_qset *)data;
+	struct sge_txq *q = &qs->txq[TXQ_OFLD];
+	struct adapter *adap = qs->netdev->priv;
+
+	spin_lock(&q->lock);
+      again:reclaim_completed_tx(adap, q);
+
+	while ((skb = skb_peek(&q->sendq)) != NULL) {
+		unsigned int gen, pidx;
+		unsigned int ndesc = skb->priority;
+
+		if (unlikely(q->size - q->in_use < ndesc)) {
+			set_bit(TXQ_OFLD, &qs->txq_stopped);
+			smp_mb__after_clear_bit();
+
+			if (should_restart_tx(q) &&
+			    test_and_clear_bit(TXQ_OFLD, &qs->txq_stopped))
+				goto again;
+			q->stops++;
+			break;
+		}
+
+		gen = q->gen;
+		q->in_use += ndesc;
+		pidx = q->pidx;
+		q->pidx += ndesc;
+		if (q->pidx >= q->size) {
+			q->pidx -= q->size;
+			q->gen ^= 1;
+		}
+		__skb_unlink(skb, &q->sendq);
+		spin_unlock(&q->lock);
+
+		write_ofld_wr(adap, skb, q, pidx, gen, ndesc);
+		spin_lock(&q->lock);
+	}
+	spin_unlock(&q->lock);
+
+#if USE_GTS
+	set_bit(TXQ_RUNNING, &q->flags);
+	set_bit(TXQ_LAST_PKT_DB, &q->flags);
+#endif
+	t3_write_reg(adap, A_SG_KDOORBELL,
+		     F_SELEGRCNTX | V_EGRCNTX(q->cntxt_id));
+}
+
+/**
+ *	queue_set - return the queue set a packet should use
+ *	@skb: the packet
+ *
+ *	Maps a packet to the SGE queue set it should use.  The desired queue
+ *	set is carried in bits 1-3 in the packet's priority.
+ */
+static inline int queue_set(const struct sk_buff *skb)
+{
+	return skb->priority >> 1;
+}
+
+/**
+ *	is_ctrl_pkt - return whether an offload packet is a control packet
+ *	@skb: the packet
+ *
+ *	Determines whether an offload packet should use an OFLD or a CTRL
+ *	Tx queue.  This is indicated by bit 0 in the packet's priority.
+ */
+static inline int is_ctrl_pkt(const struct sk_buff *skb)
+{
+	return skb->priority & 1;
+}
+
+/**
+ *	t3_offload_tx - send an offload packet
+ *	@tdev: the offload device to send to
+ *	@skb: the packet
+ *
+ *	Sends an offload packet.  We use the packet priority to select the
+ *	appropriate Tx queue as follows: bit 0 indicates whether the packet
+ *	should be sent as regular or control, bits 1-3 select the queue set.
+ */
+int t3_offload_tx(struct t3cdev *tdev, struct sk_buff *skb)
+{
+	struct adapter *adap = tdev2adap(tdev);
+	struct sge_qset *qs = &adap->sge.qs[queue_set(skb)];
+
+	if (unlikely(is_ctrl_pkt(skb)))
+		return ctrl_xmit(adap, &qs->txq[TXQ_CTRL], skb);
+
+	return ofld_xmit(adap, &qs->txq[TXQ_OFLD], skb);
+}
+
+/**
+ *	offload_enqueue - add an offload packet to an SGE offload receive queue
+ *	@q: the SGE response queue
+ *	@skb: the packet
+ *
+ *	Add a new offload packet to an SGE response queue's offload packet
+ *	queue.  If the packet is the first on the queue it schedules the RX
+ *	softirq to process the queue.
+ */
+static inline void offload_enqueue(struct sge_rspq *q, struct sk_buff *skb)
+{
+	skb->next = skb->prev = NULL;
+	if (q->rx_tail)
+		q->rx_tail->next = skb;
+	else {
+		struct sge_qset *qs = rspq_to_qset(q);
+
+		if (__netif_rx_schedule_prep(qs->netdev))
+			__netif_rx_schedule(qs->netdev);
+		q->rx_head = skb;
+	}
+	q->rx_tail = skb;
+}
+
+/**
+ *	deliver_partial_bundle - deliver a (partial) bundle of Rx offload pkts
+ *	@tdev: the offload device that will be receiving the packets
+ *	@q: the SGE response queue that assembled the bundle
+ *	@skbs: the partial bundle
+ *	@n: the number of packets in the bundle
+ *
+ *	Delivers a (partial) bundle of Rx offload packets to an offload device.
+ */
+static inline void deliver_partial_bundle(struct t3cdev *tdev,
+					  struct sge_rspq *q,
+					  struct sk_buff *skbs[], int n)
+{
+	if (n) {
+		q->offload_bundles++;
+		tdev->recv(tdev, skbs, n);
+	}
+}
+
+/**
+ *	ofld_poll - NAPI handler for offload packets in interrupt mode
+ *	@dev: the network device doing the polling
+ *	@budget: polling budget
+ *
+ *	The NAPI handler for offload packets when a response queue is serviced
+ *	by the hard interrupt handler, i.e., when it's operating in non-polling
+ *	mode.  Creates small packet batches and sends them through the offload
+ *	receive handler.  Batches need to be of modest size as we do prefetches
+ *	on the packets in each.
+ */
+static int ofld_poll(struct net_device *dev, int *budget)
+{
+	struct adapter *adapter = dev->priv;
+	struct sge_qset *qs = dev2qset(dev);
+	struct sge_rspq *q = &qs->rspq;
+	int work_done, limit = min(*budget, dev->quota), avail = limit;
+
+	while (avail) {
+		struct sk_buff *head, *tail, *skbs[RX_BUNDLE_SIZE];
+		int ngathered;
+
+		spin_lock_irq(&q->lock);
+		head = q->rx_head;
+		if (!head) {
+			work_done = limit - avail;
+			*budget -= work_done;
+			dev->quota -= work_done;
+			__netif_rx_complete(dev);
+			spin_unlock_irq(&q->lock);
+			return 0;
+		}
+
+		tail = q->rx_tail;
+		q->rx_head = q->rx_tail = NULL;
+		spin_unlock_irq(&q->lock);
+
+		for (ngathered = 0; avail && head; avail--) {
+			prefetch(head->data);
+			skbs[ngathered] = head;
+			head = head->next;
+			skbs[ngathered]->next = NULL;
+			if (++ngathered == RX_BUNDLE_SIZE) {
+				q->offload_bundles++;
+				adapter->tdev.recv(&adapter->tdev, skbs,
+						   ngathered);
+				ngathered = 0;
+			}
+		}
+		if (head) {	/* splice remaining packets back onto Rx queue */
+			spin_lock_irq(&q->lock);
+			tail->next = q->rx_head;
+			if (!q->rx_head)
+				q->rx_tail = tail;
+			q->rx_head = head;
+			spin_unlock_irq(&q->lock);
+		}
+		deliver_partial_bundle(&adapter->tdev, q, skbs, ngathered);
+	}
+	work_done = limit - avail;
+	*budget -= work_done;
+	dev->quota -= work_done;
+	return 1;
+}
+
+/**
+ *	rx_offload - process a received offload packet
+ *	@tdev: the offload device receiving the packet
+ *	@rq: the response queue that received the packet
+ *	@skb: the packet
+ *	@rx_gather: a gather list of packets if we are building a bundle
+ *	@gather_idx: index of the next available slot in the bundle
+ *
+ *	Process an ingress offload pakcet and add it to the offload ingress
+ *	queue. 	Returns the index of the next available slot in the bundle.
+ */
+static inline int rx_offload(struct t3cdev *tdev, struct sge_rspq *rq,
+			     struct sk_buff *skb, struct sk_buff *rx_gather[],
+			     unsigned int gather_idx)
+{
+	rq->offload_pkts++;
+	skb->mac.raw = skb->nh.raw = skb->h.raw = skb->data;
+
+	if (rq->polling) {
+		rx_gather[gather_idx++] = skb;
+		if (gather_idx == RX_BUNDLE_SIZE) {
+			tdev->recv(tdev, rx_gather, RX_BUNDLE_SIZE);
+			gather_idx = 0;
+			rq->offload_bundles++;
+		}
+	} else
+		offload_enqueue(rq, skb);
+
+	return gather_idx;
+}
+
+/**
+ *	update_tx_completed - update the number of processed Tx descriptors
+ *	@qs: the queue set to update
+ *	@idx: which Tx queue within the set to update
+ *	@credits: number of new processed descriptors
+ *	@tx_completed: accumulates credits for the queues
+ *
+ *	Updates the number of completed Tx descriptors for a queue set's Tx
+ *	queue.  On UP systems we updated the information immediately but on
+ *	MP we accumulate the credits locally and update the Tx queue when we
+ *	reach a threshold to avoid cache-line bouncing.
+ */
+static inline void update_tx_completed(struct sge_qset *qs, int idx,
+				       unsigned int credits,
+				       unsigned int tx_completed[])
+{
+#ifdef CONFIG_SMP
+	tx_completed[idx] += credits;
+	if (tx_completed[idx] > 32) {
+		qs->txq[idx].processed += tx_completed[idx];
+		tx_completed[idx] = 0;
+	}
+#else
+	qs->txq[idx].processed += credits;
+#endif
+}
+
+/**
+ *	restart_tx - check whether to restart suspended Tx queues
+ *	@qs: the queue set to resume
+ *
+ *	Restarts suspended Tx queues of an SGE queue set if they have enough
+ *	free resources to resume operation.
+ */
+static void restart_tx(struct sge_qset *qs)
+{
+	if (test_bit(TXQ_ETH, &qs->txq_stopped) &&
+	    should_restart_tx(&qs->txq[TXQ_ETH]) &&
+	    test_and_clear_bit(TXQ_ETH, &qs->txq_stopped)) {
+		qs->txq[TXQ_ETH].restarts++;
+		if (netif_running(qs->netdev))
+			netif_wake_queue(qs->netdev);
+	}
+
+	if (test_bit(TXQ_OFLD, &qs->txq_stopped) &&
+	    should_restart_tx(&qs->txq[TXQ_OFLD]) &&
+	    test_and_clear_bit(TXQ_OFLD, &qs->txq_stopped)) {
+		qs->txq[TXQ_OFLD].restarts++;
+		tasklet_schedule(&qs->txq[TXQ_OFLD].qresume_tsk);
+	}
+	if (test_bit(TXQ_CTRL, &qs->txq_stopped) &&
+	    should_restart_tx(&qs->txq[TXQ_CTRL]) &&
+	    test_and_clear_bit(TXQ_CTRL, &qs->txq_stopped)) {
+		qs->txq[TXQ_CTRL].restarts++;
+		tasklet_schedule(&qs->txq[TXQ_CTRL].qresume_tsk);
+	}
+}
+
+/**
+ *	rx_eth - process an ingress ethernet packet
+ *	@adap: the adapter
+ *	@rq: the response queue that received the packet
+ *	@skb: the packet
+ *	@pad: amount of padding at the start of the buffer
+ *
+ *	Process an ingress ethernet pakcet and deliver it to the stack.
+ *	The padding is 2 if the packet was delivered in an Rx buffer and 0
+ *	if it was immediate data in a response.
+ */
+static void rx_eth(struct adapter *adap, struct sge_rspq *rq,
+		   struct sk_buff *skb, int pad)
+{
+	struct cpl_rx_pkt *p = (struct cpl_rx_pkt *)(skb->data + pad);
+	struct port_info *pi;
+
+	rq->eth_pkts++;
+	skb_pull(skb, sizeof(*p) + pad);
+	skb->dev = adap->port[p->iff];
+	skb->dev->last_rx = jiffies;
+	skb->protocol = eth_type_trans(skb, skb->dev);
+	pi = netdev_priv(skb->dev);
+	if (pi->rx_csum_offload && p->csum_valid && p->csum == 0xffff &&
+	    !p->fragment) {
+		rspq_to_qset(rq)->port_stats[SGE_PSTAT_RX_CSUM_GOOD]++;
+		skb->ip_summed = CHECKSUM_UNNECESSARY;
+	} else
+		skb->ip_summed = CHECKSUM_NONE;
+
+	if (unlikely(p->vlan_valid)) {
+		struct vlan_group *grp = pi->vlan_grp;
+
+		rspq_to_qset(rq)->port_stats[SGE_PSTAT_VLANEX]++;
+		if (likely(grp))
+			__vlan_hwaccel_rx(skb, grp, ntohs(p->vlan),
+					  rq->polling);
+		else
+			dev_kfree_skb_any(skb);
+	} else if (rq->polling)
+		netif_receive_skb(skb);
+	else
+		netif_rx(skb);
+}
+
+/**
+ *	handle_rsp_cntrl_info - handles control information in a response
+ *	@qs: the queue set corresponding to the response
+ *	@flags: the response control flags
+ *	@tx_completed: accumulates completion credits for the Tx queues
+ *
+ *	Handles the control information of an SGE response, such as GTS
+ *	indications and completion credits for the queue set's Tx queues.
+ */
+static inline void handle_rsp_cntrl_info(struct sge_qset *qs, u32 flags,
+					 unsigned int tx_completed[])
+{
+	unsigned int credits;
+
+#if USE_GTS
+	if (flags & F_RSPD_TXQ0_GTS)
+		clear_bit(TXQ_RUNNING, &qs->txq[TXQ_ETH].flags);
+#endif
+
+	/* ETH credits are already coalesced, return them immediately. */
+	credits = G_RSPD_TXQ0_CR(flags);
+	if (credits)
+		qs->txq[TXQ_ETH].processed += credits;
+
+# if USE_GTS
+	if (flags & F_RSPD_TXQ1_GTS)
+		clear_bit(TXQ_RUNNING, &qs->txq[TXQ_OFLD].flags);
+# endif
+	update_tx_completed(qs, TXQ_OFLD, G_RSPD_TXQ1_CR(flags), tx_completed);
+	update_tx_completed(qs, TXQ_CTRL, G_RSPD_TXQ2_CR(flags), tx_completed);
+}
+
+/**
+ *	flush_tx_completed - returns accumulated Tx completions to Tx queues
+ *	@qs: the queue set to update
+ *	@tx_completed: pending completion credits to return to Tx queues
+ *
+ *	Updates the number of completed Tx descriptors for a queue set's Tx
+ *	queues with the credits pending in @tx_completed.  This does something
+ *	only on MP systems as on UP systems we return the credits immediately.
+ */
+static inline void flush_tx_completed(struct sge_qset *qs,
+				      unsigned int tx_completed[])
+{
+#if defined(CONFIG_SMP)
+	if (tx_completed[TXQ_OFLD])
+		qs->txq[TXQ_OFLD].processed += tx_completed[TXQ_OFLD];
+	if (tx_completed[TXQ_CTRL])
+		qs->txq[TXQ_CTRL].processed += tx_completed[TXQ_CTRL];
+#endif
+}
+
+/**
+ *	check_ring_db - check if we need to ring any doorbells
+ *	@adapter: the adapter
+ *	@qs: the queue set whose Tx queues are to be examined
+ *	@sleeping: indicates which Tx queue sent GTS
+ *
+ *	Checks if some of a queue set's Tx queues need to ring their doorbells
+ *	to resume transmission after idling while they still have unprocessed
+ *	descriptors.
+ */
+static void check_ring_db(struct adapter *adap, struct sge_qset *qs,
+			  unsigned int sleeping)
+{
+	if (sleeping & F_RSPD_TXQ0_GTS) {
+		struct sge_txq *txq = &qs->txq[TXQ_ETH];
+
+		if (txq->cleaned + txq->in_use != txq->processed &&
+		    !test_and_set_bit(TXQ_LAST_PKT_DB, &txq->flags)) {
+			set_bit(TXQ_RUNNING, &txq->flags);
+			t3_write_reg(adap, A_SG_KDOORBELL, F_SELEGRCNTX |
+				     V_EGRCNTX(txq->cntxt_id));
+		}
+	}
+
+	if (sleeping & F_RSPD_TXQ1_GTS) {
+		struct sge_txq *txq = &qs->txq[TXQ_OFLD];
+
+		if (txq->cleaned + txq->in_use != txq->processed &&
+		    !test_and_set_bit(TXQ_LAST_PKT_DB, &txq->flags)) {
+			set_bit(TXQ_RUNNING, &txq->flags);
+			t3_write_reg(adap, A_SG_KDOORBELL, F_SELEGRCNTX |
+				     V_EGRCNTX(txq->cntxt_id));
+		}
+	}
+}
+
+/**
+ *	is_new_response - check if a response is newly written
+ *	@r: the response descriptor
+ *	@q: the response queue
+ *
+ *	Returns true if a response descriptor contains a yet unprocessed
+ *	response.
+ */
+static inline int is_new_response(const struct rsp_desc *r,
+				  const struct sge_rspq *q)
+{
+	return (r->intr_gen & F_RSPD_GEN2) == q->gen;
+}
+
+#define RSPD_GTS_MASK  (F_RSPD_TXQ0_GTS | F_RSPD_TXQ1_GTS)
+#define RSPD_CTRL_MASK (RSPD_GTS_MASK | \
+			V_RSPD_TXQ0_CR(M_RSPD_TXQ0_CR) | \
+			V_RSPD_TXQ1_CR(M_RSPD_TXQ1_CR) | \
+			V_RSPD_TXQ2_CR(M_RSPD_TXQ2_CR))
+
+/* How long to delay the next interrupt in case of memory shortage, in 0.1us. */
+#define NOMEM_INTR_DELAY 2500
+
+/**
+ *	process_responses - process responses from an SGE response queue
+ *	@adap: the adapter
+ *	@qs: the queue set to which the response queue belongs
+ *	@budget: how many responses can be processed in this round
+ *
+ *	Process responses from an SGE response queue up to the supplied budget.
+ *	Responses include received packets as well as credits and other events
+ *	for the queues that belong to the response queue's queue set.
+ *	A negative budget is effectively unlimited.
+ *
+ *	Additionally choose the interrupt holdoff time for the next interrupt
+ *	on this queue.  If the system is under memory shortage use a fairly
+ *	long delay to help recovery.
+ */
+static int process_responses(struct adapter *adap, struct sge_qset *qs,
+			     int budget)
+{
+	struct sge_rspq *q = &qs->rspq;
+	struct rsp_desc *r = &q->desc[q->cidx];
+	int budget_left = budget;
+	unsigned int sleeping = 0, tx_completed[3] = { 0, 0, 0 };
+	struct sk_buff *offload_skbs[RX_BUNDLE_SIZE];
+	int ngathered = 0;
+
+	q->next_holdoff = q->holdoff_tmr;
+
+	while (likely(budget_left && is_new_response(r, q))) {
+		int eth, ethpad = 0;
+		struct sk_buff *skb = NULL;
+		u32 len, flags = ntohl(r->flags);
+		u32 rss_hi = *(const u32 *)r, rss_lo = r->rss_hdr.rss_hash_val;
+
+		eth = r->rss_hdr.opcode == CPL_RX_PKT;
+
+		if (unlikely(flags & F_RSPD_ASYNC_NOTIF)) {
+			skb = alloc_skb(AN_PKT_SIZE, GFP_ATOMIC);
+			if (!skb)
+				goto no_mem;
+
+			memcpy(__skb_put(skb, AN_PKT_SIZE), r, AN_PKT_SIZE);
+			skb->data[0] = CPL_ASYNC_NOTIF;
+			rss_hi = htonl(CPL_ASYNC_NOTIF << 24);
+			q->async_notif++;
+		} else if (flags & F_RSPD_IMM_DATA_VALID) {
+			skb = get_imm_packet(r);
+			if (unlikely(!skb)) {
+			      no_mem:
+				q->next_holdoff = NOMEM_INTR_DELAY;
+				q->nomem++;
+				/* consume one credit since we tried */
+				budget_left--;
+				break;
+			}
+			q->imm_data++;
+		} else if ((len = ntohl(r->len_cq)) != 0) {
+			struct sge_fl *fl;
+
+			fl = (len & F_RSPD_FLQ) ? &qs->fl[1] : &qs->fl[0];
+			fl->credits--;
+			skb = get_packet(adap, fl, G_RSPD_LEN(len),
+					 eth ? SGE_RX_DROP_THRES : 0);
+			if (!skb)
+				q->rx_drops++;
+			else if (r->rss_hdr.opcode == CPL_TRACE_PKT)
+				__skb_pull(skb, 2);
+			ethpad = 2;
+			if (++fl->cidx == fl->size)
+				fl->cidx = 0;
+		} else
+			q->pure_rsps++;
+
+		if (flags & RSPD_CTRL_MASK) {
+			sleeping |= flags & RSPD_GTS_MASK;
+			handle_rsp_cntrl_info(qs, flags, tx_completed);
+		}
+
+		r++;
+		if (unlikely(++q->cidx == q->size)) {
+			q->cidx = 0;
+			q->gen ^= 1;
+			r = q->desc;
+		}
+		prefetch(r);
+
+		if (++q->credits >= (q->size / 4)) {
+			refill_rspq(adap, q, q->credits);
+			q->credits = 0;
+		}
+
+		if (likely(skb != NULL)) {
+			if (eth)
+				rx_eth(adap, q, skb, ethpad);
+			else {
+				/* Preserve the RSS info in csum & priority */
+				skb->csum = rss_hi;
+				skb->priority = rss_lo;
+				ngathered = rx_offload(&adap->tdev, q, skb,
+						       offload_skbs, ngathered);
+			}
+		}
+
+		--budget_left;
+	}
+
+	flush_tx_completed(qs, tx_completed);
+	deliver_partial_bundle(&adap->tdev, q, offload_skbs, ngathered);
+	if (sleeping)
+		check_ring_db(adap, qs, sleeping);
+
+	smp_mb();		/* commit Tx queue .processed updates */
+	if (unlikely(qs->txq_stopped != 0))
+		restart_tx(qs);
+
+	budget -= budget_left;
+	return budget;
+}
+
+static inline int is_pure_response(const struct rsp_desc *r)
+{
+	u32 n = ntohl(r->flags) & (F_RSPD_ASYNC_NOTIF | F_RSPD_IMM_DATA_VALID);
+
+	return (n | r->len_cq) == 0;
+}
+
+/**
+ *	napi_rx_handler - the NAPI handler for Rx processing
+ *	@dev: the net device
+ *	@budget: how many packets we can process in this round
+ *
+ *	Handler for new data events when using NAPI.
+ */
+static int napi_rx_handler(struct net_device *dev, int *budget)
+{
+	struct adapter *adap = dev->priv;
+	struct sge_qset *qs = dev2qset(dev);
+	int effective_budget = min(*budget, dev->quota);
+
+	int work_done = process_responses(adap, qs, effective_budget);
+	*budget -= work_done;
+	dev->quota -= work_done;
+
+	if (work_done >= effective_budget)
+		return 1;
+
+	netif_rx_complete(dev);
+
+	/*
+	 * Because we don't atomically flush the following write it is
+	 * possible that in very rare cases it can reach the device in a way
+	 * that races with a new response being written plus an error interrupt
+	 * causing the NAPI interrupt handler below to return unhandled status
+	 * to the OS.  To protect against this would require flushing the write
+	 * and doing both the write and the flush with interrupts off.  Way too
+	 * expensive and unjustifiable given the rarity of the race.
+	 *
+	 * The race cannot happen at all with MSI-X.
+	 */
+	t3_write_reg(adap, A_SG_GTS, V_RSPQ(qs->rspq.cntxt_id) |
+		     V_NEWTIMER(qs->rspq.next_holdoff) |
+		     V_NEWINDEX(qs->rspq.cidx));
+	return 0;
+}
+
+/*
+ * Returns true if the device is already scheduled for polling.
+ */
+static inline int napi_is_scheduled(struct net_device *dev)
+{
+	return test_bit(__LINK_STATE_RX_SCHED, &dev->state);
+}
+
+/**
+ *	process_pure_responses - process pure responses from a response queue
+ *	@adap: the adapter
+ *	@qs: the queue set owning the response queue
+ *	@r: the first pure response to process
+ *
+ *	A simpler version of process_responses() that handles only pure (i.e.,
+ *	non data-carrying) responses.  Such respones are too light-weight to
+ *	justify calling a softirq under NAPI, so we handle them specially in
+ *	the interrupt handler.  The function is called with a pointer to a
+ *	response, which the caller must ensure is a valid pure response.
+ *
+ *	Returns 1 if it encounters a valid data-carrying response, 0 otherwise.
+ */
+static int process_pure_responses(struct adapter *adap, struct sge_qset *qs,
+				  struct rsp_desc *r)
+{
+	struct sge_rspq *q = &qs->rspq;
+	unsigned int sleeping = 0, tx_completed[3] = { 0, 0, 0 };
+
+	do {
+		u32 flags = ntohl(r->flags);
+
+		r++;
+		if (unlikely(++q->cidx == q->size)) {
+			q->cidx = 0;
+			q->gen ^= 1;
+			r = q->desc;
+		}
+		prefetch(r);
+
+		if (flags & RSPD_CTRL_MASK) {
+			sleeping |= flags & RSPD_GTS_MASK;
+			handle_rsp_cntrl_info(qs, flags, tx_completed);
+		}
+
+		q->pure_rsps++;
+		if (++q->credits >= (q->size / 4)) {
+			refill_rspq(adap, q, q->credits);
+			q->credits = 0;
+		}
+	} while (is_new_response(r, q) && is_pure_response(r));
+
+	flush_tx_completed(qs, tx_completed);
+
+	if (sleeping)
+		check_ring_db(adap, qs, sleeping);
+
+	smp_mb();		/* commit Tx queue .processed updates */
+	if (unlikely(qs->txq_stopped != 0))
+		restart_tx(qs);
+
+	return is_new_response(r, q);
+}
+
+/**
+ *	handle_responses - decide what to do with new responses in NAPI mode
+ *	@adap: the adapter
+ *	@q: the response queue
+ *
+ *	This is used by the NAPI interrupt handlers to decide what to do with
+ *	new SGE responses.  If there are no new responses it returns -1.  If
+ *	there are new responses and they are pure (i.e., non-data carrying)
+ *	it handles them straight in hard interrupt context as they are very
+ *	cheap and don't deliver any packets.  Finally, if there are any data
+ *	signaling responses it schedules the NAPI handler.  Returns 1 if it
+ *	schedules NAPI, 0 if all new responses were pure.
+ *
+ *	The caller must ascertain NAPI is not already running.
+ */
+static inline int handle_responses(struct adapter *adap, struct sge_rspq *q)
+{
+	struct sge_qset *qs = rspq_to_qset(q);
+	struct rsp_desc *r = &q->desc[q->cidx];
+
+	if (!is_new_response(r, q))
+		return -1;
+	if (is_pure_response(r) && process_pure_responses(adap, qs, r) == 0) {
+		t3_write_reg(adap, A_SG_GTS, V_RSPQ(q->cntxt_id) |
+			     V_NEWTIMER(q->holdoff_tmr) | V_NEWINDEX(q->cidx));
+		return 0;
+	}
+	if (likely(__netif_rx_schedule_prep(qs->netdev)))
+		__netif_rx_schedule(qs->netdev);
+	return 1;
+}
+
+/*
+ * The MSI-X interrupt handler for an SGE response queue for the non-NAPI case
+ * (i.e., response queue serviced in hard interrupt).
+ */
+irqreturn_t t3_sge_intr_msix(int irq, void *cookie)
+{
+	struct sge_qset *qs = cookie;
+	struct adapter *adap = qs->netdev->priv;
+	struct sge_rspq *q = &qs->rspq;
+
+	spin_lock(&q->lock);
+	if (process_responses(adap, qs, -1) == 0)
+		q->unhandled_irqs++;
+	t3_write_reg(adap, A_SG_GTS, V_RSPQ(q->cntxt_id) |
+		     V_NEWTIMER(q->next_holdoff) | V_NEWINDEX(q->cidx));
+	spin_unlock(&q->lock);
+	return IRQ_HANDLED;
+}
+
+/*
+ * The MSI-X interrupt handler for an SGE response queue for the NAPI case
+ * (i.e., response queue serviced by NAPI polling).
+ */
+irqreturn_t t3_sge_intr_msix_napi(int irq, void *cookie)
+{
+	struct sge_qset *qs = cookie;
+	struct adapter *adap = qs->netdev->priv;
+	struct sge_rspq *q = &qs->rspq;
+
+	spin_lock(&q->lock);
+	BUG_ON(napi_is_scheduled(qs->netdev));
+
+	if (handle_responses(adap, q) < 0)
+		q->unhandled_irqs++;
+	spin_unlock(&q->lock);
+	return IRQ_HANDLED;
+}
+
+/*
+ * The non-NAPI MSI interrupt handler.  This needs to handle data events from
+ * SGE response queues as well as error and other async events as they all use
+ * the same MSI vector.  We use one SGE response queue per port in this mode
+ * and protect all response queues with queue 0's lock.
+ */
+static irqreturn_t t3_intr_msi(int irq, void *cookie)
+{
+	int new_packets = 0;
+	struct adapter *adap = cookie;
+	struct sge_rspq *q = &adap->sge.qs[0].rspq;
+
+	spin_lock(&q->lock);
+
+	if (process_responses(adap, &adap->sge.qs[0], -1)) {
+		t3_write_reg(adap, A_SG_GTS, V_RSPQ(q->cntxt_id) |
+			     V_NEWTIMER(q->next_holdoff) | V_NEWINDEX(q->cidx));
+		new_packets = 1;
+	}
+
+	if (adap->params.nports == 2 &&
+	    process_responses(adap, &adap->sge.qs[1], -1)) {
+		struct sge_rspq *q1 = &adap->sge.qs[1].rspq;
+
+		t3_write_reg(adap, A_SG_GTS, V_RSPQ(q1->cntxt_id) |
+			     V_NEWTIMER(q1->next_holdoff) |
+			     V_NEWINDEX(q1->cidx));
+		new_packets = 1;
+	}
+
+	if (!new_packets && t3_slow_intr_handler(adap) == 0)
+		q->unhandled_irqs++;
+
+	spin_unlock(&q->lock);
+	return IRQ_HANDLED;
+}
+
+static int rspq_check_napi(struct net_device *dev, struct sge_rspq *q)
+{
+	if (!napi_is_scheduled(dev) && is_new_response(&q->desc[q->cidx], q)) {
+		if (likely(__netif_rx_schedule_prep(dev)))
+			__netif_rx_schedule(dev);
+		return 1;
+	}
+	return 0;
+}
+
+/*
+ * The MSI interrupt handler for the NAPI case (i.e., response queues serviced
+ * by NAPI polling).  Handles data events from SGE response queues as well as
+ * error and other async events as they all use the same MSI vector.  We use
+ * one SGE response queue per port in this mode and protect all response
+ * queues with queue 0's lock.
+ */
+irqreturn_t t3_intr_msi_napi(int irq, void *cookie)
+{
+	int new_packets;
+	struct adapter *adap = cookie;
+	struct sge_rspq *q = &adap->sge.qs[0].rspq;
+
+	spin_lock(&q->lock);
+
+	new_packets = rspq_check_napi(adap->sge.qs[0].netdev, q);
+	if (adap->params.nports == 2)
+		new_packets += rspq_check_napi(adap->sge.qs[1].netdev,
+					       &adap->sge.qs[1].rspq);
+	if (!new_packets && t3_slow_intr_handler(adap) == 0)
+		q->unhandled_irqs++;
+
+	spin_unlock(&q->lock);
+	return IRQ_HANDLED;
+}
+
+/*
+ * A helper function that processes responses and issues GTS.
+ */
+static inline int process_responses_gts(struct adapter *adap,
+					struct sge_rspq *rq)
+{
+	int work;
+
+	work = process_responses(adap, rspq_to_qset(rq), -1);
+	t3_write_reg(adap, A_SG_GTS, V_RSPQ(rq->cntxt_id) |
+		     V_NEWTIMER(rq->next_holdoff) | V_NEWINDEX(rq->cidx));
+	return work;
+}
+
+/*
+ * The legacy INTx interrupt handler.  This needs to handle data events from
+ * SGE response queues as well as error and other async events as they all use
+ * the same interrupt pin.  We use one SGE response queue per port in this mode
+ * and protect all response queues with queue 0's lock.
+ */
+static irqreturn_t t3_intr(int irq, void *cookie)
+{
+	int work_done, w0, w1;
+	struct adapter *adap = cookie;
+	struct sge_rspq *q0 = &adap->sge.qs[0].rspq;
+	struct sge_rspq *q1 = &adap->sge.qs[1].rspq;
+
+	spin_lock(&q0->lock);
+
+	w0 = is_new_response(&q0->desc[q0->cidx], q0);
+	w1 = adap->params.nports == 2 &&
+	    is_new_response(&q1->desc[q1->cidx], q1);
+
+	if (likely(w0 | w1)) {
+		t3_write_reg(adap, A_PL_CLI, 0);
+		t3_read_reg(adap, A_PL_CLI);	/* flush */
+
+		if (likely(w0))
+			process_responses_gts(adap, q0);
+
+		if (w1)
+			process_responses_gts(adap, q1);
+
+		work_done = w0 | w1;
+	} else
+		work_done = t3_slow_intr_handler(adap);
+
+	spin_unlock(&q0->lock);
+	return IRQ_RETVAL(work_done != 0);
+}
+
+/*
+ * Interrupt handler for legacy INTx interrupts for T3B-based cards.
+ * Handles data events from SGE response queues as well as error and other
+ * async events as they all use the same interrupt pin.  We use one SGE
+ * response queue per port in this mode and protect all response queues with
+ * queue 0's lock.
+ */
+static irqreturn_t t3b_intr(int irq, void *cookie)
+{
+	u32 map;
+	struct adapter *adap = cookie;
+	struct sge_rspq *q0 = &adap->sge.qs[0].rspq;
+
+	t3_write_reg(adap, A_PL_CLI, 0);
+	map = t3_read_reg(adap, A_SG_DATA_INTR);
+
+	if (unlikely(!map))	/* shared interrupt, most likely */
+		return IRQ_NONE;
+
+	spin_lock(&q0->lock);
+
+	if (unlikely(map & F_ERRINTR))
+		t3_slow_intr_handler(adap);
+
+	if (likely(map & 1))
+		process_responses_gts(adap, q0);
+
+	if (map & 2)
+		process_responses_gts(adap, &adap->sge.qs[1].rspq);
+
+	spin_unlock(&q0->lock);
+	return IRQ_HANDLED;
+}
+
+/*
+ * NAPI interrupt handler for legacy INTx interrupts for T3B-based cards.
+ * Handles data events from SGE response queues as well as error and other
+ * async events as they all use the same interrupt pin.  We use one SGE
+ * response queue per port in this mode and protect all response queues with
+ * queue 0's lock.
+ */
+static irqreturn_t t3b_intr_napi(int irq, void *cookie)
+{
+	u32 map;
+	struct net_device *dev;
+	struct adapter *adap = cookie;
+	struct sge_rspq *q0 = &adap->sge.qs[0].rspq;
+
+	t3_write_reg(adap, A_PL_CLI, 0);
+	map = t3_read_reg(adap, A_SG_DATA_INTR);
+
+	if (unlikely(!map))	/* shared interrupt, most likely */
+		return IRQ_NONE;
+
+	spin_lock(&q0->lock);
+
+	if (unlikely(map & F_ERRINTR))
+		t3_slow_intr_handler(adap);
+
+	if (likely(map & 1)) {
+		dev = adap->sge.qs[0].netdev;
+
+		BUG_ON(napi_is_scheduled(dev));
+		if (likely(__netif_rx_schedule_prep(dev)))
+			__netif_rx_schedule(dev);
+	}
+	if (map & 2) {
+		dev = adap->sge.qs[1].netdev;
+
+		BUG_ON(napi_is_scheduled(dev));
+		if (likely(__netif_rx_schedule_prep(dev)))
+			__netif_rx_schedule(dev);
+	}
+
+	spin_unlock(&q0->lock);
+	return IRQ_HANDLED;
+}
+
+/**
+ *	t3_intr_handler - select the top-level interrupt handler
+ *	@adap: the adapter
+ *	@polling: whether using NAPI to service response queues
+ *
+ *	Selects the top-level interrupt handler based on the type of interrupts
+ *	(MSI-X, MSI, or legacy) and whether NAPI will be used to service the
+ *	response queues.
+ */
+intr_handler_t t3_intr_handler(struct adapter *adap, int polling)
+{
+	if (adap->flags & USING_MSIX)
+		return polling ? t3_sge_intr_msix_napi : t3_sge_intr_msix;
+	if (adap->flags & USING_MSI)
+		return polling ? t3_intr_msi_napi : t3_intr_msi;
+	if (adap->params.rev > 0)
+		return polling ? t3b_intr_napi : t3b_intr;
+	return t3_intr;
+}
+
+/**
+ *	t3_sge_err_intr_handler - SGE async event interrupt handler
+ *	@adapter: the adapter
+ *
+ *	Interrupt handler for SGE asynchronous (non-data) events.
+ */
+void t3_sge_err_intr_handler(struct adapter *adapter)
+{
+	unsigned int v, status = t3_read_reg(adapter, A_SG_INT_CAUSE);
+
+	if (status & F_RSPQCREDITOVERFOW)
+		CH_ALERT(adapter, "SGE response queue credit overflow\n");
+
+	if (status & F_RSPQDISABLED) {
+		v = t3_read_reg(adapter, A_SG_RSPQ_FL_STATUS);
+
+		CH_ALERT(adapter,
+			 "packet delivered to disabled response queue "
+			 "(0x%x)\n", (v >> S_RSPQ0DISABLED) & 0xff);
+	}
+
+	t3_write_reg(adapter, A_SG_INT_CAUSE, status);
+	if (status & (F_RSPQCREDITOVERFOW | F_RSPQDISABLED))
+		t3_fatal_err(adapter);
+}
+
+/**
+ *	sge_timer_cb - perform periodic maintenance of an SGE qset
+ *	@data: the SGE queue set to maintain
+ *
+ *	Runs periodically from a timer to perform maintenance of an SGE queue
+ *	set.  It performs two tasks:
+ *
+ *	a) Cleans up any completed Tx descriptors that may still be pending.
+ *	Normal descriptor cleanup happens when new packets are added to a Tx
+ *	queue so this timer is relatively infrequent and does any cleanup only
+ *	if the Tx queue has not seen any new packets in a while.  We make a
+ *	best effort attempt to reclaim descriptors, in that we don't wait
+ *	around if we cannot get a queue's lock (which most likely is because
+ *	someone else is queueing new packets and so will also handle the clean
+ *	up).  Since control queues use immediate data exclusively we don't
+ *	bother cleaning them up here.
+ *
+ *	b) Replenishes Rx queues that have run out due to memory shortage.
+ *	Normally new Rx buffers are added when existing ones are consumed but
+ *	when out of memory a queue can become empty.  We try to add only a few
+ *	buffers here, the queue will be replenished fully as these new buffers
+ *	are used up if memory shortage has subsided.
+ */
+static void sge_timer_cb(unsigned long data)
+{
+	spinlock_t *lock;
+	struct sge_qset *qs = (struct sge_qset *)data;
+	struct adapter *adap = qs->netdev->priv;
+
+	if (spin_trylock(&qs->txq[TXQ_ETH].lock)) {
+		reclaim_completed_tx(adap, &qs->txq[TXQ_ETH]);
+		spin_unlock(&qs->txq[TXQ_ETH].lock);
+	}
+	if (spin_trylock(&qs->txq[TXQ_OFLD].lock)) {
+		reclaim_completed_tx(adap, &qs->txq[TXQ_OFLD]);
+		spin_unlock(&qs->txq[TXQ_OFLD].lock);
+	}
+	lock = (adap->flags & USING_MSIX) ? &qs->rspq.lock :
+	    &adap->sge.qs[0].rspq.lock;
+	if (spin_trylock_irq(lock)) {
+		if (!napi_is_scheduled(qs->netdev)) {
+			if (qs->fl[0].credits < qs->fl[0].size)
+				__refill_fl(adap, &qs->fl[0]);
+			if (qs->fl[1].credits < qs->fl[1].size)
+				__refill_fl(adap, &qs->fl[1]);
+		}
+		spin_unlock_irq(lock);
+	}
+	mod_timer(&qs->tx_reclaim_timer, jiffies + TX_RECLAIM_PERIOD);
+}
+
+/**
+ *	t3_update_qset_coalesce - update coalescing settings for a queue set
+ *	@qs: the SGE queue set
+ *	@p: new queue set parameters
+ *
+ *	Update the coalescing settings for an SGE queue set.  Nothing is done
+ *	if the queue set is not initialized yet.
+ */
+void t3_update_qset_coalesce(struct sge_qset *qs, const struct qset_params *p)
+{
+	if (!qs->netdev)
+		return;
+
+	qs->rspq.holdoff_tmr = max(p->coalesce_usecs * 10, 1U);/* can't be 0 */
+	qs->rspq.polling = p->polling;
+	qs->netdev->poll = p->polling ? napi_rx_handler : ofld_poll;
+}
+
+/**
+ *	t3_sge_alloc_qset - initialize an SGE queue set
+ *	@adapter: the adapter
+ *	@id: the queue set id
+ *	@nports: how many Ethernet ports will be using this queue set
+ *	@irq_vec_idx: the IRQ vector index for response queue interrupts
+ *	@p: configuration parameters for this queue set
+ *	@ntxq: number of Tx queues for the queue set
+ *	@netdev: net device associated with this queue set
+ *
+ *	Allocate resources and initialize an SGE queue set.  A queue set
+ *	comprises a response queue, two Rx free-buffer queues, and up to 3
+ *	Tx queues.  The Tx queues are assigned roles in the order Ethernet
+ *	queue, offload queue, and control queue.
+ */
+int t3_sge_alloc_qset(struct adapter *adapter, unsigned int id, int nports,
+		      int irq_vec_idx, const struct qset_params *p,
+		      int ntxq, struct net_device *netdev)
+{
+	int i, ret = -ENOMEM;
+	struct sge_qset *q = &adapter->sge.qs[id];
+
+	init_qset_cntxt(q, id);
+	init_timer(&q->tx_reclaim_timer);
+	q->tx_reclaim_timer.data = (unsigned long)q;
+	q->tx_reclaim_timer.function = sge_timer_cb;
+
+	q->fl[0].desc = alloc_ring(adapter->pdev, p->fl_size,
+				   sizeof(struct rx_desc),
+				   sizeof(struct rx_sw_desc),
+				   &q->fl[0].phys_addr, &q->fl[0].sdesc);
+	if (!q->fl[0].desc)
+		goto err;
+
+	q->fl[1].desc = alloc_ring(adapter->pdev, p->jumbo_size,
+				   sizeof(struct rx_desc),
+				   sizeof(struct rx_sw_desc),
+				   &q->fl[1].phys_addr, &q->fl[1].sdesc);
+	if (!q->fl[1].desc)
+		goto err;
+
+	q->rspq.desc = alloc_ring(adapter->pdev, p->rspq_size,
+				  sizeof(struct rsp_desc), 0,
+				  &q->rspq.phys_addr, NULL);
+	if (!q->rspq.desc)
+		goto err;
+
+	for (i = 0; i < ntxq; ++i) {
+		/*
+		 * The control queue always uses immediate data so does not
+		 * need to keep track of any sk_buffs.
+		 */
+		size_t sz = i == TXQ_CTRL ? 0 : sizeof(struct tx_sw_desc);
+
+		q->txq[i].desc = alloc_ring(adapter->pdev, p->txq_size[i],
+					    sizeof(struct tx_desc), sz,
+					    &q->txq[i].phys_addr,
+					    &q->txq[i].sdesc);
+		if (!q->txq[i].desc)
+			goto err;
+
+		q->txq[i].gen = 1;
+		q->txq[i].size = p->txq_size[i];
+		spin_lock_init(&q->txq[i].lock);
+		skb_queue_head_init(&q->txq[i].sendq);
+	}
+
+	tasklet_init(&q->txq[TXQ_OFLD].qresume_tsk, restart_offloadq,
+		     (unsigned long)q);
+	tasklet_init(&q->txq[TXQ_CTRL].qresume_tsk, restart_ctrlq,
+		     (unsigned long)q);
+
+	q->fl[0].gen = q->fl[1].gen = 1;
+	q->fl[0].size = p->fl_size;
+	q->fl[1].size = p->jumbo_size;
+
+	q->rspq.gen = 1;
+	q->rspq.size = p->rspq_size;
+	spin_lock_init(&q->rspq.lock);
+
+	q->txq[TXQ_ETH].stop_thres = nports *
+	    flits_to_desc(sgl_len(MAX_SKB_FRAGS + 1) + 3);
+
+	if (ntxq == 1) {
+		q->fl[0].buf_size = SGE_RX_SM_BUF_SIZE + 2 +
+		    sizeof(struct cpl_rx_pkt);
+		q->fl[1].buf_size = MAX_FRAME_SIZE + 2 +
+		    sizeof(struct cpl_rx_pkt);
+	} else {
+		q->fl[0].buf_size = SGE_RX_SM_BUF_SIZE +
+		    sizeof(struct cpl_rx_data);
+		q->fl[1].buf_size = (16 * 1024) -
+		    SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
+	}
+
+	spin_lock(&adapter->sge.reg_lock);
+
+	/* FL threshold comparison uses < */
+	ret = t3_sge_init_rspcntxt(adapter, q->rspq.cntxt_id, irq_vec_idx,
+				   q->rspq.phys_addr, q->rspq.size,
+				   q->fl[0].buf_size, 1, 0);
+	if (ret)
+		goto err_unlock;
+
+	for (i = 0; i < SGE_RXQ_PER_SET; ++i) {
+		ret = t3_sge_init_flcntxt(adapter, q->fl[i].cntxt_id, 0,
+					  q->fl[i].phys_addr, q->fl[i].size,
+					  q->fl[i].buf_size, p->cong_thres, 1,
+					  0);
+		if (ret)
+			goto err_unlock;
+	}
+
+	ret = t3_sge_init_ecntxt(adapter, q->txq[TXQ_ETH].cntxt_id, USE_GTS,
+				 SGE_CNTXT_ETH, id, q->txq[TXQ_ETH].phys_addr,
+				 q->txq[TXQ_ETH].size, q->txq[TXQ_ETH].token,
+				 1, 0);
+	if (ret)
+		goto err_unlock;
+
+	if (ntxq > 1) {
+		ret = t3_sge_init_ecntxt(adapter, q->txq[TXQ_OFLD].cntxt_id,
+					 USE_GTS, SGE_CNTXT_OFLD, id,
+					 q->txq[TXQ_OFLD].phys_addr,
+					 q->txq[TXQ_OFLD].size, 0, 1, 0);
+		if (ret)
+			goto err_unlock;
+	}
+
+	if (ntxq > 2) {
+		ret = t3_sge_init_ecntxt(adapter, q->txq[TXQ_CTRL].cntxt_id, 0,
+					 SGE_CNTXT_CTRL, id,
+					 q->txq[TXQ_CTRL].phys_addr,
+					 q->txq[TXQ_CTRL].size,
+					 q->txq[TXQ_CTRL].token, 1, 0);
+		if (ret)
+			goto err_unlock;
+	}
+
+	spin_unlock(&adapter->sge.reg_lock);
+	q->netdev = netdev;
+	t3_update_qset_coalesce(q, p);
+
+	/*
+	 * We use atalk_ptr as a backpointer to a qset.  In case a device is
+	 * associated with multiple queue sets only the first one sets
+	 * atalk_ptr.
+	 */
+	if (netdev->atalk_ptr == NULL)
+		netdev->atalk_ptr = q;
+
+	refill_fl(adapter, &q->fl[0], q->fl[0].size, GFP_KERNEL);
+	refill_fl(adapter, &q->fl[1], q->fl[1].size, GFP_KERNEL);
+	refill_rspq(adapter, &q->rspq, q->rspq.size - 1);
+
+	t3_write_reg(adapter, A_SG_GTS, V_RSPQ(q->rspq.cntxt_id) |
+		     V_NEWTIMER(q->rspq.holdoff_tmr));
+
+	mod_timer(&q->tx_reclaim_timer, jiffies + TX_RECLAIM_PERIOD);
+	return 0;
+
+      err_unlock:
+	spin_unlock(&adapter->sge.reg_lock);
+      err:
+	t3_free_qset(adapter, q);
+	return ret;
+}
+
+/**
+ *	t3_free_sge_resources - free SGE resources
+ *	@adap: the adapter
+ *
+ *	Frees resources used by the SGE queue sets.
+ */
+void t3_free_sge_resources(struct adapter *adap)
+{
+	int i;
+
+	for (i = 0; i < SGE_QSETS; ++i)
+		t3_free_qset(adap, &adap->sge.qs[i]);
+}
+
+/**
+ *	t3_sge_start - enable SGE
+ *	@adap: the adapter
+ *
+ *	Enables the SGE for DMAs.  This is the last step in starting packet
+ *	transfers.
+ */
+void t3_sge_start(struct adapter *adap)
+{
+	t3_set_reg_field(adap, A_SG_CONTROL, F_GLOBALENABLE, F_GLOBALENABLE);
+}
+
+/**
+ *	t3_sge_stop - disable SGE operation
+ *	@adap: the adapter
+ *
+ *	Disables the DMA engine.  This can be called in emeregencies (e.g.,
+ *	from error interrupts) or from normal process context.  In the latter
+ *	case it also disables any pending queue restart tasklets.  Note that
+ *	if it is called in interrupt context it cannot disable the restart
+ *	tasklets as it cannot wait, however the tasklets will have no effect
+ *	since the doorbells are disabled and the driver will call this again
+ *	later from process context, at which time the tasklets will be stopped
+ *	if they are still running.
+ */
+void t3_sge_stop(struct adapter *adap)
+{
+	t3_set_reg_field(adap, A_SG_CONTROL, F_GLOBALENABLE, 0);
+	if (!in_interrupt()) {
+		int i;
+
+		for (i = 0; i < SGE_QSETS; ++i) {
+			struct sge_qset *qs = &adap->sge.qs[i];
+
+			tasklet_kill(&qs->txq[TXQ_OFLD].qresume_tsk);
+			tasklet_kill(&qs->txq[TXQ_CTRL].qresume_tsk);
+		}
+	}
+}
+
+/**
+ *	t3_sge_init - initialize SGE
+ *	@adap: the adapter
+ *	@p: the SGE parameters
+ *
+ *	Performs SGE initialization needed every time after a chip reset.
+ *	We do not initialize any of the queue sets here, instead the driver
+ *	top-level must request those individually.  We also do not enable DMA
+ *	here, that should be done after the queues have been set up.
+ */
+void t3_sge_init(struct adapter *adap, struct sge_params *p)
+{
+	unsigned int ctrl, ups = ffs(pci_resource_len(adap->pdev, 2) >> 12);
+
+	ctrl = F_DROPPKT | V_PKTSHIFT(2) | F_FLMODE | F_AVOIDCQOVFL |
+	    F_CQCRDTCTRL |
+	    V_HOSTPAGESIZE(PAGE_SHIFT - 11) | F_BIGENDIANINGRESS |
+	    V_USERSPACESIZE(ups ? ups - 1 : 0) | F_ISCSICOALESCING;
+#if SGE_NUM_GENBITS == 1
+	ctrl |= F_EGRGENCTRL;
+#endif
+	if (adap->params.rev > 0) {
+		if (!(adap->flags & (USING_MSIX | USING_MSI)))
+			ctrl |= F_ONEINTMULTQ | F_OPTONEINTMULTQ;
+		ctrl |= F_CQCRDTCTRL | F_AVOIDCQOVFL;
+	}
+	t3_write_reg(adap, A_SG_CONTROL, ctrl);
+	t3_write_reg(adap, A_SG_EGR_RCQ_DRB_THRSH, V_HIRCQDRBTHRSH(512) |
+		     V_LORCQDRBTHRSH(512));
+	t3_write_reg(adap, A_SG_TIMER_TICK, core_ticks_per_usec(adap) / 10);
+	t3_write_reg(adap, A_SG_CMDQ_CREDIT_TH, V_THRESHOLD(32) |
+		     V_TIMEOUT(100 * core_ticks_per_usec(adap)));
+	t3_write_reg(adap, A_SG_HI_DRB_HI_THRSH, 1000);
+	t3_write_reg(adap, A_SG_HI_DRB_LO_THRSH, 256);
+	t3_write_reg(adap, A_SG_LO_DRB_HI_THRSH, 1000);
+	t3_write_reg(adap, A_SG_LO_DRB_LO_THRSH, 256);
+	t3_write_reg(adap, A_SG_OCO_BASE, V_BASE1(0xfff));
+	t3_write_reg(adap, A_SG_DRB_PRI_THRESH, 63 * 1024);
+}
+
+/**
+ *	t3_sge_prep - one-time SGE initialization
+ *	@adap: the associated adapter
+ *	@p: SGE parameters
+ *
+ *	Performs one-time initialization of SGE SW state.  Includes determining
+ *	defaults for the assorted SGE parameters, which admins can change until
+ *	they are used to initialize the SGE.
+ */
+void __devinit t3_sge_prep(struct adapter *adap, struct sge_params *p)
+{
+	int i;
+
+	p->max_pkt_size = (16 * 1024) - sizeof(struct cpl_rx_data) -
+	    SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
+
+	for (i = 0; i < SGE_QSETS; ++i) {
+		struct qset_params *q = p->qset + i;
+
+		q->polling = adap->params.rev > 0;
+		q->coalesce_usecs = 5;
+		q->rspq_size = 1024;
+		q->fl_size = 4096;
+		q->jumbo_size = 512;
+		q->txq_size[TXQ_ETH] = 1024;
+		q->txq_size[TXQ_OFLD] = 1024;
+		q->txq_size[TXQ_CTRL] = 256;
+		q->cong_thres = 0;
+	}
+
+	spin_lock_init(&adap->sge.reg_lock);
+}
+
+/**
+ *	t3_get_desc - dump an SGE descriptor for debugging purposes
+ *	@qs: the queue set
+ *	@qnum: identifies the specific queue (0..2: Tx, 3:response, 4..5: Rx)
+ *	@idx: the descriptor index in the queue
+ *	@data: where to dump the descriptor contents
+ *
+ *	Dumps the contents of a HW descriptor of an SGE queue.  Returns the
+ *	size of the descriptor.
+ */
+int t3_get_desc(const struct sge_qset *qs, unsigned int qnum, unsigned int idx,
+		unsigned char *data)
+{
+	if (qnum >= 6)
+		return -EINVAL;
+
+	if (qnum < 3) {
+		if (!qs->txq[qnum].desc || idx >= qs->txq[qnum].size)
+			return -EINVAL;
+		memcpy(data, &qs->txq[qnum].desc[idx], sizeof(struct tx_desc));
+		return sizeof(struct tx_desc);
+	}
+
+	if (qnum == 3) {
+		if (!qs->rspq.desc || idx >= qs->rspq.size)
+			return -EINVAL;
+		memcpy(data, &qs->rspq.desc[idx], sizeof(struct rsp_desc));
+		return sizeof(struct rsp_desc);
+	}
+
+	qnum -= 4;
+	if (!qs->fl[qnum].desc || idx >= qs->fl[qnum].size)
+		return -EINVAL;
+	memcpy(data, &qs->fl[qnum].desc[idx], sizeof(struct rx_desc));
+	return sizeof(struct rx_desc);
+}
diff --git a/drivers/net/cxgb3/sge_defs.h b/drivers/net/cxgb3/sge_defs.h
new file mode 100644
index 00000000000..514869e26a7
--- /dev/null
+++ b/drivers/net/cxgb3/sge_defs.h
@@ -0,0 +1,251 @@
+/*
+ * This file is automatically generated --- any changes will be lost.
+ */
+
+#ifndef _SGE_DEFS_H
+#define _SGE_DEFS_H
+
+#define S_EC_CREDITS    0
+#define M_EC_CREDITS    0x7FFF
+#define V_EC_CREDITS(x) ((x) << S_EC_CREDITS)
+#define G_EC_CREDITS(x) (((x) >> S_EC_CREDITS) & M_EC_CREDITS)
+
+#define S_EC_GTS    15
+#define V_EC_GTS(x) ((x) << S_EC_GTS)
+#define F_EC_GTS    V_EC_GTS(1U)
+
+#define S_EC_INDEX    16
+#define M_EC_INDEX    0xFFFF
+#define V_EC_INDEX(x) ((x) << S_EC_INDEX)
+#define G_EC_INDEX(x) (((x) >> S_EC_INDEX) & M_EC_INDEX)
+
+#define S_EC_SIZE    0
+#define M_EC_SIZE    0xFFFF
+#define V_EC_SIZE(x) ((x) << S_EC_SIZE)
+#define G_EC_SIZE(x) (((x) >> S_EC_SIZE) & M_EC_SIZE)
+
+#define S_EC_BASE_LO    16
+#define M_EC_BASE_LO    0xFFFF
+#define V_EC_BASE_LO(x) ((x) << S_EC_BASE_LO)
+#define G_EC_BASE_LO(x) (((x) >> S_EC_BASE_LO) & M_EC_BASE_LO)
+
+#define S_EC_BASE_HI    0
+#define M_EC_BASE_HI    0xF
+#define V_EC_BASE_HI(x) ((x) << S_EC_BASE_HI)
+#define G_EC_BASE_HI(x) (((x) >> S_EC_BASE_HI) & M_EC_BASE_HI)
+
+#define S_EC_RESPQ    4
+#define M_EC_RESPQ    0x7
+#define V_EC_RESPQ(x) ((x) << S_EC_RESPQ)
+#define G_EC_RESPQ(x) (((x) >> S_EC_RESPQ) & M_EC_RESPQ)
+
+#define S_EC_TYPE    7
+#define M_EC_TYPE    0x7
+#define V_EC_TYPE(x) ((x) << S_EC_TYPE)
+#define G_EC_TYPE(x) (((x) >> S_EC_TYPE) & M_EC_TYPE)
+
+#define S_EC_GEN    10
+#define V_EC_GEN(x) ((x) << S_EC_GEN)
+#define F_EC_GEN    V_EC_GEN(1U)
+
+#define S_EC_UP_TOKEN    11
+#define M_EC_UP_TOKEN    0xFFFFF
+#define V_EC_UP_TOKEN(x) ((x) << S_EC_UP_TOKEN)
+#define G_EC_UP_TOKEN(x) (((x) >> S_EC_UP_TOKEN) & M_EC_UP_TOKEN)
+
+#define S_EC_VALID    31
+#define V_EC_VALID(x) ((x) << S_EC_VALID)
+#define F_EC_VALID    V_EC_VALID(1U)
+
+#define S_RQ_MSI_VEC    20
+#define M_RQ_MSI_VEC    0x3F
+#define V_RQ_MSI_VEC(x) ((x) << S_RQ_MSI_VEC)
+#define G_RQ_MSI_VEC(x) (((x) >> S_RQ_MSI_VEC) & M_RQ_MSI_VEC)
+
+#define S_RQ_INTR_EN    26
+#define V_RQ_INTR_EN(x) ((x) << S_RQ_INTR_EN)
+#define F_RQ_INTR_EN    V_RQ_INTR_EN(1U)
+
+#define S_RQ_GEN    28
+#define V_RQ_GEN(x) ((x) << S_RQ_GEN)
+#define F_RQ_GEN    V_RQ_GEN(1U)
+
+#define S_CQ_INDEX    0
+#define M_CQ_INDEX    0xFFFF
+#define V_CQ_INDEX(x) ((x) << S_CQ_INDEX)
+#define G_CQ_INDEX(x) (((x) >> S_CQ_INDEX) & M_CQ_INDEX)
+
+#define S_CQ_SIZE    16
+#define M_CQ_SIZE    0xFFFF
+#define V_CQ_SIZE(x) ((x) << S_CQ_SIZE)
+#define G_CQ_SIZE(x) (((x) >> S_CQ_SIZE) & M_CQ_SIZE)
+
+#define S_CQ_BASE_HI    0
+#define M_CQ_BASE_HI    0xFFFFF
+#define V_CQ_BASE_HI(x) ((x) << S_CQ_BASE_HI)
+#define G_CQ_BASE_HI(x) (((x) >> S_CQ_BASE_HI) & M_CQ_BASE_HI)
+
+#define S_CQ_RSPQ    20
+#define M_CQ_RSPQ    0x3F
+#define V_CQ_RSPQ(x) ((x) << S_CQ_RSPQ)
+#define G_CQ_RSPQ(x) (((x) >> S_CQ_RSPQ) & M_CQ_RSPQ)
+
+#define S_CQ_ASYNC_NOTIF    26
+#define V_CQ_ASYNC_NOTIF(x) ((x) << S_CQ_ASYNC_NOTIF)
+#define F_CQ_ASYNC_NOTIF    V_CQ_ASYNC_NOTIF(1U)
+
+#define S_CQ_ARMED    27
+#define V_CQ_ARMED(x) ((x) << S_CQ_ARMED)
+#define F_CQ_ARMED    V_CQ_ARMED(1U)
+
+#define S_CQ_ASYNC_NOTIF_SOL    28
+#define V_CQ_ASYNC_NOTIF_SOL(x) ((x) << S_CQ_ASYNC_NOTIF_SOL)
+#define F_CQ_ASYNC_NOTIF_SOL    V_CQ_ASYNC_NOTIF_SOL(1U)
+
+#define S_CQ_GEN    29
+#define V_CQ_GEN(x) ((x) << S_CQ_GEN)
+#define F_CQ_GEN    V_CQ_GEN(1U)
+
+#define S_CQ_OVERFLOW_MODE    31
+#define V_CQ_OVERFLOW_MODE(x) ((x) << S_CQ_OVERFLOW_MODE)
+#define F_CQ_OVERFLOW_MODE    V_CQ_OVERFLOW_MODE(1U)
+
+#define S_CQ_CREDITS    0
+#define M_CQ_CREDITS    0xFFFF
+#define V_CQ_CREDITS(x) ((x) << S_CQ_CREDITS)
+#define G_CQ_CREDITS(x) (((x) >> S_CQ_CREDITS) & M_CQ_CREDITS)
+
+#define S_CQ_CREDIT_THRES    16
+#define M_CQ_CREDIT_THRES    0x1FFF
+#define V_CQ_CREDIT_THRES(x) ((x) << S_CQ_CREDIT_THRES)
+#define G_CQ_CREDIT_THRES(x) (((x) >> S_CQ_CREDIT_THRES) & M_CQ_CREDIT_THRES)
+
+#define S_FL_BASE_HI    0
+#define M_FL_BASE_HI    0xFFFFF
+#define V_FL_BASE_HI(x) ((x) << S_FL_BASE_HI)
+#define G_FL_BASE_HI(x) (((x) >> S_FL_BASE_HI) & M_FL_BASE_HI)
+
+#define S_FL_INDEX_LO    20
+#define M_FL_INDEX_LO    0xFFF
+#define V_FL_INDEX_LO(x) ((x) << S_FL_INDEX_LO)
+#define G_FL_INDEX_LO(x) (((x) >> S_FL_INDEX_LO) & M_FL_INDEX_LO)
+
+#define S_FL_INDEX_HI    0
+#define M_FL_INDEX_HI    0xF
+#define V_FL_INDEX_HI(x) ((x) << S_FL_INDEX_HI)
+#define G_FL_INDEX_HI(x) (((x) >> S_FL_INDEX_HI) & M_FL_INDEX_HI)
+
+#define S_FL_SIZE    4
+#define M_FL_SIZE    0xFFFF
+#define V_FL_SIZE(x) ((x) << S_FL_SIZE)
+#define G_FL_SIZE(x) (((x) >> S_FL_SIZE) & M_FL_SIZE)
+
+#define S_FL_GEN    20
+#define V_FL_GEN(x) ((x) << S_FL_GEN)
+#define F_FL_GEN    V_FL_GEN(1U)
+
+#define S_FL_ENTRY_SIZE_LO    21
+#define M_FL_ENTRY_SIZE_LO    0x7FF
+#define V_FL_ENTRY_SIZE_LO(x) ((x) << S_FL_ENTRY_SIZE_LO)
+#define G_FL_ENTRY_SIZE_LO(x) (((x) >> S_FL_ENTRY_SIZE_LO) & M_FL_ENTRY_SIZE_LO)
+
+#define S_FL_ENTRY_SIZE_HI    0
+#define M_FL_ENTRY_SIZE_HI    0x1FFFFF
+#define V_FL_ENTRY_SIZE_HI(x) ((x) << S_FL_ENTRY_SIZE_HI)
+#define G_FL_ENTRY_SIZE_HI(x) (((x) >> S_FL_ENTRY_SIZE_HI) & M_FL_ENTRY_SIZE_HI)
+
+#define S_FL_CONG_THRES    21
+#define M_FL_CONG_THRES    0x3FF
+#define V_FL_CONG_THRES(x) ((x) << S_FL_CONG_THRES)
+#define G_FL_CONG_THRES(x) (((x) >> S_FL_CONG_THRES) & M_FL_CONG_THRES)
+
+#define S_FL_GTS    31
+#define V_FL_GTS(x) ((x) << S_FL_GTS)
+#define F_FL_GTS    V_FL_GTS(1U)
+
+#define S_FLD_GEN1    31
+#define V_FLD_GEN1(x) ((x) << S_FLD_GEN1)
+#define F_FLD_GEN1    V_FLD_GEN1(1U)
+
+#define S_FLD_GEN2    0
+#define V_FLD_GEN2(x) ((x) << S_FLD_GEN2)
+#define F_FLD_GEN2    V_FLD_GEN2(1U)
+
+#define S_RSPD_TXQ1_CR    0
+#define M_RSPD_TXQ1_CR    0x7F
+#define V_RSPD_TXQ1_CR(x) ((x) << S_RSPD_TXQ1_CR)
+#define G_RSPD_TXQ1_CR(x) (((x) >> S_RSPD_TXQ1_CR) & M_RSPD_TXQ1_CR)
+
+#define S_RSPD_TXQ1_GTS    7
+#define V_RSPD_TXQ1_GTS(x) ((x) << S_RSPD_TXQ1_GTS)
+#define F_RSPD_TXQ1_GTS    V_RSPD_TXQ1_GTS(1U)
+
+#define S_RSPD_TXQ2_CR    8
+#define M_RSPD_TXQ2_CR    0x7F
+#define V_RSPD_TXQ2_CR(x) ((x) << S_RSPD_TXQ2_CR)
+#define G_RSPD_TXQ2_CR(x) (((x) >> S_RSPD_TXQ2_CR) & M_RSPD_TXQ2_CR)
+
+#define S_RSPD_TXQ2_GTS    15
+#define V_RSPD_TXQ2_GTS(x) ((x) << S_RSPD_TXQ2_GTS)
+#define F_RSPD_TXQ2_GTS    V_RSPD_TXQ2_GTS(1U)
+
+#define S_RSPD_TXQ0_CR    16
+#define M_RSPD_TXQ0_CR    0x7F
+#define V_RSPD_TXQ0_CR(x) ((x) << S_RSPD_TXQ0_CR)
+#define G_RSPD_TXQ0_CR(x) (((x) >> S_RSPD_TXQ0_CR) & M_RSPD_TXQ0_CR)
+
+#define S_RSPD_TXQ0_GTS    23
+#define V_RSPD_TXQ0_GTS(x) ((x) << S_RSPD_TXQ0_GTS)
+#define F_RSPD_TXQ0_GTS    V_RSPD_TXQ0_GTS(1U)
+
+#define S_RSPD_EOP    24
+#define V_RSPD_EOP(x) ((x) << S_RSPD_EOP)
+#define F_RSPD_EOP    V_RSPD_EOP(1U)
+
+#define S_RSPD_SOP    25
+#define V_RSPD_SOP(x) ((x) << S_RSPD_SOP)
+#define F_RSPD_SOP    V_RSPD_SOP(1U)
+
+#define S_RSPD_ASYNC_NOTIF    26
+#define V_RSPD_ASYNC_NOTIF(x) ((x) << S_RSPD_ASYNC_NOTIF)
+#define F_RSPD_ASYNC_NOTIF    V_RSPD_ASYNC_NOTIF(1U)
+
+#define S_RSPD_FL0_GTS    27
+#define V_RSPD_FL0_GTS(x) ((x) << S_RSPD_FL0_GTS)
+#define F_RSPD_FL0_GTS    V_RSPD_FL0_GTS(1U)
+
+#define S_RSPD_FL1_GTS    28
+#define V_RSPD_FL1_GTS(x) ((x) << S_RSPD_FL1_GTS)
+#define F_RSPD_FL1_GTS    V_RSPD_FL1_GTS(1U)
+
+#define S_RSPD_IMM_DATA_VALID    29
+#define V_RSPD_IMM_DATA_VALID(x) ((x) << S_RSPD_IMM_DATA_VALID)
+#define F_RSPD_IMM_DATA_VALID    V_RSPD_IMM_DATA_VALID(1U)
+
+#define S_RSPD_OFFLOAD    30
+#define V_RSPD_OFFLOAD(x) ((x) << S_RSPD_OFFLOAD)
+#define F_RSPD_OFFLOAD    V_RSPD_OFFLOAD(1U)
+
+#define S_RSPD_GEN1    31
+#define V_RSPD_GEN1(x) ((x) << S_RSPD_GEN1)
+#define F_RSPD_GEN1    V_RSPD_GEN1(1U)
+
+#define S_RSPD_LEN    0
+#define M_RSPD_LEN    0x7FFFFFFF
+#define V_RSPD_LEN(x) ((x) << S_RSPD_LEN)
+#define G_RSPD_LEN(x) (((x) >> S_RSPD_LEN) & M_RSPD_LEN)
+
+#define S_RSPD_FLQ    31
+#define V_RSPD_FLQ(x) ((x) << S_RSPD_FLQ)
+#define F_RSPD_FLQ    V_RSPD_FLQ(1U)
+
+#define S_RSPD_GEN2    0
+#define V_RSPD_GEN2(x) ((x) << S_RSPD_GEN2)
+#define F_RSPD_GEN2    V_RSPD_GEN2(1U)
+
+#define S_RSPD_INR_VEC    1
+#define M_RSPD_INR_VEC    0x7F
+#define V_RSPD_INR_VEC(x) ((x) << S_RSPD_INR_VEC)
+#define G_RSPD_INR_VEC(x) (((x) >> S_RSPD_INR_VEC) & M_RSPD_INR_VEC)
+
+#endif				/* _SGE_DEFS_H */
diff --git a/drivers/net/cxgb3/t3_cpl.h b/drivers/net/cxgb3/t3_cpl.h
new file mode 100644
index 00000000000..b0df4ba94e0
--- /dev/null
+++ b/drivers/net/cxgb3/t3_cpl.h
@@ -0,0 +1,1426 @@
+/*
+ * Definitions of the CPL 5 commands and status codes.
+ *
+ * Copyright (C) 2004-2006 Chelsio Communications.  All rights reserved.
+ *
+ * Written by Dimitris Michailidis (dm@chelsio.com)
+ *
+ * This program is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the LICENSE file included in this
+ * release for licensing terms and conditions.
+ */
+
+#ifndef T3_CPL_H
+#define T3_CPL_H
+
+#if !defined(__LITTLE_ENDIAN_BITFIELD) && !defined(__BIG_ENDIAN_BITFIELD)
+# include <asm/byteorder.h>
+#endif
+
+enum CPL_opcode {
+	CPL_PASS_OPEN_REQ = 0x1,
+	CPL_PASS_ACCEPT_RPL = 0x2,
+	CPL_ACT_OPEN_REQ = 0x3,
+	CPL_SET_TCB = 0x4,
+	CPL_SET_TCB_FIELD = 0x5,
+	CPL_GET_TCB = 0x6,
+	CPL_PCMD = 0x7,
+	CPL_CLOSE_CON_REQ = 0x8,
+	CPL_CLOSE_LISTSRV_REQ = 0x9,
+	CPL_ABORT_REQ = 0xA,
+	CPL_ABORT_RPL = 0xB,
+	CPL_TX_DATA = 0xC,
+	CPL_RX_DATA_ACK = 0xD,
+	CPL_TX_PKT = 0xE,
+	CPL_RTE_DELETE_REQ = 0xF,
+	CPL_RTE_WRITE_REQ = 0x10,
+	CPL_RTE_READ_REQ = 0x11,
+	CPL_L2T_WRITE_REQ = 0x12,
+	CPL_L2T_READ_REQ = 0x13,
+	CPL_SMT_WRITE_REQ = 0x14,
+	CPL_SMT_READ_REQ = 0x15,
+	CPL_TX_PKT_LSO = 0x16,
+	CPL_PCMD_READ = 0x17,
+	CPL_BARRIER = 0x18,
+	CPL_TID_RELEASE = 0x1A,
+
+	CPL_CLOSE_LISTSRV_RPL = 0x20,
+	CPL_ERROR = 0x21,
+	CPL_GET_TCB_RPL = 0x22,
+	CPL_L2T_WRITE_RPL = 0x23,
+	CPL_PCMD_READ_RPL = 0x24,
+	CPL_PCMD_RPL = 0x25,
+	CPL_PEER_CLOSE = 0x26,
+	CPL_RTE_DELETE_RPL = 0x27,
+	CPL_RTE_WRITE_RPL = 0x28,
+	CPL_RX_DDP_COMPLETE = 0x29,
+	CPL_RX_PHYS_ADDR = 0x2A,
+	CPL_RX_PKT = 0x2B,
+	CPL_RX_URG_NOTIFY = 0x2C,
+	CPL_SET_TCB_RPL = 0x2D,
+	CPL_SMT_WRITE_RPL = 0x2E,
+	CPL_TX_DATA_ACK = 0x2F,
+
+	CPL_ABORT_REQ_RSS = 0x30,
+	CPL_ABORT_RPL_RSS = 0x31,
+	CPL_CLOSE_CON_RPL = 0x32,
+	CPL_ISCSI_HDR = 0x33,
+	CPL_L2T_READ_RPL = 0x34,
+	CPL_RDMA_CQE = 0x35,
+	CPL_RDMA_CQE_READ_RSP = 0x36,
+	CPL_RDMA_CQE_ERR = 0x37,
+	CPL_RTE_READ_RPL = 0x38,
+	CPL_RX_DATA = 0x39,
+
+	CPL_ACT_OPEN_RPL = 0x40,
+	CPL_PASS_OPEN_RPL = 0x41,
+	CPL_RX_DATA_DDP = 0x42,
+	CPL_SMT_READ_RPL = 0x43,
+
+	CPL_ACT_ESTABLISH = 0x50,
+	CPL_PASS_ESTABLISH = 0x51,
+
+	CPL_PASS_ACCEPT_REQ = 0x70,
+
+	CPL_ASYNC_NOTIF = 0x80,	/* fake opcode for async notifications */
+
+	CPL_TX_DMA_ACK = 0xA0,
+	CPL_RDMA_READ_REQ = 0xA1,
+	CPL_RDMA_TERMINATE = 0xA2,
+	CPL_TRACE_PKT = 0xA3,
+	CPL_RDMA_EC_STATUS = 0xA5,
+
+	NUM_CPL_CMDS		/* must be last and previous entries must be sorted */
+};
+
+enum CPL_error {
+	CPL_ERR_NONE = 0,
+	CPL_ERR_TCAM_PARITY = 1,
+	CPL_ERR_TCAM_FULL = 3,
+	CPL_ERR_CONN_RESET = 20,
+	CPL_ERR_CONN_EXIST = 22,
+	CPL_ERR_ARP_MISS = 23,
+	CPL_ERR_BAD_SYN = 24,
+	CPL_ERR_CONN_TIMEDOUT = 30,
+	CPL_ERR_XMIT_TIMEDOUT = 31,
+	CPL_ERR_PERSIST_TIMEDOUT = 32,
+	CPL_ERR_FINWAIT2_TIMEDOUT = 33,
+	CPL_ERR_KEEPALIVE_TIMEDOUT = 34,
+	CPL_ERR_RTX_NEG_ADVICE = 35,
+	CPL_ERR_PERSIST_NEG_ADVICE = 36,
+	CPL_ERR_ABORT_FAILED = 42,
+	CPL_ERR_GENERAL = 99
+};
+
+enum {
+	CPL_CONN_POLICY_AUTO = 0,
+	CPL_CONN_POLICY_ASK = 1,
+	CPL_CONN_POLICY_DENY = 3
+};
+
+enum {
+	ULP_MODE_NONE = 0,
+	ULP_MODE_ISCSI = 2,
+	ULP_MODE_RDMA = 4,
+	ULP_MODE_TCPDDP = 5
+};
+
+enum {
+	ULP_CRC_HEADER = 1 << 0,
+	ULP_CRC_DATA = 1 << 1
+};
+
+enum {
+	CPL_PASS_OPEN_ACCEPT,
+	CPL_PASS_OPEN_REJECT
+};
+
+enum {
+	CPL_ABORT_SEND_RST = 0,
+	CPL_ABORT_NO_RST,
+	CPL_ABORT_POST_CLOSE_REQ = 2
+};
+
+enum {				/* TX_PKT_LSO ethernet types */
+	CPL_ETH_II,
+	CPL_ETH_II_VLAN,
+	CPL_ETH_802_3,
+	CPL_ETH_802_3_VLAN
+};
+
+enum {				/* TCP congestion control algorithms */
+	CONG_ALG_RENO,
+	CONG_ALG_TAHOE,
+	CONG_ALG_NEWRENO,
+	CONG_ALG_HIGHSPEED
+};
+
+union opcode_tid {
+	__be32 opcode_tid;
+	__u8 opcode;
+};
+
+#define S_OPCODE 24
+#define V_OPCODE(x) ((x) << S_OPCODE)
+#define G_OPCODE(x) (((x) >> S_OPCODE) & 0xFF)
+#define G_TID(x)    ((x) & 0xFFFFFF)
+
+/* tid is assumed to be 24-bits */
+#define MK_OPCODE_TID(opcode, tid) (V_OPCODE(opcode) | (tid))
+
+#define OPCODE_TID(cmd) ((cmd)->ot.opcode_tid)
+
+/* extract the TID from a CPL command */
+#define GET_TID(cmd) (G_TID(ntohl(OPCODE_TID(cmd))))
+
+struct tcp_options {
+	__be16 mss;
+	__u8 wsf;
+#if defined(__LITTLE_ENDIAN_BITFIELD)
+	 __u8:5;
+	__u8 ecn:1;
+	__u8 sack:1;
+	__u8 tstamp:1;
+#else
+	__u8 tstamp:1;
+	__u8 sack:1;
+	__u8 ecn:1;
+	 __u8:5;
+#endif
+};
+
+struct rss_header {
+	__u8 opcode;
+#if defined(__LITTLE_ENDIAN_BITFIELD)
+	__u8 cpu_idx:6;
+	__u8 hash_type:2;
+#else
+	__u8 hash_type:2;
+	__u8 cpu_idx:6;
+#endif
+	__be16 cq_idx;
+	__be32 rss_hash_val;
+};
+
+#ifndef CHELSIO_FW
+struct work_request_hdr {
+	__be32 wr_hi;
+	__be32 wr_lo;
+};
+
+/* wr_hi fields */
+#define S_WR_SGE_CREDITS    0
+#define M_WR_SGE_CREDITS    0xFF
+#define V_WR_SGE_CREDITS(x) ((x) << S_WR_SGE_CREDITS)
+#define G_WR_SGE_CREDITS(x) (((x) >> S_WR_SGE_CREDITS) & M_WR_SGE_CREDITS)
+
+#define S_WR_SGLSFLT    8
+#define M_WR_SGLSFLT    0xFF
+#define V_WR_SGLSFLT(x) ((x) << S_WR_SGLSFLT)
+#define G_WR_SGLSFLT(x) (((x) >> S_WR_SGLSFLT) & M_WR_SGLSFLT)
+
+#define S_WR_BCNTLFLT    16
+#define M_WR_BCNTLFLT    0xF
+#define V_WR_BCNTLFLT(x) ((x) << S_WR_BCNTLFLT)
+#define G_WR_BCNTLFLT(x) (((x) >> S_WR_BCNTLFLT) & M_WR_BCNTLFLT)
+
+#define S_WR_DATATYPE    20
+#define V_WR_DATATYPE(x) ((x) << S_WR_DATATYPE)
+#define F_WR_DATATYPE    V_WR_DATATYPE(1U)
+
+#define S_WR_COMPL    21
+#define V_WR_COMPL(x) ((x) << S_WR_COMPL)
+#define F_WR_COMPL    V_WR_COMPL(1U)
+
+#define S_WR_EOP    22
+#define V_WR_EOP(x) ((x) << S_WR_EOP)
+#define F_WR_EOP    V_WR_EOP(1U)
+
+#define S_WR_SOP    23
+#define V_WR_SOP(x) ((x) << S_WR_SOP)
+#define F_WR_SOP    V_WR_SOP(1U)
+
+#define S_WR_OP    24
+#define M_WR_OP    0xFF
+#define V_WR_OP(x) ((x) << S_WR_OP)
+#define G_WR_OP(x) (((x) >> S_WR_OP) & M_WR_OP)
+
+/* wr_lo fields */
+#define S_WR_LEN    0
+#define M_WR_LEN    0xFF
+#define V_WR_LEN(x) ((x) << S_WR_LEN)
+#define G_WR_LEN(x) (((x) >> S_WR_LEN) & M_WR_LEN)
+
+#define S_WR_TID    8
+#define M_WR_TID    0xFFFFF
+#define V_WR_TID(x) ((x) << S_WR_TID)
+#define G_WR_TID(x) (((x) >> S_WR_TID) & M_WR_TID)
+
+#define S_WR_CR_FLUSH    30
+#define V_WR_CR_FLUSH(x) ((x) << S_WR_CR_FLUSH)
+#define F_WR_CR_FLUSH    V_WR_CR_FLUSH(1U)
+
+#define S_WR_GEN    31
+#define V_WR_GEN(x) ((x) << S_WR_GEN)
+#define F_WR_GEN    V_WR_GEN(1U)
+
+# define WR_HDR struct work_request_hdr wr
+# define RSS_HDR
+#else
+# define WR_HDR
+# define RSS_HDR struct rss_header rss_hdr;
+#endif
+
+/* option 0 lower-half fields */
+#define S_CPL_STATUS    0
+#define M_CPL_STATUS    0xFF
+#define V_CPL_STATUS(x) ((x) << S_CPL_STATUS)
+#define G_CPL_STATUS(x) (((x) >> S_CPL_STATUS) & M_CPL_STATUS)
+
+#define S_INJECT_TIMER    6
+#define V_INJECT_TIMER(x) ((x) << S_INJECT_TIMER)
+#define F_INJECT_TIMER    V_INJECT_TIMER(1U)
+
+#define S_NO_OFFLOAD    7
+#define V_NO_OFFLOAD(x) ((x) << S_NO_OFFLOAD)
+#define F_NO_OFFLOAD    V_NO_OFFLOAD(1U)
+
+#define S_ULP_MODE    8
+#define M_ULP_MODE    0xF
+#define V_ULP_MODE(x) ((x) << S_ULP_MODE)
+#define G_ULP_MODE(x) (((x) >> S_ULP_MODE) & M_ULP_MODE)
+
+#define S_RCV_BUFSIZ    12
+#define M_RCV_BUFSIZ    0x3FFF
+#define V_RCV_BUFSIZ(x) ((x) << S_RCV_BUFSIZ)
+#define G_RCV_BUFSIZ(x) (((x) >> S_RCV_BUFSIZ) & M_RCV_BUFSIZ)
+
+#define S_TOS    26
+#define M_TOS    0x3F
+#define V_TOS(x) ((x) << S_TOS)
+#define G_TOS(x) (((x) >> S_TOS) & M_TOS)
+
+/* option 0 upper-half fields */
+#define S_DELACK    0
+#define V_DELACK(x) ((x) << S_DELACK)
+#define F_DELACK    V_DELACK(1U)
+
+#define S_NO_CONG    1
+#define V_NO_CONG(x) ((x) << S_NO_CONG)
+#define F_NO_CONG    V_NO_CONG(1U)
+
+#define S_SRC_MAC_SEL    2
+#define M_SRC_MAC_SEL    0x3
+#define V_SRC_MAC_SEL(x) ((x) << S_SRC_MAC_SEL)
+#define G_SRC_MAC_SEL(x) (((x) >> S_SRC_MAC_SEL) & M_SRC_MAC_SEL)
+
+#define S_L2T_IDX    4
+#define M_L2T_IDX    0x7FF
+#define V_L2T_IDX(x) ((x) << S_L2T_IDX)
+#define G_L2T_IDX(x) (((x) >> S_L2T_IDX) & M_L2T_IDX)
+
+#define S_TX_CHANNEL    15
+#define V_TX_CHANNEL(x) ((x) << S_TX_CHANNEL)
+#define F_TX_CHANNEL    V_TX_CHANNEL(1U)
+
+#define S_TCAM_BYPASS    16
+#define V_TCAM_BYPASS(x) ((x) << S_TCAM_BYPASS)
+#define F_TCAM_BYPASS    V_TCAM_BYPASS(1U)
+
+#define S_NAGLE    17
+#define V_NAGLE(x) ((x) << S_NAGLE)
+#define F_NAGLE    V_NAGLE(1U)
+
+#define S_WND_SCALE    18
+#define M_WND_SCALE    0xF
+#define V_WND_SCALE(x) ((x) << S_WND_SCALE)
+#define G_WND_SCALE(x) (((x) >> S_WND_SCALE) & M_WND_SCALE)
+
+#define S_KEEP_ALIVE    22
+#define V_KEEP_ALIVE(x) ((x) << S_KEEP_ALIVE)
+#define F_KEEP_ALIVE    V_KEEP_ALIVE(1U)
+
+#define S_MAX_RETRANS    23
+#define M_MAX_RETRANS    0xF
+#define V_MAX_RETRANS(x) ((x) << S_MAX_RETRANS)
+#define G_MAX_RETRANS(x) (((x) >> S_MAX_RETRANS) & M_MAX_RETRANS)
+
+#define S_MAX_RETRANS_OVERRIDE    27
+#define V_MAX_RETRANS_OVERRIDE(x) ((x) << S_MAX_RETRANS_OVERRIDE)
+#define F_MAX_RETRANS_OVERRIDE    V_MAX_RETRANS_OVERRIDE(1U)
+
+#define S_MSS_IDX    28
+#define M_MSS_IDX    0xF
+#define V_MSS_IDX(x) ((x) << S_MSS_IDX)
+#define G_MSS_IDX(x) (((x) >> S_MSS_IDX) & M_MSS_IDX)
+
+/* option 1 fields */
+#define S_RSS_ENABLE    0
+#define V_RSS_ENABLE(x) ((x) << S_RSS_ENABLE)
+#define F_RSS_ENABLE    V_RSS_ENABLE(1U)
+
+#define S_RSS_MASK_LEN    1
+#define M_RSS_MASK_LEN    0x7
+#define V_RSS_MASK_LEN(x) ((x) << S_RSS_MASK_LEN)
+#define G_RSS_MASK_LEN(x) (((x) >> S_RSS_MASK_LEN) & M_RSS_MASK_LEN)
+
+#define S_CPU_IDX    4
+#define M_CPU_IDX    0x3F
+#define V_CPU_IDX(x) ((x) << S_CPU_IDX)
+#define G_CPU_IDX(x) (((x) >> S_CPU_IDX) & M_CPU_IDX)
+
+#define S_MAC_MATCH_VALID    18
+#define V_MAC_MATCH_VALID(x) ((x) << S_MAC_MATCH_VALID)
+#define F_MAC_MATCH_VALID    V_MAC_MATCH_VALID(1U)
+
+#define S_CONN_POLICY    19
+#define M_CONN_POLICY    0x3
+#define V_CONN_POLICY(x) ((x) << S_CONN_POLICY)
+#define G_CONN_POLICY(x) (((x) >> S_CONN_POLICY) & M_CONN_POLICY)
+
+#define S_SYN_DEFENSE    21
+#define V_SYN_DEFENSE(x) ((x) << S_SYN_DEFENSE)
+#define F_SYN_DEFENSE    V_SYN_DEFENSE(1U)
+
+#define S_VLAN_PRI    22
+#define M_VLAN_PRI    0x3
+#define V_VLAN_PRI(x) ((x) << S_VLAN_PRI)
+#define G_VLAN_PRI(x) (((x) >> S_VLAN_PRI) & M_VLAN_PRI)
+
+#define S_VLAN_PRI_VALID    24
+#define V_VLAN_PRI_VALID(x) ((x) << S_VLAN_PRI_VALID)
+#define F_VLAN_PRI_VALID    V_VLAN_PRI_VALID(1U)
+
+#define S_PKT_TYPE    25
+#define M_PKT_TYPE    0x3
+#define V_PKT_TYPE(x) ((x) << S_PKT_TYPE)
+#define G_PKT_TYPE(x) (((x) >> S_PKT_TYPE) & M_PKT_TYPE)
+
+#define S_MAC_MATCH    27
+#define M_MAC_MATCH    0x1F
+#define V_MAC_MATCH(x) ((x) << S_MAC_MATCH)
+#define G_MAC_MATCH(x) (((x) >> S_MAC_MATCH) & M_MAC_MATCH)
+
+/* option 2 fields */
+#define S_CPU_INDEX    0
+#define M_CPU_INDEX    0x7F
+#define V_CPU_INDEX(x) ((x) << S_CPU_INDEX)
+#define G_CPU_INDEX(x) (((x) >> S_CPU_INDEX) & M_CPU_INDEX)
+
+#define S_CPU_INDEX_VALID    7
+#define V_CPU_INDEX_VALID(x) ((x) << S_CPU_INDEX_VALID)
+#define F_CPU_INDEX_VALID    V_CPU_INDEX_VALID(1U)
+
+#define S_RX_COALESCE    8
+#define M_RX_COALESCE    0x3
+#define V_RX_COALESCE(x) ((x) << S_RX_COALESCE)
+#define G_RX_COALESCE(x) (((x) >> S_RX_COALESCE) & M_RX_COALESCE)
+
+#define S_RX_COALESCE_VALID    10
+#define V_RX_COALESCE_VALID(x) ((x) << S_RX_COALESCE_VALID)
+#define F_RX_COALESCE_VALID    V_RX_COALESCE_VALID(1U)
+
+#define S_CONG_CONTROL_FLAVOR    11
+#define M_CONG_CONTROL_FLAVOR    0x3
+#define V_CONG_CONTROL_FLAVOR(x) ((x) << S_CONG_CONTROL_FLAVOR)
+#define G_CONG_CONTROL_FLAVOR(x) (((x) >> S_CONG_CONTROL_FLAVOR) & M_CONG_CONTROL_FLAVOR)
+
+#define S_PACING_FLAVOR    13
+#define M_PACING_FLAVOR    0x3
+#define V_PACING_FLAVOR(x) ((x) << S_PACING_FLAVOR)
+#define G_PACING_FLAVOR(x) (((x) >> S_PACING_FLAVOR) & M_PACING_FLAVOR)
+
+#define S_FLAVORS_VALID    15
+#define V_FLAVORS_VALID(x) ((x) << S_FLAVORS_VALID)
+#define F_FLAVORS_VALID    V_FLAVORS_VALID(1U)
+
+#define S_RX_FC_DISABLE    16
+#define V_RX_FC_DISABLE(x) ((x) << S_RX_FC_DISABLE)
+#define F_RX_FC_DISABLE    V_RX_FC_DISABLE(1U)
+
+#define S_RX_FC_VALID    17
+#define V_RX_FC_VALID(x) ((x) << S_RX_FC_VALID)
+#define F_RX_FC_VALID    V_RX_FC_VALID(1U)
+
+struct cpl_pass_open_req {
+	WR_HDR;
+	union opcode_tid ot;
+	__be16 local_port;
+	__be16 peer_port;
+	__be32 local_ip;
+	__be32 peer_ip;
+	__be32 opt0h;
+	__be32 opt0l;
+	__be32 peer_netmask;
+	__be32 opt1;
+};
+
+struct cpl_pass_open_rpl {
+	RSS_HDR union opcode_tid ot;
+	__be16 local_port;
+	__be16 peer_port;
+	__be32 local_ip;
+	__be32 peer_ip;
+	__u8 resvd[7];
+	__u8 status;
+};
+
+struct cpl_pass_establish {
+	RSS_HDR union opcode_tid ot;
+	__be16 local_port;
+	__be16 peer_port;
+	__be32 local_ip;
+	__be32 peer_ip;
+	__be32 tos_tid;
+	__be16 l2t_idx;
+	__be16 tcp_opt;
+	__be32 snd_isn;
+	__be32 rcv_isn;
+};
+
+/* cpl_pass_establish.tos_tid fields */
+#define S_PASS_OPEN_TID    0
+#define M_PASS_OPEN_TID    0xFFFFFF
+#define V_PASS_OPEN_TID(x) ((x) << S_PASS_OPEN_TID)
+#define G_PASS_OPEN_TID(x) (((x) >> S_PASS_OPEN_TID) & M_PASS_OPEN_TID)
+
+#define S_PASS_OPEN_TOS    24
+#define M_PASS_OPEN_TOS    0xFF
+#define V_PASS_OPEN_TOS(x) ((x) << S_PASS_OPEN_TOS)
+#define G_PASS_OPEN_TOS(x) (((x) >> S_PASS_OPEN_TOS) & M_PASS_OPEN_TOS)
+
+/* cpl_pass_establish.l2t_idx fields */
+#define S_L2T_IDX16    5
+#define M_L2T_IDX16    0x7FF
+#define V_L2T_IDX16(x) ((x) << S_L2T_IDX16)
+#define G_L2T_IDX16(x) (((x) >> S_L2T_IDX16) & M_L2T_IDX16)
+
+/* cpl_pass_establish.tcp_opt fields (also applies act_open_establish) */
+#define G_TCPOPT_WSCALE_OK(x)  (((x) >> 5) & 1)
+#define G_TCPOPT_SACK(x)       (((x) >> 6) & 1)
+#define G_TCPOPT_TSTAMP(x)     (((x) >> 7) & 1)
+#define G_TCPOPT_SND_WSCALE(x) (((x) >> 8) & 0xf)
+#define G_TCPOPT_MSS(x)        (((x) >> 12) & 0xf)
+
+struct cpl_pass_accept_req {
+	RSS_HDR union opcode_tid ot;
+	__be16 local_port;
+	__be16 peer_port;
+	__be32 local_ip;
+	__be32 peer_ip;
+	__be32 tos_tid;
+	struct tcp_options tcp_options;
+	__u8 dst_mac[6];
+	__be16 vlan_tag;
+	__u8 src_mac[6];
+#if defined(__LITTLE_ENDIAN_BITFIELD)
+	 __u8:3;
+	__u8 addr_idx:3;
+	__u8 port_idx:1;
+	__u8 exact_match:1;
+#else
+	__u8 exact_match:1;
+	__u8 port_idx:1;
+	__u8 addr_idx:3;
+	 __u8:3;
+#endif
+	__u8 rsvd;
+	__be32 rcv_isn;
+	__be32 rsvd2;
+};
+
+struct cpl_pass_accept_rpl {
+	WR_HDR;
+	union opcode_tid ot;
+	__be32 opt2;
+	__be32 rsvd;
+	__be32 peer_ip;
+	__be32 opt0h;
+	__be32 opt0l_status;
+};
+
+struct cpl_act_open_req {
+	WR_HDR;
+	union opcode_tid ot;
+	__be16 local_port;
+	__be16 peer_port;
+	__be32 local_ip;
+	__be32 peer_ip;
+	__be32 opt0h;
+	__be32 opt0l;
+	__be32 params;
+	__be32 opt2;
+};
+
+/* cpl_act_open_req.params fields */
+#define S_AOPEN_VLAN_PRI    9
+#define M_AOPEN_VLAN_PRI    0x3
+#define V_AOPEN_VLAN_PRI(x) ((x) << S_AOPEN_VLAN_PRI)
+#define G_AOPEN_VLAN_PRI(x) (((x) >> S_AOPEN_VLAN_PRI) & M_AOPEN_VLAN_PRI)
+
+#define S_AOPEN_VLAN_PRI_VALID    11
+#define V_AOPEN_VLAN_PRI_VALID(x) ((x) << S_AOPEN_VLAN_PRI_VALID)
+#define F_AOPEN_VLAN_PRI_VALID    V_AOPEN_VLAN_PRI_VALID(1U)
+
+#define S_AOPEN_PKT_TYPE    12
+#define M_AOPEN_PKT_TYPE    0x3
+#define V_AOPEN_PKT_TYPE(x) ((x) << S_AOPEN_PKT_TYPE)
+#define G_AOPEN_PKT_TYPE(x) (((x) >> S_AOPEN_PKT_TYPE) & M_AOPEN_PKT_TYPE)
+
+#define S_AOPEN_MAC_MATCH    14
+#define M_AOPEN_MAC_MATCH    0x1F
+#define V_AOPEN_MAC_MATCH(x) ((x) << S_AOPEN_MAC_MATCH)
+#define G_AOPEN_MAC_MATCH(x) (((x) >> S_AOPEN_MAC_MATCH) & M_AOPEN_MAC_MATCH)
+
+#define S_AOPEN_MAC_MATCH_VALID    19
+#define V_AOPEN_MAC_MATCH_VALID(x) ((x) << S_AOPEN_MAC_MATCH_VALID)
+#define F_AOPEN_MAC_MATCH_VALID    V_AOPEN_MAC_MATCH_VALID(1U)
+
+#define S_AOPEN_IFF_VLAN    20
+#define M_AOPEN_IFF_VLAN    0xFFF
+#define V_AOPEN_IFF_VLAN(x) ((x) << S_AOPEN_IFF_VLAN)
+#define G_AOPEN_IFF_VLAN(x) (((x) >> S_AOPEN_IFF_VLAN) & M_AOPEN_IFF_VLAN)
+
+struct cpl_act_open_rpl {
+	RSS_HDR union opcode_tid ot;
+	__be16 local_port;
+	__be16 peer_port;
+	__be32 local_ip;
+	__be32 peer_ip;
+	__be32 atid;
+	__u8 rsvd[3];
+	__u8 status;
+};
+
+struct cpl_act_establish {
+	RSS_HDR union opcode_tid ot;
+	__be16 local_port;
+	__be16 peer_port;
+	__be32 local_ip;
+	__be32 peer_ip;
+	__be32 tos_tid;
+	__be16 l2t_idx;
+	__be16 tcp_opt;
+	__be32 snd_isn;
+	__be32 rcv_isn;
+};
+
+struct cpl_get_tcb {
+	WR_HDR;
+	union opcode_tid ot;
+	__be16 cpuno;
+	__be16 rsvd;
+};
+
+struct cpl_get_tcb_rpl {
+	RSS_HDR union opcode_tid ot;
+	__u8 rsvd;
+	__u8 status;
+	__be16 len;
+};
+
+struct cpl_set_tcb {
+	WR_HDR;
+	union opcode_tid ot;
+	__u8 reply;
+	__u8 cpu_idx;
+	__be16 len;
+};
+
+/* cpl_set_tcb.reply fields */
+#define S_NO_REPLY    7
+#define V_NO_REPLY(x) ((x) << S_NO_REPLY)
+#define F_NO_REPLY    V_NO_REPLY(1U)
+
+struct cpl_set_tcb_field {
+	WR_HDR;
+	union opcode_tid ot;
+	__u8 reply;
+	__u8 cpu_idx;
+	__be16 word;
+	__be64 mask;
+	__be64 val;
+};
+
+struct cpl_set_tcb_rpl {
+	RSS_HDR union opcode_tid ot;
+	__u8 rsvd[3];
+	__u8 status;
+};
+
+struct cpl_pcmd {
+	WR_HDR;
+	union opcode_tid ot;
+	__u8 rsvd[3];
+#if defined(__LITTLE_ENDIAN_BITFIELD)
+	__u8 src:1;
+	__u8 bundle:1;
+	__u8 channel:1;
+	 __u8:5;
+#else
+	 __u8:5;
+	__u8 channel:1;
+	__u8 bundle:1;
+	__u8 src:1;
+#endif
+	__be32 pcmd_parm[2];
+};
+
+struct cpl_pcmd_reply {
+	RSS_HDR union opcode_tid ot;
+	__u8 status;
+	__u8 rsvd;
+	__be16 len;
+};
+
+struct cpl_close_con_req {
+	WR_HDR;
+	union opcode_tid ot;
+	__be32 rsvd;
+};
+
+struct cpl_close_con_rpl {
+	RSS_HDR union opcode_tid ot;
+	__u8 rsvd[3];
+	__u8 status;
+	__be32 snd_nxt;
+	__be32 rcv_nxt;
+};
+
+struct cpl_close_listserv_req {
+	WR_HDR;
+	union opcode_tid ot;
+	__u8 rsvd0;
+	__u8 cpu_idx;
+	__be16 rsvd1;
+};
+
+struct cpl_close_listserv_rpl {
+	RSS_HDR union opcode_tid ot;
+	__u8 rsvd[3];
+	__u8 status;
+};
+
+struct cpl_abort_req_rss {
+	RSS_HDR union opcode_tid ot;
+	__be32 rsvd0;
+	__u8 rsvd1;
+	__u8 status;
+	__u8 rsvd2[6];
+};
+
+struct cpl_abort_req {
+	WR_HDR;
+	union opcode_tid ot;
+	__be32 rsvd0;
+	__u8 rsvd1;
+	__u8 cmd;
+	__u8 rsvd2[6];
+};
+
+struct cpl_abort_rpl_rss {
+	RSS_HDR union opcode_tid ot;
+	__be32 rsvd0;
+	__u8 rsvd1;
+	__u8 status;
+	__u8 rsvd2[6];
+};
+
+struct cpl_abort_rpl {
+	WR_HDR;
+	union opcode_tid ot;
+	__be32 rsvd0;
+	__u8 rsvd1;
+	__u8 cmd;
+	__u8 rsvd2[6];
+};
+
+struct cpl_peer_close {
+	RSS_HDR union opcode_tid ot;
+	__be32 rcv_nxt;
+};
+
+struct tx_data_wr {
+	__be32 wr_hi;
+	__be32 wr_lo;
+	__be32 len;
+	__be32 flags;
+	__be32 sndseq;
+	__be32 param;
+};
+
+/* tx_data_wr.param fields */
+#define S_TX_PORT    0
+#define M_TX_PORT    0x7
+#define V_TX_PORT(x) ((x) << S_TX_PORT)
+#define G_TX_PORT(x) (((x) >> S_TX_PORT) & M_TX_PORT)
+
+#define S_TX_MSS    4
+#define M_TX_MSS    0xF
+#define V_TX_MSS(x) ((x) << S_TX_MSS)
+#define G_TX_MSS(x) (((x) >> S_TX_MSS) & M_TX_MSS)
+
+#define S_TX_QOS    8
+#define M_TX_QOS    0xFF
+#define V_TX_QOS(x) ((x) << S_TX_QOS)
+#define G_TX_QOS(x) (((x) >> S_TX_QOS) & M_TX_QOS)
+
+#define S_TX_SNDBUF 16
+#define M_TX_SNDBUF 0xFFFF
+#define V_TX_SNDBUF(x) ((x) << S_TX_SNDBUF)
+#define G_TX_SNDBUF(x) (((x) >> S_TX_SNDBUF) & M_TX_SNDBUF)
+
+struct cpl_tx_data {
+	union opcode_tid ot;
+	__be32 len;
+	__be32 rsvd;
+	__be16 urg;
+	__be16 flags;
+};
+
+/* cpl_tx_data.flags fields */
+#define S_TX_ULP_SUBMODE    6
+#define M_TX_ULP_SUBMODE    0xF
+#define V_TX_ULP_SUBMODE(x) ((x) << S_TX_ULP_SUBMODE)
+#define G_TX_ULP_SUBMODE(x) (((x) >> S_TX_ULP_SUBMODE) & M_TX_ULP_SUBMODE)
+
+#define S_TX_ULP_MODE    10
+#define M_TX_ULP_MODE    0xF
+#define V_TX_ULP_MODE(x) ((x) << S_TX_ULP_MODE)
+#define G_TX_ULP_MODE(x) (((x) >> S_TX_ULP_MODE) & M_TX_ULP_MODE)
+
+#define S_TX_SHOVE    14
+#define V_TX_SHOVE(x) ((x) << S_TX_SHOVE)
+#define F_TX_SHOVE    V_TX_SHOVE(1U)
+
+#define S_TX_MORE    15
+#define V_TX_MORE(x) ((x) << S_TX_MORE)
+#define F_TX_MORE    V_TX_MORE(1U)
+
+/* additional tx_data_wr.flags fields */
+#define S_TX_CPU_IDX    0
+#define M_TX_CPU_IDX    0x3F
+#define V_TX_CPU_IDX(x) ((x) << S_TX_CPU_IDX)
+#define G_TX_CPU_IDX(x) (((x) >> S_TX_CPU_IDX) & M_TX_CPU_IDX)
+
+#define S_TX_URG    16
+#define V_TX_URG(x) ((x) << S_TX_URG)
+#define F_TX_URG    V_TX_URG(1U)
+
+#define S_TX_CLOSE    17
+#define V_TX_CLOSE(x) ((x) << S_TX_CLOSE)
+#define F_TX_CLOSE    V_TX_CLOSE(1U)
+
+#define S_TX_INIT    18
+#define V_TX_INIT(x) ((x) << S_TX_INIT)
+#define F_TX_INIT    V_TX_INIT(1U)
+
+#define S_TX_IMM_ACK    19
+#define V_TX_IMM_ACK(x) ((x) << S_TX_IMM_ACK)
+#define F_TX_IMM_ACK    V_TX_IMM_ACK(1U)
+
+#define S_TX_IMM_DMA    20
+#define V_TX_IMM_DMA(x) ((x) << S_TX_IMM_DMA)
+#define F_TX_IMM_DMA    V_TX_IMM_DMA(1U)
+
+struct cpl_tx_data_ack {
+	RSS_HDR union opcode_tid ot;
+	__be32 ack_seq;
+};
+
+struct cpl_wr_ack {
+	RSS_HDR union opcode_tid ot;
+	__be16 credits;
+	__be16 rsvd;
+	__be32 snd_nxt;
+	__be32 snd_una;
+};
+
+struct cpl_rdma_ec_status {
+	RSS_HDR union opcode_tid ot;
+	__u8 rsvd[3];
+	__u8 status;
+};
+
+struct mngt_pktsched_wr {
+	__be32 wr_hi;
+	__be32 wr_lo;
+	__u8 mngt_opcode;
+	__u8 rsvd[7];
+	__u8 sched;
+	__u8 idx;
+	__u8 min;
+	__u8 max;
+	__u8 binding;
+	__u8 rsvd1[3];
+};
+
+struct cpl_iscsi_hdr {
+	RSS_HDR union opcode_tid ot;
+	__be16 pdu_len_ddp;
+	__be16 len;
+	__be32 seq;
+	__be16 urg;
+	__u8 rsvd;
+	__u8 status;
+};
+
+/* cpl_iscsi_hdr.pdu_len_ddp fields */
+#define S_ISCSI_PDU_LEN    0
+#define M_ISCSI_PDU_LEN    0x7FFF
+#define V_ISCSI_PDU_LEN(x) ((x) << S_ISCSI_PDU_LEN)
+#define G_ISCSI_PDU_LEN(x) (((x) >> S_ISCSI_PDU_LEN) & M_ISCSI_PDU_LEN)
+
+#define S_ISCSI_DDP    15
+#define V_ISCSI_DDP(x) ((x) << S_ISCSI_DDP)
+#define F_ISCSI_DDP    V_ISCSI_DDP(1U)
+
+struct cpl_rx_data {
+	RSS_HDR union opcode_tid ot;
+	__be16 rsvd;
+	__be16 len;
+	__be32 seq;
+	__be16 urg;
+#if defined(__LITTLE_ENDIAN_BITFIELD)
+	__u8 dack_mode:2;
+	__u8 psh:1;
+	__u8 heartbeat:1;
+	 __u8:4;
+#else
+	 __u8:4;
+	__u8 heartbeat:1;
+	__u8 psh:1;
+	__u8 dack_mode:2;
+#endif
+	__u8 status;
+};
+
+struct cpl_rx_data_ack {
+	WR_HDR;
+	union opcode_tid ot;
+	__be32 credit_dack;
+};
+
+/* cpl_rx_data_ack.ack_seq fields */
+#define S_RX_CREDITS    0
+#define M_RX_CREDITS    0x7FFFFFF
+#define V_RX_CREDITS(x) ((x) << S_RX_CREDITS)
+#define G_RX_CREDITS(x) (((x) >> S_RX_CREDITS) & M_RX_CREDITS)
+
+#define S_RX_MODULATE    27
+#define V_RX_MODULATE(x) ((x) << S_RX_MODULATE)
+#define F_RX_MODULATE    V_RX_MODULATE(1U)
+
+#define S_RX_FORCE_ACK    28
+#define V_RX_FORCE_ACK(x) ((x) << S_RX_FORCE_ACK)
+#define F_RX_FORCE_ACK    V_RX_FORCE_ACK(1U)
+
+#define S_RX_DACK_MODE    29
+#define M_RX_DACK_MODE    0x3
+#define V_RX_DACK_MODE(x) ((x) << S_RX_DACK_MODE)
+#define G_RX_DACK_MODE(x) (((x) >> S_RX_DACK_MODE) & M_RX_DACK_MODE)
+
+#define S_RX_DACK_CHANGE    31
+#define V_RX_DACK_CHANGE(x) ((x) << S_RX_DACK_CHANGE)
+#define F_RX_DACK_CHANGE    V_RX_DACK_CHANGE(1U)
+
+struct cpl_rx_urg_notify {
+	RSS_HDR union opcode_tid ot;
+	__be32 seq;
+};
+
+struct cpl_rx_ddp_complete {
+	RSS_HDR union opcode_tid ot;
+	__be32 ddp_report;
+};
+
+struct cpl_rx_data_ddp {
+	RSS_HDR union opcode_tid ot;
+	__be16 urg;
+	__be16 len;
+	__be32 seq;
+	union {
+		__be32 nxt_seq;
+		__be32 ddp_report;
+	};
+	__be32 ulp_crc;
+	__be32 ddpvld_status;
+};
+
+/* cpl_rx_data_ddp.ddpvld_status fields */
+#define S_DDP_STATUS    0
+#define M_DDP_STATUS    0xFF
+#define V_DDP_STATUS(x) ((x) << S_DDP_STATUS)
+#define G_DDP_STATUS(x) (((x) >> S_DDP_STATUS) & M_DDP_STATUS)
+
+#define S_DDP_VALID    15
+#define M_DDP_VALID    0x1FFFF
+#define V_DDP_VALID(x) ((x) << S_DDP_VALID)
+#define G_DDP_VALID(x) (((x) >> S_DDP_VALID) & M_DDP_VALID)
+
+#define S_DDP_PPOD_MISMATCH    15
+#define V_DDP_PPOD_MISMATCH(x) ((x) << S_DDP_PPOD_MISMATCH)
+#define F_DDP_PPOD_MISMATCH    V_DDP_PPOD_MISMATCH(1U)
+
+#define S_DDP_PDU    16
+#define V_DDP_PDU(x) ((x) << S_DDP_PDU)
+#define F_DDP_PDU    V_DDP_PDU(1U)
+
+#define S_DDP_LLIMIT_ERR    17
+#define V_DDP_LLIMIT_ERR(x) ((x) << S_DDP_LLIMIT_ERR)
+#define F_DDP_LLIMIT_ERR    V_DDP_LLIMIT_ERR(1U)
+
+#define S_DDP_PPOD_PARITY_ERR    18
+#define V_DDP_PPOD_PARITY_ERR(x) ((x) << S_DDP_PPOD_PARITY_ERR)
+#define F_DDP_PPOD_PARITY_ERR    V_DDP_PPOD_PARITY_ERR(1U)
+
+#define S_DDP_PADDING_ERR    19
+#define V_DDP_PADDING_ERR(x) ((x) << S_DDP_PADDING_ERR)
+#define F_DDP_PADDING_ERR    V_DDP_PADDING_ERR(1U)
+
+#define S_DDP_HDRCRC_ERR    20
+#define V_DDP_HDRCRC_ERR(x) ((x) << S_DDP_HDRCRC_ERR)
+#define F_DDP_HDRCRC_ERR    V_DDP_HDRCRC_ERR(1U)
+
+#define S_DDP_DATACRC_ERR    21
+#define V_DDP_DATACRC_ERR(x) ((x) << S_DDP_DATACRC_ERR)
+#define F_DDP_DATACRC_ERR    V_DDP_DATACRC_ERR(1U)
+
+#define S_DDP_INVALID_TAG    22
+#define V_DDP_INVALID_TAG(x) ((x) << S_DDP_INVALID_TAG)
+#define F_DDP_INVALID_TAG    V_DDP_INVALID_TAG(1U)
+
+#define S_DDP_ULIMIT_ERR    23
+#define V_DDP_ULIMIT_ERR(x) ((x) << S_DDP_ULIMIT_ERR)
+#define F_DDP_ULIMIT_ERR    V_DDP_ULIMIT_ERR(1U)
+
+#define S_DDP_OFFSET_ERR    24
+#define V_DDP_OFFSET_ERR(x) ((x) << S_DDP_OFFSET_ERR)
+#define F_DDP_OFFSET_ERR    V_DDP_OFFSET_ERR(1U)
+
+#define S_DDP_COLOR_ERR    25
+#define V_DDP_COLOR_ERR(x) ((x) << S_DDP_COLOR_ERR)
+#define F_DDP_COLOR_ERR    V_DDP_COLOR_ERR(1U)
+
+#define S_DDP_TID_MISMATCH    26
+#define V_DDP_TID_MISMATCH(x) ((x) << S_DDP_TID_MISMATCH)
+#define F_DDP_TID_MISMATCH    V_DDP_TID_MISMATCH(1U)
+
+#define S_DDP_INVALID_PPOD    27
+#define V_DDP_INVALID_PPOD(x) ((x) << S_DDP_INVALID_PPOD)
+#define F_DDP_INVALID_PPOD    V_DDP_INVALID_PPOD(1U)
+
+#define S_DDP_ULP_MODE    28
+#define M_DDP_ULP_MODE    0xF
+#define V_DDP_ULP_MODE(x) ((x) << S_DDP_ULP_MODE)
+#define G_DDP_ULP_MODE(x) (((x) >> S_DDP_ULP_MODE) & M_DDP_ULP_MODE)
+
+/* cpl_rx_data_ddp.ddp_report fields */
+#define S_DDP_OFFSET    0
+#define M_DDP_OFFSET    0x3FFFFF
+#define V_DDP_OFFSET(x) ((x) << S_DDP_OFFSET)
+#define G_DDP_OFFSET(x) (((x) >> S_DDP_OFFSET) & M_DDP_OFFSET)
+
+#define S_DDP_URG    24
+#define V_DDP_URG(x) ((x) << S_DDP_URG)
+#define F_DDP_URG    V_DDP_URG(1U)
+
+#define S_DDP_PSH    25
+#define V_DDP_PSH(x) ((x) << S_DDP_PSH)
+#define F_DDP_PSH    V_DDP_PSH(1U)
+
+#define S_DDP_BUF_COMPLETE    26
+#define V_DDP_BUF_COMPLETE(x) ((x) << S_DDP_BUF_COMPLETE)
+#define F_DDP_BUF_COMPLETE    V_DDP_BUF_COMPLETE(1U)
+
+#define S_DDP_BUF_TIMED_OUT    27
+#define V_DDP_BUF_TIMED_OUT(x) ((x) << S_DDP_BUF_TIMED_OUT)
+#define F_DDP_BUF_TIMED_OUT    V_DDP_BUF_TIMED_OUT(1U)
+
+#define S_DDP_BUF_IDX    28
+#define V_DDP_BUF_IDX(x) ((x) << S_DDP_BUF_IDX)
+#define F_DDP_BUF_IDX    V_DDP_BUF_IDX(1U)
+
+struct cpl_tx_pkt {
+	WR_HDR;
+	__be32 cntrl;
+	__be32 len;
+};
+
+struct cpl_tx_pkt_lso {
+	WR_HDR;
+	__be32 cntrl;
+	__be32 len;
+
+	__be32 rsvd;
+	__be32 lso_info;
+};
+
+/* cpl_tx_pkt*.cntrl fields */
+#define S_TXPKT_VLAN    0
+#define M_TXPKT_VLAN    0xFFFF
+#define V_TXPKT_VLAN(x) ((x) << S_TXPKT_VLAN)
+#define G_TXPKT_VLAN(x) (((x) >> S_TXPKT_VLAN) & M_TXPKT_VLAN)
+
+#define S_TXPKT_INTF    16
+#define M_TXPKT_INTF    0xF
+#define V_TXPKT_INTF(x) ((x) << S_TXPKT_INTF)
+#define G_TXPKT_INTF(x) (((x) >> S_TXPKT_INTF) & M_TXPKT_INTF)
+
+#define S_TXPKT_IPCSUM_DIS    20
+#define V_TXPKT_IPCSUM_DIS(x) ((x) << S_TXPKT_IPCSUM_DIS)
+#define F_TXPKT_IPCSUM_DIS    V_TXPKT_IPCSUM_DIS(1U)
+
+#define S_TXPKT_L4CSUM_DIS    21
+#define V_TXPKT_L4CSUM_DIS(x) ((x) << S_TXPKT_L4CSUM_DIS)
+#define F_TXPKT_L4CSUM_DIS    V_TXPKT_L4CSUM_DIS(1U)
+
+#define S_TXPKT_VLAN_VLD    22
+#define V_TXPKT_VLAN_VLD(x) ((x) << S_TXPKT_VLAN_VLD)
+#define F_TXPKT_VLAN_VLD    V_TXPKT_VLAN_VLD(1U)
+
+#define S_TXPKT_LOOPBACK    23
+#define V_TXPKT_LOOPBACK(x) ((x) << S_TXPKT_LOOPBACK)
+#define F_TXPKT_LOOPBACK    V_TXPKT_LOOPBACK(1U)
+
+#define S_TXPKT_OPCODE    24
+#define M_TXPKT_OPCODE    0xFF
+#define V_TXPKT_OPCODE(x) ((x) << S_TXPKT_OPCODE)
+#define G_TXPKT_OPCODE(x) (((x) >> S_TXPKT_OPCODE) & M_TXPKT_OPCODE)
+
+/* cpl_tx_pkt_lso.lso_info fields */
+#define S_LSO_MSS    0
+#define M_LSO_MSS    0x3FFF
+#define V_LSO_MSS(x) ((x) << S_LSO_MSS)
+#define G_LSO_MSS(x) (((x) >> S_LSO_MSS) & M_LSO_MSS)
+
+#define S_LSO_ETH_TYPE    14
+#define M_LSO_ETH_TYPE    0x3
+#define V_LSO_ETH_TYPE(x) ((x) << S_LSO_ETH_TYPE)
+#define G_LSO_ETH_TYPE(x) (((x) >> S_LSO_ETH_TYPE) & M_LSO_ETH_TYPE)
+
+#define S_LSO_TCPHDR_WORDS    16
+#define M_LSO_TCPHDR_WORDS    0xF
+#define V_LSO_TCPHDR_WORDS(x) ((x) << S_LSO_TCPHDR_WORDS)
+#define G_LSO_TCPHDR_WORDS(x) (((x) >> S_LSO_TCPHDR_WORDS) & M_LSO_TCPHDR_WORDS)
+
+#define S_LSO_IPHDR_WORDS    20
+#define M_LSO_IPHDR_WORDS    0xF
+#define V_LSO_IPHDR_WORDS(x) ((x) << S_LSO_IPHDR_WORDS)
+#define G_LSO_IPHDR_WORDS(x) (((x) >> S_LSO_IPHDR_WORDS) & M_LSO_IPHDR_WORDS)
+
+#define S_LSO_IPV6    24
+#define V_LSO_IPV6(x) ((x) << S_LSO_IPV6)
+#define F_LSO_IPV6    V_LSO_IPV6(1U)
+
+struct cpl_trace_pkt {
+#ifdef CHELSIO_FW
+	__u8 rss_opcode;
+#if defined(__LITTLE_ENDIAN_BITFIELD)
+	__u8 err:1;
+	 __u8:7;
+#else
+	 __u8:7;
+	__u8 err:1;
+#endif
+	__u8 rsvd0;
+#if defined(__LITTLE_ENDIAN_BITFIELD)
+	__u8 qid:4;
+	 __u8:4;
+#else
+	 __u8:4;
+	__u8 qid:4;
+#endif
+	__be32 tstamp;
+#endif				/* CHELSIO_FW */
+
+	__u8 opcode;
+#if defined(__LITTLE_ENDIAN_BITFIELD)
+	__u8 iff:4;
+	 __u8:4;
+#else
+	 __u8:4;
+	__u8 iff:4;
+#endif
+	__u8 rsvd[4];
+	__be16 len;
+};
+
+struct cpl_rx_pkt {
+	RSS_HDR __u8 opcode;
+#if defined(__LITTLE_ENDIAN_BITFIELD)
+	__u8 iff:4;
+	__u8 csum_valid:1;
+	__u8 ipmi_pkt:1;
+	__u8 vlan_valid:1;
+	__u8 fragment:1;
+#else
+	__u8 fragment:1;
+	__u8 vlan_valid:1;
+	__u8 ipmi_pkt:1;
+	__u8 csum_valid:1;
+	__u8 iff:4;
+#endif
+	__be16 csum;
+	__be16 vlan;
+	__be16 len;
+};
+
+struct cpl_l2t_write_req {
+	WR_HDR;
+	union opcode_tid ot;
+	__be32 params;
+	__u8 rsvd[2];
+	__u8 dst_mac[6];
+};
+
+/* cpl_l2t_write_req.params fields */
+#define S_L2T_W_IDX    0
+#define M_L2T_W_IDX    0x7FF
+#define V_L2T_W_IDX(x) ((x) << S_L2T_W_IDX)
+#define G_L2T_W_IDX(x) (((x) >> S_L2T_W_IDX) & M_L2T_W_IDX)
+
+#define S_L2T_W_VLAN    11
+#define M_L2T_W_VLAN    0xFFF
+#define V_L2T_W_VLAN(x) ((x) << S_L2T_W_VLAN)
+#define G_L2T_W_VLAN(x) (((x) >> S_L2T_W_VLAN) & M_L2T_W_VLAN)
+
+#define S_L2T_W_IFF    23
+#define M_L2T_W_IFF    0xF
+#define V_L2T_W_IFF(x) ((x) << S_L2T_W_IFF)
+#define G_L2T_W_IFF(x) (((x) >> S_L2T_W_IFF) & M_L2T_W_IFF)
+
+#define S_L2T_W_PRIO    27
+#define M_L2T_W_PRIO    0x7
+#define V_L2T_W_PRIO(x) ((x) << S_L2T_W_PRIO)
+#define G_L2T_W_PRIO(x) (((x) >> S_L2T_W_PRIO) & M_L2T_W_PRIO)
+
+struct cpl_l2t_write_rpl {
+	RSS_HDR union opcode_tid ot;
+	__u8 status;
+	__u8 rsvd[3];
+};
+
+struct cpl_l2t_read_req {
+	WR_HDR;
+	union opcode_tid ot;
+	__be16 rsvd;
+	__be16 l2t_idx;
+};
+
+struct cpl_l2t_read_rpl {
+	RSS_HDR union opcode_tid ot;
+	__be32 params;
+	__u8 rsvd[2];
+	__u8 dst_mac[6];
+};
+
+/* cpl_l2t_read_rpl.params fields */
+#define S_L2T_R_PRIO    0
+#define M_L2T_R_PRIO    0x7
+#define V_L2T_R_PRIO(x) ((x) << S_L2T_R_PRIO)
+#define G_L2T_R_PRIO(x) (((x) >> S_L2T_R_PRIO) & M_L2T_R_PRIO)
+
+#define S_L2T_R_VLAN    8
+#define M_L2T_R_VLAN    0xFFF
+#define V_L2T_R_VLAN(x) ((x) << S_L2T_R_VLAN)
+#define G_L2T_R_VLAN(x) (((x) >> S_L2T_R_VLAN) & M_L2T_R_VLAN)
+
+#define S_L2T_R_IFF    20
+#define M_L2T_R_IFF    0xF
+#define V_L2T_R_IFF(x) ((x) << S_L2T_R_IFF)
+#define G_L2T_R_IFF(x) (((x) >> S_L2T_R_IFF) & M_L2T_R_IFF)
+
+#define S_L2T_STATUS    24
+#define M_L2T_STATUS    0xFF
+#define V_L2T_STATUS(x) ((x) << S_L2T_STATUS)
+#define G_L2T_STATUS(x) (((x) >> S_L2T_STATUS) & M_L2T_STATUS)
+
+struct cpl_smt_write_req {
+	WR_HDR;
+	union opcode_tid ot;
+	__u8 rsvd0;
+#if defined(__LITTLE_ENDIAN_BITFIELD)
+	__u8 mtu_idx:4;
+	__u8 iff:4;
+#else
+	__u8 iff:4;
+	__u8 mtu_idx:4;
+#endif
+	__be16 rsvd2;
+	__be16 rsvd3;
+	__u8 src_mac1[6];
+	__be16 rsvd4;
+	__u8 src_mac0[6];
+};
+
+struct cpl_smt_write_rpl {
+	RSS_HDR union opcode_tid ot;
+	__u8 status;
+	__u8 rsvd[3];
+};
+
+struct cpl_smt_read_req {
+	WR_HDR;
+	union opcode_tid ot;
+	__u8 rsvd0;
+#if defined(__LITTLE_ENDIAN_BITFIELD)
+	 __u8:4;
+	__u8 iff:4;
+#else
+	__u8 iff:4;
+	 __u8:4;
+#endif
+	__be16 rsvd2;
+};
+
+struct cpl_smt_read_rpl {
+	RSS_HDR union opcode_tid ot;
+	__u8 status;
+#if defined(__LITTLE_ENDIAN_BITFIELD)
+	__u8 mtu_idx:4;
+	 __u8:4;
+#else
+	 __u8:4;
+	__u8 mtu_idx:4;
+#endif
+	__be16 rsvd2;
+	__be16 rsvd3;
+	__u8 src_mac1[6];
+	__be16 rsvd4;
+	__u8 src_mac0[6];
+};
+
+struct cpl_rte_delete_req {
+	WR_HDR;
+	union opcode_tid ot;
+	__be32 params;
+};
+
+/* { cpl_rte_delete_req, cpl_rte_read_req }.params fields */
+#define S_RTE_REQ_LUT_IX    8
+#define M_RTE_REQ_LUT_IX    0x7FF
+#define V_RTE_REQ_LUT_IX(x) ((x) << S_RTE_REQ_LUT_IX)
+#define G_RTE_REQ_LUT_IX(x) (((x) >> S_RTE_REQ_LUT_IX) & M_RTE_REQ_LUT_IX)
+
+#define S_RTE_REQ_LUT_BASE    19
+#define M_RTE_REQ_LUT_BASE    0x7FF
+#define V_RTE_REQ_LUT_BASE(x) ((x) << S_RTE_REQ_LUT_BASE)
+#define G_RTE_REQ_LUT_BASE(x) (((x) >> S_RTE_REQ_LUT_BASE) & M_RTE_REQ_LUT_BASE)
+
+#define S_RTE_READ_REQ_SELECT    31
+#define V_RTE_READ_REQ_SELECT(x) ((x) << S_RTE_READ_REQ_SELECT)
+#define F_RTE_READ_REQ_SELECT    V_RTE_READ_REQ_SELECT(1U)
+
+struct cpl_rte_delete_rpl {
+	RSS_HDR union opcode_tid ot;
+	__u8 status;
+	__u8 rsvd[3];
+};
+
+struct cpl_rte_write_req {
+	WR_HDR;
+	union opcode_tid ot;
+#if defined(__LITTLE_ENDIAN_BITFIELD)
+	 __u8:6;
+	__u8 write_tcam:1;
+	__u8 write_l2t_lut:1;
+#else
+	__u8 write_l2t_lut:1;
+	__u8 write_tcam:1;
+	 __u8:6;
+#endif
+	__u8 rsvd[3];
+	__be32 lut_params;
+	__be16 rsvd2;
+	__be16 l2t_idx;
+	__be32 netmask;
+	__be32 faddr;
+};
+
+/* cpl_rte_write_req.lut_params fields */
+#define S_RTE_WRITE_REQ_LUT_IX    10
+#define M_RTE_WRITE_REQ_LUT_IX    0x7FF
+#define V_RTE_WRITE_REQ_LUT_IX(x) ((x) << S_RTE_WRITE_REQ_LUT_IX)
+#define G_RTE_WRITE_REQ_LUT_IX(x) (((x) >> S_RTE_WRITE_REQ_LUT_IX) & M_RTE_WRITE_REQ_LUT_IX)
+
+#define S_RTE_WRITE_REQ_LUT_BASE    21
+#define M_RTE_WRITE_REQ_LUT_BASE    0x7FF
+#define V_RTE_WRITE_REQ_LUT_BASE(x) ((x) << S_RTE_WRITE_REQ_LUT_BASE)
+#define G_RTE_WRITE_REQ_LUT_BASE(x) (((x) >> S_RTE_WRITE_REQ_LUT_BASE) & M_RTE_WRITE_REQ_LUT_BASE)
+
+struct cpl_rte_write_rpl {
+	RSS_HDR union opcode_tid ot;
+	__u8 status;
+	__u8 rsvd[3];
+};
+
+struct cpl_rte_read_req {
+	WR_HDR;
+	union opcode_tid ot;
+	__be32 params;
+};
+
+struct cpl_rte_read_rpl {
+	RSS_HDR union opcode_tid ot;
+	__u8 status;
+	__u8 rsvd0;
+	__be16 l2t_idx;
+#if defined(__LITTLE_ENDIAN_BITFIELD)
+	 __u8:7;
+	__u8 select:1;
+#else
+	__u8 select:1;
+	 __u8:7;
+#endif
+	__u8 rsvd2[3];
+	__be32 addr;
+};
+
+struct cpl_tid_release {
+	WR_HDR;
+	union opcode_tid ot;
+	__be32 rsvd;
+};
+
+struct cpl_barrier {
+	WR_HDR;
+	__u8 opcode;
+	__u8 rsvd[7];
+};
+
+struct cpl_rdma_read_req {
+	__u8 opcode;
+	__u8 rsvd[15];
+};
+
+struct cpl_rdma_terminate {
+#ifdef CHELSIO_FW
+	__u8 opcode;
+	__u8 rsvd[2];
+#if defined(__LITTLE_ENDIAN_BITFIELD)
+	__u8 rspq:3;
+	 __u8:5;
+#else
+	 __u8:5;
+	__u8 rspq:3;
+#endif
+	__be32 tid_len;
+#endif
+	__be32 msn;
+	__be32 mo;
+	__u8 data[0];
+};
+
+/* cpl_rdma_terminate.tid_len fields */
+#define S_FLIT_CNT    0
+#define M_FLIT_CNT    0xFF
+#define V_FLIT_CNT(x) ((x) << S_FLIT_CNT)
+#define G_FLIT_CNT(x) (((x) >> S_FLIT_CNT) & M_FLIT_CNT)
+
+#define S_TERM_TID    8
+#define M_TERM_TID    0xFFFFF
+#define V_TERM_TID(x) ((x) << S_TERM_TID)
+#define G_TERM_TID(x) (((x) >> S_TERM_TID) & M_TERM_TID)
+#endif				/* T3_CPL_H */
diff --git a/drivers/net/cxgb3/t3_hw.c b/drivers/net/cxgb3/t3_hw.c
new file mode 100644
index 00000000000..a4e2e57e146
--- /dev/null
+++ b/drivers/net/cxgb3/t3_hw.c
@@ -0,0 +1,3354 @@
+/*
+ * This file is part of the Chelsio T3 Ethernet driver.
+ *
+ * Copyright (C) 2003-2006 Chelsio Communications.  All rights reserved.
+ *
+ * This program is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the LICENSE file included in this
+ * release for licensing terms and conditions.
+ */
+
+#include "common.h"
+#include "regs.h"
+#include "sge_defs.h"
+#include "firmware_exports.h"
+
+ /**
+  *	t3_wait_op_done_val - wait until an operation is completed
+  *	@adapter: the adapter performing the operation
+  *	@reg: the register to check for completion
+  *	@mask: a single-bit field within @reg that indicates completion
+  *	@polarity: the value of the field when the operation is completed
+  *	@attempts: number of check iterations
+  *	@delay: delay in usecs between iterations
+  *	@valp: where to store the value of the register at completion time
+  *
+  *	Wait until an operation is completed by checking a bit in a register
+  *	up to @attempts times.  If @valp is not NULL the value of the register
+  *	at the time it indicated completion is stored there.  Returns 0 if the
+  *	operation completes and -EAGAIN otherwise.
+  */
+
+int t3_wait_op_done_val(struct adapter *adapter, int reg, u32 mask,
+			int polarity, int attempts, int delay, u32 *valp)
+{
+	while (1) {
+		u32 val = t3_read_reg(adapter, reg);
+
+		if (!!(val & mask) == polarity) {
+			if (valp)
+				*valp = val;
+			return 0;
+		}
+		if (--attempts == 0)
+ 			return -EAGAIN;
+		if (delay)
+			udelay(delay);
+	}
+}
+
+/**
+ *	t3_write_regs - write a bunch of registers
+ *	@adapter: the adapter to program
+ *	@p: an array of register address/register value pairs
+ *	@n: the number of address/value pairs
+ *	@offset: register address offset
+ *
+ *	Takes an array of register address/register value pairs and writes each
+ *	value to the corresponding register.  Register addresses are adjusted
+ *	by the supplied offset.
+ */
+void t3_write_regs(struct adapter *adapter, const struct addr_val_pair *p,
+		   int n, unsigned int offset)
+{
+	while (n--) {
+		t3_write_reg(adapter, p->reg_addr + offset, p->val);
+		p++;
+	}
+}
+
+/**
+ *	t3_set_reg_field - set a register field to a value
+ *	@adapter: the adapter to program
+ *	@addr: the register address
+ *	@mask: specifies the portion of the register to modify
+ *	@val: the new value for the register field
+ *
+ *	Sets a register field specified by the supplied mask to the
+ *	given value.
+ */
+void t3_set_reg_field(struct adapter *adapter, unsigned int addr, u32 mask,
+		      u32 val)
+{
+	u32 v = t3_read_reg(adapter, addr) & ~mask;
+
+	t3_write_reg(adapter, addr, v | val);
+	t3_read_reg(adapter, addr);	/* flush */
+}
+
+/**
+ *	t3_read_indirect - read indirectly addressed registers
+ *	@adap: the adapter
+ *	@addr_reg: register holding the indirect address
+ *	@data_reg: register holding the value of the indirect register
+ *	@vals: where the read register values are stored
+ *	@start_idx: index of first indirect register to read
+ *	@nregs: how many indirect registers to read
+ *
+ *	Reads registers that are accessed indirectly through an address/data
+ *	register pair.
+ */
+void t3_read_indirect(struct adapter *adap, unsigned int addr_reg,
+		      unsigned int data_reg, u32 *vals, unsigned int nregs,
+		      unsigned int start_idx)
+{
+	while (nregs--) {
+		t3_write_reg(adap, addr_reg, start_idx);
+		*vals++ = t3_read_reg(adap, data_reg);
+		start_idx++;
+	}
+}
+
+/**
+ *	t3_mc7_bd_read - read from MC7 through backdoor accesses
+ *	@mc7: identifies MC7 to read from
+ *	@start: index of first 64-bit word to read
+ *	@n: number of 64-bit words to read
+ *	@buf: where to store the read result
+ *
+ *	Read n 64-bit words from MC7 starting at word start, using backdoor
+ *	accesses.
+ */
+int t3_mc7_bd_read(struct mc7 *mc7, unsigned int start, unsigned int n,
+		   u64 *buf)
+{
+	static const int shift[] = { 0, 0, 16, 24 };
+	static const int step[] = { 0, 32, 16, 8 };
+
+	unsigned int size64 = mc7->size / 8;	/* # of 64-bit words */
+	struct adapter *adap = mc7->adapter;
+
+	if (start >= size64 || start + n > size64)
+		return -EINVAL;
+
+	start *= (8 << mc7->width);
+	while (n--) {
+		int i;
+		u64 val64 = 0;
+
+		for (i = (1 << mc7->width) - 1; i >= 0; --i) {
+			int attempts = 10;
+			u32 val;
+
+			t3_write_reg(adap, mc7->offset + A_MC7_BD_ADDR, start);
+			t3_write_reg(adap, mc7->offset + A_MC7_BD_OP, 0);
+			val = t3_read_reg(adap, mc7->offset + A_MC7_BD_OP);
+			while ((val & F_BUSY) && attempts--)
+				val = t3_read_reg(adap,
+						  mc7->offset + A_MC7_BD_OP);
+			if (val & F_BUSY)
+				return -EIO;
+
+			val = t3_read_reg(adap, mc7->offset + A_MC7_BD_DATA1);
+			if (mc7->width == 0) {
+				val64 = t3_read_reg(adap,
+						    mc7->offset +
+						    A_MC7_BD_DATA0);
+				val64 |= (u64) val << 32;
+			} else {
+				if (mc7->width > 1)
+					val >>= shift[mc7->width];
+				val64 |= (u64) val << (step[mc7->width] * i);
+			}
+			start += 8;
+		}
+		*buf++ = val64;
+	}
+	return 0;
+}
+
+/*
+ * Initialize MI1.
+ */
+static void mi1_init(struct adapter *adap, const struct adapter_info *ai)
+{
+	u32 clkdiv = adap->params.vpd.cclk / (2 * adap->params.vpd.mdc) - 1;
+	u32 val = F_PREEN | V_MDIINV(ai->mdiinv) | V_MDIEN(ai->mdien) |
+	    V_CLKDIV(clkdiv);
+
+	if (!(ai->caps & SUPPORTED_10000baseT_Full))
+		val |= V_ST(1);
+	t3_write_reg(adap, A_MI1_CFG, val);
+}
+
+#define MDIO_ATTEMPTS 10
+
+/*
+ * MI1 read/write operations for direct-addressed PHYs.
+ */
+static int mi1_read(struct adapter *adapter, int phy_addr, int mmd_addr,
+		    int reg_addr, unsigned int *valp)
+{
+	int ret;
+	u32 addr = V_REGADDR(reg_addr) | V_PHYADDR(phy_addr);
+
+	if (mmd_addr)
+		return -EINVAL;
+
+	mutex_lock(&adapter->mdio_lock);
+	t3_write_reg(adapter, A_MI1_ADDR, addr);
+	t3_write_reg(adapter, A_MI1_OP, V_MDI_OP(2));
+	ret = t3_wait_op_done(adapter, A_MI1_OP, F_BUSY, 0, MDIO_ATTEMPTS, 20);
+	if (!ret)
+		*valp = t3_read_reg(adapter, A_MI1_DATA);
+	mutex_unlock(&adapter->mdio_lock);
+	return ret;
+}
+
+static int mi1_write(struct adapter *adapter, int phy_addr, int mmd_addr,
+		     int reg_addr, unsigned int val)
+{
+	int ret;
+	u32 addr = V_REGADDR(reg_addr) | V_PHYADDR(phy_addr);
+
+	if (mmd_addr)
+		return -EINVAL;
+
+	mutex_lock(&adapter->mdio_lock);
+	t3_write_reg(adapter, A_MI1_ADDR, addr);
+	t3_write_reg(adapter, A_MI1_DATA, val);
+	t3_write_reg(adapter, A_MI1_OP, V_MDI_OP(1));
+	ret = t3_wait_op_done(adapter, A_MI1_OP, F_BUSY, 0, MDIO_ATTEMPTS, 20);
+	mutex_unlock(&adapter->mdio_lock);
+	return ret;
+}
+
+static const struct mdio_ops mi1_mdio_ops = {
+	mi1_read,
+	mi1_write
+};
+
+/*
+ * MI1 read/write operations for indirect-addressed PHYs.
+ */
+static int mi1_ext_read(struct adapter *adapter, int phy_addr, int mmd_addr,
+			int reg_addr, unsigned int *valp)
+{
+	int ret;
+	u32 addr = V_REGADDR(mmd_addr) | V_PHYADDR(phy_addr);
+
+	mutex_lock(&adapter->mdio_lock);
+	t3_write_reg(adapter, A_MI1_ADDR, addr);
+	t3_write_reg(adapter, A_MI1_DATA, reg_addr);
+	t3_write_reg(adapter, A_MI1_OP, V_MDI_OP(0));
+	ret = t3_wait_op_done(adapter, A_MI1_OP, F_BUSY, 0, MDIO_ATTEMPTS, 20);
+	if (!ret) {
+		t3_write_reg(adapter, A_MI1_OP, V_MDI_OP(3));
+		ret = t3_wait_op_done(adapter, A_MI1_OP, F_BUSY, 0,
+				      MDIO_ATTEMPTS, 20);
+		if (!ret)
+			*valp = t3_read_reg(adapter, A_MI1_DATA);
+	}
+	mutex_unlock(&adapter->mdio_lock);
+	return ret;
+}
+
+static int mi1_ext_write(struct adapter *adapter, int phy_addr, int mmd_addr,
+			 int reg_addr, unsigned int val)
+{
+	int ret;
+	u32 addr = V_REGADDR(mmd_addr) | V_PHYADDR(phy_addr);
+
+	mutex_lock(&adapter->mdio_lock);
+	t3_write_reg(adapter, A_MI1_ADDR, addr);
+	t3_write_reg(adapter, A_MI1_DATA, reg_addr);
+	t3_write_reg(adapter, A_MI1_OP, V_MDI_OP(0));
+	ret = t3_wait_op_done(adapter, A_MI1_OP, F_BUSY, 0, MDIO_ATTEMPTS, 20);
+	if (!ret) {
+		t3_write_reg(adapter, A_MI1_DATA, val);
+		t3_write_reg(adapter, A_MI1_OP, V_MDI_OP(1));
+		ret = t3_wait_op_done(adapter, A_MI1_OP, F_BUSY, 0,
+				      MDIO_ATTEMPTS, 20);
+	}
+	mutex_unlock(&adapter->mdio_lock);
+	return ret;
+}
+
+static const struct mdio_ops mi1_mdio_ext_ops = {
+	mi1_ext_read,
+	mi1_ext_write
+};
+
+/**
+ *	t3_mdio_change_bits - modify the value of a PHY register
+ *	@phy: the PHY to operate on
+ *	@mmd: the device address
+ *	@reg: the register address
+ *	@clear: what part of the register value to mask off
+ *	@set: what part of the register value to set
+ *
+ *	Changes the value of a PHY register by applying a mask to its current
+ *	value and ORing the result with a new value.
+ */
+int t3_mdio_change_bits(struct cphy *phy, int mmd, int reg, unsigned int clear,
+			unsigned int set)
+{
+	int ret;
+	unsigned int val;
+
+	ret = mdio_read(phy, mmd, reg, &val);
+	if (!ret) {
+		val &= ~clear;
+		ret = mdio_write(phy, mmd, reg, val | set);
+	}
+	return ret;
+}
+
+/**
+ *	t3_phy_reset - reset a PHY block
+ *	@phy: the PHY to operate on
+ *	@mmd: the device address of the PHY block to reset
+ *	@wait: how long to wait for the reset to complete in 1ms increments
+ *
+ *	Resets a PHY block and optionally waits for the reset to complete.
+ *	@mmd should be 0 for 10/100/1000 PHYs and the device address to reset
+ *	for 10G PHYs.
+ */
+int t3_phy_reset(struct cphy *phy, int mmd, int wait)
+{
+	int err;
+	unsigned int ctl;
+
+	err = t3_mdio_change_bits(phy, mmd, MII_BMCR, BMCR_PDOWN, BMCR_RESET);
+	if (err || !wait)
+		return err;
+
+	do {
+		err = mdio_read(phy, mmd, MII_BMCR, &ctl);
+		if (err)
+			return err;
+		ctl &= BMCR_RESET;
+		if (ctl)
+			msleep(1);
+	} while (ctl && --wait);
+
+	return ctl ? -1 : 0;
+}
+
+/**
+ *	t3_phy_advertise - set the PHY advertisement registers for autoneg
+ *	@phy: the PHY to operate on
+ *	@advert: bitmap of capabilities the PHY should advertise
+ *
+ *	Sets a 10/100/1000 PHY's advertisement registers to advertise the
+ *	requested capabilities.
+ */
+int t3_phy_advertise(struct cphy *phy, unsigned int advert)
+{
+	int err;
+	unsigned int val = 0;
+
+	err = mdio_read(phy, 0, MII_CTRL1000, &val);
+	if (err)
+		return err;
+
+	val &= ~(ADVERTISE_1000HALF | ADVERTISE_1000FULL);
+	if (advert & ADVERTISED_1000baseT_Half)
+		val |= ADVERTISE_1000HALF;
+	if (advert & ADVERTISED_1000baseT_Full)
+		val |= ADVERTISE_1000FULL;
+
+	err = mdio_write(phy, 0, MII_CTRL1000, val);
+	if (err)
+		return err;
+
+	val = 1;
+	if (advert & ADVERTISED_10baseT_Half)
+		val |= ADVERTISE_10HALF;
+	if (advert & ADVERTISED_10baseT_Full)
+		val |= ADVERTISE_10FULL;
+	if (advert & ADVERTISED_100baseT_Half)
+		val |= ADVERTISE_100HALF;
+	if (advert & ADVERTISED_100baseT_Full)
+		val |= ADVERTISE_100FULL;
+	if (advert & ADVERTISED_Pause)
+		val |= ADVERTISE_PAUSE_CAP;
+	if (advert & ADVERTISED_Asym_Pause)
+		val |= ADVERTISE_PAUSE_ASYM;
+	return mdio_write(phy, 0, MII_ADVERTISE, val);
+}
+
+/**
+ *	t3_set_phy_speed_duplex - force PHY speed and duplex
+ *	@phy: the PHY to operate on
+ *	@speed: requested PHY speed
+ *	@duplex: requested PHY duplex
+ *
+ *	Force a 10/100/1000 PHY's speed and duplex.  This also disables
+ *	auto-negotiation except for GigE, where auto-negotiation is mandatory.
+ */
+int t3_set_phy_speed_duplex(struct cphy *phy, int speed, int duplex)
+{
+	int err;
+	unsigned int ctl;
+
+	err = mdio_read(phy, 0, MII_BMCR, &ctl);
+	if (err)
+		return err;
+
+	if (speed >= 0) {
+		ctl &= ~(BMCR_SPEED100 | BMCR_SPEED1000 | BMCR_ANENABLE);
+		if (speed == SPEED_100)
+			ctl |= BMCR_SPEED100;
+		else if (speed == SPEED_1000)
+			ctl |= BMCR_SPEED1000;
+	}
+	if (duplex >= 0) {
+		ctl &= ~(BMCR_FULLDPLX | BMCR_ANENABLE);
+		if (duplex == DUPLEX_FULL)
+			ctl |= BMCR_FULLDPLX;
+	}
+	if (ctl & BMCR_SPEED1000) /* auto-negotiation required for GigE */
+		ctl |= BMCR_ANENABLE;
+	return mdio_write(phy, 0, MII_BMCR, ctl);
+}
+
+static const struct adapter_info t3_adap_info[] = {
+	{2, 0, 0, 0,
+	 F_GPIO2_OEN | F_GPIO4_OEN |
+	 F_GPIO2_OUT_VAL | F_GPIO4_OUT_VAL, F_GPIO3 | F_GPIO5,
+	 SUPPORTED_OFFLOAD,
+	 &mi1_mdio_ops, "Chelsio PE9000"},
+	{2, 0, 0, 0,
+	 F_GPIO2_OEN | F_GPIO4_OEN |
+	 F_GPIO2_OUT_VAL | F_GPIO4_OUT_VAL, F_GPIO3 | F_GPIO5,
+	 SUPPORTED_OFFLOAD,
+	 &mi1_mdio_ops, "Chelsio T302"},
+	{1, 0, 0, 0,
+	 F_GPIO1_OEN | F_GPIO6_OEN | F_GPIO7_OEN | F_GPIO10_OEN |
+	 F_GPIO1_OUT_VAL | F_GPIO6_OUT_VAL | F_GPIO10_OUT_VAL, 0,
+	 SUPPORTED_10000baseT_Full | SUPPORTED_AUI | SUPPORTED_OFFLOAD,
+	 &mi1_mdio_ext_ops, "Chelsio T310"},
+	{2, 0, 0, 0,
+	 F_GPIO1_OEN | F_GPIO2_OEN | F_GPIO4_OEN | F_GPIO5_OEN | F_GPIO6_OEN |
+	 F_GPIO7_OEN | F_GPIO10_OEN | F_GPIO11_OEN | F_GPIO1_OUT_VAL |
+	 F_GPIO5_OUT_VAL | F_GPIO6_OUT_VAL | F_GPIO10_OUT_VAL, 0,
+	 SUPPORTED_10000baseT_Full | SUPPORTED_AUI | SUPPORTED_OFFLOAD,
+	 &mi1_mdio_ext_ops, "Chelsio T320"},
+};
+
+/*
+ * Return the adapter_info structure with a given index.  Out-of-range indices
+ * return NULL.
+ */
+const struct adapter_info *t3_get_adapter_info(unsigned int id)
+{
+	return id < ARRAY_SIZE(t3_adap_info) ? &t3_adap_info[id] : NULL;
+}
+
+#define CAPS_1G (SUPPORTED_10baseT_Full | SUPPORTED_100baseT_Full | \
+		 SUPPORTED_1000baseT_Full | SUPPORTED_Autoneg | SUPPORTED_MII)
+#define CAPS_10G (SUPPORTED_10000baseT_Full | SUPPORTED_AUI)
+
+static const struct port_type_info port_types[] = {
+	{NULL},
+	{t3_ael1002_phy_prep, CAPS_10G | SUPPORTED_FIBRE,
+	 "10GBASE-XR"},
+	{t3_vsc8211_phy_prep, CAPS_1G | SUPPORTED_TP | SUPPORTED_IRQ,
+	 "10/100/1000BASE-T"},
+	{NULL, CAPS_1G | SUPPORTED_TP | SUPPORTED_IRQ,
+	 "10/100/1000BASE-T"},
+	{t3_xaui_direct_phy_prep, CAPS_10G | SUPPORTED_TP, "10GBASE-CX4"},
+	{NULL, CAPS_10G, "10GBASE-KX4"},
+	{t3_qt2045_phy_prep, CAPS_10G | SUPPORTED_TP, "10GBASE-CX4"},
+	{t3_ael1006_phy_prep, CAPS_10G | SUPPORTED_FIBRE,
+	 "10GBASE-SR"},
+	{NULL, CAPS_10G | SUPPORTED_TP, "10GBASE-CX4"},
+};
+
+#undef CAPS_1G
+#undef CAPS_10G
+
+#define VPD_ENTRY(name, len) \
+	u8 name##_kword[2]; u8 name##_len; u8 name##_data[len]
+
+/*
+ * Partial EEPROM Vital Product Data structure.  Includes only the ID and
+ * VPD-R sections.
+ */
+struct t3_vpd {
+	u8 id_tag;
+	u8 id_len[2];
+	u8 id_data[16];
+	u8 vpdr_tag;
+	u8 vpdr_len[2];
+	VPD_ENTRY(pn, 16);	/* part number */
+	VPD_ENTRY(ec, 16);	/* EC level */
+	VPD_ENTRY(sn, 16);	/* serial number */
+	VPD_ENTRY(na, 12);	/* MAC address base */
+	VPD_ENTRY(cclk, 6);	/* core clock */
+	VPD_ENTRY(mclk, 6);	/* mem clock */
+	VPD_ENTRY(uclk, 6);	/* uP clk */
+	VPD_ENTRY(mdc, 6);	/* MDIO clk */
+	VPD_ENTRY(mt, 2);	/* mem timing */
+	VPD_ENTRY(xaui0cfg, 6);	/* XAUI0 config */
+	VPD_ENTRY(xaui1cfg, 6);	/* XAUI1 config */
+	VPD_ENTRY(port0, 2);	/* PHY0 complex */
+	VPD_ENTRY(port1, 2);	/* PHY1 complex */
+	VPD_ENTRY(port2, 2);	/* PHY2 complex */
+	VPD_ENTRY(port3, 2);	/* PHY3 complex */
+	VPD_ENTRY(rv, 1);	/* csum */
+	u32 pad;		/* for multiple-of-4 sizing and alignment */
+};
+
+#define EEPROM_MAX_POLL   4
+#define EEPROM_STAT_ADDR  0x4000
+#define VPD_BASE          0xc00
+
+/**
+ *	t3_seeprom_read - read a VPD EEPROM location
+ *	@adapter: adapter to read
+ *	@addr: EEPROM address
+ *	@data: where to store the read data
+ *
+ *	Read a 32-bit word from a location in VPD EEPROM using the card's PCI
+ *	VPD ROM capability.  A zero is written to the flag bit when the
+ *	addres is written to the control register.  The hardware device will
+ *	set the flag to 1 when 4 bytes have been read into the data register.
+ */
+int t3_seeprom_read(struct adapter *adapter, u32 addr, u32 *data)
+{
+	u16 val;
+	int attempts = EEPROM_MAX_POLL;
+	unsigned int base = adapter->params.pci.vpd_cap_addr;
+
+	if ((addr >= EEPROMSIZE && addr != EEPROM_STAT_ADDR) || (addr & 3))
+		return -EINVAL;
+
+	pci_write_config_word(adapter->pdev, base + PCI_VPD_ADDR, addr);
+	do {
+		udelay(10);
+		pci_read_config_word(adapter->pdev, base + PCI_VPD_ADDR, &val);
+	} while (!(val & PCI_VPD_ADDR_F) && --attempts);
+
+	if (!(val & PCI_VPD_ADDR_F)) {
+		CH_ERR(adapter, "reading EEPROM address 0x%x failed\n", addr);
+		return -EIO;
+	}
+	pci_read_config_dword(adapter->pdev, base + PCI_VPD_DATA, data);
+	*data = le32_to_cpu(*data);
+	return 0;
+}
+
+/**
+ *	t3_seeprom_write - write a VPD EEPROM location
+ *	@adapter: adapter to write
+ *	@addr: EEPROM address
+ *	@data: value to write
+ *
+ *	Write a 32-bit word to a location in VPD EEPROM using the card's PCI
+ *	VPD ROM capability.
+ */
+int t3_seeprom_write(struct adapter *adapter, u32 addr, u32 data)
+{
+	u16 val;
+	int attempts = EEPROM_MAX_POLL;
+	unsigned int base = adapter->params.pci.vpd_cap_addr;
+
+	if ((addr >= EEPROMSIZE && addr != EEPROM_STAT_ADDR) || (addr & 3))
+		return -EINVAL;
+
+	pci_write_config_dword(adapter->pdev, base + PCI_VPD_DATA,
+			       cpu_to_le32(data));
+	pci_write_config_word(adapter->pdev,base + PCI_VPD_ADDR,
+			      addr | PCI_VPD_ADDR_F);
+	do {
+		msleep(1);
+		pci_read_config_word(adapter->pdev, base + PCI_VPD_ADDR, &val);
+	} while ((val & PCI_VPD_ADDR_F) && --attempts);
+
+	if (val & PCI_VPD_ADDR_F) {
+		CH_ERR(adapter, "write to EEPROM address 0x%x failed\n", addr);
+		return -EIO;
+	}
+	return 0;
+}
+
+/**
+ *	t3_seeprom_wp - enable/disable EEPROM write protection
+ *	@adapter: the adapter
+ *	@enable: 1 to enable write protection, 0 to disable it
+ *
+ *	Enables or disables write protection on the serial EEPROM.
+ */
+int t3_seeprom_wp(struct adapter *adapter, int enable)
+{
+	return t3_seeprom_write(adapter, EEPROM_STAT_ADDR, enable ? 0xc : 0);
+}
+
+/*
+ * Convert a character holding a hex digit to a number.
+ */
+static unsigned int hex2int(unsigned char c)
+{
+	return isdigit(c) ? c - '0' : toupper(c) - 'A' + 10;
+}
+
+/**
+ *	get_vpd_params - read VPD parameters from VPD EEPROM
+ *	@adapter: adapter to read
+ *	@p: where to store the parameters
+ *
+ *	Reads card parameters stored in VPD EEPROM.
+ */
+static int get_vpd_params(struct adapter *adapter, struct vpd_params *p)
+{
+	int i, addr, ret;
+	struct t3_vpd vpd;
+
+	/*
+	 * Card information is normally at VPD_BASE but some early cards had
+	 * it at 0.
+	 */
+	ret = t3_seeprom_read(adapter, VPD_BASE, (u32 *)&vpd);
+	if (ret)
+		return ret;
+	addr = vpd.id_tag == 0x82 ? VPD_BASE : 0;
+
+	for (i = 0; i < sizeof(vpd); i += 4) {
+		ret = t3_seeprom_read(adapter, addr + i,
+				      (u32 *)((u8 *)&vpd + i));
+		if (ret)
+			return ret;
+	}
+
+	p->cclk = simple_strtoul(vpd.cclk_data, NULL, 10);
+	p->mclk = simple_strtoul(vpd.mclk_data, NULL, 10);
+	p->uclk = simple_strtoul(vpd.uclk_data, NULL, 10);
+	p->mdc = simple_strtoul(vpd.mdc_data, NULL, 10);
+	p->mem_timing = simple_strtoul(vpd.mt_data, NULL, 10);
+
+	/* Old eeproms didn't have port information */
+	if (adapter->params.rev == 0 && !vpd.port0_data[0]) {
+		p->port_type[0] = uses_xaui(adapter) ? 1 : 2;
+		p->port_type[1] = uses_xaui(adapter) ? 6 : 2;
+	} else {
+		p->port_type[0] = hex2int(vpd.port0_data[0]);
+		p->port_type[1] = hex2int(vpd.port1_data[0]);
+		p->xauicfg[0] = simple_strtoul(vpd.xaui0cfg_data, NULL, 16);
+		p->xauicfg[1] = simple_strtoul(vpd.xaui1cfg_data, NULL, 16);
+	}
+
+	for (i = 0; i < 6; i++)
+		p->eth_base[i] = hex2int(vpd.na_data[2 * i]) * 16 +
+				 hex2int(vpd.na_data[2 * i + 1]);
+	return 0;
+}
+
+/* serial flash and firmware constants */
+enum {
+	SF_ATTEMPTS = 5,	/* max retries for SF1 operations */
+	SF_SEC_SIZE = 64 * 1024,	/* serial flash sector size */
+	SF_SIZE = SF_SEC_SIZE * 8,	/* serial flash size */
+
+	/* flash command opcodes */
+	SF_PROG_PAGE = 2,	/* program page */
+	SF_WR_DISABLE = 4,	/* disable writes */
+	SF_RD_STATUS = 5,	/* read status register */
+	SF_WR_ENABLE = 6,	/* enable writes */
+	SF_RD_DATA_FAST = 0xb,	/* read flash */
+	SF_ERASE_SECTOR = 0xd8,	/* erase sector */
+
+	FW_FLASH_BOOT_ADDR = 0x70000,	/* start address of FW in flash */
+	FW_VERS_ADDR = 0x77ffc	/* flash address holding FW version */
+};
+
+/**
+ *	sf1_read - read data from the serial flash
+ *	@adapter: the adapter
+ *	@byte_cnt: number of bytes to read
+ *	@cont: whether another operation will be chained
+ *	@valp: where to store the read data
+ *
+ *	Reads up to 4 bytes of data from the serial flash.  The location of
+ *	the read needs to be specified prior to calling this by issuing the
+ *	appropriate commands to the serial flash.
+ */
+static int sf1_read(struct adapter *adapter, unsigned int byte_cnt, int cont,
+		    u32 *valp)
+{
+	int ret;
+
+	if (!byte_cnt || byte_cnt > 4)
+		return -EINVAL;
+	if (t3_read_reg(adapter, A_SF_OP) & F_BUSY)
+		return -EBUSY;
+	t3_write_reg(adapter, A_SF_OP, V_CONT(cont) | V_BYTECNT(byte_cnt - 1));
+	ret = t3_wait_op_done(adapter, A_SF_OP, F_BUSY, 0, SF_ATTEMPTS, 10);
+	if (!ret)
+		*valp = t3_read_reg(adapter, A_SF_DATA);
+	return ret;
+}
+
+/**
+ *	sf1_write - write data to the serial flash
+ *	@adapter: the adapter
+ *	@byte_cnt: number of bytes to write
+ *	@cont: whether another operation will be chained
+ *	@val: value to write
+ *
+ *	Writes up to 4 bytes of data to the serial flash.  The location of
+ *	the write needs to be specified prior to calling this by issuing the
+ *	appropriate commands to the serial flash.
+ */
+static int sf1_write(struct adapter *adapter, unsigned int byte_cnt, int cont,
+		     u32 val)
+{
+	if (!byte_cnt || byte_cnt > 4)
+		return -EINVAL;
+	if (t3_read_reg(adapter, A_SF_OP) & F_BUSY)
+		return -EBUSY;
+	t3_write_reg(adapter, A_SF_DATA, val);
+	t3_write_reg(adapter, A_SF_OP,
+		     V_CONT(cont) | V_BYTECNT(byte_cnt - 1) | V_OP(1));
+	return t3_wait_op_done(adapter, A_SF_OP, F_BUSY, 0, SF_ATTEMPTS, 10);
+}
+
+/**
+ *	flash_wait_op - wait for a flash operation to complete
+ *	@adapter: the adapter
+ *	@attempts: max number of polls of the status register
+ *	@delay: delay between polls in ms
+ *
+ *	Wait for a flash operation to complete by polling the status register.
+ */
+static int flash_wait_op(struct adapter *adapter, int attempts, int delay)
+{
+	int ret;
+	u32 status;
+
+	while (1) {
+		if ((ret = sf1_write(adapter, 1, 1, SF_RD_STATUS)) != 0 ||
+		    (ret = sf1_read(adapter, 1, 0, &status)) != 0)
+			return ret;
+		if (!(status & 1))
+			return 0;
+		if (--attempts == 0)
+			return -EAGAIN;
+		if (delay)
+			msleep(delay);
+	}
+}
+
+/**
+ *	t3_read_flash - read words from serial flash
+ *	@adapter: the adapter
+ *	@addr: the start address for the read
+ *	@nwords: how many 32-bit words to read
+ *	@data: where to store the read data
+ *	@byte_oriented: whether to store data as bytes or as words
+ *
+ *	Read the specified number of 32-bit words from the serial flash.
+ *	If @byte_oriented is set the read data is stored as a byte array
+ *	(i.e., big-endian), otherwise as 32-bit words in the platform's
+ *	natural endianess.
+ */
+int t3_read_flash(struct adapter *adapter, unsigned int addr,
+		  unsigned int nwords, u32 *data, int byte_oriented)
+{
+	int ret;
+
+	if (addr + nwords * sizeof(u32) > SF_SIZE || (addr & 3))
+		return -EINVAL;
+
+	addr = swab32(addr) | SF_RD_DATA_FAST;
+
+	if ((ret = sf1_write(adapter, 4, 1, addr)) != 0 ||
+	    (ret = sf1_read(adapter, 1, 1, data)) != 0)
+		return ret;
+
+	for (; nwords; nwords--, data++) {
+		ret = sf1_read(adapter, 4, nwords > 1, data);
+		if (ret)
+			return ret;
+		if (byte_oriented)
+			*data = htonl(*data);
+	}
+	return 0;
+}
+
+/**
+ *	t3_write_flash - write up to a page of data to the serial flash
+ *	@adapter: the adapter
+ *	@addr: the start address to write
+ *	@n: length of data to write
+ *	@data: the data to write
+ *
+ *	Writes up to a page of data (256 bytes) to the serial flash starting
+ *	at the given address.
+ */
+static int t3_write_flash(struct adapter *adapter, unsigned int addr,
+			  unsigned int n, const u8 *data)
+{
+	int ret;
+	u32 buf[64];
+	unsigned int i, c, left, val, offset = addr & 0xff;
+
+	if (addr + n > SF_SIZE || offset + n > 256)
+		return -EINVAL;
+
+	val = swab32(addr) | SF_PROG_PAGE;
+
+	if ((ret = sf1_write(adapter, 1, 0, SF_WR_ENABLE)) != 0 ||
+	    (ret = sf1_write(adapter, 4, 1, val)) != 0)
+		return ret;
+
+	for (left = n; left; left -= c) {
+		c = min(left, 4U);
+		for (val = 0, i = 0; i < c; ++i)
+			val = (val << 8) + *data++;
+
+		ret = sf1_write(adapter, c, c != left, val);
+		if (ret)
+			return ret;
+	}
+	if ((ret = flash_wait_op(adapter, 5, 1)) != 0)
+		return ret;
+
+	/* Read the page to verify the write succeeded */
+	ret = t3_read_flash(adapter, addr & ~0xff, ARRAY_SIZE(buf), buf, 1);
+	if (ret)
+		return ret;
+
+	if (memcmp(data - n, (u8 *) buf + offset, n))
+		return -EIO;
+	return 0;
+}
+
+/**
+ *	t3_get_fw_version - read the firmware version
+ *	@adapter: the adapter
+ *	@vers: where to place the version
+ *
+ *	Reads the FW version from flash.
+ */
+int t3_get_fw_version(struct adapter *adapter, u32 *vers)
+{
+	return t3_read_flash(adapter, FW_VERS_ADDR, 1, vers, 0);
+}
+
+/**
+ *	t3_check_fw_version - check if the FW is compatible with this driver
+ *	@adapter: the adapter
+ *
+ *	Checks if an adapter's FW is compatible with the driver.  Returns 0
+ *	if the versions are compatible, a negative error otherwise.
+ */
+int t3_check_fw_version(struct adapter *adapter)
+{
+	int ret;
+	u32 vers;
+
+	ret = t3_get_fw_version(adapter, &vers);
+	if (ret)
+		return ret;
+
+	/* Minor 0xfff means the FW is an internal development-only version. */
+	if ((vers & 0xfff) == 0xfff)
+		return 0;
+
+	if (vers == 0x1002009)
+		return 0;
+
+	CH_ERR(adapter, "found wrong FW version, driver needs version 2.9\n");
+	return -EINVAL;
+}
+
+/**
+ *	t3_flash_erase_sectors - erase a range of flash sectors
+ *	@adapter: the adapter
+ *	@start: the first sector to erase
+ *	@end: the last sector to erase
+ *
+ *	Erases the sectors in the given range.
+ */
+static int t3_flash_erase_sectors(struct adapter *adapter, int start, int end)
+{
+	while (start <= end) {
+		int ret;
+
+		if ((ret = sf1_write(adapter, 1, 0, SF_WR_ENABLE)) != 0 ||
+		    (ret = sf1_write(adapter, 4, 0,
+				     SF_ERASE_SECTOR | (start << 8))) != 0 ||
+		    (ret = flash_wait_op(adapter, 5, 500)) != 0)
+			return ret;
+		start++;
+	}
+	return 0;
+}
+
+/*
+ *	t3_load_fw - download firmware
+ *	@adapter: the adapter
+ *	@fw_data: the firrware image to write
+ *	@size: image size
+ *
+ *	Write the supplied firmware image to the card's serial flash.
+ *	The FW image has the following sections: @size - 8 bytes of code and
+ *	data, followed by 4 bytes of FW version, followed by the 32-bit
+ *	1's complement checksum of the whole image.
+ */
+int t3_load_fw(struct adapter *adapter, const u8 *fw_data, unsigned int size)
+{
+	u32 csum;
+	unsigned int i;
+	const u32 *p = (const u32 *)fw_data;
+	int ret, addr, fw_sector = FW_FLASH_BOOT_ADDR >> 16;
+
+	if (size & 3)
+		return -EINVAL;
+	if (size > FW_VERS_ADDR + 8 - FW_FLASH_BOOT_ADDR)
+		return -EFBIG;
+
+	for (csum = 0, i = 0; i < size / sizeof(csum); i++)
+		csum += ntohl(p[i]);
+	if (csum != 0xffffffff) {
+		CH_ERR(adapter, "corrupted firmware image, checksum %u\n",
+		       csum);
+		return -EINVAL;
+	}
+
+	ret = t3_flash_erase_sectors(adapter, fw_sector, fw_sector);
+	if (ret)
+		goto out;
+
+	size -= 8;		/* trim off version and checksum */
+	for (addr = FW_FLASH_BOOT_ADDR; size;) {
+		unsigned int chunk_size = min(size, 256U);
+
+		ret = t3_write_flash(adapter, addr, chunk_size, fw_data);
+		if (ret)
+			goto out;
+
+		addr += chunk_size;
+		fw_data += chunk_size;
+		size -= chunk_size;
+	}
+
+	ret = t3_write_flash(adapter, FW_VERS_ADDR, 4, fw_data);
+out:
+	if (ret)
+		CH_ERR(adapter, "firmware download failed, error %d\n", ret);
+	return ret;
+}
+
+#define CIM_CTL_BASE 0x2000
+
+/**
+ *      t3_cim_ctl_blk_read - read a block from CIM control region
+ *
+ *      @adap: the adapter
+ *      @addr: the start address within the CIM control region
+ *      @n: number of words to read
+ *      @valp: where to store the result
+ *
+ *      Reads a block of 4-byte words from the CIM control region.
+ */
+int t3_cim_ctl_blk_read(struct adapter *adap, unsigned int addr,
+			unsigned int n, unsigned int *valp)
+{
+	int ret = 0;
+
+	if (t3_read_reg(adap, A_CIM_HOST_ACC_CTRL) & F_HOSTBUSY)
+		return -EBUSY;
+
+	for ( ; !ret && n--; addr += 4) {
+		t3_write_reg(adap, A_CIM_HOST_ACC_CTRL, CIM_CTL_BASE + addr);
+		ret = t3_wait_op_done(adap, A_CIM_HOST_ACC_CTRL, F_HOSTBUSY,
+				      0, 5, 2);
+		if (!ret)
+			*valp++ = t3_read_reg(adap, A_CIM_HOST_ACC_DATA);
+	}
+	return ret;
+}
+
+
+/**
+ *	t3_link_changed - handle interface link changes
+ *	@adapter: the adapter
+ *	@port_id: the port index that changed link state
+ *
+ *	Called when a port's link settings change to propagate the new values
+ *	to the associated PHY and MAC.  After performing the common tasks it
+ *	invokes an OS-specific handler.
+ */
+void t3_link_changed(struct adapter *adapter, int port_id)
+{
+	int link_ok, speed, duplex, fc;
+	struct port_info *pi = adap2pinfo(adapter, port_id);
+	struct cphy *phy = &pi->phy;
+	struct cmac *mac = &pi->mac;
+	struct link_config *lc = &pi->link_config;
+
+	phy->ops->get_link_status(phy, &link_ok, &speed, &duplex, &fc);
+
+	if (link_ok != lc->link_ok && adapter->params.rev > 0 &&
+	    uses_xaui(adapter)) {
+		if (link_ok)
+			t3b_pcs_reset(mac);
+		t3_write_reg(adapter, A_XGM_XAUI_ACT_CTRL + mac->offset,
+			     link_ok ? F_TXACTENABLE | F_RXEN : 0);
+	}
+	lc->link_ok = link_ok;
+	lc->speed = speed < 0 ? SPEED_INVALID : speed;
+	lc->duplex = duplex < 0 ? DUPLEX_INVALID : duplex;
+	if (lc->requested_fc & PAUSE_AUTONEG)
+		fc &= lc->requested_fc;
+	else
+		fc = lc->requested_fc & (PAUSE_RX | PAUSE_TX);
+
+	if (link_ok && speed >= 0 && lc->autoneg == AUTONEG_ENABLE) {
+		/* Set MAC speed, duplex, and flow control to match PHY. */
+		t3_mac_set_speed_duplex_fc(mac, speed, duplex, fc);
+		lc->fc = fc;
+	}
+
+	t3_os_link_changed(adapter, port_id, link_ok, speed, duplex, fc);
+}
+
+/**
+ *	t3_link_start - apply link configuration to MAC/PHY
+ *	@phy: the PHY to setup
+ *	@mac: the MAC to setup
+ *	@lc: the requested link configuration
+ *
+ *	Set up a port's MAC and PHY according to a desired link configuration.
+ *	- If the PHY can auto-negotiate first decide what to advertise, then
+ *	  enable/disable auto-negotiation as desired, and reset.
+ *	- If the PHY does not auto-negotiate just reset it.
+ *	- If auto-negotiation is off set the MAC to the proper speed/duplex/FC,
+ *	  otherwise do it later based on the outcome of auto-negotiation.
+ */
+int t3_link_start(struct cphy *phy, struct cmac *mac, struct link_config *lc)
+{
+	unsigned int fc = lc->requested_fc & (PAUSE_RX | PAUSE_TX);
+
+	lc->link_ok = 0;
+	if (lc->supported & SUPPORTED_Autoneg) {
+		lc->advertising &= ~(ADVERTISED_Asym_Pause | ADVERTISED_Pause);
+		if (fc) {
+			lc->advertising |= ADVERTISED_Asym_Pause;
+			if (fc & PAUSE_RX)
+				lc->advertising |= ADVERTISED_Pause;
+		}
+		phy->ops->advertise(phy, lc->advertising);
+
+		if (lc->autoneg == AUTONEG_DISABLE) {
+			lc->speed = lc->requested_speed;
+			lc->duplex = lc->requested_duplex;
+			lc->fc = (unsigned char)fc;
+			t3_mac_set_speed_duplex_fc(mac, lc->speed, lc->duplex,
+						   fc);
+			/* Also disables autoneg */
+			phy->ops->set_speed_duplex(phy, lc->speed, lc->duplex);
+			phy->ops->reset(phy, 0);
+		} else
+			phy->ops->autoneg_enable(phy);
+	} else {
+		t3_mac_set_speed_duplex_fc(mac, -1, -1, fc);
+		lc->fc = (unsigned char)fc;
+		phy->ops->reset(phy, 0);
+	}
+	return 0;
+}
+
+/**
+ *	t3_set_vlan_accel - control HW VLAN extraction
+ *	@adapter: the adapter
+ *	@ports: bitmap of adapter ports to operate on
+ *	@on: enable (1) or disable (0) HW VLAN extraction
+ *
+ *	Enables or disables HW extraction of VLAN tags for the given port.
+ */
+void t3_set_vlan_accel(struct adapter *adapter, unsigned int ports, int on)
+{
+	t3_set_reg_field(adapter, A_TP_OUT_CONFIG,
+			 ports << S_VLANEXTRACTIONENABLE,
+			 on ? (ports << S_VLANEXTRACTIONENABLE) : 0);
+}
+
+struct intr_info {
+	unsigned int mask;	/* bits to check in interrupt status */
+	const char *msg;	/* message to print or NULL */
+	short stat_idx;		/* stat counter to increment or -1 */
+	unsigned short fatal:1;	/* whether the condition reported is fatal */
+};
+
+/**
+ *	t3_handle_intr_status - table driven interrupt handler
+ *	@adapter: the adapter that generated the interrupt
+ *	@reg: the interrupt status register to process
+ *	@mask: a mask to apply to the interrupt status
+ *	@acts: table of interrupt actions
+ *	@stats: statistics counters tracking interrupt occurences
+ *
+ *	A table driven interrupt handler that applies a set of masks to an
+ *	interrupt status word and performs the corresponding actions if the
+ *	interrupts described by the mask have occured.  The actions include
+ *	optionally printing a warning or alert message, and optionally
+ *	incrementing a stat counter.  The table is terminated by an entry
+ *	specifying mask 0.  Returns the number of fatal interrupt conditions.
+ */
+static int t3_handle_intr_status(struct adapter *adapter, unsigned int reg,
+				 unsigned int mask,
+				 const struct intr_info *acts,
+				 unsigned long *stats)
+{
+	int fatal = 0;
+	unsigned int status = t3_read_reg(adapter, reg) & mask;
+
+	for (; acts->mask; ++acts) {
+		if (!(status & acts->mask))
+			continue;
+		if (acts->fatal) {
+			fatal++;
+			CH_ALERT(adapter, "%s (0x%x)\n",
+				 acts->msg, status & acts->mask);
+		} else if (acts->msg)
+			CH_WARN(adapter, "%s (0x%x)\n",
+				acts->msg, status & acts->mask);
+		if (acts->stat_idx >= 0)
+			stats[acts->stat_idx]++;
+	}
+	if (status)		/* clear processed interrupts */
+		t3_write_reg(adapter, reg, status);
+	return fatal;
+}
+
+#define SGE_INTR_MASK (F_RSPQDISABLED)
+#define MC5_INTR_MASK (F_PARITYERR | F_ACTRGNFULL | F_UNKNOWNCMD | \
+		       F_REQQPARERR | F_DISPQPARERR | F_DELACTEMPTY | \
+		       F_NFASRCHFAIL)
+#define MC7_INTR_MASK (F_AE | F_UE | F_CE | V_PE(M_PE))
+#define XGM_INTR_MASK (V_TXFIFO_PRTY_ERR(M_TXFIFO_PRTY_ERR) | \
+		       V_RXFIFO_PRTY_ERR(M_RXFIFO_PRTY_ERR) | \
+		       F_TXFIFO_UNDERRUN | F_RXFIFO_OVERFLOW)
+#define PCIX_INTR_MASK (F_MSTDETPARERR | F_SIGTARABT | F_RCVTARABT | \
+			F_RCVMSTABT | F_SIGSYSERR | F_DETPARERR | \
+			F_SPLCMPDIS | F_UNXSPLCMP | F_RCVSPLCMPERR | \
+			F_DETCORECCERR | F_DETUNCECCERR | F_PIOPARERR | \
+			V_WFPARERR(M_WFPARERR) | V_RFPARERR(M_RFPARERR) | \
+			V_CFPARERR(M_CFPARERR) /* | V_MSIXPARERR(M_MSIXPARERR) */)
+#define PCIE_INTR_MASK (F_UNXSPLCPLERRR | F_UNXSPLCPLERRC | F_PCIE_PIOPARERR |\
+			F_PCIE_WFPARERR | F_PCIE_RFPARERR | F_PCIE_CFPARERR | \
+			/* V_PCIE_MSIXPARERR(M_PCIE_MSIXPARERR) | */ \
+			V_BISTERR(M_BISTERR) | F_PEXERR)
+#define ULPRX_INTR_MASK F_PARERR
+#define ULPTX_INTR_MASK 0
+#define CPLSW_INTR_MASK (F_TP_FRAMING_ERROR | \
+			 F_SGE_FRAMING_ERROR | F_CIM_FRAMING_ERROR | \
+			 F_ZERO_SWITCH_ERROR)
+#define CIM_INTR_MASK (F_BLKWRPLINT | F_BLKRDPLINT | F_BLKWRCTLINT | \
+		       F_BLKRDCTLINT | F_BLKWRFLASHINT | F_BLKRDFLASHINT | \
+		       F_SGLWRFLASHINT | F_WRBLKFLASHINT | F_BLKWRBOOTINT | \
+	 	       F_FLASHRANGEINT | F_SDRAMRANGEINT | F_RSVDSPACEINT)
+#define PMTX_INTR_MASK (F_ZERO_C_CMD_ERROR | ICSPI_FRM_ERR | OESPI_FRM_ERR | \
+			V_ICSPI_PAR_ERROR(M_ICSPI_PAR_ERROR) | \
+			V_OESPI_PAR_ERROR(M_OESPI_PAR_ERROR))
+#define PMRX_INTR_MASK (F_ZERO_E_CMD_ERROR | IESPI_FRM_ERR | OCSPI_FRM_ERR | \
+			V_IESPI_PAR_ERROR(M_IESPI_PAR_ERROR) | \
+			V_OCSPI_PAR_ERROR(M_OCSPI_PAR_ERROR))
+#define MPS_INTR_MASK (V_TX0TPPARERRENB(M_TX0TPPARERRENB) | \
+		       V_TX1TPPARERRENB(M_TX1TPPARERRENB) | \
+		       V_RXTPPARERRENB(M_RXTPPARERRENB) | \
+		       V_MCAPARERRENB(M_MCAPARERRENB))
+#define PL_INTR_MASK (F_T3DBG | F_XGMAC0_0 | F_XGMAC0_1 | F_MC5A | F_PM1_TX | \
+		      F_PM1_RX | F_ULP2_TX | F_ULP2_RX | F_TP1 | F_CIM | \
+		      F_MC7_CM | F_MC7_PMTX | F_MC7_PMRX | F_SGE3 | F_PCIM0 | \
+		      F_MPS0 | F_CPL_SWITCH)
+
+/*
+ * Interrupt handler for the PCIX1 module.
+ */
+static void pci_intr_handler(struct adapter *adapter)
+{
+	static const struct intr_info pcix1_intr_info[] = {
+		{ F_PEXERR, "PCI PEX error", -1, 1 },
+		{F_MSTDETPARERR, "PCI master detected parity error", -1, 1},
+		{F_SIGTARABT, "PCI signaled target abort", -1, 1},
+		{F_RCVTARABT, "PCI received target abort", -1, 1},
+		{F_RCVMSTABT, "PCI received master abort", -1, 1},
+		{F_SIGSYSERR, "PCI signaled system error", -1, 1},
+		{F_DETPARERR, "PCI detected parity error", -1, 1},
+		{F_SPLCMPDIS, "PCI split completion discarded", -1, 1},
+		{F_UNXSPLCMP, "PCI unexpected split completion error", -1, 1},
+		{F_RCVSPLCMPERR, "PCI received split completion error", -1,
+		 1},
+		{F_DETCORECCERR, "PCI correctable ECC error",
+		 STAT_PCI_CORR_ECC, 0},
+		{F_DETUNCECCERR, "PCI uncorrectable ECC error", -1, 1},
+		{F_PIOPARERR, "PCI PIO FIFO parity error", -1, 1},
+		{V_WFPARERR(M_WFPARERR), "PCI write FIFO parity error", -1,
+		 1},
+		{V_RFPARERR(M_RFPARERR), "PCI read FIFO parity error", -1,
+		 1},
+		{V_CFPARERR(M_CFPARERR), "PCI command FIFO parity error", -1,
+		 1},
+		{V_MSIXPARERR(M_MSIXPARERR), "PCI MSI-X table/PBA parity "
+		 "error", -1, 1},
+		{0}
+	};
+
+	if (t3_handle_intr_status(adapter, A_PCIX_INT_CAUSE, PCIX_INTR_MASK,
+				  pcix1_intr_info, adapter->irq_stats))
+		t3_fatal_err(adapter);
+}
+
+/*
+ * Interrupt handler for the PCIE module.
+ */
+static void pcie_intr_handler(struct adapter *adapter)
+{
+	static const struct intr_info pcie_intr_info[] = {
+		{F_UNXSPLCPLERRR,
+		 "PCI unexpected split completion DMA read error", -1, 1},
+		{F_UNXSPLCPLERRC,
+		 "PCI unexpected split completion DMA command error", -1, 1},
+		{F_PCIE_PIOPARERR, "PCI PIO FIFO parity error", -1, 1},
+		{F_PCIE_WFPARERR, "PCI write FIFO parity error", -1, 1},
+		{F_PCIE_RFPARERR, "PCI read FIFO parity error", -1, 1},
+		{F_PCIE_CFPARERR, "PCI command FIFO parity error", -1, 1},
+		{V_PCIE_MSIXPARERR(M_PCIE_MSIXPARERR),
+		 "PCI MSI-X table/PBA parity error", -1, 1},
+		{V_BISTERR(M_BISTERR), "PCI BIST error", -1, 1},
+		{0}
+	};
+
+	if (t3_handle_intr_status(adapter, A_PCIE_INT_CAUSE, PCIE_INTR_MASK,
+				  pcie_intr_info, adapter->irq_stats))
+		t3_fatal_err(adapter);
+}
+
+/*
+ * TP interrupt handler.
+ */
+static void tp_intr_handler(struct adapter *adapter)
+{
+	static const struct intr_info tp_intr_info[] = {
+		{0xffffff, "TP parity error", -1, 1},
+		{0x1000000, "TP out of Rx pages", -1, 1},
+		{0x2000000, "TP out of Tx pages", -1, 1},
+		{0}
+	};
+
+	if (t3_handle_intr_status(adapter, A_TP_INT_CAUSE, 0xffffffff,
+				  tp_intr_info, NULL))
+		t3_fatal_err(adapter);
+}
+
+/*
+ * CIM interrupt handler.
+ */
+static void cim_intr_handler(struct adapter *adapter)
+{
+	static const struct intr_info cim_intr_info[] = {
+		{F_RSVDSPACEINT, "CIM reserved space write", -1, 1},
+		{F_SDRAMRANGEINT, "CIM SDRAM address out of range", -1, 1},
+		{F_FLASHRANGEINT, "CIM flash address out of range", -1, 1},
+		{F_BLKWRBOOTINT, "CIM block write to boot space", -1, 1},
+		{F_WRBLKFLASHINT, "CIM write to cached flash space", -1, 1},
+		{F_SGLWRFLASHINT, "CIM single write to flash space", -1, 1},
+		{F_BLKRDFLASHINT, "CIM block read from flash space", -1, 1},
+		{F_BLKWRFLASHINT, "CIM block write to flash space", -1, 1},
+		{F_BLKRDCTLINT, "CIM block read from CTL space", -1, 1},
+		{F_BLKWRCTLINT, "CIM block write to CTL space", -1, 1},
+		{F_BLKRDPLINT, "CIM block read from PL space", -1, 1},
+		{F_BLKWRPLINT, "CIM block write to PL space", -1, 1},
+		{0}
+	};
+
+	if (t3_handle_intr_status(adapter, A_CIM_HOST_INT_CAUSE, 0xffffffff,
+				  cim_intr_info, NULL))
+		t3_fatal_err(adapter);
+}
+
+/*
+ * ULP RX interrupt handler.
+ */
+static void ulprx_intr_handler(struct adapter *adapter)
+{
+	static const struct intr_info ulprx_intr_info[] = {
+		{F_PARERR, "ULP RX parity error", -1, 1},
+		{0}
+	};
+
+	if (t3_handle_intr_status(adapter, A_ULPRX_INT_CAUSE, 0xffffffff,
+				  ulprx_intr_info, NULL))
+		t3_fatal_err(adapter);
+}
+
+/*
+ * ULP TX interrupt handler.
+ */
+static void ulptx_intr_handler(struct adapter *adapter)
+{
+	static const struct intr_info ulptx_intr_info[] = {
+		{F_PBL_BOUND_ERR_CH0, "ULP TX channel 0 PBL out of bounds",
+		 STAT_ULP_CH0_PBL_OOB, 0},
+		{F_PBL_BOUND_ERR_CH1, "ULP TX channel 1 PBL out of bounds",
+		 STAT_ULP_CH1_PBL_OOB, 0},
+		{0}
+	};
+
+	if (t3_handle_intr_status(adapter, A_ULPTX_INT_CAUSE, 0xffffffff,
+				  ulptx_intr_info, adapter->irq_stats))
+		t3_fatal_err(adapter);
+}
+
+#define ICSPI_FRM_ERR (F_ICSPI0_FIFO2X_RX_FRAMING_ERROR | \
+	F_ICSPI1_FIFO2X_RX_FRAMING_ERROR | F_ICSPI0_RX_FRAMING_ERROR | \
+	F_ICSPI1_RX_FRAMING_ERROR | F_ICSPI0_TX_FRAMING_ERROR | \
+	F_ICSPI1_TX_FRAMING_ERROR)
+#define OESPI_FRM_ERR (F_OESPI0_RX_FRAMING_ERROR | \
+	F_OESPI1_RX_FRAMING_ERROR | F_OESPI0_TX_FRAMING_ERROR | \
+	F_OESPI1_TX_FRAMING_ERROR | F_OESPI0_OFIFO2X_TX_FRAMING_ERROR | \
+	F_OESPI1_OFIFO2X_TX_FRAMING_ERROR)
+
+/*
+ * PM TX interrupt handler.
+ */
+static void pmtx_intr_handler(struct adapter *adapter)
+{
+	static const struct intr_info pmtx_intr_info[] = {
+		{F_ZERO_C_CMD_ERROR, "PMTX 0-length pcmd", -1, 1},
+		{ICSPI_FRM_ERR, "PMTX ispi framing error", -1, 1},
+		{OESPI_FRM_ERR, "PMTX ospi framing error", -1, 1},
+		{V_ICSPI_PAR_ERROR(M_ICSPI_PAR_ERROR),
+		 "PMTX ispi parity error", -1, 1},
+		{V_OESPI_PAR_ERROR(M_OESPI_PAR_ERROR),
+		 "PMTX ospi parity error", -1, 1},
+		{0}
+	};
+
+	if (t3_handle_intr_status(adapter, A_PM1_TX_INT_CAUSE, 0xffffffff,
+				  pmtx_intr_info, NULL))
+		t3_fatal_err(adapter);
+}
+
+#define IESPI_FRM_ERR (F_IESPI0_FIFO2X_RX_FRAMING_ERROR | \
+	F_IESPI1_FIFO2X_RX_FRAMING_ERROR | F_IESPI0_RX_FRAMING_ERROR | \
+	F_IESPI1_RX_FRAMING_ERROR | F_IESPI0_TX_FRAMING_ERROR | \
+	F_IESPI1_TX_FRAMING_ERROR)
+#define OCSPI_FRM_ERR (F_OCSPI0_RX_FRAMING_ERROR | \
+	F_OCSPI1_RX_FRAMING_ERROR | F_OCSPI0_TX_FRAMING_ERROR | \
+	F_OCSPI1_TX_FRAMING_ERROR | F_OCSPI0_OFIFO2X_TX_FRAMING_ERROR | \
+	F_OCSPI1_OFIFO2X_TX_FRAMING_ERROR)
+
+/*
+ * PM RX interrupt handler.
+ */
+static void pmrx_intr_handler(struct adapter *adapter)
+{
+	static const struct intr_info pmrx_intr_info[] = {
+		{F_ZERO_E_CMD_ERROR, "PMRX 0-length pcmd", -1, 1},
+		{IESPI_FRM_ERR, "PMRX ispi framing error", -1, 1},
+		{OCSPI_FRM_ERR, "PMRX ospi framing error", -1, 1},
+		{V_IESPI_PAR_ERROR(M_IESPI_PAR_ERROR),
+		 "PMRX ispi parity error", -1, 1},
+		{V_OCSPI_PAR_ERROR(M_OCSPI_PAR_ERROR),
+		 "PMRX ospi parity error", -1, 1},
+		{0}
+	};
+
+	if (t3_handle_intr_status(adapter, A_PM1_RX_INT_CAUSE, 0xffffffff,
+				  pmrx_intr_info, NULL))
+		t3_fatal_err(adapter);
+}
+
+/*
+ * CPL switch interrupt handler.
+ */
+static void cplsw_intr_handler(struct adapter *adapter)
+{
+	static const struct intr_info cplsw_intr_info[] = {
+/*		{ F_CIM_OVFL_ERROR, "CPL switch CIM overflow", -1, 1 }, */
+		{F_TP_FRAMING_ERROR, "CPL switch TP framing error", -1, 1},
+		{F_SGE_FRAMING_ERROR, "CPL switch SGE framing error", -1, 1},
+		{F_CIM_FRAMING_ERROR, "CPL switch CIM framing error", -1, 1},
+		{F_ZERO_SWITCH_ERROR, "CPL switch no-switch error", -1, 1},
+		{0}
+	};
+
+	if (t3_handle_intr_status(adapter, A_CPL_INTR_CAUSE, 0xffffffff,
+				  cplsw_intr_info, NULL))
+		t3_fatal_err(adapter);
+}
+
+/*
+ * MPS interrupt handler.
+ */
+static void mps_intr_handler(struct adapter *adapter)
+{
+	static const struct intr_info mps_intr_info[] = {
+		{0x1ff, "MPS parity error", -1, 1},
+		{0}
+	};
+
+	if (t3_handle_intr_status(adapter, A_MPS_INT_CAUSE, 0xffffffff,
+				  mps_intr_info, NULL))
+		t3_fatal_err(adapter);
+}
+
+#define MC7_INTR_FATAL (F_UE | V_PE(M_PE) | F_AE)
+
+/*
+ * MC7 interrupt handler.
+ */
+static void mc7_intr_handler(struct mc7 *mc7)
+{
+	struct adapter *adapter = mc7->adapter;
+	u32 cause = t3_read_reg(adapter, mc7->offset + A_MC7_INT_CAUSE);
+
+	if (cause & F_CE) {
+		mc7->stats.corr_err++;
+		CH_WARN(adapter, "%s MC7 correctable error at addr 0x%x, "
+			"data 0x%x 0x%x 0x%x\n", mc7->name,
+			t3_read_reg(adapter, mc7->offset + A_MC7_CE_ADDR),
+			t3_read_reg(adapter, mc7->offset + A_MC7_CE_DATA0),
+			t3_read_reg(adapter, mc7->offset + A_MC7_CE_DATA1),
+			t3_read_reg(adapter, mc7->offset + A_MC7_CE_DATA2));
+	}
+
+	if (cause & F_UE) {
+		mc7->stats.uncorr_err++;
+		CH_ALERT(adapter, "%s MC7 uncorrectable error at addr 0x%x, "
+			 "data 0x%x 0x%x 0x%x\n", mc7->name,
+			 t3_read_reg(adapter, mc7->offset + A_MC7_UE_ADDR),
+			 t3_read_reg(adapter, mc7->offset + A_MC7_UE_DATA0),
+			 t3_read_reg(adapter, mc7->offset + A_MC7_UE_DATA1),
+			 t3_read_reg(adapter, mc7->offset + A_MC7_UE_DATA2));
+	}
+
+	if (G_PE(cause)) {
+		mc7->stats.parity_err++;
+		CH_ALERT(adapter, "%s MC7 parity error 0x%x\n",
+			 mc7->name, G_PE(cause));
+	}
+
+	if (cause & F_AE) {
+		u32 addr = 0;
+
+		if (adapter->params.rev > 0)
+			addr = t3_read_reg(adapter,
+					   mc7->offset + A_MC7_ERR_ADDR);
+		mc7->stats.addr_err++;
+		CH_ALERT(adapter, "%s MC7 address error: 0x%x\n",
+			 mc7->name, addr);
+	}
+
+	if (cause & MC7_INTR_FATAL)
+		t3_fatal_err(adapter);
+
+	t3_write_reg(adapter, mc7->offset + A_MC7_INT_CAUSE, cause);
+}
+
+#define XGM_INTR_FATAL (V_TXFIFO_PRTY_ERR(M_TXFIFO_PRTY_ERR) | \
+			V_RXFIFO_PRTY_ERR(M_RXFIFO_PRTY_ERR))
+/*
+ * XGMAC interrupt handler.
+ */
+static int mac_intr_handler(struct adapter *adap, unsigned int idx)
+{
+	struct cmac *mac = &adap2pinfo(adap, idx)->mac;
+	u32 cause = t3_read_reg(adap, A_XGM_INT_CAUSE + mac->offset);
+
+	if (cause & V_TXFIFO_PRTY_ERR(M_TXFIFO_PRTY_ERR)) {
+		mac->stats.tx_fifo_parity_err++;
+		CH_ALERT(adap, "port%d: MAC TX FIFO parity error\n", idx);
+	}
+	if (cause & V_RXFIFO_PRTY_ERR(M_RXFIFO_PRTY_ERR)) {
+		mac->stats.rx_fifo_parity_err++;
+		CH_ALERT(adap, "port%d: MAC RX FIFO parity error\n", idx);
+	}
+	if (cause & F_TXFIFO_UNDERRUN)
+		mac->stats.tx_fifo_urun++;
+	if (cause & F_RXFIFO_OVERFLOW)
+		mac->stats.rx_fifo_ovfl++;
+	if (cause & V_SERDES_LOS(M_SERDES_LOS))
+		mac->stats.serdes_signal_loss++;
+	if (cause & F_XAUIPCSCTCERR)
+		mac->stats.xaui_pcs_ctc_err++;
+	if (cause & F_XAUIPCSALIGNCHANGE)
+		mac->stats.xaui_pcs_align_change++;
+
+	t3_write_reg(adap, A_XGM_INT_CAUSE + mac->offset, cause);
+	if (cause & XGM_INTR_FATAL)
+		t3_fatal_err(adap);
+	return cause != 0;
+}
+
+/*
+ * Interrupt handler for PHY events.
+ */
+int t3_phy_intr_handler(struct adapter *adapter)
+{
+	static const int intr_gpio_bits[] = { 8, 0x20 };
+
+	u32 i, cause = t3_read_reg(adapter, A_T3DBG_INT_CAUSE);
+
+	for_each_port(adapter, i) {
+		if (cause & intr_gpio_bits[i]) {
+			struct cphy *phy = &adap2pinfo(adapter, i)->phy;
+			int phy_cause = phy->ops->intr_handler(phy);
+
+			if (phy_cause & cphy_cause_link_change)
+				t3_link_changed(adapter, i);
+			if (phy_cause & cphy_cause_fifo_error)
+				phy->fifo_errors++;
+		}
+	}
+
+	t3_write_reg(adapter, A_T3DBG_INT_CAUSE, cause);
+	return 0;
+}
+
+/*
+ * T3 slow path (non-data) interrupt handler.
+ */
+int t3_slow_intr_handler(struct adapter *adapter)
+{
+	u32 cause = t3_read_reg(adapter, A_PL_INT_CAUSE0);
+
+	cause &= adapter->slow_intr_mask;
+	if (!cause)
+		return 0;
+	if (cause & F_PCIM0) {
+		if (is_pcie(adapter))
+			pcie_intr_handler(adapter);
+		else
+			pci_intr_handler(adapter);
+	}
+	if (cause & F_SGE3)
+		t3_sge_err_intr_handler(adapter);
+	if (cause & F_MC7_PMRX)
+		mc7_intr_handler(&adapter->pmrx);
+	if (cause & F_MC7_PMTX)
+		mc7_intr_handler(&adapter->pmtx);
+	if (cause & F_MC7_CM)
+		mc7_intr_handler(&adapter->cm);
+	if (cause & F_CIM)
+		cim_intr_handler(adapter);
+	if (cause & F_TP1)
+		tp_intr_handler(adapter);
+	if (cause & F_ULP2_RX)
+		ulprx_intr_handler(adapter);
+	if (cause & F_ULP2_TX)
+		ulptx_intr_handler(adapter);
+	if (cause & F_PM1_RX)
+		pmrx_intr_handler(adapter);
+	if (cause & F_PM1_TX)
+		pmtx_intr_handler(adapter);
+	if (cause & F_CPL_SWITCH)
+		cplsw_intr_handler(adapter);
+	if (cause & F_MPS0)
+		mps_intr_handler(adapter);
+	if (cause & F_MC5A)
+		t3_mc5_intr_handler(&adapter->mc5);
+	if (cause & F_XGMAC0_0)
+		mac_intr_handler(adapter, 0);
+	if (cause & F_XGMAC0_1)
+		mac_intr_handler(adapter, 1);
+	if (cause & F_T3DBG)
+		t3_os_ext_intr_handler(adapter);
+
+	/* Clear the interrupts just processed. */
+	t3_write_reg(adapter, A_PL_INT_CAUSE0, cause);
+	t3_read_reg(adapter, A_PL_INT_CAUSE0);	/* flush */
+	return 1;
+}
+
+/**
+ *	t3_intr_enable - enable interrupts
+ *	@adapter: the adapter whose interrupts should be enabled
+ *
+ *	Enable interrupts by setting the interrupt enable registers of the
+ *	various HW modules and then enabling the top-level interrupt
+ *	concentrator.
+ */
+void t3_intr_enable(struct adapter *adapter)
+{
+	static const struct addr_val_pair intr_en_avp[] = {
+		{A_SG_INT_ENABLE, SGE_INTR_MASK},
+		{A_MC7_INT_ENABLE, MC7_INTR_MASK},
+		{A_MC7_INT_ENABLE - MC7_PMRX_BASE_ADDR + MC7_PMTX_BASE_ADDR,
+		 MC7_INTR_MASK},
+		{A_MC7_INT_ENABLE - MC7_PMRX_BASE_ADDR + MC7_CM_BASE_ADDR,
+		 MC7_INTR_MASK},
+		{A_MC5_DB_INT_ENABLE, MC5_INTR_MASK},
+		{A_ULPRX_INT_ENABLE, ULPRX_INTR_MASK},
+		{A_TP_INT_ENABLE, 0x3bfffff},
+		{A_PM1_TX_INT_ENABLE, PMTX_INTR_MASK},
+		{A_PM1_RX_INT_ENABLE, PMRX_INTR_MASK},
+		{A_CIM_HOST_INT_ENABLE, CIM_INTR_MASK},
+		{A_MPS_INT_ENABLE, MPS_INTR_MASK},
+	};
+
+	adapter->slow_intr_mask = PL_INTR_MASK;
+
+	t3_write_regs(adapter, intr_en_avp, ARRAY_SIZE(intr_en_avp), 0);
+
+	if (adapter->params.rev > 0) {
+		t3_write_reg(adapter, A_CPL_INTR_ENABLE,
+			     CPLSW_INTR_MASK | F_CIM_OVFL_ERROR);
+		t3_write_reg(adapter, A_ULPTX_INT_ENABLE,
+			     ULPTX_INTR_MASK | F_PBL_BOUND_ERR_CH0 |
+			     F_PBL_BOUND_ERR_CH1);
+	} else {
+		t3_write_reg(adapter, A_CPL_INTR_ENABLE, CPLSW_INTR_MASK);
+		t3_write_reg(adapter, A_ULPTX_INT_ENABLE, ULPTX_INTR_MASK);
+	}
+
+	t3_write_reg(adapter, A_T3DBG_GPIO_ACT_LOW,
+		     adapter_info(adapter)->gpio_intr);
+	t3_write_reg(adapter, A_T3DBG_INT_ENABLE,
+		     adapter_info(adapter)->gpio_intr);
+	if (is_pcie(adapter))
+		t3_write_reg(adapter, A_PCIE_INT_ENABLE, PCIE_INTR_MASK);
+	else
+		t3_write_reg(adapter, A_PCIX_INT_ENABLE, PCIX_INTR_MASK);
+	t3_write_reg(adapter, A_PL_INT_ENABLE0, adapter->slow_intr_mask);
+	t3_read_reg(adapter, A_PL_INT_ENABLE0);	/* flush */
+}
+
+/**
+ *	t3_intr_disable - disable a card's interrupts
+ *	@adapter: the adapter whose interrupts should be disabled
+ *
+ *	Disable interrupts.  We only disable the top-level interrupt
+ *	concentrator and the SGE data interrupts.
+ */
+void t3_intr_disable(struct adapter *adapter)
+{
+	t3_write_reg(adapter, A_PL_INT_ENABLE0, 0);
+	t3_read_reg(adapter, A_PL_INT_ENABLE0);	/* flush */
+	adapter->slow_intr_mask = 0;
+}
+
+/**
+ *	t3_intr_clear - clear all interrupts
+ *	@adapter: the adapter whose interrupts should be cleared
+ *
+ *	Clears all interrupts.
+ */
+void t3_intr_clear(struct adapter *adapter)
+{
+	static const unsigned int cause_reg_addr[] = {
+		A_SG_INT_CAUSE,
+		A_SG_RSPQ_FL_STATUS,
+		A_PCIX_INT_CAUSE,
+		A_MC7_INT_CAUSE,
+		A_MC7_INT_CAUSE - MC7_PMRX_BASE_ADDR + MC7_PMTX_BASE_ADDR,
+		A_MC7_INT_CAUSE - MC7_PMRX_BASE_ADDR + MC7_CM_BASE_ADDR,
+		A_CIM_HOST_INT_CAUSE,
+		A_TP_INT_CAUSE,
+		A_MC5_DB_INT_CAUSE,
+		A_ULPRX_INT_CAUSE,
+		A_ULPTX_INT_CAUSE,
+		A_CPL_INTR_CAUSE,
+		A_PM1_TX_INT_CAUSE,
+		A_PM1_RX_INT_CAUSE,
+		A_MPS_INT_CAUSE,
+		A_T3DBG_INT_CAUSE,
+	};
+	unsigned int i;
+
+	/* Clear PHY and MAC interrupts for each port. */
+	for_each_port(adapter, i)
+	    t3_port_intr_clear(adapter, i);
+
+	for (i = 0; i < ARRAY_SIZE(cause_reg_addr); ++i)
+		t3_write_reg(adapter, cause_reg_addr[i], 0xffffffff);
+
+	t3_write_reg(adapter, A_PL_INT_CAUSE0, 0xffffffff);
+	t3_read_reg(adapter, A_PL_INT_CAUSE0);	/* flush */
+}
+
+/**
+ *	t3_port_intr_enable - enable port-specific interrupts
+ *	@adapter: associated adapter
+ *	@idx: index of port whose interrupts should be enabled
+ *
+ *	Enable port-specific (i.e., MAC and PHY) interrupts for the given
+ *	adapter port.
+ */
+void t3_port_intr_enable(struct adapter *adapter, int idx)
+{
+	struct cphy *phy = &adap2pinfo(adapter, idx)->phy;
+
+	t3_write_reg(adapter, XGM_REG(A_XGM_INT_ENABLE, idx), XGM_INTR_MASK);
+	t3_read_reg(adapter, XGM_REG(A_XGM_INT_ENABLE, idx)); /* flush */
+	phy->ops->intr_enable(phy);
+}
+
+/**
+ *	t3_port_intr_disable - disable port-specific interrupts
+ *	@adapter: associated adapter
+ *	@idx: index of port whose interrupts should be disabled
+ *
+ *	Disable port-specific (i.e., MAC and PHY) interrupts for the given
+ *	adapter port.
+ */
+void t3_port_intr_disable(struct adapter *adapter, int idx)
+{
+	struct cphy *phy = &adap2pinfo(adapter, idx)->phy;
+
+	t3_write_reg(adapter, XGM_REG(A_XGM_INT_ENABLE, idx), 0);
+	t3_read_reg(adapter, XGM_REG(A_XGM_INT_ENABLE, idx)); /* flush */
+	phy->ops->intr_disable(phy);
+}
+
+/**
+ *	t3_port_intr_clear - clear port-specific interrupts
+ *	@adapter: associated adapter
+ *	@idx: index of port whose interrupts to clear
+ *
+ *	Clear port-specific (i.e., MAC and PHY) interrupts for the given
+ *	adapter port.
+ */
+void t3_port_intr_clear(struct adapter *adapter, int idx)
+{
+	struct cphy *phy = &adap2pinfo(adapter, idx)->phy;
+
+	t3_write_reg(adapter, XGM_REG(A_XGM_INT_CAUSE, idx), 0xffffffff);
+	t3_read_reg(adapter, XGM_REG(A_XGM_INT_CAUSE, idx)); /* flush */
+	phy->ops->intr_clear(phy);
+}
+
+/**
+ * 	t3_sge_write_context - write an SGE context
+ * 	@adapter: the adapter
+ * 	@id: the context id
+ * 	@type: the context type
+ *
+ * 	Program an SGE context with the values already loaded in the
+ * 	CONTEXT_DATA? registers.
+ */
+static int t3_sge_write_context(struct adapter *adapter, unsigned int id,
+				unsigned int type)
+{
+	t3_write_reg(adapter, A_SG_CONTEXT_MASK0, 0xffffffff);
+	t3_write_reg(adapter, A_SG_CONTEXT_MASK1, 0xffffffff);
+	t3_write_reg(adapter, A_SG_CONTEXT_MASK2, 0xffffffff);
+	t3_write_reg(adapter, A_SG_CONTEXT_MASK3, 0xffffffff);
+	t3_write_reg(adapter, A_SG_CONTEXT_CMD,
+		     V_CONTEXT_CMD_OPCODE(1) | type | V_CONTEXT(id));
+	return t3_wait_op_done(adapter, A_SG_CONTEXT_CMD, F_CONTEXT_CMD_BUSY,
+			       0, 5, 1);
+}
+
+/**
+ *	t3_sge_init_ecntxt - initialize an SGE egress context
+ *	@adapter: the adapter to configure
+ *	@id: the context id
+ *	@gts_enable: whether to enable GTS for the context
+ *	@type: the egress context type
+ *	@respq: associated response queue
+ *	@base_addr: base address of queue
+ *	@size: number of queue entries
+ *	@token: uP token
+ *	@gen: initial generation value for the context
+ *	@cidx: consumer pointer
+ *
+ *	Initialize an SGE egress context and make it ready for use.  If the
+ *	platform allows concurrent context operations, the caller is
+ *	responsible for appropriate locking.
+ */
+int t3_sge_init_ecntxt(struct adapter *adapter, unsigned int id, int gts_enable,
+		       enum sge_context_type type, int respq, u64 base_addr,
+		       unsigned int size, unsigned int token, int gen,
+		       unsigned int cidx)
+{
+	unsigned int credits = type == SGE_CNTXT_OFLD ? 0 : FW_WR_NUM;
+
+	if (base_addr & 0xfff)	/* must be 4K aligned */
+		return -EINVAL;
+	if (t3_read_reg(adapter, A_SG_CONTEXT_CMD) & F_CONTEXT_CMD_BUSY)
+		return -EBUSY;
+
+	base_addr >>= 12;
+	t3_write_reg(adapter, A_SG_CONTEXT_DATA0, V_EC_INDEX(cidx) |
+		     V_EC_CREDITS(credits) | V_EC_GTS(gts_enable));
+	t3_write_reg(adapter, A_SG_CONTEXT_DATA1, V_EC_SIZE(size) |
+		     V_EC_BASE_LO(base_addr & 0xffff));
+	base_addr >>= 16;
+	t3_write_reg(adapter, A_SG_CONTEXT_DATA2, base_addr);
+	base_addr >>= 32;
+	t3_write_reg(adapter, A_SG_CONTEXT_DATA3,
+		     V_EC_BASE_HI(base_addr & 0xf) | V_EC_RESPQ(respq) |
+		     V_EC_TYPE(type) | V_EC_GEN(gen) | V_EC_UP_TOKEN(token) |
+		     F_EC_VALID);
+	return t3_sge_write_context(adapter, id, F_EGRESS);
+}
+
+/**
+ *	t3_sge_init_flcntxt - initialize an SGE free-buffer list context
+ *	@adapter: the adapter to configure
+ *	@id: the context id
+ *	@gts_enable: whether to enable GTS for the context
+ *	@base_addr: base address of queue
+ *	@size: number of queue entries
+ *	@bsize: size of each buffer for this queue
+ *	@cong_thres: threshold to signal congestion to upstream producers
+ *	@gen: initial generation value for the context
+ *	@cidx: consumer pointer
+ *
+ *	Initialize an SGE free list context and make it ready for use.  The
+ *	caller is responsible for ensuring only one context operation occurs
+ *	at a time.
+ */
+int t3_sge_init_flcntxt(struct adapter *adapter, unsigned int id,
+			int gts_enable, u64 base_addr, unsigned int size,
+			unsigned int bsize, unsigned int cong_thres, int gen,
+			unsigned int cidx)
+{
+	if (base_addr & 0xfff)	/* must be 4K aligned */
+		return -EINVAL;
+	if (t3_read_reg(adapter, A_SG_CONTEXT_CMD) & F_CONTEXT_CMD_BUSY)
+		return -EBUSY;
+
+	base_addr >>= 12;
+	t3_write_reg(adapter, A_SG_CONTEXT_DATA0, base_addr);
+	base_addr >>= 32;
+	t3_write_reg(adapter, A_SG_CONTEXT_DATA1,
+		     V_FL_BASE_HI((u32) base_addr) |
+		     V_FL_INDEX_LO(cidx & M_FL_INDEX_LO));
+	t3_write_reg(adapter, A_SG_CONTEXT_DATA2, V_FL_SIZE(size) |
+		     V_FL_GEN(gen) | V_FL_INDEX_HI(cidx >> 12) |
+		     V_FL_ENTRY_SIZE_LO(bsize & M_FL_ENTRY_SIZE_LO));
+	t3_write_reg(adapter, A_SG_CONTEXT_DATA3,
+		     V_FL_ENTRY_SIZE_HI(bsize >> (32 - S_FL_ENTRY_SIZE_LO)) |
+		     V_FL_CONG_THRES(cong_thres) | V_FL_GTS(gts_enable));
+	return t3_sge_write_context(adapter, id, F_FREELIST);
+}
+
+/**
+ *	t3_sge_init_rspcntxt - initialize an SGE response queue context
+ *	@adapter: the adapter to configure
+ *	@id: the context id
+ *	@irq_vec_idx: MSI-X interrupt vector index, 0 if no MSI-X, -1 if no IRQ
+ *	@base_addr: base address of queue
+ *	@size: number of queue entries
+ *	@fl_thres: threshold for selecting the normal or jumbo free list
+ *	@gen: initial generation value for the context
+ *	@cidx: consumer pointer
+ *
+ *	Initialize an SGE response queue context and make it ready for use.
+ *	The caller is responsible for ensuring only one context operation
+ *	occurs at a time.
+ */
+int t3_sge_init_rspcntxt(struct adapter *adapter, unsigned int id,
+			 int irq_vec_idx, u64 base_addr, unsigned int size,
+			 unsigned int fl_thres, int gen, unsigned int cidx)
+{
+	unsigned int intr = 0;
+
+	if (base_addr & 0xfff)	/* must be 4K aligned */
+		return -EINVAL;
+	if (t3_read_reg(adapter, A_SG_CONTEXT_CMD) & F_CONTEXT_CMD_BUSY)
+		return -EBUSY;
+
+	base_addr >>= 12;
+	t3_write_reg(adapter, A_SG_CONTEXT_DATA0, V_CQ_SIZE(size) |
+		     V_CQ_INDEX(cidx));
+	t3_write_reg(adapter, A_SG_CONTEXT_DATA1, base_addr);
+	base_addr >>= 32;
+	if (irq_vec_idx >= 0)
+		intr = V_RQ_MSI_VEC(irq_vec_idx) | F_RQ_INTR_EN;
+	t3_write_reg(adapter, A_SG_CONTEXT_DATA2,
+		     V_CQ_BASE_HI((u32) base_addr) | intr | V_RQ_GEN(gen));
+	t3_write_reg(adapter, A_SG_CONTEXT_DATA3, fl_thres);
+	return t3_sge_write_context(adapter, id, F_RESPONSEQ);
+}
+
+/**
+ *	t3_sge_init_cqcntxt - initialize an SGE completion queue context
+ *	@adapter: the adapter to configure
+ *	@id: the context id
+ *	@base_addr: base address of queue
+ *	@size: number of queue entries
+ *	@rspq: response queue for async notifications
+ *	@ovfl_mode: CQ overflow mode
+ *	@credits: completion queue credits
+ *	@credit_thres: the credit threshold
+ *
+ *	Initialize an SGE completion queue context and make it ready for use.
+ *	The caller is responsible for ensuring only one context operation
+ *	occurs at a time.
+ */
+int t3_sge_init_cqcntxt(struct adapter *adapter, unsigned int id, u64 base_addr,
+			unsigned int size, int rspq, int ovfl_mode,
+			unsigned int credits, unsigned int credit_thres)
+{
+	if (base_addr & 0xfff)	/* must be 4K aligned */
+		return -EINVAL;
+	if (t3_read_reg(adapter, A_SG_CONTEXT_CMD) & F_CONTEXT_CMD_BUSY)
+		return -EBUSY;
+
+	base_addr >>= 12;
+	t3_write_reg(adapter, A_SG_CONTEXT_DATA0, V_CQ_SIZE(size));
+	t3_write_reg(adapter, A_SG_CONTEXT_DATA1, base_addr);
+	base_addr >>= 32;
+	t3_write_reg(adapter, A_SG_CONTEXT_DATA2,
+		     V_CQ_BASE_HI((u32) base_addr) | V_CQ_RSPQ(rspq) |
+		     V_CQ_GEN(1) | V_CQ_OVERFLOW_MODE(ovfl_mode));
+	t3_write_reg(adapter, A_SG_CONTEXT_DATA3, V_CQ_CREDITS(credits) |
+		     V_CQ_CREDIT_THRES(credit_thres));
+	return t3_sge_write_context(adapter, id, F_CQ);
+}
+
+/**
+ *	t3_sge_enable_ecntxt - enable/disable an SGE egress context
+ *	@adapter: the adapter
+ *	@id: the egress context id
+ *	@enable: enable (1) or disable (0) the context
+ *
+ *	Enable or disable an SGE egress context.  The caller is responsible for
+ *	ensuring only one context operation occurs at a time.
+ */
+int t3_sge_enable_ecntxt(struct adapter *adapter, unsigned int id, int enable)
+{
+	if (t3_read_reg(adapter, A_SG_CONTEXT_CMD) & F_CONTEXT_CMD_BUSY)
+		return -EBUSY;
+
+	t3_write_reg(adapter, A_SG_CONTEXT_MASK0, 0);
+	t3_write_reg(adapter, A_SG_CONTEXT_MASK1, 0);
+	t3_write_reg(adapter, A_SG_CONTEXT_MASK2, 0);
+	t3_write_reg(adapter, A_SG_CONTEXT_MASK3, F_EC_VALID);
+	t3_write_reg(adapter, A_SG_CONTEXT_DATA3, V_EC_VALID(enable));
+	t3_write_reg(adapter, A_SG_CONTEXT_CMD,
+		     V_CONTEXT_CMD_OPCODE(1) | F_EGRESS | V_CONTEXT(id));
+	return t3_wait_op_done(adapter, A_SG_CONTEXT_CMD, F_CONTEXT_CMD_BUSY,
+			       0, 5, 1);
+}
+
+/**
+ *	t3_sge_disable_fl - disable an SGE free-buffer list
+ *	@adapter: the adapter
+ *	@id: the free list context id
+ *
+ *	Disable an SGE free-buffer list.  The caller is responsible for
+ *	ensuring only one context operation occurs at a time.
+ */
+int t3_sge_disable_fl(struct adapter *adapter, unsigned int id)
+{
+	if (t3_read_reg(adapter, A_SG_CONTEXT_CMD) & F_CONTEXT_CMD_BUSY)
+		return -EBUSY;
+
+	t3_write_reg(adapter, A_SG_CONTEXT_MASK0, 0);
+	t3_write_reg(adapter, A_SG_CONTEXT_MASK1, 0);
+	t3_write_reg(adapter, A_SG_CONTEXT_MASK2, V_FL_SIZE(M_FL_SIZE));
+	t3_write_reg(adapter, A_SG_CONTEXT_MASK3, 0);
+	t3_write_reg(adapter, A_SG_CONTEXT_DATA2, 0);
+	t3_write_reg(adapter, A_SG_CONTEXT_CMD,
+		     V_CONTEXT_CMD_OPCODE(1) | F_FREELIST | V_CONTEXT(id));
+	return t3_wait_op_done(adapter, A_SG_CONTEXT_CMD, F_CONTEXT_CMD_BUSY,
+			       0, 5, 1);
+}
+
+/**
+ *	t3_sge_disable_rspcntxt - disable an SGE response queue
+ *	@adapter: the adapter
+ *	@id: the response queue context id
+ *
+ *	Disable an SGE response queue.  The caller is responsible for
+ *	ensuring only one context operation occurs at a time.
+ */
+int t3_sge_disable_rspcntxt(struct adapter *adapter, unsigned int id)
+{
+	if (t3_read_reg(adapter, A_SG_CONTEXT_CMD) & F_CONTEXT_CMD_BUSY)
+		return -EBUSY;
+
+	t3_write_reg(adapter, A_SG_CONTEXT_MASK0, V_CQ_SIZE(M_CQ_SIZE));
+	t3_write_reg(adapter, A_SG_CONTEXT_MASK1, 0);
+	t3_write_reg(adapter, A_SG_CONTEXT_MASK2, 0);
+	t3_write_reg(adapter, A_SG_CONTEXT_MASK3, 0);
+	t3_write_reg(adapter, A_SG_CONTEXT_DATA0, 0);
+	t3_write_reg(adapter, A_SG_CONTEXT_CMD,
+		     V_CONTEXT_CMD_OPCODE(1) | F_RESPONSEQ | V_CONTEXT(id));
+	return t3_wait_op_done(adapter, A_SG_CONTEXT_CMD, F_CONTEXT_CMD_BUSY,
+			       0, 5, 1);
+}
+
+/**
+ *	t3_sge_disable_cqcntxt - disable an SGE completion queue
+ *	@adapter: the adapter
+ *	@id: the completion queue context id
+ *
+ *	Disable an SGE completion queue.  The caller is responsible for
+ *	ensuring only one context operation occurs at a time.
+ */
+int t3_sge_disable_cqcntxt(struct adapter *adapter, unsigned int id)
+{
+	if (t3_read_reg(adapter, A_SG_CONTEXT_CMD) & F_CONTEXT_CMD_BUSY)
+		return -EBUSY;
+
+	t3_write_reg(adapter, A_SG_CONTEXT_MASK0, V_CQ_SIZE(M_CQ_SIZE));
+	t3_write_reg(adapter, A_SG_CONTEXT_MASK1, 0);
+	t3_write_reg(adapter, A_SG_CONTEXT_MASK2, 0);
+	t3_write_reg(adapter, A_SG_CONTEXT_MASK3, 0);
+	t3_write_reg(adapter, A_SG_CONTEXT_DATA0, 0);
+	t3_write_reg(adapter, A_SG_CONTEXT_CMD,
+		     V_CONTEXT_CMD_OPCODE(1) | F_CQ | V_CONTEXT(id));
+	return t3_wait_op_done(adapter, A_SG_CONTEXT_CMD, F_CONTEXT_CMD_BUSY,
+			       0, 5, 1);
+}
+
+/**
+ *	t3_sge_cqcntxt_op - perform an operation on a completion queue context
+ *	@adapter: the adapter
+ *	@id: the context id
+ *	@op: the operation to perform
+ *
+ *	Perform the selected operation on an SGE completion queue context.
+ *	The caller is responsible for ensuring only one context operation
+ *	occurs at a time.
+ */
+int t3_sge_cqcntxt_op(struct adapter *adapter, unsigned int id, unsigned int op,
+		      unsigned int credits)
+{
+	u32 val;
+
+	if (t3_read_reg(adapter, A_SG_CONTEXT_CMD) & F_CONTEXT_CMD_BUSY)
+		return -EBUSY;
+
+	t3_write_reg(adapter, A_SG_CONTEXT_DATA0, credits << 16);
+	t3_write_reg(adapter, A_SG_CONTEXT_CMD, V_CONTEXT_CMD_OPCODE(op) |
+		     V_CONTEXT(id) | F_CQ);
+	if (t3_wait_op_done_val(adapter, A_SG_CONTEXT_CMD, F_CONTEXT_CMD_BUSY,
+				0, 5, 1, &val))
+		return -EIO;
+
+	if (op >= 2 && op < 7) {
+		if (adapter->params.rev > 0)
+			return G_CQ_INDEX(val);
+
+		t3_write_reg(adapter, A_SG_CONTEXT_CMD,
+			     V_CONTEXT_CMD_OPCODE(0) | F_CQ | V_CONTEXT(id));
+		if (t3_wait_op_done(adapter, A_SG_CONTEXT_CMD,
+				    F_CONTEXT_CMD_BUSY, 0, 5, 1))
+			return -EIO;
+		return G_CQ_INDEX(t3_read_reg(adapter, A_SG_CONTEXT_DATA0));
+	}
+	return 0;
+}
+
+/**
+ * 	t3_sge_read_context - read an SGE context
+ * 	@type: the context type
+ * 	@adapter: the adapter
+ * 	@id: the context id
+ * 	@data: holds the retrieved context
+ *
+ * 	Read an SGE egress context.  The caller is responsible for ensuring
+ * 	only one context operation occurs at a time.
+ */
+static int t3_sge_read_context(unsigned int type, struct adapter *adapter,
+			       unsigned int id, u32 data[4])
+{
+	if (t3_read_reg(adapter, A_SG_CONTEXT_CMD) & F_CONTEXT_CMD_BUSY)
+		return -EBUSY;
+
+	t3_write_reg(adapter, A_SG_CONTEXT_CMD,
+		     V_CONTEXT_CMD_OPCODE(0) | type | V_CONTEXT(id));
+	if (t3_wait_op_done(adapter, A_SG_CONTEXT_CMD, F_CONTEXT_CMD_BUSY, 0,
+			    5, 1))
+		return -EIO;
+	data[0] = t3_read_reg(adapter, A_SG_CONTEXT_DATA0);
+	data[1] = t3_read_reg(adapter, A_SG_CONTEXT_DATA1);
+	data[2] = t3_read_reg(adapter, A_SG_CONTEXT_DATA2);
+	data[3] = t3_read_reg(adapter, A_SG_CONTEXT_DATA3);
+	return 0;
+}
+
+/**
+ * 	t3_sge_read_ecntxt - read an SGE egress context
+ * 	@adapter: the adapter
+ * 	@id: the context id
+ * 	@data: holds the retrieved context
+ *
+ * 	Read an SGE egress context.  The caller is responsible for ensuring
+ * 	only one context operation occurs at a time.
+ */
+int t3_sge_read_ecntxt(struct adapter *adapter, unsigned int id, u32 data[4])
+{
+	if (id >= 65536)
+		return -EINVAL;
+	return t3_sge_read_context(F_EGRESS, adapter, id, data);
+}
+
+/**
+ * 	t3_sge_read_cq - read an SGE CQ context
+ * 	@adapter: the adapter
+ * 	@id: the context id
+ * 	@data: holds the retrieved context
+ *
+ * 	Read an SGE CQ context.  The caller is responsible for ensuring
+ * 	only one context operation occurs at a time.
+ */
+int t3_sge_read_cq(struct adapter *adapter, unsigned int id, u32 data[4])
+{
+	if (id >= 65536)
+		return -EINVAL;
+	return t3_sge_read_context(F_CQ, adapter, id, data);
+}
+
+/**
+ * 	t3_sge_read_fl - read an SGE free-list context
+ * 	@adapter: the adapter
+ * 	@id: the context id
+ * 	@data: holds the retrieved context
+ *
+ * 	Read an SGE free-list context.  The caller is responsible for ensuring
+ * 	only one context operation occurs at a time.
+ */
+int t3_sge_read_fl(struct adapter *adapter, unsigned int id, u32 data[4])
+{
+	if (id >= SGE_QSETS * 2)
+		return -EINVAL;
+	return t3_sge_read_context(F_FREELIST, adapter, id, data);
+}
+
+/**
+ * 	t3_sge_read_rspq - read an SGE response queue context
+ * 	@adapter: the adapter
+ * 	@id: the context id
+ * 	@data: holds the retrieved context
+ *
+ * 	Read an SGE response queue context.  The caller is responsible for
+ * 	ensuring only one context operation occurs at a time.
+ */
+int t3_sge_read_rspq(struct adapter *adapter, unsigned int id, u32 data[4])
+{
+	if (id >= SGE_QSETS)
+		return -EINVAL;
+	return t3_sge_read_context(F_RESPONSEQ, adapter, id, data);
+}
+
+/**
+ *	t3_config_rss - configure Rx packet steering
+ *	@adapter: the adapter
+ *	@rss_config: RSS settings (written to TP_RSS_CONFIG)
+ *	@cpus: values for the CPU lookup table (0xff terminated)
+ *	@rspq: values for the response queue lookup table (0xffff terminated)
+ *
+ *	Programs the receive packet steering logic.  @cpus and @rspq provide
+ *	the values for the CPU and response queue lookup tables.  If they
+ *	provide fewer values than the size of the tables the supplied values
+ *	are used repeatedly until the tables are fully populated.
+ */
+void t3_config_rss(struct adapter *adapter, unsigned int rss_config,
+		   const u8 * cpus, const u16 *rspq)
+{
+	int i, j, cpu_idx = 0, q_idx = 0;
+
+	if (cpus)
+		for (i = 0; i < RSS_TABLE_SIZE; ++i) {
+			u32 val = i << 16;
+
+			for (j = 0; j < 2; ++j) {
+				val |= (cpus[cpu_idx++] & 0x3f) << (8 * j);
+				if (cpus[cpu_idx] == 0xff)
+					cpu_idx = 0;
+			}
+			t3_write_reg(adapter, A_TP_RSS_LKP_TABLE, val);
+		}
+
+	if (rspq)
+		for (i = 0; i < RSS_TABLE_SIZE; ++i) {
+			t3_write_reg(adapter, A_TP_RSS_MAP_TABLE,
+				     (i << 16) | rspq[q_idx++]);
+			if (rspq[q_idx] == 0xffff)
+				q_idx = 0;
+		}
+
+	t3_write_reg(adapter, A_TP_RSS_CONFIG, rss_config);
+}
+
+/**
+ *	t3_read_rss - read the contents of the RSS tables
+ *	@adapter: the adapter
+ *	@lkup: holds the contents of the RSS lookup table
+ *	@map: holds the contents of the RSS map table
+ *
+ *	Reads the contents of the receive packet steering tables.
+ */
+int t3_read_rss(struct adapter *adapter, u8 * lkup, u16 *map)
+{
+	int i;
+	u32 val;
+
+	if (lkup)
+		for (i = 0; i < RSS_TABLE_SIZE; ++i) {
+			t3_write_reg(adapter, A_TP_RSS_LKP_TABLE,
+				     0xffff0000 | i);
+			val = t3_read_reg(adapter, A_TP_RSS_LKP_TABLE);
+			if (!(val & 0x80000000))
+				return -EAGAIN;
+			*lkup++ = val;
+			*lkup++ = (val >> 8);
+		}
+
+	if (map)
+		for (i = 0; i < RSS_TABLE_SIZE; ++i) {
+			t3_write_reg(adapter, A_TP_RSS_MAP_TABLE,
+				     0xffff0000 | i);
+			val = t3_read_reg(adapter, A_TP_RSS_MAP_TABLE);
+			if (!(val & 0x80000000))
+				return -EAGAIN;
+			*map++ = val;
+		}
+	return 0;
+}
+
+/**
+ *	t3_tp_set_offload_mode - put TP in NIC/offload mode
+ *	@adap: the adapter
+ *	@enable: 1 to select offload mode, 0 for regular NIC
+ *
+ *	Switches TP to NIC/offload mode.
+ */
+void t3_tp_set_offload_mode(struct adapter *adap, int enable)
+{
+	if (is_offload(adap) || !enable)
+		t3_set_reg_field(adap, A_TP_IN_CONFIG, F_NICMODE,
+				 V_NICMODE(!enable));
+}
+
+/**
+ *	pm_num_pages - calculate the number of pages of the payload memory
+ *	@mem_size: the size of the payload memory
+ *	@pg_size: the size of each payload memory page
+ *
+ *	Calculate the number of pages, each of the given size, that fit in a
+ *	memory of the specified size, respecting the HW requirement that the
+ *	number of pages must be a multiple of 24.
+ */
+static inline unsigned int pm_num_pages(unsigned int mem_size,
+					unsigned int pg_size)
+{
+	unsigned int n = mem_size / pg_size;
+
+	return n - n % 24;
+}
+
+#define mem_region(adap, start, size, reg) \
+	t3_write_reg((adap), A_ ## reg, (start)); \
+	start += size
+
+/*
+ *	partition_mem - partition memory and configure TP memory settings
+ *	@adap: the adapter
+ *	@p: the TP parameters
+ *
+ *	Partitions context and payload memory and configures TP's memory
+ *	registers.
+ */
+static void partition_mem(struct adapter *adap, const struct tp_params *p)
+{
+	unsigned int m, pstructs, tids = t3_mc5_size(&adap->mc5);
+	unsigned int timers = 0, timers_shift = 22;
+
+	if (adap->params.rev > 0) {
+		if (tids <= 16 * 1024) {
+			timers = 1;
+			timers_shift = 16;
+		} else if (tids <= 64 * 1024) {
+			timers = 2;
+			timers_shift = 18;
+		} else if (tids <= 256 * 1024) {
+			timers = 3;
+			timers_shift = 20;
+		}
+	}
+
+	t3_write_reg(adap, A_TP_PMM_SIZE,
+		     p->chan_rx_size | (p->chan_tx_size >> 16));
+
+	t3_write_reg(adap, A_TP_PMM_TX_BASE, 0);
+	t3_write_reg(adap, A_TP_PMM_TX_PAGE_SIZE, p->tx_pg_size);
+	t3_write_reg(adap, A_TP_PMM_TX_MAX_PAGE, p->tx_num_pgs);
+	t3_set_reg_field(adap, A_TP_PARA_REG3, V_TXDATAACKIDX(M_TXDATAACKIDX),
+			 V_TXDATAACKIDX(fls(p->tx_pg_size) - 12));
+
+	t3_write_reg(adap, A_TP_PMM_RX_BASE, 0);
+	t3_write_reg(adap, A_TP_PMM_RX_PAGE_SIZE, p->rx_pg_size);
+	t3_write_reg(adap, A_TP_PMM_RX_MAX_PAGE, p->rx_num_pgs);
+
+	pstructs = p->rx_num_pgs + p->tx_num_pgs;
+	/* Add a bit of headroom and make multiple of 24 */
+	pstructs += 48;
+	pstructs -= pstructs % 24;
+	t3_write_reg(adap, A_TP_CMM_MM_MAX_PSTRUCT, pstructs);
+
+	m = tids * TCB_SIZE;
+	mem_region(adap, m, (64 << 10) * 64, SG_EGR_CNTX_BADDR);
+	mem_region(adap, m, (64 << 10) * 64, SG_CQ_CONTEXT_BADDR);
+	t3_write_reg(adap, A_TP_CMM_TIMER_BASE, V_CMTIMERMAXNUM(timers) | m);
+	m += ((p->ntimer_qs - 1) << timers_shift) + (1 << 22);
+	mem_region(adap, m, pstructs * 64, TP_CMM_MM_BASE);
+	mem_region(adap, m, 64 * (pstructs / 24), TP_CMM_MM_PS_FLST_BASE);
+	mem_region(adap, m, 64 * (p->rx_num_pgs / 24), TP_CMM_MM_RX_FLST_BASE);
+	mem_region(adap, m, 64 * (p->tx_num_pgs / 24), TP_CMM_MM_TX_FLST_BASE);
+
+	m = (m + 4095) & ~0xfff;
+	t3_write_reg(adap, A_CIM_SDRAM_BASE_ADDR, m);
+	t3_write_reg(adap, A_CIM_SDRAM_ADDR_SIZE, p->cm_size - m);
+
+	tids = (p->cm_size - m - (3 << 20)) / 3072 - 32;
+	m = t3_mc5_size(&adap->mc5) - adap->params.mc5.nservers -
+	    adap->params.mc5.nfilters - adap->params.mc5.nroutes;
+	if (tids < m)
+		adap->params.mc5.nservers += m - tids;
+}
+
+static inline void tp_wr_indirect(struct adapter *adap, unsigned int addr,
+				  u32 val)
+{
+	t3_write_reg(adap, A_TP_PIO_ADDR, addr);
+	t3_write_reg(adap, A_TP_PIO_DATA, val);
+}
+
+static void tp_config(struct adapter *adap, const struct tp_params *p)
+{
+	unsigned int v;
+
+	t3_write_reg(adap, A_TP_GLOBAL_CONFIG, F_TXPACINGENABLE | F_PATHMTU |
+		     F_IPCHECKSUMOFFLOAD | F_UDPCHECKSUMOFFLOAD |
+		     F_TCPCHECKSUMOFFLOAD | V_IPTTL(64));
+	t3_write_reg(adap, A_TP_TCP_OPTIONS, V_MTUDEFAULT(576) |
+		     F_MTUENABLE | V_WINDOWSCALEMODE(1) |
+		     V_TIMESTAMPSMODE(1) | V_SACKMODE(1) | V_SACKRX(1));
+	t3_write_reg(adap, A_TP_DACK_CONFIG, V_AUTOSTATE3(1) |
+		     V_AUTOSTATE2(1) | V_AUTOSTATE1(0) |
+		     V_BYTETHRESHOLD(16384) | V_MSSTHRESHOLD(2) |
+		     F_AUTOCAREFUL | F_AUTOENABLE | V_DACK_MODE(1));
+	t3_set_reg_field(adap, A_TP_IN_CONFIG, F_IPV6ENABLE | F_NICMODE,
+			 F_IPV6ENABLE | F_NICMODE);
+	t3_write_reg(adap, A_TP_TX_RESOURCE_LIMIT, 0x18141814);
+	t3_write_reg(adap, A_TP_PARA_REG4, 0x5050105);
+	t3_set_reg_field(adap, A_TP_PARA_REG6,
+			 adap->params.rev > 0 ? F_ENABLEESND : F_T3A_ENABLEESND,
+			 0);
+
+	v = t3_read_reg(adap, A_TP_PC_CONFIG);
+	v &= ~(F_ENABLEEPCMDAFULL | F_ENABLEOCSPIFULL);
+	t3_write_reg(adap, A_TP_PC_CONFIG, v | F_TXDEFERENABLE |
+		     F_MODULATEUNIONMODE | F_HEARBEATDACK |
+		     F_TXCONGESTIONMODE | F_RXCONGESTIONMODE);
+
+	v = t3_read_reg(adap, A_TP_PC_CONFIG2);
+	v &= ~F_CHDRAFULL;
+	t3_write_reg(adap, A_TP_PC_CONFIG2, v);
+
+	if (adap->params.rev > 0) {
+		tp_wr_indirect(adap, A_TP_EGRESS_CONFIG, F_REWRITEFORCETOSIZE);
+		t3_set_reg_field(adap, A_TP_PARA_REG3, F_TXPACEAUTO,
+				 F_TXPACEAUTO);
+		t3_set_reg_field(adap, A_TP_PC_CONFIG, F_LOCKTID, F_LOCKTID);
+		t3_set_reg_field(adap, A_TP_PARA_REG3, 0, F_TXPACEAUTOSTRICT);
+	} else
+		t3_set_reg_field(adap, A_TP_PARA_REG3, 0, F_TXPACEFIXED);
+
+	t3_write_reg(adap, A_TP_TX_MOD_QUEUE_WEIGHT1, 0x12121212);
+	t3_write_reg(adap, A_TP_TX_MOD_QUEUE_WEIGHT0, 0x12121212);
+	t3_write_reg(adap, A_TP_MOD_CHANNEL_WEIGHT, 0x1212);
+}
+
+/* Desired TP timer resolution in usec */
+#define TP_TMR_RES 50
+
+/* TCP timer values in ms */
+#define TP_DACK_TIMER 50
+#define TP_RTO_MIN    250
+
+/**
+ *	tp_set_timers - set TP timing parameters
+ *	@adap: the adapter to set
+ *	@core_clk: the core clock frequency in Hz
+ *
+ *	Set TP's timing parameters, such as the various timer resolutions and
+ *	the TCP timer values.
+ */
+static void tp_set_timers(struct adapter *adap, unsigned int core_clk)
+{
+	unsigned int tre = fls(core_clk / (1000000 / TP_TMR_RES)) - 1;
+	unsigned int dack_re = fls(core_clk / 5000) - 1;	/* 200us */
+	unsigned int tstamp_re = fls(core_clk / 1000);	/* 1ms, at least */
+	unsigned int tps = core_clk >> tre;
+
+	t3_write_reg(adap, A_TP_TIMER_RESOLUTION, V_TIMERRESOLUTION(tre) |
+		     V_DELAYEDACKRESOLUTION(dack_re) |
+		     V_TIMESTAMPRESOLUTION(tstamp_re));
+	t3_write_reg(adap, A_TP_DACK_TIMER,
+		     (core_clk >> dack_re) / (1000 / TP_DACK_TIMER));
+	t3_write_reg(adap, A_TP_TCP_BACKOFF_REG0, 0x3020100);
+	t3_write_reg(adap, A_TP_TCP_BACKOFF_REG1, 0x7060504);
+	t3_write_reg(adap, A_TP_TCP_BACKOFF_REG2, 0xb0a0908);
+	t3_write_reg(adap, A_TP_TCP_BACKOFF_REG3, 0xf0e0d0c);
+	t3_write_reg(adap, A_TP_SHIFT_CNT, V_SYNSHIFTMAX(6) |
+		     V_RXTSHIFTMAXR1(4) | V_RXTSHIFTMAXR2(15) |
+		     V_PERSHIFTBACKOFFMAX(8) | V_PERSHIFTMAX(8) |
+		     V_KEEPALIVEMAX(9));
+
+#define SECONDS * tps
+
+	t3_write_reg(adap, A_TP_MSL, adap->params.rev > 0 ? 0 : 2 SECONDS);
+	t3_write_reg(adap, A_TP_RXT_MIN, tps / (1000 / TP_RTO_MIN));
+	t3_write_reg(adap, A_TP_RXT_MAX, 64 SECONDS);
+	t3_write_reg(adap, A_TP_PERS_MIN, 5 SECONDS);
+	t3_write_reg(adap, A_TP_PERS_MAX, 64 SECONDS);
+	t3_write_reg(adap, A_TP_KEEP_IDLE, 7200 SECONDS);
+	t3_write_reg(adap, A_TP_KEEP_INTVL, 75 SECONDS);
+	t3_write_reg(adap, A_TP_INIT_SRTT, 3 SECONDS);
+	t3_write_reg(adap, A_TP_FINWAIT2_TIMER, 600 SECONDS);
+
+#undef SECONDS
+}
+
+/**
+ *	t3_tp_set_coalescing_size - set receive coalescing size
+ *	@adap: the adapter
+ *	@size: the receive coalescing size
+ *	@psh: whether a set PSH bit should deliver coalesced data
+ *
+ *	Set the receive coalescing size and PSH bit handling.
+ */
+int t3_tp_set_coalescing_size(struct adapter *adap, unsigned int size, int psh)
+{
+	u32 val;
+
+	if (size > MAX_RX_COALESCING_LEN)
+		return -EINVAL;
+
+	val = t3_read_reg(adap, A_TP_PARA_REG3);
+	val &= ~(F_RXCOALESCEENABLE | F_RXCOALESCEPSHEN);
+
+	if (size) {
+		val |= F_RXCOALESCEENABLE;
+		if (psh)
+			val |= F_RXCOALESCEPSHEN;
+		t3_write_reg(adap, A_TP_PARA_REG2, V_RXCOALESCESIZE(size) |
+			     V_MAXRXDATA(MAX_RX_COALESCING_LEN));
+	}
+	t3_write_reg(adap, A_TP_PARA_REG3, val);
+	return 0;
+}
+
+/**
+ *	t3_tp_set_max_rxsize - set the max receive size
+ *	@adap: the adapter
+ *	@size: the max receive size
+ *
+ *	Set TP's max receive size.  This is the limit that applies when
+ *	receive coalescing is disabled.
+ */
+void t3_tp_set_max_rxsize(struct adapter *adap, unsigned int size)
+{
+	t3_write_reg(adap, A_TP_PARA_REG7,
+		     V_PMMAXXFERLEN0(size) | V_PMMAXXFERLEN1(size));
+}
+
+static void __devinit init_mtus(unsigned short mtus[])
+{
+	/*
+	 * See draft-mathis-plpmtud-00.txt for the values.  The min is 88 so
+	 * it can accomodate max size TCP/IP headers when SACK and timestamps
+	 * are enabled and still have at least 8 bytes of payload.
+	 */
+	mtus[0] = 88;
+	mtus[1] = 256;
+	mtus[2] = 512;
+	mtus[3] = 576;
+	mtus[4] = 808;
+	mtus[5] = 1024;
+	mtus[6] = 1280;
+	mtus[7] = 1492;
+	mtus[8] = 1500;
+	mtus[9] = 2002;
+	mtus[10] = 2048;
+	mtus[11] = 4096;
+	mtus[12] = 4352;
+	mtus[13] = 8192;
+	mtus[14] = 9000;
+	mtus[15] = 9600;
+}
+
+/*
+ * Initial congestion control parameters.
+ */
+static void __devinit init_cong_ctrl(unsigned short *a, unsigned short *b)
+{
+	a[0] = a[1] = a[2] = a[3] = a[4] = a[5] = a[6] = a[7] = a[8] = 1;
+	a[9] = 2;
+	a[10] = 3;
+	a[11] = 4;
+	a[12] = 5;
+	a[13] = 6;
+	a[14] = 7;
+	a[15] = 8;
+	a[16] = 9;
+	a[17] = 10;
+	a[18] = 14;
+	a[19] = 17;
+	a[20] = 21;
+	a[21] = 25;
+	a[22] = 30;
+	a[23] = 35;
+	a[24] = 45;
+	a[25] = 60;
+	a[26] = 80;
+	a[27] = 100;
+	a[28] = 200;
+	a[29] = 300;
+	a[30] = 400;
+	a[31] = 500;
+
+	b[0] = b[1] = b[2] = b[3] = b[4] = b[5] = b[6] = b[7] = b[8] = 0;
+	b[9] = b[10] = 1;
+	b[11] = b[12] = 2;
+	b[13] = b[14] = b[15] = b[16] = 3;
+	b[17] = b[18] = b[19] = b[20] = b[21] = 4;
+	b[22] = b[23] = b[24] = b[25] = b[26] = b[27] = 5;
+	b[28] = b[29] = 6;
+	b[30] = b[31] = 7;
+}
+
+/* The minimum additive increment value for the congestion control table */
+#define CC_MIN_INCR 2U
+
+/**
+ *	t3_load_mtus - write the MTU and congestion control HW tables
+ *	@adap: the adapter
+ *	@mtus: the unrestricted values for the MTU table
+ *	@alphs: the values for the congestion control alpha parameter
+ *	@beta: the values for the congestion control beta parameter
+ *	@mtu_cap: the maximum permitted effective MTU
+ *
+ *	Write the MTU table with the supplied MTUs capping each at &mtu_cap.
+ *	Update the high-speed congestion control table with the supplied alpha,
+ * 	beta, and MTUs.
+ */
+void t3_load_mtus(struct adapter *adap, unsigned short mtus[NMTUS],
+		  unsigned short alpha[NCCTRL_WIN],
+		  unsigned short beta[NCCTRL_WIN], unsigned short mtu_cap)
+{
+	static const unsigned int avg_pkts[NCCTRL_WIN] = {
+		2, 6, 10, 14, 20, 28, 40, 56, 80, 112, 160, 224, 320, 448, 640,
+		896, 1281, 1792, 2560, 3584, 5120, 7168, 10240, 14336, 20480,
+		28672, 40960, 57344, 81920, 114688, 163840, 229376
+	};
+
+	unsigned int i, w;
+
+	for (i = 0; i < NMTUS; ++i) {
+		unsigned int mtu = min(mtus[i], mtu_cap);
+		unsigned int log2 = fls(mtu);
+
+		if (!(mtu & ((1 << log2) >> 2)))	/* round */
+			log2--;
+		t3_write_reg(adap, A_TP_MTU_TABLE,
+			     (i << 24) | (log2 << 16) | mtu);
+
+		for (w = 0; w < NCCTRL_WIN; ++w) {
+			unsigned int inc;
+
+			inc = max(((mtu - 40) * alpha[w]) / avg_pkts[w],
+				  CC_MIN_INCR);
+
+			t3_write_reg(adap, A_TP_CCTRL_TABLE, (i << 21) |
+				     (w << 16) | (beta[w] << 13) | inc);
+		}
+	}
+}
+
+/**
+ *	t3_read_hw_mtus - returns the values in the HW MTU table
+ *	@adap: the adapter
+ *	@mtus: where to store the HW MTU values
+ *
+ *	Reads the HW MTU table.
+ */
+void t3_read_hw_mtus(struct adapter *adap, unsigned short mtus[NMTUS])
+{
+	int i;
+
+	for (i = 0; i < NMTUS; ++i) {
+		unsigned int val;
+
+		t3_write_reg(adap, A_TP_MTU_TABLE, 0xff000000 | i);
+		val = t3_read_reg(adap, A_TP_MTU_TABLE);
+		mtus[i] = val & 0x3fff;
+	}
+}
+
+/**
+ *	t3_get_cong_cntl_tab - reads the congestion control table
+ *	@adap: the adapter
+ *	@incr: where to store the alpha values
+ *
+ *	Reads the additive increments programmed into the HW congestion
+ *	control table.
+ */
+void t3_get_cong_cntl_tab(struct adapter *adap,
+			  unsigned short incr[NMTUS][NCCTRL_WIN])
+{
+	unsigned int mtu, w;
+
+	for (mtu = 0; mtu < NMTUS; ++mtu)
+		for (w = 0; w < NCCTRL_WIN; ++w) {
+			t3_write_reg(adap, A_TP_CCTRL_TABLE,
+				     0xffff0000 | (mtu << 5) | w);
+			incr[mtu][w] = t3_read_reg(adap, A_TP_CCTRL_TABLE) &
+				       0x1fff;
+		}
+}
+
+/**
+ *	t3_tp_get_mib_stats - read TP's MIB counters
+ *	@adap: the adapter
+ *	@tps: holds the returned counter values
+ *
+ *	Returns the values of TP's MIB counters.
+ */
+void t3_tp_get_mib_stats(struct adapter *adap, struct tp_mib_stats *tps)
+{
+	t3_read_indirect(adap, A_TP_MIB_INDEX, A_TP_MIB_RDATA, (u32 *) tps,
+			 sizeof(*tps) / sizeof(u32), 0);
+}
+
+#define ulp_region(adap, name, start, len) \
+	t3_write_reg((adap), A_ULPRX_ ## name ## _LLIMIT, (start)); \
+	t3_write_reg((adap), A_ULPRX_ ## name ## _ULIMIT, \
+		     (start) + (len) - 1); \
+	start += len
+
+#define ulptx_region(adap, name, start, len) \
+	t3_write_reg((adap), A_ULPTX_ ## name ## _LLIMIT, (start)); \
+	t3_write_reg((adap), A_ULPTX_ ## name ## _ULIMIT, \
+		     (start) + (len) - 1)
+
+static void ulp_config(struct adapter *adap, const struct tp_params *p)
+{
+	unsigned int m = p->chan_rx_size;
+
+	ulp_region(adap, ISCSI, m, p->chan_rx_size / 8);
+	ulp_region(adap, TDDP, m, p->chan_rx_size / 8);
+	ulptx_region(adap, TPT, m, p->chan_rx_size / 4);
+	ulp_region(adap, STAG, m, p->chan_rx_size / 4);
+	ulp_region(adap, RQ, m, p->chan_rx_size / 4);
+	ulptx_region(adap, PBL, m, p->chan_rx_size / 4);
+	ulp_region(adap, PBL, m, p->chan_rx_size / 4);
+	t3_write_reg(adap, A_ULPRX_TDDP_TAGMASK, 0xffffffff);
+}
+
+void t3_config_trace_filter(struct adapter *adapter,
+			    const struct trace_params *tp, int filter_index,
+			    int invert, int enable)
+{
+	u32 addr, key[4], mask[4];
+
+	key[0] = tp->sport | (tp->sip << 16);
+	key[1] = (tp->sip >> 16) | (tp->dport << 16);
+	key[2] = tp->dip;
+	key[3] = tp->proto | (tp->vlan << 8) | (tp->intf << 20);
+
+	mask[0] = tp->sport_mask | (tp->sip_mask << 16);
+	mask[1] = (tp->sip_mask >> 16) | (tp->dport_mask << 16);
+	mask[2] = tp->dip_mask;
+	mask[3] = tp->proto_mask | (tp->vlan_mask << 8) | (tp->intf_mask << 20);
+
+	if (invert)
+		key[3] |= (1 << 29);
+	if (enable)
+		key[3] |= (1 << 28);
+
+	addr = filter_index ? A_TP_RX_TRC_KEY0 : A_TP_TX_TRC_KEY0;
+	tp_wr_indirect(adapter, addr++, key[0]);
+	tp_wr_indirect(adapter, addr++, mask[0]);
+	tp_wr_indirect(adapter, addr++, key[1]);
+	tp_wr_indirect(adapter, addr++, mask[1]);
+	tp_wr_indirect(adapter, addr++, key[2]);
+	tp_wr_indirect(adapter, addr++, mask[2]);
+	tp_wr_indirect(adapter, addr++, key[3]);
+	tp_wr_indirect(adapter, addr, mask[3]);
+	t3_read_reg(adapter, A_TP_PIO_DATA);
+}
+
+/**
+ *	t3_config_sched - configure a HW traffic scheduler
+ *	@adap: the adapter
+ *	@kbps: target rate in Kbps
+ *	@sched: the scheduler index
+ *
+ *	Configure a HW scheduler for the target rate
+ */
+int t3_config_sched(struct adapter *adap, unsigned int kbps, int sched)
+{
+	unsigned int v, tps, cpt, bpt, delta, mindelta = ~0;
+	unsigned int clk = adap->params.vpd.cclk * 1000;
+	unsigned int selected_cpt = 0, selected_bpt = 0;
+
+	if (kbps > 0) {
+		kbps *= 125;	/* -> bytes */
+		for (cpt = 1; cpt <= 255; cpt++) {
+			tps = clk / cpt;
+			bpt = (kbps + tps / 2) / tps;
+			if (bpt > 0 && bpt <= 255) {
+				v = bpt * tps;
+				delta = v >= kbps ? v - kbps : kbps - v;
+				if (delta <= mindelta) {
+					mindelta = delta;
+					selected_cpt = cpt;
+					selected_bpt = bpt;
+				}
+			} else if (selected_cpt)
+				break;
+		}
+		if (!selected_cpt)
+			return -EINVAL;
+	}
+	t3_write_reg(adap, A_TP_TM_PIO_ADDR,
+		     A_TP_TX_MOD_Q1_Q0_RATE_LIMIT - sched / 2);
+	v = t3_read_reg(adap, A_TP_TM_PIO_DATA);
+	if (sched & 1)
+		v = (v & 0xffff) | (selected_cpt << 16) | (selected_bpt << 24);
+	else
+		v = (v & 0xffff0000) | selected_cpt | (selected_bpt << 8);
+	t3_write_reg(adap, A_TP_TM_PIO_DATA, v);
+	return 0;
+}
+
+static int tp_init(struct adapter *adap, const struct tp_params *p)
+{
+	int busy = 0;
+
+	tp_config(adap, p);
+	t3_set_vlan_accel(adap, 3, 0);
+
+	if (is_offload(adap)) {
+		tp_set_timers(adap, adap->params.vpd.cclk * 1000);
+		t3_write_reg(adap, A_TP_RESET, F_FLSTINITENABLE);
+		busy = t3_wait_op_done(adap, A_TP_RESET, F_FLSTINITENABLE,
+				       0, 1000, 5);
+		if (busy)
+			CH_ERR(adap, "TP initialization timed out\n");
+	}
+
+	if (!busy)
+		t3_write_reg(adap, A_TP_RESET, F_TPRESET);
+	return busy;
+}
+
+int t3_mps_set_active_ports(struct adapter *adap, unsigned int port_mask)
+{
+	if (port_mask & ~((1 << adap->params.nports) - 1))
+		return -EINVAL;
+	t3_set_reg_field(adap, A_MPS_CFG, F_PORT1ACTIVE | F_PORT0ACTIVE,
+			 port_mask << S_PORT0ACTIVE);
+	return 0;
+}
+
+/*
+ * Perform the bits of HW initialization that are dependent on the number
+ * of available ports.
+ */
+static void init_hw_for_avail_ports(struct adapter *adap, int nports)
+{
+	int i;
+
+	if (nports == 1) {
+		t3_set_reg_field(adap, A_ULPRX_CTL, F_ROUND_ROBIN, 0);
+		t3_set_reg_field(adap, A_ULPTX_CONFIG, F_CFG_RR_ARB, 0);
+		t3_write_reg(adap, A_MPS_CFG, F_TPRXPORTEN | F_TPTXPORT0EN |
+			     F_PORT0ACTIVE | F_ENFORCEPKT);
+		t3_write_reg(adap, A_PM1_TX_CFG, 0xc000c000);
+	} else {
+		t3_set_reg_field(adap, A_ULPRX_CTL, 0, F_ROUND_ROBIN);
+		t3_set_reg_field(adap, A_ULPTX_CONFIG, 0, F_CFG_RR_ARB);
+		t3_write_reg(adap, A_ULPTX_DMA_WEIGHT,
+			     V_D1_WEIGHT(16) | V_D0_WEIGHT(16));
+		t3_write_reg(adap, A_MPS_CFG, F_TPTXPORT0EN | F_TPTXPORT1EN |
+			     F_TPRXPORTEN | F_PORT0ACTIVE | F_PORT1ACTIVE |
+			     F_ENFORCEPKT);
+		t3_write_reg(adap, A_PM1_TX_CFG, 0x80008000);
+		t3_set_reg_field(adap, A_TP_PC_CONFIG, 0, F_TXTOSQUEUEMAPMODE);
+		t3_write_reg(adap, A_TP_TX_MOD_QUEUE_REQ_MAP,
+			     V_TX_MOD_QUEUE_REQ_MAP(0xaa));
+		for (i = 0; i < 16; i++)
+			t3_write_reg(adap, A_TP_TX_MOD_QUE_TABLE,
+				     (i << 16) | 0x1010);
+	}
+}
+
+static int calibrate_xgm(struct adapter *adapter)
+{
+	if (uses_xaui(adapter)) {
+		unsigned int v, i;
+
+		for (i = 0; i < 5; ++i) {
+			t3_write_reg(adapter, A_XGM_XAUI_IMP, 0);
+			t3_read_reg(adapter, A_XGM_XAUI_IMP);
+			msleep(1);
+			v = t3_read_reg(adapter, A_XGM_XAUI_IMP);
+			if (!(v & (F_XGM_CALFAULT | F_CALBUSY))) {
+				t3_write_reg(adapter, A_XGM_XAUI_IMP,
+					     V_XAUIIMP(G_CALIMP(v) >> 2));
+				return 0;
+			}
+		}
+		CH_ERR(adapter, "MAC calibration failed\n");
+		return -1;
+	} else {
+		t3_write_reg(adapter, A_XGM_RGMII_IMP,
+			     V_RGMIIIMPPD(2) | V_RGMIIIMPPU(3));
+		t3_set_reg_field(adapter, A_XGM_RGMII_IMP, F_XGM_IMPSETUPDATE,
+				 F_XGM_IMPSETUPDATE);
+	}
+	return 0;
+}
+
+static void calibrate_xgm_t3b(struct adapter *adapter)
+{
+	if (!uses_xaui(adapter)) {
+		t3_write_reg(adapter, A_XGM_RGMII_IMP, F_CALRESET |
+			     F_CALUPDATE | V_RGMIIIMPPD(2) | V_RGMIIIMPPU(3));
+		t3_set_reg_field(adapter, A_XGM_RGMII_IMP, F_CALRESET, 0);
+		t3_set_reg_field(adapter, A_XGM_RGMII_IMP, 0,
+				 F_XGM_IMPSETUPDATE);
+		t3_set_reg_field(adapter, A_XGM_RGMII_IMP, F_XGM_IMPSETUPDATE,
+				 0);
+		t3_set_reg_field(adapter, A_XGM_RGMII_IMP, F_CALUPDATE, 0);
+		t3_set_reg_field(adapter, A_XGM_RGMII_IMP, 0, F_CALUPDATE);
+	}
+}
+
+struct mc7_timing_params {
+	unsigned char ActToPreDly;
+	unsigned char ActToRdWrDly;
+	unsigned char PreCyc;
+	unsigned char RefCyc[5];
+	unsigned char BkCyc;
+	unsigned char WrToRdDly;
+	unsigned char RdToWrDly;
+};
+
+/*
+ * Write a value to a register and check that the write completed.  These
+ * writes normally complete in a cycle or two, so one read should suffice.
+ * The very first read exists to flush the posted write to the device.
+ */
+static int wrreg_wait(struct adapter *adapter, unsigned int addr, u32 val)
+{
+	t3_write_reg(adapter, addr, val);
+	t3_read_reg(adapter, addr);	/* flush */
+	if (!(t3_read_reg(adapter, addr) & F_BUSY))
+		return 0;
+	CH_ERR(adapter, "write to MC7 register 0x%x timed out\n", addr);
+	return -EIO;
+}
+
+static int mc7_init(struct mc7 *mc7, unsigned int mc7_clock, int mem_type)
+{
+	static const unsigned int mc7_mode[] = {
+		0x632, 0x642, 0x652, 0x432, 0x442
+	};
+	static const struct mc7_timing_params mc7_timings[] = {
+		{12, 3, 4, {20, 28, 34, 52, 0}, 15, 6, 4},
+		{12, 4, 5, {20, 28, 34, 52, 0}, 16, 7, 4},
+		{12, 5, 6, {20, 28, 34, 52, 0}, 17, 8, 4},
+		{9, 3, 4, {15, 21, 26, 39, 0}, 12, 6, 4},
+		{9, 4, 5, {15, 21, 26, 39, 0}, 13, 7, 4}
+	};
+
+	u32 val;
+	unsigned int width, density, slow, attempts;
+	struct adapter *adapter = mc7->adapter;
+	const struct mc7_timing_params *p = &mc7_timings[mem_type];
+
+	val = t3_read_reg(adapter, mc7->offset + A_MC7_CFG);
+	slow = val & F_SLOW;
+	width = G_WIDTH(val);
+	density = G_DEN(val);
+
+	t3_write_reg(adapter, mc7->offset + A_MC7_CFG, val | F_IFEN);
+	val = t3_read_reg(adapter, mc7->offset + A_MC7_CFG);	/* flush */
+	msleep(1);
+
+	if (!slow) {
+		t3_write_reg(adapter, mc7->offset + A_MC7_CAL, F_SGL_CAL_EN);
+		t3_read_reg(adapter, mc7->offset + A_MC7_CAL);
+		msleep(1);
+		if (t3_read_reg(adapter, mc7->offset + A_MC7_CAL) &
+		    (F_BUSY | F_SGL_CAL_EN | F_CAL_FAULT)) {
+			CH_ERR(adapter, "%s MC7 calibration timed out\n",
+			       mc7->name);
+			goto out_fail;
+		}
+	}
+
+	t3_write_reg(adapter, mc7->offset + A_MC7_PARM,
+		     V_ACTTOPREDLY(p->ActToPreDly) |
+		     V_ACTTORDWRDLY(p->ActToRdWrDly) | V_PRECYC(p->PreCyc) |
+		     V_REFCYC(p->RefCyc[density]) | V_BKCYC(p->BkCyc) |
+		     V_WRTORDDLY(p->WrToRdDly) | V_RDTOWRDLY(p->RdToWrDly));
+
+	t3_write_reg(adapter, mc7->offset + A_MC7_CFG,
+		     val | F_CLKEN | F_TERM150);
+	t3_read_reg(adapter, mc7->offset + A_MC7_CFG);	/* flush */
+
+	if (!slow)
+		t3_set_reg_field(adapter, mc7->offset + A_MC7_DLL, F_DLLENB,
+				 F_DLLENB);
+	udelay(1);
+
+	val = slow ? 3 : 6;
+	if (wrreg_wait(adapter, mc7->offset + A_MC7_PRE, 0) ||
+	    wrreg_wait(adapter, mc7->offset + A_MC7_EXT_MODE2, 0) ||
+	    wrreg_wait(adapter, mc7->offset + A_MC7_EXT_MODE3, 0) ||
+	    wrreg_wait(adapter, mc7->offset + A_MC7_EXT_MODE1, val))
+		goto out_fail;
+
+	if (!slow) {
+		t3_write_reg(adapter, mc7->offset + A_MC7_MODE, 0x100);
+		t3_set_reg_field(adapter, mc7->offset + A_MC7_DLL, F_DLLRST, 0);
+		udelay(5);
+	}
+
+	if (wrreg_wait(adapter, mc7->offset + A_MC7_PRE, 0) ||
+	    wrreg_wait(adapter, mc7->offset + A_MC7_REF, 0) ||
+	    wrreg_wait(adapter, mc7->offset + A_MC7_REF, 0) ||
+	    wrreg_wait(adapter, mc7->offset + A_MC7_MODE,
+		       mc7_mode[mem_type]) ||
+	    wrreg_wait(adapter, mc7->offset + A_MC7_EXT_MODE1, val | 0x380) ||
+	    wrreg_wait(adapter, mc7->offset + A_MC7_EXT_MODE1, val))
+		goto out_fail;
+
+	/* clock value is in KHz */
+	mc7_clock = mc7_clock * 7812 + mc7_clock / 2;	/* ns */
+	mc7_clock /= 1000000;	/* KHz->MHz, ns->us */
+
+	t3_write_reg(adapter, mc7->offset + A_MC7_REF,
+		     F_PERREFEN | V_PREREFDIV(mc7_clock));
+	t3_read_reg(adapter, mc7->offset + A_MC7_REF);	/* flush */
+
+	t3_write_reg(adapter, mc7->offset + A_MC7_ECC, F_ECCGENEN | F_ECCCHKEN);
+	t3_write_reg(adapter, mc7->offset + A_MC7_BIST_DATA, 0);
+	t3_write_reg(adapter, mc7->offset + A_MC7_BIST_ADDR_BEG, 0);
+	t3_write_reg(adapter, mc7->offset + A_MC7_BIST_ADDR_END,
+		     (mc7->size << width) - 1);
+	t3_write_reg(adapter, mc7->offset + A_MC7_BIST_OP, V_OP(1));
+	t3_read_reg(adapter, mc7->offset + A_MC7_BIST_OP);	/* flush */
+
+	attempts = 50;
+	do {
+		msleep(250);
+		val = t3_read_reg(adapter, mc7->offset + A_MC7_BIST_OP);
+	} while ((val & F_BUSY) && --attempts);
+	if (val & F_BUSY) {
+		CH_ERR(adapter, "%s MC7 BIST timed out\n", mc7->name);
+		goto out_fail;
+	}
+
+	/* Enable normal memory accesses. */
+	t3_set_reg_field(adapter, mc7->offset + A_MC7_CFG, 0, F_RDY);
+	return 0;
+
+out_fail:
+	return -1;
+}
+
+static void config_pcie(struct adapter *adap)
+{
+	static const u16 ack_lat[4][6] = {
+		{237, 416, 559, 1071, 2095, 4143},
+		{128, 217, 289, 545, 1057, 2081},
+		{73, 118, 154, 282, 538, 1050},
+		{67, 107, 86, 150, 278, 534}
+	};
+	static const u16 rpl_tmr[4][6] = {
+		{711, 1248, 1677, 3213, 6285, 12429},
+		{384, 651, 867, 1635, 3171, 6243},
+		{219, 354, 462, 846, 1614, 3150},
+		{201, 321, 258, 450, 834, 1602}
+	};
+
+	u16 val;
+	unsigned int log2_width, pldsize;
+	unsigned int fst_trn_rx, fst_trn_tx, acklat, rpllmt;
+
+	pci_read_config_word(adap->pdev,
+			     adap->params.pci.pcie_cap_addr + PCI_EXP_DEVCTL,
+			     &val);
+	pldsize = (val & PCI_EXP_DEVCTL_PAYLOAD) >> 5;
+	pci_read_config_word(adap->pdev,
+			     adap->params.pci.pcie_cap_addr + PCI_EXP_LNKCTL,
+			     &val);
+
+	fst_trn_tx = G_NUMFSTTRNSEQ(t3_read_reg(adap, A_PCIE_PEX_CTRL0));
+	fst_trn_rx = adap->params.rev == 0 ? fst_trn_tx :
+	    G_NUMFSTTRNSEQRX(t3_read_reg(adap, A_PCIE_MODE));
+	log2_width = fls(adap->params.pci.width) - 1;
+	acklat = ack_lat[log2_width][pldsize];
+	if (val & 1)		/* check LOsEnable */
+		acklat += fst_trn_tx * 4;
+	rpllmt = rpl_tmr[log2_width][pldsize] + fst_trn_rx * 4;
+
+	if (adap->params.rev == 0)
+		t3_set_reg_field(adap, A_PCIE_PEX_CTRL1,
+				 V_T3A_ACKLAT(M_T3A_ACKLAT),
+				 V_T3A_ACKLAT(acklat));
+	else
+		t3_set_reg_field(adap, A_PCIE_PEX_CTRL1, V_ACKLAT(M_ACKLAT),
+				 V_ACKLAT(acklat));
+
+	t3_set_reg_field(adap, A_PCIE_PEX_CTRL0, V_REPLAYLMT(M_REPLAYLMT),
+			 V_REPLAYLMT(rpllmt));
+
+	t3_write_reg(adap, A_PCIE_PEX_ERR, 0xffffffff);
+	t3_set_reg_field(adap, A_PCIE_CFG, F_PCIE_CLIDECEN, F_PCIE_CLIDECEN);
+}
+
+/*
+ * Initialize and configure T3 HW modules.  This performs the
+ * initialization steps that need to be done once after a card is reset.
+ * MAC and PHY initialization is handled separarely whenever a port is enabled.
+ *
+ * fw_params are passed to FW and their value is platform dependent.  Only the
+ * top 8 bits are available for use, the rest must be 0.
+ */
+int t3_init_hw(struct adapter *adapter, u32 fw_params)
+{
+	int err = -EIO, attempts = 100;
+	const struct vpd_params *vpd = &adapter->params.vpd;
+
+	if (adapter->params.rev > 0)
+		calibrate_xgm_t3b(adapter);
+	else if (calibrate_xgm(adapter))
+		goto out_err;
+
+	if (vpd->mclk) {
+		partition_mem(adapter, &adapter->params.tp);
+
+		if (mc7_init(&adapter->pmrx, vpd->mclk, vpd->mem_timing) ||
+		    mc7_init(&adapter->pmtx, vpd->mclk, vpd->mem_timing) ||
+		    mc7_init(&adapter->cm, vpd->mclk, vpd->mem_timing) ||
+		    t3_mc5_init(&adapter->mc5, adapter->params.mc5.nservers,
+				adapter->params.mc5.nfilters,
+				adapter->params.mc5.nroutes))
+			goto out_err;
+	}
+
+	if (tp_init(adapter, &adapter->params.tp))
+		goto out_err;
+
+	t3_tp_set_coalescing_size(adapter,
+				  min(adapter->params.sge.max_pkt_size,
+				      MAX_RX_COALESCING_LEN), 1);
+	t3_tp_set_max_rxsize(adapter,
+			     min(adapter->params.sge.max_pkt_size, 16384U));
+	ulp_config(adapter, &adapter->params.tp);
+
+	if (is_pcie(adapter))
+		config_pcie(adapter);
+	else
+		t3_set_reg_field(adapter, A_PCIX_CFG, 0, F_CLIDECEN);
+
+	t3_write_reg(adapter, A_PM1_RX_CFG, 0xf000f000);
+	init_hw_for_avail_ports(adapter, adapter->params.nports);
+	t3_sge_init(adapter, &adapter->params.sge);
+
+	t3_write_reg(adapter, A_CIM_HOST_ACC_DATA, vpd->uclk | fw_params);
+	t3_write_reg(adapter, A_CIM_BOOT_CFG,
+		     V_BOOTADDR(FW_FLASH_BOOT_ADDR >> 2));
+	t3_read_reg(adapter, A_CIM_BOOT_CFG);	/* flush */
+
+	do {			/* wait for uP to initialize */
+		msleep(20);
+	} while (t3_read_reg(adapter, A_CIM_HOST_ACC_DATA) && --attempts);
+	if (!attempts)
+		goto out_err;
+
+	err = 0;
+out_err:
+	return err;
+}
+
+/**
+ *	get_pci_mode - determine a card's PCI mode
+ *	@adapter: the adapter
+ *	@p: where to store the PCI settings
+ *
+ *	Determines a card's PCI mode and associated parameters, such as speed
+ *	and width.
+ */
+static void __devinit get_pci_mode(struct adapter *adapter,
+				   struct pci_params *p)
+{
+	static unsigned short speed_map[] = { 33, 66, 100, 133 };
+	u32 pci_mode, pcie_cap;
+
+	pcie_cap = pci_find_capability(adapter->pdev, PCI_CAP_ID_EXP);
+	if (pcie_cap) {
+		u16 val;
+
+		p->variant = PCI_VARIANT_PCIE;
+		p->pcie_cap_addr = pcie_cap;
+		pci_read_config_word(adapter->pdev, pcie_cap + PCI_EXP_LNKSTA,
+					&val);
+		p->width = (val >> 4) & 0x3f;
+		return;
+	}
+
+	pci_mode = t3_read_reg(adapter, A_PCIX_MODE);
+	p->speed = speed_map[G_PCLKRANGE(pci_mode)];
+	p->width = (pci_mode & F_64BIT) ? 64 : 32;
+	pci_mode = G_PCIXINITPAT(pci_mode);
+	if (pci_mode == 0)
+		p->variant = PCI_VARIANT_PCI;
+	else if (pci_mode < 4)
+		p->variant = PCI_VARIANT_PCIX_MODE1_PARITY;
+	else if (pci_mode < 8)
+		p->variant = PCI_VARIANT_PCIX_MODE1_ECC;
+	else
+		p->variant = PCI_VARIANT_PCIX_266_MODE2;
+}
+
+/**
+ *	init_link_config - initialize a link's SW state
+ *	@lc: structure holding the link state
+ *	@ai: information about the current card
+ *
+ *	Initializes the SW state maintained for each link, including the link's
+ *	capabilities and default speed/duplex/flow-control/autonegotiation
+ *	settings.
+ */
+static void __devinit init_link_config(struct link_config *lc,
+				       unsigned int caps)
+{
+	lc->supported = caps;
+	lc->requested_speed = lc->speed = SPEED_INVALID;
+	lc->requested_duplex = lc->duplex = DUPLEX_INVALID;
+	lc->requested_fc = lc->fc = PAUSE_RX | PAUSE_TX;
+	if (lc->supported & SUPPORTED_Autoneg) {
+		lc->advertising = lc->supported;
+		lc->autoneg = AUTONEG_ENABLE;
+		lc->requested_fc |= PAUSE_AUTONEG;
+	} else {
+		lc->advertising = 0;
+		lc->autoneg = AUTONEG_DISABLE;
+	}
+}
+
+/**
+ *	mc7_calc_size - calculate MC7 memory size
+ *	@cfg: the MC7 configuration
+ *
+ *	Calculates the size of an MC7 memory in bytes from the value of its
+ *	configuration register.
+ */
+static unsigned int __devinit mc7_calc_size(u32 cfg)
+{
+	unsigned int width = G_WIDTH(cfg);
+	unsigned int banks = !!(cfg & F_BKS) + 1;
+	unsigned int org = !!(cfg & F_ORG) + 1;
+	unsigned int density = G_DEN(cfg);
+	unsigned int MBs = ((256 << density) * banks) / (org << width);
+
+	return MBs << 20;
+}
+
+static void __devinit mc7_prep(struct adapter *adapter, struct mc7 *mc7,
+			       unsigned int base_addr, const char *name)
+{
+	u32 cfg;
+
+	mc7->adapter = adapter;
+	mc7->name = name;
+	mc7->offset = base_addr - MC7_PMRX_BASE_ADDR;
+	cfg = t3_read_reg(adapter, mc7->offset + A_MC7_CFG);
+	mc7->size = mc7_calc_size(cfg);
+	mc7->width = G_WIDTH(cfg);
+}
+
+void mac_prep(struct cmac *mac, struct adapter *adapter, int index)
+{
+	mac->adapter = adapter;
+	mac->offset = (XGMAC0_1_BASE_ADDR - XGMAC0_0_BASE_ADDR) * index;
+	mac->nucast = 1;
+
+	if (adapter->params.rev == 0 && uses_xaui(adapter)) {
+		t3_write_reg(adapter, A_XGM_SERDES_CTRL + mac->offset,
+			     is_10G(adapter) ? 0x2901c04 : 0x2301c04);
+		t3_set_reg_field(adapter, A_XGM_PORT_CFG + mac->offset,
+				 F_ENRGMII, 0);
+	}
+}
+
+void early_hw_init(struct adapter *adapter, const struct adapter_info *ai)
+{
+	u32 val = V_PORTSPEED(is_10G(adapter) ? 3 : 2);
+
+	mi1_init(adapter, ai);
+	t3_write_reg(adapter, A_I2C_CFG,	/* set for 80KHz */
+		     V_I2C_CLKDIV(adapter->params.vpd.cclk / 80 - 1));
+	t3_write_reg(adapter, A_T3DBG_GPIO_EN,
+		     ai->gpio_out | F_GPIO0_OEN | F_GPIO0_OUT_VAL);
+
+	if (adapter->params.rev == 0 || !uses_xaui(adapter))
+		val |= F_ENRGMII;
+
+	/* Enable MAC clocks so we can access the registers */
+	t3_write_reg(adapter, A_XGM_PORT_CFG, val);
+	t3_read_reg(adapter, A_XGM_PORT_CFG);
+
+	val |= F_CLKDIVRESET_;
+	t3_write_reg(adapter, A_XGM_PORT_CFG, val);
+	t3_read_reg(adapter, A_XGM_PORT_CFG);
+	t3_write_reg(adapter, XGM_REG(A_XGM_PORT_CFG, 1), val);
+	t3_read_reg(adapter, A_XGM_PORT_CFG);
+}
+
+/*
+ * Reset the adapter.  PCIe cards lose their config space during reset, PCI-X
+ * ones don't.
+ */
+int t3_reset_adapter(struct adapter *adapter)
+{
+	int i;
+	uint16_t devid = 0;
+
+	if (is_pcie(adapter))
+		pci_save_state(adapter->pdev);
+	t3_write_reg(adapter, A_PL_RST, F_CRSTWRM | F_CRSTWRMMODE);
+
+	/*
+	 * Delay. Give Some time to device to reset fully.
+	 * XXX The delay time should be modified.
+	 */
+	for (i = 0; i < 10; i++) {
+		msleep(50);
+		pci_read_config_word(adapter->pdev, 0x00, &devid);
+		if (devid == 0x1425)
+			break;
+	}
+
+	if (devid != 0x1425)
+		return -1;
+
+	if (is_pcie(adapter))
+		pci_restore_state(adapter->pdev);
+	return 0;
+}
+
+/*
+ * Initialize adapter SW state for the various HW modules, set initial values
+ * for some adapter tunables, take PHYs out of reset, and initialize the MDIO
+ * interface.
+ */
+int __devinit t3_prep_adapter(struct adapter *adapter,
+			      const struct adapter_info *ai, int reset)
+{
+	int ret;
+	unsigned int i, j = 0;
+
+	get_pci_mode(adapter, &adapter->params.pci);
+
+	adapter->params.info = ai;
+	adapter->params.nports = ai->nports;
+	adapter->params.rev = t3_read_reg(adapter, A_PL_REV);
+	adapter->params.linkpoll_period = 0;
+	adapter->params.stats_update_period = is_10G(adapter) ?
+	    MAC_STATS_ACCUM_SECS : (MAC_STATS_ACCUM_SECS * 10);
+	adapter->params.pci.vpd_cap_addr =
+	    pci_find_capability(adapter->pdev, PCI_CAP_ID_VPD);
+	ret = get_vpd_params(adapter, &adapter->params.vpd);
+	if (ret < 0)
+		return ret;
+
+	if (reset && t3_reset_adapter(adapter))
+		return -1;
+
+	t3_sge_prep(adapter, &adapter->params.sge);
+
+	if (adapter->params.vpd.mclk) {
+		struct tp_params *p = &adapter->params.tp;
+
+		mc7_prep(adapter, &adapter->pmrx, MC7_PMRX_BASE_ADDR, "PMRX");
+		mc7_prep(adapter, &adapter->pmtx, MC7_PMTX_BASE_ADDR, "PMTX");
+		mc7_prep(adapter, &adapter->cm, MC7_CM_BASE_ADDR, "CM");
+
+		p->nchan = ai->nports;
+		p->pmrx_size = t3_mc7_size(&adapter->pmrx);
+		p->pmtx_size = t3_mc7_size(&adapter->pmtx);
+		p->cm_size = t3_mc7_size(&adapter->cm);
+		p->chan_rx_size = p->pmrx_size / 2;	/* only 1 Rx channel */
+		p->chan_tx_size = p->pmtx_size / p->nchan;
+		p->rx_pg_size = 64 * 1024;
+		p->tx_pg_size = is_10G(adapter) ? 64 * 1024 : 16 * 1024;
+		p->rx_num_pgs = pm_num_pages(p->chan_rx_size, p->rx_pg_size);
+		p->tx_num_pgs = pm_num_pages(p->chan_tx_size, p->tx_pg_size);
+		p->ntimer_qs = p->cm_size >= (128 << 20) ||
+		    adapter->params.rev > 0 ? 12 : 6;
+
+		adapter->params.mc5.nservers = DEFAULT_NSERVERS;
+		adapter->params.mc5.nfilters = adapter->params.rev > 0 ?
+		    DEFAULT_NFILTERS : 0;
+		adapter->params.mc5.nroutes = 0;
+		t3_mc5_prep(adapter, &adapter->mc5, MC5_MODE_144_BIT);
+
+		init_mtus(adapter->params.mtus);
+		init_cong_ctrl(adapter->params.a_wnd, adapter->params.b_wnd);
+	}
+
+	early_hw_init(adapter, ai);
+
+	for_each_port(adapter, i) {
+		u8 hw_addr[6];
+		struct port_info *p = adap2pinfo(adapter, i);
+
+		while (!adapter->params.vpd.port_type[j])
+			++j;
+
+		p->port_type = &port_types[adapter->params.vpd.port_type[j]];
+		p->port_type->phy_prep(&p->phy, adapter, ai->phy_base_addr + j,
+				       ai->mdio_ops);
+		mac_prep(&p->mac, adapter, j);
+		++j;
+
+		/*
+		 * The VPD EEPROM stores the base Ethernet address for the
+		 * card.  A port's address is derived from the base by adding
+		 * the port's index to the base's low octet.
+		 */
+		memcpy(hw_addr, adapter->params.vpd.eth_base, 5);
+		hw_addr[5] = adapter->params.vpd.eth_base[5] + i;
+
+		memcpy(adapter->port[i]->dev_addr, hw_addr,
+		       ETH_ALEN);
+		memcpy(adapter->port[i]->perm_addr, hw_addr,
+		       ETH_ALEN);
+		init_link_config(&p->link_config, p->port_type->caps);
+		p->phy.ops->power_down(&p->phy, 1);
+		if (!(p->port_type->caps & SUPPORTED_IRQ))
+			adapter->params.linkpoll_period = 10;
+	}
+
+	return 0;
+}
+
+void t3_led_ready(struct adapter *adapter)
+{
+	t3_set_reg_field(adapter, A_T3DBG_GPIO_EN, F_GPIO0_OUT_VAL,
+			 F_GPIO0_OUT_VAL);
+}
diff --git a/drivers/net/cxgb3/t3cdev.h b/drivers/net/cxgb3/t3cdev.h
new file mode 100644
index 00000000000..359584e7d58
--- /dev/null
+++ b/drivers/net/cxgb3/t3cdev.h
@@ -0,0 +1,72 @@
+/*
+ * Copyright (C) 2003-2006 Chelsio Communications.  All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef _T3CDEV_H_
+#define _T3CDEV_H_
+
+#include <linux/list.h>
+#include <asm/atomic.h>
+#include <asm/semaphore.h>
+#include <linux/netdevice.h>
+#include <linux/proc_fs.h>
+#include <linux/skbuff.h>
+#include <net/neighbour.h>
+
+#define T3CNAMSIZ 16
+
+/* Get the t3cdev associated with a net_device */
+#define T3CDEV(netdev) (struct t3cdev *)(netdev->priv)
+
+struct cxgb3_client;
+
+enum t3ctype {
+	T3A = 0,
+	T3B
+};
+
+struct t3cdev {
+	char name[T3CNAMSIZ];	/* T3C device name */
+	enum t3ctype type;
+	struct list_head ofld_dev_list;	/* for list linking */
+	struct net_device *lldev;	/* LL dev associated with T3C messages */
+	struct proc_dir_entry *proc_dir;	/* root of proc dir for this T3C */
+	int (*send)(struct t3cdev *dev, struct sk_buff *skb);
+	int (*recv)(struct t3cdev *dev, struct sk_buff **skb, int n);
+	int (*ctl)(struct t3cdev *dev, unsigned int req, void *data);
+	void (*neigh_update)(struct t3cdev *dev, struct neighbour *neigh);
+	void *priv;		/* driver private data */
+	void *l2opt;		/* optional layer 2 data */
+	void *l3opt;		/* optional layer 3 data */
+	void *l4opt;		/* optional layer 4 data */
+	void *ulp;		/* ulp stuff */
+};
+
+#endif				/* _T3CDEV_H_ */
diff --git a/drivers/net/cxgb3/version.h b/drivers/net/cxgb3/version.h
new file mode 100644
index 00000000000..1413ea37f41
--- /dev/null
+++ b/drivers/net/cxgb3/version.h
@@ -0,0 +1,24 @@
+/*****************************************************************************
+ *                                                                           *
+ * File:                                                                     *
+ *  version.h                                                                *
+ *                                                                           *
+ * Description:                                                              *
+ *  Chelsio driver version defines.                                          *
+ *                                                                           *
+ * Copyright (c) 2003 - 2006 Chelsio Communications, Inc.                    *
+ * All rights reserved.                                                      *
+ *                                                                           *
+ * Maintainers: maintainers@chelsio.com                                      *
+ *                                                                           *
+ * http://www.chelsio.com                                                    *
+ *                                                                           *
+ ****************************************************************************/
+/* $Date: 2006/10/31 18:57:51 $ $RCSfile: version.h,v $ $Revision: 1.3 $ */
+#ifndef __CHELSIO_VERSION_H
+#define __CHELSIO_VERSION_H
+#define DRV_DESC "Chelsio T3 Network Driver"
+#define DRV_NAME "cxgb3"
+/* Driver version */
+#define DRV_VERSION "1.0"
+#endif				/* __CHELSIO_VERSION_H */
diff --git a/drivers/net/cxgb3/vsc8211.c b/drivers/net/cxgb3/vsc8211.c
new file mode 100644
index 00000000000..6a0a815b89c
--- /dev/null
+++ b/drivers/net/cxgb3/vsc8211.c
@@ -0,0 +1,208 @@
+/*
+ * This file is part of the Chelsio T3 Ethernet driver.
+ *
+ * Copyright (C) 2005-2006 Chelsio Communications.  All rights reserved.
+ *
+ * This program is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the LICENSE file included in this
+ * release for licensing terms and conditions.
+ */
+
+#include "common.h"
+
+/* VSC8211 PHY specific registers. */
+enum {
+	VSC8211_INTR_ENABLE = 25,
+	VSC8211_INTR_STATUS = 26,
+	VSC8211_AUX_CTRL_STAT = 28,
+};
+
+enum {
+	VSC_INTR_RX_ERR = 1 << 0,
+	VSC_INTR_MS_ERR = 1 << 1,	/* master/slave resolution error */
+	VSC_INTR_CABLE = 1 << 2,	/* cable impairment */
+	VSC_INTR_FALSE_CARR = 1 << 3,	/* false carrier */
+	VSC_INTR_MEDIA_CHG = 1 << 4,	/* AMS media change */
+	VSC_INTR_RX_FIFO = 1 << 5,	/* Rx FIFO over/underflow */
+	VSC_INTR_TX_FIFO = 1 << 6,	/* Tx FIFO over/underflow */
+	VSC_INTR_DESCRAMBL = 1 << 7,	/* descrambler lock-lost */
+	VSC_INTR_SYMBOL_ERR = 1 << 8,	/* symbol error */
+	VSC_INTR_NEG_DONE = 1 << 10,	/* autoneg done */
+	VSC_INTR_NEG_ERR = 1 << 11,	/* autoneg error */
+	VSC_INTR_LINK_CHG = 1 << 13,	/* link change */
+	VSC_INTR_ENABLE = 1 << 15,	/* interrupt enable */
+};
+
+#define CFG_CHG_INTR_MASK (VSC_INTR_LINK_CHG | VSC_INTR_NEG_ERR | \
+	 		   VSC_INTR_NEG_DONE)
+#define INTR_MASK (CFG_CHG_INTR_MASK | VSC_INTR_TX_FIFO | VSC_INTR_RX_FIFO | \
+		   VSC_INTR_ENABLE)
+
+/* PHY specific auxiliary control & status register fields */
+#define S_ACSR_ACTIPHY_TMR    0
+#define M_ACSR_ACTIPHY_TMR    0x3
+#define V_ACSR_ACTIPHY_TMR(x) ((x) << S_ACSR_ACTIPHY_TMR)
+
+#define S_ACSR_SPEED    3
+#define M_ACSR_SPEED    0x3
+#define G_ACSR_SPEED(x) (((x) >> S_ACSR_SPEED) & M_ACSR_SPEED)
+
+#define S_ACSR_DUPLEX 5
+#define F_ACSR_DUPLEX (1 << S_ACSR_DUPLEX)
+
+#define S_ACSR_ACTIPHY 6
+#define F_ACSR_ACTIPHY (1 << S_ACSR_ACTIPHY)
+
+/*
+ * Reset the PHY.  This PHY completes reset immediately so we never wait.
+ */
+static int vsc8211_reset(struct cphy *cphy, int wait)
+{
+	return t3_phy_reset(cphy, 0, 0);
+}
+
+static int vsc8211_intr_enable(struct cphy *cphy)
+{
+	return mdio_write(cphy, 0, VSC8211_INTR_ENABLE, INTR_MASK);
+}
+
+static int vsc8211_intr_disable(struct cphy *cphy)
+{
+	return mdio_write(cphy, 0, VSC8211_INTR_ENABLE, 0);
+}
+
+static int vsc8211_intr_clear(struct cphy *cphy)
+{
+	u32 val;
+
+	/* Clear PHY interrupts by reading the register. */
+	return mdio_read(cphy, 0, VSC8211_INTR_STATUS, &val);
+}
+
+static int vsc8211_autoneg_enable(struct cphy *cphy)
+{
+	return t3_mdio_change_bits(cphy, 0, MII_BMCR, BMCR_PDOWN | BMCR_ISOLATE,
+				   BMCR_ANENABLE | BMCR_ANRESTART);
+}
+
+static int vsc8211_autoneg_restart(struct cphy *cphy)
+{
+	return t3_mdio_change_bits(cphy, 0, MII_BMCR, BMCR_PDOWN | BMCR_ISOLATE,
+				   BMCR_ANRESTART);
+}
+
+static int vsc8211_get_link_status(struct cphy *cphy, int *link_ok,
+				   int *speed, int *duplex, int *fc)
+{
+	unsigned int bmcr, status, lpa, adv;
+	int err, sp = -1, dplx = -1, pause = 0;
+
+	err = mdio_read(cphy, 0, MII_BMCR, &bmcr);
+	if (!err)
+		err = mdio_read(cphy, 0, MII_BMSR, &status);
+	if (err)
+		return err;
+
+	if (link_ok) {
+		/*
+		 * BMSR_LSTATUS is latch-low, so if it is 0 we need to read it
+		 * once more to get the current link state.
+		 */
+		if (!(status & BMSR_LSTATUS))
+			err = mdio_read(cphy, 0, MII_BMSR, &status);
+		if (err)
+			return err;
+		*link_ok = (status & BMSR_LSTATUS) != 0;
+	}
+	if (!(bmcr & BMCR_ANENABLE)) {
+		dplx = (bmcr & BMCR_FULLDPLX) ? DUPLEX_FULL : DUPLEX_HALF;
+		if (bmcr & BMCR_SPEED1000)
+			sp = SPEED_1000;
+		else if (bmcr & BMCR_SPEED100)
+			sp = SPEED_100;
+		else
+			sp = SPEED_10;
+	} else if (status & BMSR_ANEGCOMPLETE) {
+		err = mdio_read(cphy, 0, VSC8211_AUX_CTRL_STAT, &status);
+		if (err)
+			return err;
+
+		dplx = (status & F_ACSR_DUPLEX) ? DUPLEX_FULL : DUPLEX_HALF;
+		sp = G_ACSR_SPEED(status);
+		if (sp == 0)
+			sp = SPEED_10;
+		else if (sp == 1)
+			sp = SPEED_100;
+		else
+			sp = SPEED_1000;
+
+		if (fc && dplx == DUPLEX_FULL) {
+			err = mdio_read(cphy, 0, MII_LPA, &lpa);
+			if (!err)
+				err = mdio_read(cphy, 0, MII_ADVERTISE, &adv);
+			if (err)
+				return err;
+
+			if (lpa & adv & ADVERTISE_PAUSE_CAP)
+				pause = PAUSE_RX | PAUSE_TX;
+			else if ((lpa & ADVERTISE_PAUSE_CAP) &&
+				 (lpa & ADVERTISE_PAUSE_ASYM) &&
+				 (adv & ADVERTISE_PAUSE_ASYM))
+				pause = PAUSE_TX;
+			else if ((lpa & ADVERTISE_PAUSE_ASYM) &&
+				 (adv & ADVERTISE_PAUSE_CAP))
+				pause = PAUSE_RX;
+		}
+	}
+	if (speed)
+		*speed = sp;
+	if (duplex)
+		*duplex = dplx;
+	if (fc)
+		*fc = pause;
+	return 0;
+}
+
+static int vsc8211_power_down(struct cphy *cphy, int enable)
+{
+	return t3_mdio_change_bits(cphy, 0, MII_BMCR, BMCR_PDOWN,
+				   enable ? BMCR_PDOWN : 0);
+}
+
+static int vsc8211_intr_handler(struct cphy *cphy)
+{
+	unsigned int cause;
+	int err, cphy_cause = 0;
+
+	err = mdio_read(cphy, 0, VSC8211_INTR_STATUS, &cause);
+	if (err)
+		return err;
+
+	cause &= INTR_MASK;
+	if (cause & CFG_CHG_INTR_MASK)
+		cphy_cause |= cphy_cause_link_change;
+	if (cause & (VSC_INTR_RX_FIFO | VSC_INTR_TX_FIFO))
+		cphy_cause |= cphy_cause_fifo_error;
+	return cphy_cause;
+}
+
+static struct cphy_ops vsc8211_ops = {
+	.reset = vsc8211_reset,
+	.intr_enable = vsc8211_intr_enable,
+	.intr_disable = vsc8211_intr_disable,
+	.intr_clear = vsc8211_intr_clear,
+	.intr_handler = vsc8211_intr_handler,
+	.autoneg_enable = vsc8211_autoneg_enable,
+	.autoneg_restart = vsc8211_autoneg_restart,
+	.advertise = t3_phy_advertise,
+	.set_speed_duplex = t3_set_phy_speed_duplex,
+	.get_link_status = vsc8211_get_link_status,
+	.power_down = vsc8211_power_down,
+};
+
+void t3_vsc8211_phy_prep(struct cphy *phy, struct adapter *adapter,
+			 int phy_addr, const struct mdio_ops *mdio_ops)
+{
+	cphy_init(phy, adapter, phy_addr, &vsc8211_ops, mdio_ops);
+}
diff --git a/drivers/net/cxgb3/xgmac.c b/drivers/net/cxgb3/xgmac.c
new file mode 100644
index 00000000000..0f209c77692
--- /dev/null
+++ b/drivers/net/cxgb3/xgmac.c
@@ -0,0 +1,389 @@
+/*
+ * This file is part of the Chelsio T3 Ethernet driver.
+ *
+ * Copyright (C) 2005-2006 Chelsio Communications.  All rights reserved.
+ *
+ * This program is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the LICENSE file included in this
+ * release for licensing terms and conditions.
+ */
+
+#include "common.h"
+#include "regs.h"
+
+/*
+ * # of exact address filters.  The first one is used for the station address,
+ * the rest are available for multicast addresses.
+ */
+#define EXACT_ADDR_FILTERS 8
+
+static inline int macidx(const struct cmac *mac)
+{
+	return mac->offset / (XGMAC0_1_BASE_ADDR - XGMAC0_0_BASE_ADDR);
+}
+
+static void xaui_serdes_reset(struct cmac *mac)
+{
+	static const unsigned int clear[] = {
+		F_PWRDN0 | F_PWRDN1, F_RESETPLL01, F_RESET0 | F_RESET1,
+		F_PWRDN2 | F_PWRDN3, F_RESETPLL23, F_RESET2 | F_RESET3
+	};
+
+	int i;
+	struct adapter *adap = mac->adapter;
+	u32 ctrl = A_XGM_SERDES_CTRL0 + mac->offset;
+
+	t3_write_reg(adap, ctrl, adap->params.vpd.xauicfg[macidx(mac)] |
+		     F_RESET3 | F_RESET2 | F_RESET1 | F_RESET0 |
+		     F_PWRDN3 | F_PWRDN2 | F_PWRDN1 | F_PWRDN0 |
+		     F_RESETPLL23 | F_RESETPLL01);
+	t3_read_reg(adap, ctrl);
+	udelay(15);
+
+	for (i = 0; i < ARRAY_SIZE(clear); i++) {
+		t3_set_reg_field(adap, ctrl, clear[i], 0);
+		udelay(15);
+	}
+}
+
+void t3b_pcs_reset(struct cmac *mac)
+{
+	t3_set_reg_field(mac->adapter, A_XGM_RESET_CTRL + mac->offset,
+			 F_PCS_RESET_, 0);
+	udelay(20);
+	t3_set_reg_field(mac->adapter, A_XGM_RESET_CTRL + mac->offset, 0,
+			 F_PCS_RESET_);
+}
+
+int t3_mac_reset(struct cmac *mac)
+{
+	static const struct addr_val_pair mac_reset_avp[] = {
+		{A_XGM_TX_CTRL, 0},
+		{A_XGM_RX_CTRL, 0},
+		{A_XGM_RX_CFG, F_DISPAUSEFRAMES | F_EN1536BFRAMES |
+		 F_RMFCS | F_ENJUMBO | F_ENHASHMCAST},
+		{A_XGM_RX_HASH_LOW, 0},
+		{A_XGM_RX_HASH_HIGH, 0},
+		{A_XGM_RX_EXACT_MATCH_LOW_1, 0},
+		{A_XGM_RX_EXACT_MATCH_LOW_2, 0},
+		{A_XGM_RX_EXACT_MATCH_LOW_3, 0},
+		{A_XGM_RX_EXACT_MATCH_LOW_4, 0},
+		{A_XGM_RX_EXACT_MATCH_LOW_5, 0},
+		{A_XGM_RX_EXACT_MATCH_LOW_6, 0},
+		{A_XGM_RX_EXACT_MATCH_LOW_7, 0},
+		{A_XGM_RX_EXACT_MATCH_LOW_8, 0},
+		{A_XGM_STAT_CTRL, F_CLRSTATS}
+	};
+	u32 val;
+	struct adapter *adap = mac->adapter;
+	unsigned int oft = mac->offset;
+
+	t3_write_reg(adap, A_XGM_RESET_CTRL + oft, F_MAC_RESET_);
+	t3_read_reg(adap, A_XGM_RESET_CTRL + oft);	/* flush */
+
+	t3_write_regs(adap, mac_reset_avp, ARRAY_SIZE(mac_reset_avp), oft);
+	t3_set_reg_field(adap, A_XGM_RXFIFO_CFG + oft,
+			 F_RXSTRFRWRD | F_DISERRFRAMES,
+			 uses_xaui(adap) ? 0 : F_RXSTRFRWRD);
+
+	if (uses_xaui(adap)) {
+		if (adap->params.rev == 0) {
+			t3_set_reg_field(adap, A_XGM_SERDES_CTRL + oft, 0,
+					 F_RXENABLE | F_TXENABLE);
+			if (t3_wait_op_done(adap, A_XGM_SERDES_STATUS1 + oft,
+					    F_CMULOCK, 1, 5, 2)) {
+				CH_ERR(adap,
+				       "MAC %d XAUI SERDES CMU lock failed\n",
+				       macidx(mac));
+				return -1;
+			}
+			t3_set_reg_field(adap, A_XGM_SERDES_CTRL + oft, 0,
+					 F_SERDESRESET_);
+		} else
+			xaui_serdes_reset(mac);
+	}
+
+	if (adap->params.rev > 0)
+		t3_write_reg(adap, A_XGM_PAUSE_TIMER + oft, 0xf000);
+
+	val = F_MAC_RESET_;
+	if (is_10G(adap))
+		val |= F_PCS_RESET_;
+	else if (uses_xaui(adap))
+		val |= F_PCS_RESET_ | F_XG2G_RESET_;
+	else
+		val |= F_RGMII_RESET_ | F_XG2G_RESET_;
+	t3_write_reg(adap, A_XGM_RESET_CTRL + oft, val);
+	t3_read_reg(adap, A_XGM_RESET_CTRL + oft);	/* flush */
+	if ((val & F_PCS_RESET_) && adap->params.rev) {
+		msleep(1);
+		t3b_pcs_reset(mac);
+	}
+
+	memset(&mac->stats, 0, sizeof(mac->stats));
+	return 0;
+}
+
+/*
+ * Set the exact match register 'idx' to recognize the given Ethernet address.
+ */
+static void set_addr_filter(struct cmac *mac, int idx, const u8 * addr)
+{
+	u32 addr_lo, addr_hi;
+	unsigned int oft = mac->offset + idx * 8;
+
+	addr_lo = (addr[3] << 24) | (addr[2] << 16) | (addr[1] << 8) | addr[0];
+	addr_hi = (addr[5] << 8) | addr[4];
+
+	t3_write_reg(mac->adapter, A_XGM_RX_EXACT_MATCH_LOW_1 + oft, addr_lo);
+	t3_write_reg(mac->adapter, A_XGM_RX_EXACT_MATCH_HIGH_1 + oft, addr_hi);
+}
+
+/* Set one of the station's unicast MAC addresses. */
+int t3_mac_set_address(struct cmac *mac, unsigned int idx, u8 addr[6])
+{
+	if (idx >= mac->nucast)
+		return -EINVAL;
+	set_addr_filter(mac, idx, addr);
+	return 0;
+}
+
+/*
+ * Specify the number of exact address filters that should be reserved for
+ * unicast addresses.  Caller should reload the unicast and multicast addresses
+ * after calling this.
+ */
+int t3_mac_set_num_ucast(struct cmac *mac, int n)
+{
+	if (n > EXACT_ADDR_FILTERS)
+		return -EINVAL;
+	mac->nucast = n;
+	return 0;
+}
+
+/* Calculate the RX hash filter index of an Ethernet address */
+static int hash_hw_addr(const u8 * addr)
+{
+	int hash = 0, octet, bit, i = 0, c;
+
+	for (octet = 0; octet < 6; ++octet)
+		for (c = addr[octet], bit = 0; bit < 8; c >>= 1, ++bit) {
+			hash ^= (c & 1) << i;
+			if (++i == 6)
+				i = 0;
+		}
+	return hash;
+}
+
+int t3_mac_set_rx_mode(struct cmac *mac, struct t3_rx_mode *rm)
+{
+	u32 val, hash_lo, hash_hi;
+	struct adapter *adap = mac->adapter;
+	unsigned int oft = mac->offset;
+
+	val = t3_read_reg(adap, A_XGM_RX_CFG + oft) & ~F_COPYALLFRAMES;
+	if (rm->dev->flags & IFF_PROMISC)
+		val |= F_COPYALLFRAMES;
+	t3_write_reg(adap, A_XGM_RX_CFG + oft, val);
+
+	if (rm->dev->flags & IFF_ALLMULTI)
+		hash_lo = hash_hi = 0xffffffff;
+	else {
+		u8 *addr;
+		int exact_addr_idx = mac->nucast;
+
+		hash_lo = hash_hi = 0;
+		while ((addr = t3_get_next_mcaddr(rm)))
+			if (exact_addr_idx < EXACT_ADDR_FILTERS)
+				set_addr_filter(mac, exact_addr_idx++, addr);
+			else {
+				int hash = hash_hw_addr(addr);
+
+				if (hash < 32)
+					hash_lo |= (1 << hash);
+				else
+					hash_hi |= (1 << (hash - 32));
+			}
+	}
+
+	t3_write_reg(adap, A_XGM_RX_HASH_LOW + oft, hash_lo);
+	t3_write_reg(adap, A_XGM_RX_HASH_HIGH + oft, hash_hi);
+	return 0;
+}
+
+int t3_mac_set_mtu(struct cmac *mac, unsigned int mtu)
+{
+	int hwm, lwm;
+	unsigned int thres, v;
+	struct adapter *adap = mac->adapter;
+
+	/*
+	 * MAX_FRAME_SIZE inludes header + FCS, mtu doesn't.  The HW max
+	 * packet size register includes header, but not FCS.
+	 */
+	mtu += 14;
+	if (mtu > MAX_FRAME_SIZE - 4)
+		return -EINVAL;
+	t3_write_reg(adap, A_XGM_RX_MAX_PKT_SIZE + mac->offset, mtu);
+
+	/*
+	 * Adjust the PAUSE frame watermarks.  We always set the LWM, and the
+	 * HWM only if flow-control is enabled.
+	 */
+	hwm = max(MAC_RXFIFO_SIZE - 3 * mtu, MAC_RXFIFO_SIZE / 2U);
+	hwm = min(hwm, 3 * MAC_RXFIFO_SIZE / 4 + 1024);
+	lwm = hwm - 1024;
+	v = t3_read_reg(adap, A_XGM_RXFIFO_CFG + mac->offset);
+	v &= ~V_RXFIFOPAUSELWM(M_RXFIFOPAUSELWM);
+	v |= V_RXFIFOPAUSELWM(lwm / 8);
+	if (G_RXFIFOPAUSEHWM(v))
+		v = (v & ~V_RXFIFOPAUSEHWM(M_RXFIFOPAUSEHWM)) |
+		    V_RXFIFOPAUSEHWM(hwm / 8);
+	t3_write_reg(adap, A_XGM_RXFIFO_CFG + mac->offset, v);
+
+	/* Adjust the TX FIFO threshold based on the MTU */
+	thres = (adap->params.vpd.cclk * 1000) / 15625;
+	thres = (thres * mtu) / 1000;
+	if (is_10G(adap))
+		thres /= 10;
+	thres = mtu > thres ? (mtu - thres + 7) / 8 : 0;
+	thres = max(thres, 8U);	/* need at least 8 */
+	t3_set_reg_field(adap, A_XGM_TXFIFO_CFG + mac->offset,
+			 V_TXFIFOTHRESH(M_TXFIFOTHRESH), V_TXFIFOTHRESH(thres));
+	return 0;
+}
+
+int t3_mac_set_speed_duplex_fc(struct cmac *mac, int speed, int duplex, int fc)
+{
+	u32 val;
+	struct adapter *adap = mac->adapter;
+	unsigned int oft = mac->offset;
+
+	if (duplex >= 0 && duplex != DUPLEX_FULL)
+		return -EINVAL;
+	if (speed >= 0) {
+		if (speed == SPEED_10)
+			val = V_PORTSPEED(0);
+		else if (speed == SPEED_100)
+			val = V_PORTSPEED(1);
+		else if (speed == SPEED_1000)
+			val = V_PORTSPEED(2);
+		else if (speed == SPEED_10000)
+			val = V_PORTSPEED(3);
+		else
+			return -EINVAL;
+
+		t3_set_reg_field(adap, A_XGM_PORT_CFG + oft,
+				 V_PORTSPEED(M_PORTSPEED), val);
+	}
+
+	val = t3_read_reg(adap, A_XGM_RXFIFO_CFG + oft);
+	val &= ~V_RXFIFOPAUSEHWM(M_RXFIFOPAUSEHWM);
+	if (fc & PAUSE_TX)
+		val |= V_RXFIFOPAUSEHWM(G_RXFIFOPAUSELWM(val) + 128);	/* +1KB */
+	t3_write_reg(adap, A_XGM_RXFIFO_CFG + oft, val);
+
+	t3_set_reg_field(adap, A_XGM_TX_CFG + oft, F_TXPAUSEEN,
+			 (fc & PAUSE_RX) ? F_TXPAUSEEN : 0);
+	return 0;
+}
+
+int t3_mac_enable(struct cmac *mac, int which)
+{
+	int idx = macidx(mac);
+	struct adapter *adap = mac->adapter;
+	unsigned int oft = mac->offset;
+
+	if (which & MAC_DIRECTION_TX) {
+		t3_write_reg(adap, A_XGM_TX_CTRL + oft, F_TXEN);
+		t3_write_reg(adap, A_TP_PIO_ADDR, A_TP_TX_DROP_CFG_CH0 + idx);
+		t3_write_reg(adap, A_TP_PIO_DATA, 0xbf000001);
+		t3_write_reg(adap, A_TP_PIO_ADDR, A_TP_TX_DROP_MODE);
+		t3_set_reg_field(adap, A_TP_PIO_DATA, 1 << idx, 1 << idx);
+	}
+	if (which & MAC_DIRECTION_RX)
+		t3_write_reg(adap, A_XGM_RX_CTRL + oft, F_RXEN);
+	return 0;
+}
+
+int t3_mac_disable(struct cmac *mac, int which)
+{
+	int idx = macidx(mac);
+	struct adapter *adap = mac->adapter;
+
+	if (which & MAC_DIRECTION_TX) {
+		t3_write_reg(adap, A_XGM_TX_CTRL + mac->offset, 0);
+		t3_write_reg(adap, A_TP_PIO_ADDR, A_TP_TX_DROP_CFG_CH0 + idx);
+		t3_write_reg(adap, A_TP_PIO_DATA, 0xc000001f);
+		t3_write_reg(adap, A_TP_PIO_ADDR, A_TP_TX_DROP_MODE);
+		t3_set_reg_field(adap, A_TP_PIO_DATA, 1 << idx, 0);
+	}
+	if (which & MAC_DIRECTION_RX)
+		t3_write_reg(adap, A_XGM_RX_CTRL + mac->offset, 0);
+	return 0;
+}
+
+/*
+ * This function is called periodically to accumulate the current values of the
+ * RMON counters into the port statistics.  Since the packet counters are only
+ * 32 bits they can overflow in ~286 secs at 10G, so the function should be
+ * called more frequently than that.  The byte counters are 45-bit wide, they
+ * would overflow in ~7.8 hours.
+ */
+const struct mac_stats *t3_mac_update_stats(struct cmac *mac)
+{
+#define RMON_READ(mac, addr) t3_read_reg(mac->adapter, addr + mac->offset)
+#define RMON_UPDATE(mac, name, reg) \
+	(mac)->stats.name += (u64)RMON_READ(mac, A_XGM_STAT_##reg)
+#define RMON_UPDATE64(mac, name, reg_lo, reg_hi) \
+	(mac)->stats.name += RMON_READ(mac, A_XGM_STAT_##reg_lo) + \
+			     ((u64)RMON_READ(mac, A_XGM_STAT_##reg_hi) << 32)
+
+	u32 v, lo;
+
+	RMON_UPDATE64(mac, rx_octets, RX_BYTES_LOW, RX_BYTES_HIGH);
+	RMON_UPDATE64(mac, rx_frames, RX_FRAMES_LOW, RX_FRAMES_HIGH);
+	RMON_UPDATE(mac, rx_mcast_frames, RX_MCAST_FRAMES);
+	RMON_UPDATE(mac, rx_bcast_frames, RX_BCAST_FRAMES);
+	RMON_UPDATE(mac, rx_fcs_errs, RX_CRC_ERR_FRAMES);
+	RMON_UPDATE(mac, rx_pause, RX_PAUSE_FRAMES);
+	RMON_UPDATE(mac, rx_jabber, RX_JABBER_FRAMES);
+	RMON_UPDATE(mac, rx_short, RX_SHORT_FRAMES);
+	RMON_UPDATE(mac, rx_symbol_errs, RX_SYM_CODE_ERR_FRAMES);
+
+	RMON_UPDATE(mac, rx_too_long, RX_OVERSIZE_FRAMES);
+	mac->stats.rx_too_long += RMON_READ(mac, A_XGM_RX_MAX_PKT_SIZE_ERR_CNT);
+
+	RMON_UPDATE(mac, rx_frames_64, RX_64B_FRAMES);
+	RMON_UPDATE(mac, rx_frames_65_127, RX_65_127B_FRAMES);
+	RMON_UPDATE(mac, rx_frames_128_255, RX_128_255B_FRAMES);
+	RMON_UPDATE(mac, rx_frames_256_511, RX_256_511B_FRAMES);
+	RMON_UPDATE(mac, rx_frames_512_1023, RX_512_1023B_FRAMES);
+	RMON_UPDATE(mac, rx_frames_1024_1518, RX_1024_1518B_FRAMES);
+	RMON_UPDATE(mac, rx_frames_1519_max, RX_1519_MAXB_FRAMES);
+
+	RMON_UPDATE64(mac, tx_octets, TX_BYTE_LOW, TX_BYTE_HIGH);
+	RMON_UPDATE64(mac, tx_frames, TX_FRAME_LOW, TX_FRAME_HIGH);
+	RMON_UPDATE(mac, tx_mcast_frames, TX_MCAST);
+	RMON_UPDATE(mac, tx_bcast_frames, TX_BCAST);
+	RMON_UPDATE(mac, tx_pause, TX_PAUSE);
+	/* This counts error frames in general (bad FCS, underrun, etc). */
+	RMON_UPDATE(mac, tx_underrun, TX_ERR_FRAMES);
+
+	RMON_UPDATE(mac, tx_frames_64, TX_64B_FRAMES);
+	RMON_UPDATE(mac, tx_frames_65_127, TX_65_127B_FRAMES);
+	RMON_UPDATE(mac, tx_frames_128_255, TX_128_255B_FRAMES);
+	RMON_UPDATE(mac, tx_frames_256_511, TX_256_511B_FRAMES);
+	RMON_UPDATE(mac, tx_frames_512_1023, TX_512_1023B_FRAMES);
+	RMON_UPDATE(mac, tx_frames_1024_1518, TX_1024_1518B_FRAMES);
+	RMON_UPDATE(mac, tx_frames_1519_max, TX_1519_MAXB_FRAMES);
+
+	/* The next stat isn't clear-on-read. */
+	t3_write_reg(mac->adapter, A_TP_MIB_INDEX, mac->offset ? 51 : 50);
+	v = t3_read_reg(mac->adapter, A_TP_MIB_RDATA);
+	lo = (u32) mac->stats.rx_cong_drops;
+	mac->stats.rx_cong_drops += (u64) (v - lo);
+
+	return &mac->stats;
+}