Sophie

Sophie

distrib > Scientific%20Linux > 5x > x86_64 > by-pkgid > 27922b4260f65d317aabda37e42bbbff > files > 2551

kernel-2.6.18-238.el5.src.rpm

From: Andy Gospodarek <gospo@redhat.com>
Date: Tue, 9 Sep 2008 18:00:22 -0400
Subject: [net] cxgb3: updates and lro fixes
Message-id: 20080909220021.GL5762@gospo.rdu.redhat.com
O-Subject: [RHEL5.3 PATCH] cxgb3: updates and lro fixes
Bugzilla: 441959
RH-Acked-by: David Miller <davem@redhat.com>

This is an update to the cxgb3 driver from upstream.  There are several
important fixes that help with performance and one late breaking fix
that helps LRO performance.

It has been tested by Chelsio without complaints and passed iWARP and
network tests done by Doug Ledford.  It is also required so that Doug's
OFED updates work correctly.

This will resolve BZ 252351 and BZ 441959.

diff --git a/drivers/net/Kconfig b/drivers/net/Kconfig
index 12ec53f..46b940c 100644
--- a/drivers/net/Kconfig
+++ b/drivers/net/Kconfig
@@ -2376,7 +2376,8 @@ config CHELSIO_T1
 config CHELSIO_T3
         tristate "Chelsio Communications T3 10Gb Ethernet support"
         depends on PCI
-        help
+	select INET_LRO
+	---help---
           This driver supports Chelsio T3-based gigabit and 10Gb Ethernet
           adapters.
 
diff --git a/drivers/net/cxgb3/adapter.h b/drivers/net/cxgb3/adapter.h
index 16e3824..1b6ff1c 100644
--- a/drivers/net/cxgb3/adapter.h
+++ b/drivers/net/cxgb3/adapter.h
@@ -41,9 +41,8 @@
 #include <linux/timer.h>
 #include <linux/cache.h>
 #include <linux/mutex.h>
+#include <linux/inet_lro.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 pt_regs *);
@@ -72,6 +71,7 @@ enum {				/* adapter flags */
 	USING_MSIX = (1 << 2),
 	QUEUES_BOUND = (1 << 3),
 	TP_PARITY_INIT = (1 << 4),
+	NAPI_INIT = (1 << 5),
 };
 
 struct fl_pg_chunk {
@@ -92,6 +92,7 @@ struct sge_fl {                     /* SGE per free-buffer list state */
 	unsigned int gen;           /* free list generation */
 	struct fl_pg_chunk pg_chunk;/* page chunk cache */
 	unsigned int use_pages;     /* whether FL uses pages or sk_buffs */
+	unsigned int order;	    /* order of page allocations */
 	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 */
@@ -116,12 +117,15 @@ struct sge_rspq {		/* state for an SGE response queue */
 	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 */
+	unsigned int rx_recycle_buf; /* whether recycling occurred
+					within current sop-eop */
 	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 */
+	struct sk_buff *pg_skb; /* used to build frag list in napi handler */
 
 	unsigned long offload_pkts;
 	unsigned long offload_bundles;
@@ -169,14 +173,28 @@ enum {				/* per port SGE statistics */
 	SGE_PSTAT_TX_CSUM,	/* # of TX checksum offloads */
 	SGE_PSTAT_VLANEX,	/* # of VLAN tag extractions */
 	SGE_PSTAT_VLANINS,	/* # of VLAN tag insertions */
+	SGE_PSTAT_LRO_AGGR,	/* # of page chunks added to LRO sessions */
+	SGE_PSTAT_LRO_FLUSHED,	/* # of flushed LRO sessions */
+	SGE_PSTAT_LRO_NO_DESC,	/* # of overflown LRO sessions */
 
 	SGE_PSTAT_MAX		/* must be last */
 };
 
+#define T3_MAX_LRO_SES 8
+#define T3_MAX_LRO_MAX_PKTS 64
+
 struct sge_qset {		/* an SGE queue set */
+	struct adapter *adap;
 	struct sge_rspq rspq;
 	struct sge_fl fl[SGE_RXQ_PER_SET];
 	struct sge_txq txq[SGE_TXQ_PER_SET];
+	struct net_lro_mgr lro_mgr;
+	struct net_lro_desc lro_desc[T3_MAX_LRO_SES];
+	struct skb_frag_struct *lro_frag_tbl;
+	int lro_nfrags;
+	int lro_enabled;
+	int lro_frag_len;
+	void *lro_va;
 	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 */
diff --git a/drivers/net/cxgb3/common.h b/drivers/net/cxgb3/common.h
index 99c75d3..9ecf8a6 100644
--- a/drivers/net/cxgb3/common.h
+++ b/drivers/net/cxgb3/common.h
@@ -351,6 +351,7 @@ struct tp_params {
 
 struct qset_params {		/* SGE queue set parameters */
 	unsigned int polling;	/* polling/interrupt service for rspq */
+	unsigned int lro;	/* large receive offload */
 	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 */
@@ -681,13 +682,14 @@ 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_read(struct adapter *adapter, u32 addr, __le32 *data);
+int t3_seeprom_write(struct adapter *adapter, u32 addr, __le32 data);
 int t3_seeprom_wp(struct adapter *adapter, int enable);
 int t3_get_tp_version(struct adapter *adapter, u32 *vers);
 int t3_check_tpsram_version(struct adapter *adapter, int *must_load);
-int t3_check_tpsram(struct adapter *adapter, u8 *tp_ram, unsigned int size);
-int t3_set_proto_sram(struct adapter *adap, u8 *data);
+int t3_check_tpsram(struct adapter *adapter, const u8 *tp_ram,
+		    unsigned int size);
+int t3_set_proto_sram(struct adapter *adap, const u8 *data);
 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);
@@ -698,6 +700,7 @@ 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);
+int t3_replay_prep_adapter(struct adapter *adapter);
 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);
diff --git a/drivers/net/cxgb3/cxgb3_ctl_defs.h b/drivers/net/cxgb3/cxgb3_ctl_defs.h
index 6c4f320..6ad9240 100644
--- a/drivers/net/cxgb3/cxgb3_ctl_defs.h
+++ b/drivers/net/cxgb3/cxgb3_ctl_defs.h
@@ -54,6 +54,7 @@ enum {
 	RDMA_CQ_DISABLE		= 16,
 	RDMA_CTRL_QP_SETUP	= 17,
 	RDMA_GET_MEM		= 18,
+	RDMA_GET_MIB		= 19,
 
 	GET_RX_PAGE_INFO	= 50,
 };
@@ -110,10 +111,7 @@ struct ulp_iscsi_info {
 	unsigned int llimit;
 	unsigned int ulimit;
 	unsigned int tagmask;
-	unsigned int pgsz3;
-	unsigned int pgsz2;
-	unsigned int pgsz1;
-	unsigned int pgsz0;
+	u8 pgsz_factor[4];
 	unsigned int max_rxsz;
 	unsigned int max_txsz;
 	struct pci_dev *pdev;
diff --git a/drivers/net/cxgb3/cxgb3_ioctl.h b/drivers/net/cxgb3/cxgb3_ioctl.h
index 0a82fcd..68200a1 100644
--- a/drivers/net/cxgb3/cxgb3_ioctl.h
+++ b/drivers/net/cxgb3/cxgb3_ioctl.h
@@ -90,6 +90,7 @@ struct ch_qset_params {
 	int32_t fl_size[2];
 	int32_t intr_lat;
 	int32_t polling;
+	int32_t lro;
 	int32_t cong_thres;
 };
 
diff --git a/drivers/net/cxgb3/cxgb3_main.c b/drivers/net/cxgb3/cxgb3_main.c
index 49d2ed7..e4b0bda 100644
--- a/drivers/net/cxgb3/cxgb3_main.c
+++ b/drivers/net/cxgb3/cxgb3_main.c
@@ -168,6 +168,7 @@ static void link_report(struct net_device *dev)
 		printk(KERN_INFO "%s: link up, %s, %s-duplex\n", dev->name, s,
 		       p->link_config.duplex == DUPLEX_FULL ? "full" : "half");
 	}
+
 }
 
 /**
@@ -445,7 +446,13 @@ static int init_dummy_netdevs(struct adapter *adap)
 			strcpy(adap->dummy_netdev[dummy_idx]->name, dev->name);
 			dummy_idx++;
 		}
+		/* 	
+		 * Be careful not to call init_dummy_netdevs a second
+		 * time, e.g., during EEH recovery, by making a note of
+		 * it here */
+        	adap->flags |= NAPI_INIT;
 	}
+	
 	return 0;
 
 free_all:
@@ -918,9 +925,11 @@ static int cxgb_up(struct adapter *adap)
 				goto out;
 		}
 
-		err = init_dummy_netdevs(adap);
-		if (err)
-			goto out;
+		if (!(adap->flags & NAPI_INIT)) {
+			err = init_dummy_netdevs(adap);
+			if (err)
+				goto out;
+		}
 
 		err = t3_init_hw(adap, 0);
 		if (err)
@@ -934,6 +943,7 @@ static int cxgb_up(struct adapter *adap)
 			goto out;
 
 		setup_rss(adap);
+
 		adap->flags |= FULL_INIT_DONE;
 	}
 
@@ -1035,7 +1045,7 @@ static int offload_open(struct net_device *dev)
 		return 0;
 
 	if (!adap_up && (err = cxgb_up(adapter)) < 0)
-		return err;
+		goto out;
 
 	t3_tp_set_offload_mode(adapter, 1);
 	tdev->lldev = adapter->port[0];
@@ -1051,7 +1061,8 @@ static int offload_open(struct net_device *dev)
 	init_smt(adapter);
 
 	/* Never mind if the next step fails */
-	sysfs_create_group(&tdev->lldev->class_dev.kobj, &offload_attr_group);
+	if (sysfs_create_group(&tdev->lldev->class_dev.kobj, &offload_attr_group))
+		dev_dbg(dev, "cannot create sysfs group\n");
 
 	/* Call back all registered clients */
 	cxgb3_add_clients(tdev);
@@ -1240,6 +1251,9 @@ static char stats_strings[][ETH_GSTRING_LEN] = {
 	"VLANinsertions     ",
 	"TxCsumOffload      ",
 	"RxCsumGood         ",
+	"LroAggregated      ",
+	"LroFlushed         ",
+	"LroNoDesc          ",
 	"RxDrops            ",
 
 	"CheckTXEnToggled   ",
@@ -1363,6 +1377,9 @@ static void get_stats(struct net_device *dev, struct ethtool_stats *stats,
 	*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++ = collect_sge_port_stats(adapter, pi, SGE_PSTAT_LRO_AGGR);
+	*data++ = collect_sge_port_stats(adapter, pi, SGE_PSTAT_LRO_FLUSHED);
+	*data++ = collect_sge_port_stats(adapter, pi, SGE_PSTAT_LRO_NO_DESC);
 	*data++ = s->rx_cong_drops;
 
 	*data++ = s->num_toggled;
@@ -1581,6 +1598,13 @@ static int set_rx_csum(struct net_device *dev, u32 data)
 	struct port_info *p = netdev_priv(dev);
 
 	p->rx_csum_offload = data;
+	if (!data) {
+		struct adapter *adap = p->adapter;
+		int i;
+
+		for (i = p->first_qset; i < p->first_qset + p->nqsets; i++)
+			adap->sge.qs[i].lro_enabled = 0;
+	}
 	return 0;
 }
 
@@ -1671,7 +1695,7 @@ static int get_eeprom(struct net_device *dev, struct ethtool_eeprom *e,
 
 	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]);
+		err = t3_seeprom_read(adapter, i, (__le32 *) & buf[i]);
 
 	if (!err)
 		memcpy(data, buf + e->offset, e->len);
@@ -1684,7 +1708,8 @@ static int set_eeprom(struct net_device *dev, struct ethtool_eeprom *eeprom,
 {
 	struct port_info *pi = netdev_priv(dev);
 	struct adapter *adapter = pi->adapter;
-	u32 aligned_offset, aligned_len, *p;
+	u32 aligned_offset, aligned_len;
+	__le32 *p;
 	u8 *buf;
 	int err;
 
@@ -1698,11 +1723,11 @@ static int set_eeprom(struct net_device *dev, struct ethtool_eeprom *eeprom,
 		buf = kmalloc(aligned_len, GFP_KERNEL);
 		if (!buf)
 			return -ENOMEM;
-		err = t3_seeprom_read(adapter, aligned_offset, (u32 *) buf);
+		err = t3_seeprom_read(adapter, aligned_offset, (__le32 *) buf);
 		if (!err && aligned_len > 4)
 			err = t3_seeprom_read(adapter,
 					      aligned_offset + aligned_len - 4,
-					      (u32 *) & buf[aligned_len - 4]);
+					      (__le32 *) & buf[aligned_len - 4]);
 		if (err)
 			goto out;
 		memcpy(buf + (eeprom->offset & 3), data, eeprom->len);
@@ -1713,7 +1738,7 @@ static int set_eeprom(struct net_device *dev, struct ethtool_eeprom *eeprom,
 	if (err)
 		goto out;
 
-	for (p = (u32 *) buf; !err && aligned_len; aligned_len -= 4, p++) {
+	for (p = (__le32 *) buf; !err && aligned_len; aligned_len -= 4, p++) {
 		err = t3_seeprom_write(adapter, aligned_offset, *p);
 		aligned_offset += 4;
 	}
@@ -1856,6 +1881,11 @@ static int cxgb_extension_ioctl(struct net_device *dev, void __user *useraddr)
 				}
 			}
 		}
+		if (t.lro >= 0) {
+			struct sge_qset *qs = &adapter->sge.qs[t.qset_idx];
+			q->lro = t.lro;
+			qs->lro_enabled = t.lro;
+		}
 		break;
 	}
 	case CHELSIO_GET_QSET_PARAMS:{
@@ -1875,6 +1905,7 @@ static int cxgb_extension_ioctl(struct net_device *dev, void __user *useraddr)
 		t.fl_size[0] = q->fl_size;
 		t.fl_size[1] = q->jumbo_size;
 		t.polling = q->polling;
+		t.lro = q->lro;
 		t.intr_lat = q->coalesce_usecs;
 		t.cong_thres = q->cong_thres;
 
@@ -1926,11 +1957,11 @@ static int cxgb_extension_ioctl(struct net_device *dev, void __user *useraddr)
 		u8 *fw_data;
 		struct ch_mem_range t;
 
-		if (!capable(CAP_NET_ADMIN))
+		if (!capable(CAP_SYS_RAWIO))
 			return -EPERM;
 		if (copy_from_user(&t, useraddr, sizeof(t)))
 			return -EFAULT;
-
+		/* Check t.len sanity ? */
 		fw_data = kmalloc(t.len, GFP_KERNEL);
 		if (!fw_data)
 			return -ENOMEM;
@@ -2458,14 +2489,11 @@ static pci_ers_result_t t3_io_error_detected(struct pci_dev *pdev,
 	    test_bit(OFFLOAD_DEVMAP_BIT, &adapter->open_device_map))
 		offload_close(&adapter->tdev);
 
-	/* Free sge resources */
-	t3_free_sge_resources(adapter);
-
 	adapter->flags &= ~FULL_INIT_DONE;
 
 	pci_disable_device(pdev);
 
-	/* Request a slot slot reset. */
+	/* Request a slot reset. */
 	return PCI_ERS_RESULT_NEED_RESET;
 }
 
@@ -2482,13 +2510,20 @@ static pci_ers_result_t t3_io_slot_reset(struct pci_dev *pdev)
 	if (pci_enable_device(pdev)) {
 		dev_err(&pdev->dev,
 			"Cannot re-enable PCI device after reset.\n");
-		return PCI_ERS_RESULT_DISCONNECT;
+		goto err;
 	}
 	pci_set_master(pdev);
+	pci_restore_state(pdev);
+
+	/* Free sge resources */
+	t3_free_sge_resources(adapter);
 
-	t3_prep_adapter(adapter, adapter->params.info, 1);
+	if (t3_replay_prep_adapter(adapter))
+		goto err;
 
 	return PCI_ERS_RESULT_RECOVERED;
+err:
+	return PCI_ERS_RESULT_DISCONNECT;
 }
 
 /**
@@ -2517,13 +2552,6 @@ static void t3_io_resume(struct pci_dev *pdev)
 			netif_device_attach(netdev);
 		}
 	}
-
-	if (is_offload(adapter)) {
-		__set_bit(OFFLOAD_DEVMAP_BIT, &adapter->registered_device_map);
-		if (offload_open(adapter->port[0]))
-			printk(KERN_WARNING
-			       "Could not bring back offload capabilities\n");
-	}
 }
 
 static struct pci_error_handlers t3_err_handler = {
@@ -2642,6 +2670,7 @@ static int __devinit init_one(struct pci_dev *pdev,
 	}
 
 	pci_set_master(pdev);
+	pci_save_state(pdev);
 
 	mmio_start = pci_resource_start(pdev, 0);
 	mmio_len = pci_resource_len(pdev, 0);
diff --git a/drivers/net/cxgb3/cxgb3_offload.c b/drivers/net/cxgb3/cxgb3_offload.c
index fcc53b6..b67fdda 100644
--- a/drivers/net/cxgb3/cxgb3_offload.c
+++ b/drivers/net/cxgb3/cxgb3_offload.c
@@ -205,6 +205,17 @@ static int cxgb_ulp_iscsi_ctl(struct adapter *adapter, unsigned int req,
 		break;
 	case ULP_ISCSI_SET_PARAMS:
 		t3_write_reg(adapter, A_ULPRX_ISCSI_TAGMASK, uiip->tagmask);
+		/* set MaxRxData and MaxCoalesceSize to 16224 */
+		t3_write_reg(adapter, A_TP_PARA_REG2, 0x3f603f60);
+		/* program the ddp page sizes */
+		{
+			int i;
+			unsigned int val = 0;
+			for (i = 0; i < 4; i++)
+				val |= (uiip->pgsz_factor[i] & 0xF) << (8 * i);
+			if (val)
+				t3_write_reg(adapter, A_ULPRX_ISCSI_PSZ, val);
+		}
 		break;
 	default:
 		ret = -EOPNOTSUPP;
@@ -301,6 +312,12 @@ static int cxgb_rdma_ctl(struct adapter *adapter, unsigned int req, void *data)
 		spin_unlock_irq(&adapter->sge.reg_lock);
 		break;
 	}
+	case RDMA_GET_MIB: {
+		spin_lock(&adapter->stats_lock);
+		t3_tp_get_mib_stats(adapter, (struct tp_mib_stats *)data);
+		spin_unlock(&adapter->stats_lock);
+		break;
+	}
 	default:
 		ret = -EOPNOTSUPP;
 	}
@@ -379,6 +396,7 @@ static int cxgb_offload_ctl(struct t3cdev *tdev, unsigned int req, void *data)
 	case RDMA_CQ_DISABLE:
 	case RDMA_CTRL_QP_SETUP:
 	case RDMA_GET_MEM:
+	case RDMA_GET_MIB:
 		if (!offload_running(adapter))
 			return -EAGAIN;
 		return cxgb_rdma_ctl(adapter, req, data);
@@ -830,10 +848,26 @@ static int do_trace(struct t3cdev *dev, struct sk_buff *skb)
 	return 0;
 }
 
+/*
+ * That skb would better have come from process_responses() where we abuse
+ * ->priority and ->csum to carry our data.  NB: if we get to per-arch
+ * ->csum, the things might get really interesting here.
+ */
+
+static inline u32 get_hwtid(struct sk_buff *skb)
+{
+	return ntohl((__force __be32)skb->priority) >> 8 & 0xfffff;
+}
+
+static inline u32 get_opcode(struct sk_buff *skb)
+{
+	return G_OPCODE(ntohl((__force __be32)skb->csum));
+}
+
 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));
+	unsigned int hwtid = get_hwtid(skb);
+	unsigned int opcode = get_opcode(skb);
 	struct t3c_tid_entry *t3c_tid;
 
 	t3c_tid = lookup_tid(&(T3C_DATA(dev))->tid_maps, hwtid);
@@ -911,7 +945,7 @@ 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));
+		unsigned int opcode = get_opcode(skb);
 		int ret = cpl_handlers[opcode] (dev, skb);
 
 #if VALIDATE_TID
@@ -1229,6 +1263,25 @@ static inline void unregister_tdev(struct t3cdev *tdev)
 	mutex_unlock(&cxgb3_db_lock);
 }
 
+static inline int adap2type(struct adapter *adapter)
+{
+	int type = 0;
+
+	switch (adapter->params.rev) {
+	case T3_REV_A:
+		type = T3A;
+		break;
+	case T3_REV_B:
+	case T3_REV_B2:
+		type = T3B;
+		break;
+	case T3_REV_C:
+		type = T3C;
+		break;
+	}
+	return type;
+}
+
 void __devinit cxgb3_adapter_ofld(struct adapter *adapter)
 {
 	struct t3cdev *tdev = &adapter->tdev;
@@ -1238,7 +1291,7 @@ void __devinit cxgb3_adapter_ofld(struct adapter *adapter)
 	cxgb3_set_dummy_ops(tdev);
 	tdev->send = t3_offload_tx;
 	tdev->ctl = cxgb_offload_ctl;
-	tdev->type = adapter->params.rev == 0 ? T3A : T3B;
+	tdev->type = adap2type(adapter);
 
 	register_tdev(tdev);
 }
diff --git a/drivers/net/cxgb3/l2t.c b/drivers/net/cxgb3/l2t.c
index d80bbdb..655a4b0 100644
--- a/drivers/net/cxgb3/l2t.c
+++ b/drivers/net/cxgb3/l2t.c
@@ -407,7 +407,7 @@ found:
 			} else if (neigh->nud_state & (NUD_CONNECTED|NUD_STALE))
 				setup_l2e_send_pending(dev, NULL, e);
 		} else {
-			e->state = neigh_is_connected(neigh) ?
+			e->state = neigh->nud_state & NUD_CONNECTED ?
 			    L2T_STATE_VALID : L2T_STATE_STALE;
 			if (memcmp(e->dmac, neigh->ha, 6))
 				setup_l2e_send_pending(dev, NULL, e);
diff --git a/drivers/net/cxgb3/mc5.c b/drivers/net/cxgb3/mc5.c
index 84c1ffa..4c4d6e8 100644
--- a/drivers/net/cxgb3/mc5.c
+++ b/drivers/net/cxgb3/mc5.c
@@ -452,7 +452,7 @@ void t3_mc5_intr_handler(struct mc5 *mc5)
 	t3_write_reg(adap, A_MC5_DB_INT_CAUSE, cause);
 }
 
-void __devinit t3_mc5_prep(struct adapter *adapter, struct mc5 *mc5, int mode)
+void t3_mc5_prep(struct adapter *adapter, struct mc5 *mc5, int mode)
 {
 #define K * 1024
 
diff --git a/drivers/net/cxgb3/regs.h b/drivers/net/cxgb3/regs.h
index 02dbbb3..4bda27c 100644
--- a/drivers/net/cxgb3/regs.h
+++ b/drivers/net/cxgb3/regs.h
@@ -444,6 +444,14 @@
 
 #define A_PCIE_CFG 0x88
 
+#define S_ENABLELINKDWNDRST    21
+#define V_ENABLELINKDWNDRST(x) ((x) << S_ENABLELINKDWNDRST)
+#define F_ENABLELINKDWNDRST    V_ENABLELINKDWNDRST(1U)
+
+#define S_ENABLELINKDOWNRST    20
+#define V_ENABLELINKDOWNRST(x) ((x) << S_ENABLELINKDOWNRST)
+#define F_ENABLELINKDOWNRST    V_ENABLELINKDOWNRST(1U)
+
 #define S_PCIE_CLIDECEN    16
 #define V_PCIE_CLIDECEN(x) ((x) << S_PCIE_CLIDECEN)
 #define F_PCIE_CLIDECEN    V_PCIE_CLIDECEN(1U)
@@ -1509,16 +1517,18 @@
 
 #define A_ULPRX_ISCSI_TAGMASK 0x514
 
-#define S_HPZ0    0
-#define M_HPZ0    0xf
-#define V_HPZ0(x) ((x) << S_HPZ0)
-#define G_HPZ0(x) (((x) >> S_HPZ0) & M_HPZ0)
+#define A_ULPRX_ISCSI_PSZ 0x518
 
 #define A_ULPRX_TDDP_LLIMIT 0x51c
 
 #define A_ULPRX_TDDP_ULIMIT 0x520
 #define A_ULPRX_TDDP_PSZ 0x528
 
+#define S_HPZ0    0
+#define M_HPZ0    0xf
+#define V_HPZ0(x) ((x) << S_HPZ0)
+#define G_HPZ0(x) (((x) >> S_HPZ0) & M_HPZ0)
+
 #define A_ULPRX_STAG_LLIMIT 0x52c
 
 #define A_ULPRX_STAG_ULIMIT 0x530
diff --git a/drivers/net/cxgb3/sge.c b/drivers/net/cxgb3/sge.c
index 92a4926..d69e7ab 100644
--- a/drivers/net/cxgb3/sge.c
+++ b/drivers/net/cxgb3/sge.c
@@ -55,6 +55,9 @@
  * directly.
  */
 #define FL0_PG_CHUNK_SIZE  2048
+#define FL0_PG_ORDER 0
+#define FL1_PG_CHUNK_SIZE (PAGE_SIZE > 8192 ? 16384 : 8192)
+#define FL1_PG_ORDER (PAGE_SIZE > 8192 ? 0 : 1)
 
 #define SGE_RX_DROP_THRES 16
 
@@ -359,7 +362,7 @@ static void free_rx_bufs(struct pci_dev *pdev, struct sge_fl *q)
 	}
 
 	if (q->pg_chunk.page) {
-		__free_page(q->pg_chunk.page);
+		__free_pages(q->pg_chunk.page, q->order);
 		q->pg_chunk.page = NULL;
 	}
 }
@@ -376,13 +379,16 @@ static void free_rx_bufs(struct pci_dev *pdev, struct sge_fl *q)
  *	Add a buffer of the given length to the supplied HW and SW Rx
  *	descriptors.
  */
-static inline void add_one_rx_buf(void *va, unsigned int len,
-				  struct rx_desc *d, struct rx_sw_desc *sd,
-				  unsigned int gen, struct pci_dev *pdev)
+static inline int add_one_rx_buf(void *va, unsigned int len,
+				 struct rx_desc *d, struct rx_sw_desc *sd,
+				 unsigned int gen, struct pci_dev *pdev)
 {
 	dma_addr_t mapping;
 
 	mapping = pci_map_single(pdev, va, len, PCI_DMA_FROMDEVICE);
+	if (unlikely(pci_dma_mapping_error(mapping)))
+		return -ENOMEM;
+
 	pci_unmap_addr_set(sd, dma_addr, mapping);
 
 	d->addr_lo = cpu_to_be32(mapping);
@@ -390,12 +396,14 @@ static inline void add_one_rx_buf(void *va, unsigned int len,
 	wmb();
 	d->len_gen = cpu_to_be32(V_FLD_GEN1(gen));
 	d->gen2 = cpu_to_be32(V_FLD_GEN2(gen));
+	return 0;
 }
 
-static int alloc_pg_chunk(struct sge_fl *q, struct rx_sw_desc *sd, gfp_t gfp)
+static int alloc_pg_chunk(struct sge_fl *q, struct rx_sw_desc *sd, gfp_t gfp,
+			  unsigned int order)
 {
 	if (!q->pg_chunk.page) {
-		q->pg_chunk.page = alloc_page(gfp);
+		q->pg_chunk.page = alloc_pages(gfp, order);
 		if (unlikely(!q->pg_chunk.page))
 			return -ENOMEM;
 		q->pg_chunk.va = page_address(q->pg_chunk.page);
@@ -404,7 +412,7 @@ static int alloc_pg_chunk(struct sge_fl *q, struct rx_sw_desc *sd, gfp_t gfp)
 	sd->pg_chunk = q->pg_chunk;
 
 	q->pg_chunk.offset += q->buf_size;
-	if (q->pg_chunk.offset == PAGE_SIZE)
+	if (q->pg_chunk.offset == (PAGE_SIZE << order))
 		q->pg_chunk.page = NULL;
 	else {
 		q->pg_chunk.va += q->buf_size;
@@ -424,15 +432,18 @@ static int alloc_pg_chunk(struct sge_fl *q, struct rx_sw_desc *sd, gfp_t gfp)
  *	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)
+static int refill_fl(struct adapter *adap, struct sge_fl *q, int n, gfp_t gfp)
 {
 	void *buf_start;
 	struct rx_sw_desc *sd = &q->sdesc[q->pidx];
 	struct rx_desc *d = &q->desc[q->pidx];
+	unsigned int count = 0;
 
 	while (n--) {
+		int err;
+
 		if (q->use_pages) {
-			if (unlikely(alloc_pg_chunk(q, sd, gfp))) {
+			if (unlikely(alloc_pg_chunk(q, sd, gfp, q->order))) {
 nomem:				q->alloc_failed++;
 				break;
 			}
@@ -447,8 +458,16 @@ nomem:				q->alloc_failed++;
 			buf_start = skb->data;
 		}
 
-		add_one_rx_buf(buf_start, q->buf_size, d, sd, q->gen,
-			       adap->pdev);
+		err = add_one_rx_buf(buf_start, q->buf_size, d, sd, q->gen,
+				     adap->pdev);
+		if (unlikely(err)) {
+			if (!q->use_pages) {
+				kfree_skb(sd->skb);
+				sd->skb = NULL;
+			}
+			break;
+		}
+
 		d++;
 		sd++;
 		if (++q->pidx == q->size) {
@@ -458,14 +477,19 @@ nomem:				q->alloc_failed++;
 			d = q->desc;
 		}
 		q->credits++;
+		count++;
 	}
 	wmb();
-	t3_write_reg(adap, A_SG_KDOORBELL, V_EGRCNTX(q->cntxt_id));
+	if (likely(count))
+		t3_write_reg(adap, A_SG_KDOORBELL, V_EGRCNTX(q->cntxt_id));
+
+	return count;
 }
 
 static inline void __refill_fl(struct adapter *adap, struct sge_fl *fl)
 {
-	refill_fl(adap, fl, min(16U, fl->size - fl->credits), GFP_ATOMIC);
+	refill_fl(adap, fl, min(16U, fl->size - fl->credits),
+		  GFP_ATOMIC | __GFP_COMP);
 }
 
 /**
@@ -539,6 +563,33 @@ static void *alloc_ring(struct pci_dev *pdev, size_t nelem, size_t elem_size,
 }
 
 /**
+ *	t3_reset_qset - reset a sge qset
+ *	@q: the queue set
+ *
+ *	Reset the qset structure.
+ *	the NAPI structure is preserved in the event of
+ *	the qset's reincarnation, for example during EEH recovery.
+ */
+static void t3_reset_qset(struct sge_qset *q)
+{
+	if (q->adap &&
+	    !(q->adap->flags & NAPI_INIT)) {
+		memset(q, 0, sizeof(*q));
+		return;
+	}
+
+	q->adap = NULL;
+	memset(&q->rspq, 0, sizeof(q->rspq));
+	memset(q->fl, 0, sizeof(struct sge_fl) * SGE_RXQ_PER_SET);
+	memset(q->txq, 0, sizeof(struct sge_txq) * SGE_TXQ_PER_SET);
+	q->txq_stopped = 0;
+	memset(&q->tx_reclaim_timer, 0, sizeof(q->tx_reclaim_timer));
+	kfree(q->lro_frag_tbl);
+	q->lro_nfrags = q->lro_frag_len = 0;
+}
+
+
+/**
  *	free_qset - free the resources of an SGE queue set
  *	@adapter: the adapter owning the queue set
  *	@q: the queue set
@@ -557,9 +608,9 @@ static void t3_free_qset(struct adapter *adapter, struct sge_qset *q)
 
 	for (i = 0; i < SGE_RXQ_PER_SET; ++i)
 		if (q->fl[i].desc) {
-			spin_lock(&adapter->sge.reg_lock);
+			spin_lock_irq(&adapter->sge.reg_lock);
 			t3_sge_disable_fl(adapter, q->fl[i].cntxt_id);
-			spin_unlock(&adapter->sge.reg_lock);
+			spin_unlock_irq(&adapter->sge.reg_lock);
 			free_rx_bufs(pdev, &q->fl[i]);
 			kfree(q->fl[i].sdesc);
 			dma_free_coherent(&pdev->dev,
@@ -570,9 +621,9 @@ static void t3_free_qset(struct adapter *adapter, struct sge_qset *q)
 
 	for (i = 0; i < SGE_TXQ_PER_SET; ++i)
 		if (q->txq[i].desc) {
-			spin_lock(&adapter->sge.reg_lock);
+			spin_lock_irq(&adapter->sge.reg_lock);
 			t3_sge_enable_ecntxt(adapter, q->txq[i].cntxt_id, 0);
-			spin_unlock(&adapter->sge.reg_lock);
+			spin_unlock_irq(&adapter->sge.reg_lock);
 			if (q->txq[i].sdesc) {
 				free_tx_desc(adapter, &q->txq[i],
 					     q->txq[i].in_use);
@@ -586,9 +637,9 @@ static void t3_free_qset(struct adapter *adapter, struct sge_qset *q)
 		}
 
 	if (q->rspq.desc) {
-		spin_lock(&adapter->sge.reg_lock);
+		spin_lock_irq(&adapter->sge.reg_lock);
 		t3_sge_disable_rspcntxt(adapter, q->rspq.cntxt_id);
-		spin_unlock(&adapter->sge.reg_lock);
+		spin_unlock_irq(&adapter->sge.reg_lock);
 		dma_free_coherent(&pdev->dev,
 				  q->rspq.size * sizeof(struct rsp_desc),
 				  q->rspq.desc, q->rspq.phys_addr);
@@ -597,7 +648,7 @@ static void t3_free_qset(struct adapter *adapter, struct sge_qset *q)
 	if (q->netdev)
 		q->netdev->atalk_ptr = NULL;
 
-	memset(q, 0, sizeof(*q));
+	t3_reset_qset(q);
 }
 
 /**
@@ -718,19 +769,22 @@ use_orig_buf:
  * 	that are page chunks rather than sk_buffs.
  */
 static struct sk_buff *get_packet_pg(struct adapter *adap, struct sge_fl *fl,
-				     unsigned int len, unsigned int drop_thres)
+				     struct sge_rspq *q, unsigned int len,
+				     unsigned int drop_thres)
 {
-	struct sk_buff *skb = NULL;
+	struct sk_buff *newskb, *skb;
 	struct rx_sw_desc *sd = &fl->sdesc[fl->cidx];
 
-	if (len <= SGE_RX_COPY_THRES) {
-		skb = alloc_skb(len, GFP_ATOMIC);
-		if (likely(skb != NULL)) {
-			__skb_put(skb, len);
+	newskb = skb = q->pg_skb;
+
+	if (!skb && (len <= SGE_RX_COPY_THRES)) {
+		newskb = alloc_skb(len, GFP_ATOMIC);
+		if (likely(newskb != NULL)) {
+			__skb_put(newskb, len);
 			pci_dma_sync_single_for_cpu(adap->pdev,
 					    pci_unmap_addr(sd, dma_addr), len,
 					    PCI_DMA_FROMDEVICE);
-			memcpy(skb->data, sd->pg_chunk.va, len);
+			memcpy(newskb->data, sd->pg_chunk.va, len);
 			pci_dma_sync_single_for_device(adap->pdev,
 					    pci_unmap_addr(sd, dma_addr), len,
 					    PCI_DMA_FROMDEVICE);
@@ -739,14 +793,16 @@ static struct sk_buff *get_packet_pg(struct adapter *adap, struct sge_fl *fl,
 recycle:
 		fl->credits--;
 		recycle_rx_buf(adap, fl, fl->cidx);
-		return skb;
+		q->rx_recycle_buf++;
+		return newskb;
 	}
 
-	if (unlikely(fl->credits <= drop_thres))
+	if (unlikely(q->rx_recycle_buf || (!skb && fl->credits <= drop_thres)))
 		goto recycle;
 
-	skb = alloc_skb(SGE_RX_PULL_LEN, GFP_ATOMIC);
-	if (unlikely(!skb)) {
+	if (!skb)
+		newskb = alloc_skb(SGE_RX_PULL_LEN, GFP_ATOMIC);
+	if (unlikely(!newskb)) {
 		if (!drop_thres)
 			return NULL;
 		goto recycle;
@@ -754,21 +810,29 @@ recycle:
 
 	pci_unmap_single(adap->pdev, pci_unmap_addr(sd, dma_addr),
 			 fl->buf_size, PCI_DMA_FROMDEVICE);
-	__skb_put(skb, SGE_RX_PULL_LEN);
-	memcpy(skb->data, sd->pg_chunk.va, SGE_RX_PULL_LEN);
-	skb_fill_page_desc(skb, 0, sd->pg_chunk.page,
-			   sd->pg_chunk.offset + SGE_RX_PULL_LEN,
-			   len - SGE_RX_PULL_LEN);
-	skb->len = len;
-	skb->data_len = len - SGE_RX_PULL_LEN;
-	skb->truesize += skb->data_len;
+	if (!skb) {
+		__skb_put(newskb, SGE_RX_PULL_LEN);
+		memcpy(newskb->data, sd->pg_chunk.va, SGE_RX_PULL_LEN);
+		skb_fill_page_desc(newskb, 0, sd->pg_chunk.page,
+				   sd->pg_chunk.offset + SGE_RX_PULL_LEN,
+				   len - SGE_RX_PULL_LEN);
+		newskb->len = len;
+		newskb->data_len = len - SGE_RX_PULL_LEN;
+	} else {
+		skb_fill_page_desc(newskb, skb_shinfo(newskb)->nr_frags,
+				   sd->pg_chunk.page,
+				   sd->pg_chunk.offset, len);
+		newskb->len += len;
+		newskb->data_len += len;
+	}
+	newskb->truesize += newskb->data_len;
 
 	fl->credits--;
 	/*
 	 * We do not refill FLs here, we let the caller do it to overlap a
 	 * prefetch.
 	 */
-	return skb;
+	return newskb;
 }
 
 /**
@@ -1061,6 +1125,14 @@ static void write_tx_pkt_wr(struct adapter *adap, struct sk_buff *skb,
 			 htonl(V_WR_TID(q->token)));
 }
 
+static inline void t3_stop_queue(struct net_device *dev, struct sge_qset *qs,
+				 struct sge_txq *q)
+{
+	netif_stop_queue(dev);
+	set_bit(TXQ_ETH, &qs->txq_stopped);
+	q->stops++;
+}
+
 /**
  *	eth_xmit - add a packet to the Ethernet Tx queue
  *	@skb: the packet
@@ -1092,30 +1164,23 @@ int t3_eth_xmit(struct sk_buff *skb, struct net_device *dev)
 	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);
-		}
+		t3_stop_queue(dev, qs, q);
+		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
+		t3_stop_queue(dev, qs, q);
+
 		if (should_restart_tx(q) &&
 		    test_and_clear_bit(TXQ_ETH, &qs->txq_stopped)) {
 			q->restarts++;
 			netif_wake_queue(dev);
 		}
-#endif
 	}
 
 	gen = q->gen;
@@ -1367,7 +1432,7 @@ static void restart_ctrlq(unsigned long data)
  */
 int t3_mgmt_tx(struct adapter *adap, struct sk_buff *skb)
 {
-	int ret; 
+	int ret;
 	local_bh_disable();
 	ret = ctrl_xmit(adap, &adap->sge.qs[0].txq[TXQ_CTRL], skb);
 	local_bh_enable();
@@ -1811,9 +1876,10 @@ static void restart_tx(struct sge_qset *qs)
  *	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 sk_buff *skb, int pad, int lro)
 {
 	struct cpl_rx_pkt *p = (struct cpl_rx_pkt *)(skb->data + pad);
+	struct sge_qset *qs = rspq_to_qset(rq);
 	struct port_info *pi;
 
 	skb_pull(skb, sizeof(*p) + pad);
@@ -1821,7 +1887,7 @@ static void rx_eth(struct adapter *adap, struct sge_rspq *rq,
 	skb->protocol = eth_type_trans(skb, skb->dev);
 	skb->dev->last_rx = jiffies;
 	pi = netdev_priv(skb->dev);
-	if (pi->rx_csum_offload && p->csum_valid && p->csum == 0xffff &&
+	if (pi->rx_csum_offload && p->csum_valid && p->csum == htons(0xffff) &&
 	    !p->fragment) {
 		rspq_to_qset(rq)->port_stats[SGE_PSTAT_RX_CSUM_GOOD]++;
 		skb->ip_summed = CHECKSUM_UNNECESSARY;
@@ -1831,18 +1897,167 @@ static void rx_eth(struct adapter *adap, struct sge_rspq *rq,
 	if (unlikely(p->vlan_valid)) {
 		struct vlan_group *grp = pi->vlan_grp;
 
-		rspq_to_qset(rq)->port_stats[SGE_PSTAT_VLANEX]++;
+		qs->port_stats[SGE_PSTAT_VLANEX]++;
 		if (likely(grp))
-			__vlan_hwaccel_rx(skb, grp, ntohs(p->vlan),
-					  rq->polling);
+			if (lro)
+				lro_vlan_hwaccel_receive_skb(&qs->lro_mgr, skb,
+							     grp,
+							     ntohs(p->vlan),
+							     p);
+			else
+				__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
+	} else if (rq->polling) {
+		if (lro)
+			lro_receive_skb(&qs->lro_mgr, skb, p);
+		else
+			netif_receive_skb(skb);
+	} else
 		netif_rx(skb);
 }
 
+static inline int is_eth_tcp(u32 rss)
+{
+	return G_HASHTYPE(ntohl(rss)) == RSS_HASH_4_TUPLE;
+}
+
+/**
+ *	lro_frame_ok - check if an ingress packet is eligible for LRO
+ *	@p: the CPL header of the packet
+ *
+ *	Returns true if a received packet is eligible for LRO.
+ *	The following conditions must be true:
+ *	- packet is TCP/IP Ethernet II (checked elsewhere)
+ *	- not an IP fragment
+ *	- no IP options
+ *	- TCP/IP checksums are correct
+ *	- the packet is for this host
+ */
+static inline int lro_frame_ok(const struct cpl_rx_pkt *p)
+{
+	const struct ethhdr *eh = (struct ethhdr *)(p + 1);
+	const struct iphdr *ih = (struct iphdr *)(eh + 1);
+
+	return (*((u8 *)p + 1) & 0x90) == 0x10 && p->csum == htons(0xffff) &&
+		eh->h_proto == htons(ETH_P_IP) && ih->ihl == (sizeof(*ih) >> 2);
+}
+
+static int t3_get_lro_header(void **eh,  void **iph, void **tcph,
+			     u64 *hdr_flags, void *priv)
+{
+	const struct cpl_rx_pkt *cpl = priv;
+
+	if (!lro_frame_ok(cpl))
+		return -1;
+
+	*eh = (struct ethhdr *)(cpl + 1);
+	*iph = (struct iphdr *)((struct ethhdr *)*eh + 1);
+	*tcph = (struct tcphdr *)((struct iphdr *)*iph + 1);
+
+	*hdr_flags = LRO_IPV4 | LRO_TCP;
+	return 0;
+}
+
+static int t3_get_skb_header(struct sk_buff *skb,
+			      void **iph, void **tcph, u64 *hdr_flags,
+			      void *priv)
+{
+	void *eh;
+
+	return t3_get_lro_header(&eh, iph, tcph, hdr_flags, priv);
+}
+
+static int t3_get_frag_header(struct skb_frag_struct *frag, void **eh,
+			      void **iph, void **tcph, u64 *hdr_flags,
+			      void *priv)
+{
+	return t3_get_lro_header(eh, iph, tcph, hdr_flags, priv);
+}
+
+/**
+ *	lro_add_page - add a page chunk to an LRO session
+ *	@adap: the adapter
+ *	@qs: the associated queue set
+ *	@fl: the free list containing the page chunk to add
+ *	@len: packet length
+ *	@complete: Indicates the last fragment of a frame
+ *
+ *	Add a received packet contained in a page chunk to an existing LRO
+ *	session.
+ */
+static void lro_add_page(struct adapter *adap, struct sge_qset *qs,
+			 struct sge_fl *fl, int len, int complete)
+{
+	struct rx_sw_desc *sd = &fl->sdesc[fl->cidx];
+	struct cpl_rx_pkt *cpl;
+	struct skb_frag_struct *rx_frag = qs->lro_frag_tbl;
+	int nr_frags = qs->lro_nfrags, frag_len = qs->lro_frag_len;
+	int offset = 0;
+
+	if (!nr_frags) {
+		offset = 2 + sizeof(struct cpl_rx_pkt);
+		qs->lro_va = cpl = sd->pg_chunk.va + 2;
+	}
+
+	fl->credits--;
+
+	len -= offset;
+	pci_unmap_single(adap->pdev, pci_unmap_addr(sd, dma_addr),
+			 fl->buf_size, PCI_DMA_FROMDEVICE);
+
+	rx_frag += nr_frags;
+	rx_frag->page = sd->pg_chunk.page;
+	rx_frag->page_offset = sd->pg_chunk.offset + offset;
+	rx_frag->size = len;
+	frag_len += len;
+	qs->lro_nfrags++;
+	qs->lro_frag_len = frag_len;
+
+	if (!complete)
+		return;
+
+	qs->lro_nfrags = qs->lro_frag_len = 0;
+	cpl = qs->lro_va;
+
+	if (unlikely(cpl->vlan_valid)) {
+		struct net_device *dev = qs->netdev;
+		struct port_info *pi = netdev_priv(dev);
+		struct vlan_group *grp = pi->vlan_grp;
+
+		if (likely(grp != NULL)) {
+			lro_vlan_hwaccel_receive_frags(&qs->lro_mgr,
+						       qs->lro_frag_tbl,
+						       frag_len, frag_len,
+						       grp, ntohs(cpl->vlan),
+						       cpl, 0);
+			return;
+		}
+	}
+	lro_receive_frags(&qs->lro_mgr, qs->lro_frag_tbl,
+			  frag_len, frag_len, cpl, 0);
+}
+
+/**
+ *	init_lro_mgr - initialize a LRO manager object
+ *	@lro_mgr: the LRO manager object
+ */
+static void init_lro_mgr(struct sge_qset *qs, struct net_lro_mgr *lro_mgr)
+{
+	lro_mgr->dev = qs->netdev;
+	lro_mgr->features = LRO_F_NAPI;
+	lro_mgr->ip_summed = CHECKSUM_UNNECESSARY;
+	lro_mgr->ip_summed_aggr = CHECKSUM_UNNECESSARY;
+	lro_mgr->max_desc = T3_MAX_LRO_SES;
+	lro_mgr->lro_arr = qs->lro_desc;
+	lro_mgr->get_frag_header = t3_get_frag_header;
+	lro_mgr->get_skb_header = t3_get_skb_header;
+	lro_mgr->max_aggr = T3_MAX_LRO_MAX_PKTS;
+	if (lro_mgr->max_aggr > MAX_SKB_FRAGS)
+		lro_mgr->max_aggr = MAX_SKB_FRAGS;
+}
+
 /**
  *	handle_rsp_cntrl_info - handles control information in a response
  *	@qs: the queue set corresponding to the response
@@ -1928,6 +2143,12 @@ static inline int is_new_response(const struct rsp_desc *r,
 	return (r->intr_gen & F_RSPD_GEN2) == q->gen;
 }
 
+static inline void clear_rspq_bufstate(struct sge_rspq * const q)
+{
+	q->pg_skb = NULL;
+	q->rx_recycle_buf = 0;
+}
+
 #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) | \
@@ -1965,10 +2186,11 @@ static int process_responses(struct adapter *adap, struct sge_qset *qs,
 	q->next_holdoff = q->holdoff_tmr;
 
 	while (likely(budget_left && is_new_response(r, q))) {
-		int eth, ethpad = 2;
+		int packet_complete, eth, ethpad = 2, lro = qs->lro_enabled;
 		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;
+		__be32 rss_hi = *(const __be32 *)r,
+		       rss_lo = r->rss_hdr.rss_hash_val;
 
 		eth = r->rss_hdr.opcode == CPL_RX_PKT;
 
@@ -1996,6 +2218,9 @@ no_mem:
 		} else if ((len = ntohl(r->len_cq)) != 0) {
 			struct sge_fl *fl;
 
+			if (eth)
+				lro = qs->lro_enabled && is_eth_tcp(rss_hi);
+
 			fl = (len & F_RSPD_FLQ) ? &qs->fl[1] : &qs->fl[0];
 			if (fl->use_pages) {
 				void *addr = fl->sdesc[fl->cidx].pg_chunk.va;
@@ -2005,9 +2230,18 @@ no_mem:
 				prefetch(addr + L1_CACHE_BYTES);
 #endif
 				__refill_fl(adap, fl);
+				if (lro > 0) {
+					lro_add_page(adap, qs, fl,
+						     G_RSPD_LEN(len),
+						     flags & F_RSPD_EOP);
+					 goto next_fl;
+				}
 
-				skb = get_packet_pg(adap, fl, G_RSPD_LEN(len),
-						 eth ? SGE_RX_DROP_THRES : 0);
+				skb = get_packet_pg(adap, fl, q,
+						    G_RSPD_LEN(len),
+						    eth ?
+						    SGE_RX_DROP_THRES : 0);
+				q->pg_skb = skb;
 			} else
 				skb = get_packet(adap, fl, G_RSPD_LEN(len),
 						 eth ? SGE_RX_DROP_THRES : 0);
@@ -2017,7 +2251,7 @@ no_mem:
 				q->rx_drops++;
 			} else if (unlikely(r->rss_hdr.opcode == CPL_TRACE_PKT))
 				__skb_pull(skb, 2);
-
+next_fl:
 			if (++fl->cidx == fl->size)
 				fl->cidx = 0;
 		} else
@@ -2041,9 +2275,13 @@ no_mem:
 			q->credits = 0;
 		}
 
-		if (likely(skb != NULL)) {
+		packet_complete = flags &
+				  (F_RSPD_EOP | F_RSPD_IMM_DATA_VALID |
+				   F_RSPD_ASYNC_NOTIF);
+
+		if (skb != NULL && packet_complete) {
 			if (eth)
-				rx_eth(adap, q, skb, ethpad);
+				rx_eth(adap, q, skb, ethpad, lro);
 			else {
 				q->offload_pkts++;
 				/* Preserve the RSS info in csum & priority */
@@ -2053,11 +2291,19 @@ no_mem:
 						       offload_skbs,
 						       ngathered);
 			}
+
+			if (flags & F_RSPD_EOP)
+				clear_rspq_bufstate(q);
 		}
 		--budget_left;
 	}
 
 	deliver_partial_bundle(&adap->tdev, q, offload_skbs, ngathered);
+	lro_flush_all(&qs->lro_mgr);
+	qs->port_stats[SGE_PSTAT_LRO_AGGR] = qs->lro_mgr.stats.aggregated;
+	qs->port_stats[SGE_PSTAT_LRO_FLUSHED] = qs->lro_mgr.stats.flushed;
+	qs->port_stats[SGE_PSTAT_LRO_NO_DESC] = qs->lro_mgr.stats.no_desc;
+
 	if (sleeping)
 		check_ring_db(adap, qs, sleeping);
 
@@ -2616,8 +2862,9 @@ 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;
+	int i, avail, ret = -ENOMEM;
 	struct sge_qset *q = &adapter->sge.qs[id];
+	struct net_lro_mgr *lro_mgr = &q->lro_mgr;
 
 	init_qset_cntxt(q, id);
 	init_timer(&q->tx_reclaim_timer);
@@ -2685,12 +2932,24 @@ int t3_sge_alloc_qset(struct adapter *adapter, unsigned int id, int nports,
 #else
 	q->fl[0].buf_size = SGE_RX_SM_BUF_SIZE + sizeof(struct cpl_rx_data);
 #endif
-	q->fl[0].use_pages = FL0_PG_CHUNK_SIZE > 0;
+#if FL1_PG_CHUNK_SIZE > 0
+	q->fl[1].buf_size = FL1_PG_CHUNK_SIZE;
+#else
 	q->fl[1].buf_size = is_offload(adapter) ?
 		(16 * 1024) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) :
 		MAX_FRAME_SIZE + 2 + sizeof(struct cpl_rx_pkt);
+#endif
 
-	spin_lock(&adapter->sge.reg_lock);
+	q->fl[0].use_pages = FL0_PG_CHUNK_SIZE > 0;
+	q->fl[1].use_pages = FL1_PG_CHUNK_SIZE > 0;
+	q->fl[0].order = FL0_PG_ORDER;
+	q->fl[1].order = FL1_PG_ORDER;
+
+	q->lro_frag_tbl = kcalloc(MAX_FRAME_SIZE / FL1_PG_CHUNK_SIZE + 1,
+				  sizeof(struct skb_frag_struct),
+				  GFP_KERNEL);
+	q->lro_nfrags = q->lro_frag_len = 0;
+	spin_lock_irq(&adapter->sge.reg_lock);
 
 	/* FL threshold comparison uses < */
 	ret = t3_sge_init_rspcntxt(adapter, q->rspq.cntxt_id, irq_vec_idx,
@@ -2734,7 +2993,9 @@ int t3_sge_alloc_qset(struct adapter *adapter, unsigned int id, int nports,
 			goto err_unlock;
 	}
 
-	spin_unlock(&adapter->sge.reg_lock);
+	spin_unlock_irq(&adapter->sge.reg_lock);
+
+	q->adap = adapter;
 	q->netdev = netdev;
 	t3_update_qset_coalesce(q, p);
 
@@ -2746,8 +3007,23 @@ int t3_sge_alloc_qset(struct adapter *adapter, unsigned int id, int nports,
 	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);
+	init_lro_mgr(q, lro_mgr);
+
+	avail = refill_fl(adapter, &q->fl[0], q->fl[0].size,
+			  GFP_KERNEL | __GFP_COMP);
+	if (!avail) {
+		CH_ALERT(adapter, "free list queue 0 initialization failed\n");
+		goto err;
+	}
+	if (avail < q->fl[0].size)
+		CH_WARN(adapter, "free list queue 0 enabled with %d credits\n",
+			avail);
+
+	avail = refill_fl(adapter, &q->fl[1], q->fl[1].size,
+			  GFP_KERNEL | __GFP_COMP);
+	if (avail < q->fl[1].size)
+		CH_WARN(adapter, "free list queue 1 enabled with %d credits\n",
+			avail);
 	refill_rspq(adapter, &q->rspq, q->rspq.size - 1);
 
 	t3_write_reg(adapter, A_SG_GTS, V_RSPQ(q->rspq.cntxt_id) |
@@ -2756,9 +3032,9 @@ int t3_sge_alloc_qset(struct adapter *adapter, unsigned int id, int nports,
 	mod_timer(&q->tx_reclaim_timer, jiffies + TX_RECLAIM_PERIOD);
 	return 0;
 
-      err_unlock:
-	spin_unlock(&adapter->sge.reg_lock);
-      err:
+err_unlock:
+	spin_unlock_irq(&adapter->sge.reg_lock);
+err:
 	t3_free_qset(adapter, q);
 	return ret;
 }
@@ -2866,7 +3142,7 @@ void t3_sge_init(struct adapter *adap, struct sge_params *p)
  *	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)
+void t3_sge_prep(struct adapter *adap, struct sge_params *p)
 {
 	int i;
 
@@ -2880,7 +3156,7 @@ void __devinit t3_sge_prep(struct adapter *adap, struct sge_params *p)
 		q->coalesce_usecs = 5;
 		q->rspq_size = 1024;
 		q->fl_size = 1024;
-		q->jumbo_size = 512;
+ 		q->jumbo_size = 512;
 		q->txq_size[TXQ_ETH] = 1024;
 		q->txq_size[TXQ_OFLD] = 1024;
 		q->txq_size[TXQ_CTRL] = 256;
diff --git a/drivers/net/cxgb3/t3_cpl.h b/drivers/net/cxgb3/t3_cpl.h
index b7a1a31..917970e 100644
--- a/drivers/net/cxgb3/t3_cpl.h
+++ b/drivers/net/cxgb3/t3_cpl.h
@@ -174,6 +174,13 @@ enum {				/* TCP congestion control algorithms */
 	CONG_ALG_HIGHSPEED
 };
 
+enum {			/* RSS hash type */
+	RSS_HASH_NONE = 0,
+	RSS_HASH_2_TUPLE = 1,
+	RSS_HASH_4_TUPLE = 2,
+	RSS_HASH_TCPV6 = 3
+};
+
 union opcode_tid {
 	__be32 opcode_tid;
 	__u8 opcode;
@@ -184,6 +191,13 @@ union opcode_tid {
 #define G_OPCODE(x) (((x) >> S_OPCODE) & 0xFF)
 #define G_TID(x)    ((x) & 0xFFFFFF)
 
+#define S_QNUM 0
+#define G_QNUM(x) (((x) >> S_QNUM) & 0xFFFF)
+
+#define S_HASHTYPE 22
+#define M_HASHTYPE 0x3
+#define G_HASHTYPE(x) (((x) >> S_HASHTYPE) & M_HASHTYPE)
+
 /* tid is assumed to be 24-bits */
 #define MK_OPCODE_TID(opcode, tid) (V_OPCODE(opcode) | (tid))
 
@@ -768,6 +782,12 @@ struct tx_data_wr {
 	__be32 param;
 };
 
+/* tx_data_wr.flags fields */
+#define S_TX_ACK_PAGES	21
+#define M_TX_ACK_PAGES	0x7
+#define V_TX_ACK_PAGES(x) ((x) << S_TX_ACK_PAGES)
+#define G_TX_ACK_PAGES(x) (((x) >> S_TX_ACK_PAGES) & M_TX_ACK_PAGES)
+
 /* tx_data_wr.param fields */
 #define S_TX_PORT    0
 #define M_TX_PORT    0x7
@@ -1441,4 +1461,35 @@ struct cpl_rdma_terminate {
 #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)
+
+/* ULP_TX opcodes */
+enum { ULP_MEM_READ = 2, ULP_MEM_WRITE = 3, ULP_TXPKT = 4 };
+
+#define S_ULPTX_CMD	28
+#define M_ULPTX_CMD	0xF
+#define V_ULPTX_CMD(x)	((x) << S_ULPTX_CMD)
+
+#define S_ULPTX_NFLITS	0
+#define M_ULPTX_NFLITS	0xFF
+#define V_ULPTX_NFLITS(x) ((x) << S_ULPTX_NFLITS)
+
+struct ulp_mem_io {
+	WR_HDR;
+	__be32 cmd_lock_addr;
+	__be32 len;
+};
+
+/* ulp_mem_io.cmd_lock_addr fields */
+#define S_ULP_MEMIO_ADDR	0
+#define M_ULP_MEMIO_ADDR	0x7FFFFFF
+#define V_ULP_MEMIO_ADDR(x)	((x) << S_ULP_MEMIO_ADDR)
+#define S_ULP_MEMIO_LOCK	27
+#define V_ULP_MEMIO_LOCK(x)	((x) << S_ULP_MEMIO_LOCK)
+#define F_ULP_MEMIO_LOCK	V_ULP_MEMIO_LOCK(1U)
+
+/* ulp_mem_io.len fields */
+#define S_ULP_MEMIO_DATA_LEN	28
+#define M_ULP_MEMIO_DATA_LEN	0xF
+#define V_ULP_MEMIO_DATA_LEN(x)	((x) << S_ULP_MEMIO_DATA_LEN)
+
 #endif				/* T3_CPL_H */
diff --git a/drivers/net/cxgb3/t3_hw.c b/drivers/net/cxgb3/t3_hw.c
index c39b430..dce82ac 100644
--- a/drivers/net/cxgb3/t3_hw.c
+++ b/drivers/net/cxgb3/t3_hw.c
@@ -537,10 +537,11 @@ struct t3_vpd {
  *	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)
+int t3_seeprom_read(struct adapter *adapter, u32 addr, __le32 *data)
 {
 	u16 val;
 	int attempts = EEPROM_MAX_POLL;
+	u32 v;
 	unsigned int base = adapter->params.pci.vpd_cap_addr;
 
 	if ((addr >= EEPROMSIZE && addr != EEPROM_STAT_ADDR) || (addr & 3))
@@ -556,8 +557,8 @@ int t3_seeprom_read(struct adapter *adapter, u32 addr, u32 *data)
 		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);
+	pci_read_config_dword(adapter->pdev, base + PCI_VPD_DATA, &v);
+	*data = cpu_to_le32(v);
 	return 0;
 }
 
@@ -570,7 +571,7 @@ int t3_seeprom_read(struct adapter *adapter, u32 addr, u32 *data)
  *	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)
+int t3_seeprom_write(struct adapter *adapter, u32 addr, __le32 data)
 {
 	u16 val;
 	int attempts = EEPROM_MAX_POLL;
@@ -580,7 +581,7 @@ int t3_seeprom_write(struct adapter *adapter, u32 addr, u32 data)
 		return -EINVAL;
 
 	pci_write_config_dword(adapter->pdev, base + PCI_VPD_DATA,
-			       cpu_to_le32(data));
+			       le32_to_cpu(data));
 	pci_write_config_word(adapter->pdev,base + PCI_VPD_ADDR,
 			      addr | PCI_VPD_ADDR_F);
 	do {
@@ -631,14 +632,14 @@ static int get_vpd_params(struct adapter *adapter, struct vpd_params *p)
 	 * Card information is normally at VPD_BASE but some early cards had
 	 * it at 0.
 	 */
-	ret = t3_seeprom_read(adapter, VPD_BASE, (u32 *)&vpd);
+	ret = t3_seeprom_read(adapter, VPD_BASE, (__le32 *)&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));
+				      (__le32 *)((u8 *)&vpd + i));
 		if (ret)
 			return ret;
 	}
@@ -922,11 +923,12 @@ int t3_check_tpsram_version(struct adapter *adapter, int *must_load)
  *	Checks if an adapter's tp sram is compatible with the driver.
  *	Returns 0 if the versions are compatible, a negative error otherwise.
  */
-int t3_check_tpsram(struct adapter *adapter, u8 *tp_sram, unsigned int size)
+int t3_check_tpsram(struct adapter *adapter, const u8 *tp_sram,
+		    unsigned int size)
 {
 	u32 csum;
 	unsigned int i;
-	const u32 *p = (const u32 *)tp_sram;
+	const __be32 *p = (const __be32 *)tp_sram;
 
 	/* Verify checksum */
 	for (csum = 0, i = 0; i < size / sizeof(csum); i++)
@@ -1040,7 +1042,7 @@ 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;
+	const __be32 *p = (const __be32 *)fw_data;
 	int ret, addr, fw_sector = FW_FLASH_BOOT_ADDR >> 16;
 
 	if ((size & 3) || size < FW_MIN_SIZE)
@@ -2674,7 +2676,7 @@ void t3_tp_set_max_rxsize(struct adapter *adap, unsigned int size)
 		     V_PMMAXXFERLEN0(size) | V_PMMAXXFERLEN1(size));
 }
 
-static void __devinit init_mtus(unsigned short mtus[])
+static void init_mtus(unsigned short mtus[])
 {
 	/*
 	 * See draft-mathis-plpmtud-00.txt for the values.  The min is 88 so
@@ -2702,7 +2704,7 @@ static void __devinit init_mtus(unsigned short mtus[])
 /*
  * Initial congestion control parameters.
  */
-static void __devinit init_cong_ctrl(unsigned short *a, unsigned short *b)
+static void 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;
@@ -2874,17 +2876,17 @@ static void ulp_config(struct adapter *adap, const struct tp_params *p)
  *
  *	Write the contents of the protocol SRAM.
  */
-int t3_set_proto_sram(struct adapter *adap, u8 *data)
+int t3_set_proto_sram(struct adapter *adap, const u8 *data)
 {
 	int i;
-	u32 *buf = (u32 *)data;
+	const __be32 *buf = (const __be32 *)data;
 
 	for (i = 0; i < PROTO_SRAM_LINES; i++) {
-		t3_write_reg(adap, A_TP_EMBED_OP_FIELD5, cpu_to_be32(*buf++));
-		t3_write_reg(adap, A_TP_EMBED_OP_FIELD4, cpu_to_be32(*buf++));
-		t3_write_reg(adap, A_TP_EMBED_OP_FIELD3, cpu_to_be32(*buf++));
-		t3_write_reg(adap, A_TP_EMBED_OP_FIELD2, cpu_to_be32(*buf++));
-		t3_write_reg(adap, A_TP_EMBED_OP_FIELD1, cpu_to_be32(*buf++));
+		t3_write_reg(adap, A_TP_EMBED_OP_FIELD5, be32_to_cpu(*buf++));
+		t3_write_reg(adap, A_TP_EMBED_OP_FIELD4, be32_to_cpu(*buf++));
+		t3_write_reg(adap, A_TP_EMBED_OP_FIELD3, be32_to_cpu(*buf++));
+		t3_write_reg(adap, A_TP_EMBED_OP_FIELD2, be32_to_cpu(*buf++));
+		t3_write_reg(adap, A_TP_EMBED_OP_FIELD1, be32_to_cpu(*buf++));
 
 		t3_write_reg(adap, A_TP_EMBED_OP_FIELD0, i << 1 | 1 << 31);
 		if (t3_wait_op_done(adap, A_TP_EMBED_OP_FIELD0, 1, 1, 5, 1))
@@ -3263,6 +3265,7 @@ static void config_pcie(struct adapter *adap)
 
 	t3_write_reg(adap, A_PCIE_PEX_ERR, 0xffffffff);
 	t3_set_reg_field(adap, A_PCIE_CFG, 0,
+			 F_ENABLELINKDWNDRST | F_ENABLELINKDOWNRST |
 			 F_PCIE_DMASTOPEN | F_PCIE_CLIDECEN);
 }
 
@@ -3355,8 +3358,7 @@ out_err:
  *	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 void get_pci_mode(struct adapter *adapter, struct pci_params *p)
 {
 	static unsigned short speed_map[] = { 33, 66, 100, 133 };
 	u32 pci_mode, pcie_cap;
@@ -3396,8 +3398,7 @@ static void __devinit get_pci_mode(struct adapter *adapter,
  *	capabilities and default speed/duplex/flow-control/autonegotiation
  *	settings.
  */
-static void __devinit init_link_config(struct link_config *lc,
-				       unsigned int caps)
+static void init_link_config(struct link_config *lc, unsigned int caps)
 {
 	lc->supported = caps;
 	lc->requested_speed = lc->speed = SPEED_INVALID;
@@ -3420,7 +3421,7 @@ static void __devinit init_link_config(struct link_config *lc,
  *	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)
+static unsigned int mc7_calc_size(u32 cfg)
 {
 	unsigned int width = G_WIDTH(cfg);
 	unsigned int banks = !!(cfg & F_BKS) + 1;
@@ -3431,8 +3432,8 @@ static unsigned int __devinit mc7_calc_size(u32 cfg)
 	return MBs << 20;
 }
 
-static void __devinit mc7_prep(struct adapter *adapter, struct mc7 *mc7,
-			       unsigned int base_addr, const char *name)
+static void mc7_prep(struct adapter *adapter, struct mc7 *mc7,
+		     unsigned int base_addr, const char *name)
 {
 	u32 cfg;
 
@@ -3518,7 +3519,7 @@ static int t3_reset_adapter(struct adapter *adapter)
 	return 0;
 }
 
-static int __devinit init_parity(struct adapter *adap)
+static int init_parity(struct adapter *adap)
 {
 		int i, err, addr;
 
@@ -3553,8 +3554,8 @@ static int __devinit init_parity(struct adapter *adap)
  * 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 t3_prep_adapter(struct adapter *adapter, const struct adapter_info *ai,
+		    int reset)
 {
 	int ret;
 	unsigned int i, j = 0;
@@ -3658,3 +3659,30 @@ void t3_led_ready(struct adapter *adapter)
 	t3_set_reg_field(adapter, A_T3DBG_GPIO_EN, F_GPIO0_OUT_VAL,
 			 F_GPIO0_OUT_VAL);
 }
+
+int t3_replay_prep_adapter(struct adapter *adapter)
+{
+	const struct adapter_info *ai = adapter->params.info;
+	unsigned int i, j = 0;
+	int ret;
+
+	early_hw_init(adapter, ai);
+	ret = init_parity(adapter);
+	if (ret)
+		return ret;
+
+	for_each_port(adapter, i) {
+		struct port_info *p = adap2pinfo(adapter, i);
+		while (!adapter->params.vpd.port_type[j])
+			++j;
+
+		p->port_type->phy_prep(&p->phy, adapter, ai->phy_base_addr + j,
+					ai->mdio_ops);
+
+		p->phy.ops->power_down(&p->phy, 1);
+		++j;
+	}
+
+	return 0;
+}
+
diff --git a/drivers/net/cxgb3/t3cdev.h b/drivers/net/cxgb3/t3cdev.h
index 77fcc1a..0a21cfb 100644
--- a/drivers/net/cxgb3/t3cdev.h
+++ b/drivers/net/cxgb3/t3cdev.h
@@ -34,7 +34,6 @@
 
 #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>
@@ -46,7 +45,8 @@ struct cxgb3_client;
 
 enum t3ctype {
 	T3A = 0,
-	T3B
+	T3B,
+	T3C,
 };
 
 struct t3cdev {
@@ -64,6 +64,7 @@ struct t3cdev {
 	void *l3opt;		/* optional layer 3 data */
 	void *l4opt;		/* optional layer 4 data */
 	void *ulp;		/* ulp stuff */
+	void *ulp_iscsi;	/* ulp iscsi */
 };
 
 #endif				/* _T3CDEV_H_ */
diff --git a/drivers/net/cxgb3/version.h b/drivers/net/cxgb3/version.h
index 229303f..a0177fc 100644
--- a/drivers/net/cxgb3/version.h
+++ b/drivers/net/cxgb3/version.h
@@ -38,7 +38,7 @@
 #define DRV_VERSION "1.0-ko"
 
 /* Firmware version */
-#define FW_VERSION_MAJOR 5
+#define FW_VERSION_MAJOR 6
 #define FW_VERSION_MINOR 0
 #define FW_VERSION_MICRO 0
 #endif				/* __CHELSIO_VERSION_H */