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

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

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

drivers/net/cxgb3/Makefile

+#
+# 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

drivers/net/cxgb3/adapter.h

+/*
+ * 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__ */

drivers/net/cxgb3/ael1002.c

+/*
+ * 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);
+}

drivers/net/cxgb3/cxgb3_ctl_defs.h

+/*
+ * 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 */

drivers/net/cxgb3/cxgb3_defs.h

+/*
+ * 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

drivers/net/cxgb3/cxgb3_ioctl.h

+/*
+ * 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

drivers/net/cxgb3/cxgb3_offload.h

+/*
+ * 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

drivers/net/cxgb3/firmware_exports.h

+/* 
+ * ----------------------------------------------------------------------------
+ * >>>>>>>>>>>>>>>>>>>>>>>>>>>>> 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_ */

drivers/net/cxgb3/l2t.c

+/*
+ * 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);
+}
+

drivers/net/cxgb3/t3cdev.h

+/*
+ * 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_ */

drivers/net/cxgb3/version.h

+/*****************************************************************************
+ *                                                                           *
+ * 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 */