DPDK patches and discussions
 help / color / mirror / Atom feed
From: Rahul Lakkireddy <rahul.lakkireddy@chelsio.com>
To: dev@dpdk.org
Cc: Kumar Sanghvi <kumaras@chelsio.com>,
	Felix Marti <felix@chelsio.com>,
	Nirranjan Kirubaharan <nirranjan@chelsio.com>
Subject: [dpdk-dev] [PATCH 2/5] cxgbe: add cxgbe poll mode driver.
Date: Fri, 22 May 2015 18:54:20 +0530	[thread overview]
Message-ID: <25f06823ef87d05bf57d996df387478e7fbc966f.1432300701.git.rahul.lakkireddy@chelsio.com> (raw)
In-Reply-To: <cover.1432300701.git.rahul.lakkireddy@chelsio.com>
In-Reply-To: <cover.1432300701.git.rahul.lakkireddy@chelsio.com>

Adds cxgbe poll mode driver for DPDK under lib/librte_pmd_cxgbe directory. This
patch:

1. Adds the Makefile to compile cxgbe pmd.
2. Registers and initializes the cxgbe pmd driver.
3. Implements the necessary eth_dev_ops.

Signed-off-by: Rahul Lakkireddy <rahul.lakkireddy@chelsio.com>
Signed-off-by: Kumar Sanghvi <kumaras@chelsio.com>
---
 lib/librte_pmd_cxgbe/Makefile       |   74 ++
 lib/librte_pmd_cxgbe/cxgbe.h        |   60 +
 lib/librte_pmd_cxgbe/cxgbe_compat.h |  290 +++++
 lib/librte_pmd_cxgbe/cxgbe_ethdev.c |  796 +++++++++++++
 lib/librte_pmd_cxgbe/cxgbe_main.c   | 1219 +++++++++++++++++++
 lib/librte_pmd_cxgbe/sge.c          | 2250 +++++++++++++++++++++++++++++++++++
 6 files changed, 4689 insertions(+), 0 deletions(-)
 create mode 100644 lib/librte_pmd_cxgbe/Makefile
 create mode 100644 lib/librte_pmd_cxgbe/cxgbe.h
 create mode 100644 lib/librte_pmd_cxgbe/cxgbe_compat.h
 create mode 100644 lib/librte_pmd_cxgbe/cxgbe_ethdev.c
 create mode 100644 lib/librte_pmd_cxgbe/cxgbe_main.c
 create mode 100644 lib/librte_pmd_cxgbe/sge.c

diff --git a/lib/librte_pmd_cxgbe/Makefile b/lib/librte_pmd_cxgbe/Makefile
new file mode 100644
index 0000000..945f196
--- /dev/null
+++ b/lib/librte_pmd_cxgbe/Makefile
@@ -0,0 +1,74 @@
+#   BSD LICENSE
+#
+#   Copyright(c) 2014-2015 Chelsio Communications.
+#   All rights reserved.
+#
+#   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.
+#     * Neither the name of Chelsio Communications nor the names of its
+#       contributors may be used to endorse or promote products derived
+#       from this software without specific prior written permission.
+#
+#   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+#   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+#   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+#   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+#   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+#   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+#   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+#   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+#   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+#   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+#   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+include $(RTE_SDK)/mk/rte.vars.mk
+
+#
+# library name
+#
+LIB = librte_pmd_cxgbe.a
+
+CFLAGS += -I$(RTE_SDK)/lib/librte_pmd_cxgbe/cxgbe/
+CFLAGS += -I$(RTE_SDK)/lib/librte_pmd_cxgbe/
+CFLAGS += -O3
+CFLAGS += $(WERROR_FLAGS)
+
+ifeq ($(CC), icc)
+#
+# CFLAGS for icc
+#
+CFLAGS_BASE_DRIVER = -wd174 -wd593 -wd869 -wd981 -wd2259
+else
+#
+# CFLAGS for gcc
+#
+ifeq ($(shell test $(GCC_VERSION) -ge 44 && echo 1), 1)
+CFLAGS     += -Wno-deprecated
+endif
+CFLAGS_BASE_DRIVER = -Wno-unused-parameter -Wno-unused-value
+CFLAGS_BASE_DRIVER += -Wno-strict-aliasing -Wno-format-extra-args
+
+endif
+
+#
+# all source are stored in SRCS-y
+#
+SRCS-$(CONFIG_RTE_LIBRTE_CXGBE_PMD) += cxgbe_ethdev.c
+SRCS-$(CONFIG_RTE_LIBRTE_CXGBE_PMD) += cxgbe_main.c
+SRCS-$(CONFIG_RTE_LIBRTE_CXGBE_PMD) += sge.c
+SRCS-$(CONFIG_RTE_LIBRTE_CXGBE_PMD) += cxgbe/t4_hw.c
+
+# this lib depends upon:
+DEPDIRS-$(CONFIG_RTE_LIBRTE_CXGBE_PMD) += lib/librte_eal lib/librte_ether
+DEPDIRS-$(CONFIG_RTE_LIBRTE_CXGBE_PMD) += lib/librte_mempool lib/librte_mbuf
+DEPDIRS-$(CONFIG_RTE_LIBRTE_CXGBE_PMD) += lib/librte_net lib/librte_malloc
+
+include $(RTE_SDK)/mk/rte.lib.mk
diff --git a/lib/librte_pmd_cxgbe/cxgbe.h b/lib/librte_pmd_cxgbe/cxgbe.h
new file mode 100644
index 0000000..97c37d2
--- /dev/null
+++ b/lib/librte_pmd_cxgbe/cxgbe.h
@@ -0,0 +1,60 @@
+/*-
+ *   BSD LICENSE
+ *
+ *   Copyright(c) 2014-2015 Chelsio Communications.
+ *   All rights reserved.
+ *
+ *   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.
+ *     * Neither the name of Chelsio Communications nor the names of its
+ *       contributors may be used to endorse or promote products derived
+ *       from this software without specific prior written permission.
+ *
+ *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#ifndef _CXGBE_H_
+#define _CXGBE_H_
+
+#include "common.h"
+#include "t4_regs.h"
+
+#define CXGBE_MIN_RING_DESC_SIZE      1024 /* Min TX/RX descriptor ring size */
+#define CXGBE_MAX_RING_DESC_SIZE      4096 /* Max TX/RX descriptor ring size */
+
+#define CXGBE_DEFAULT_TX_DESC_SIZE    1024 /* Default TX ring size */
+#define CXGBE_DEFAULT_RX_DESC_SIZE    1024 /* Default RX ring size */
+
+int cxgbe_probe(struct adapter *adapter);
+int cxgbe_up(struct adapter *adap);
+int cxgbe_down(struct port_info *pi);
+void cxgbe_close(struct adapter *adapter);
+void cxgbe_stats_get(struct port_info *pi, struct port_stats *stats);
+void cxgbe_stats_reset(struct port_info *pi);
+int link_start(struct port_info *pi);
+void init_rspq(struct adapter *adap, struct sge_rspq *q, unsigned int us,
+	       unsigned int cnt, unsigned int size, unsigned int iqe_size);
+int setup_sge_fwevtq(struct adapter *adapter);
+void cfg_queues(struct rte_eth_dev *eth_dev);
+int cfg_queue_count(struct rte_eth_dev *eth_dev);
+int setup_rss(struct port_info *pi);
+
+#endif /* _CXGBE_H_ */
diff --git a/lib/librte_pmd_cxgbe/cxgbe_compat.h b/lib/librte_pmd_cxgbe/cxgbe_compat.h
new file mode 100644
index 0000000..1526659
--- /dev/null
+++ b/lib/librte_pmd_cxgbe/cxgbe_compat.h
@@ -0,0 +1,290 @@
+/*-
+ *   BSD LICENSE
+ *
+ *   Copyright(c) 2014-2015 Chelsio Communications.
+ *   All rights reserved.
+ *
+ *   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.
+ *     * Neither the name of Chelsio Communications nor the names of its
+ *       contributors may be used to endorse or promote products derived
+ *       from this software without specific prior written permission.
+ *
+ *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#ifndef _CXGBE_COMPAT_H_
+#define _CXGBE_COMPAT_H_
+
+#include <string.h>
+#include <stdint.h>
+#include <stdio.h>
+#include <stdarg.h>
+
+#include <rte_common.h>
+#include <rte_memcpy.h>
+#include <rte_byteorder.h>
+#include <rte_cycles.h>
+#include <rte_spinlock.h>
+#include <rte_log.h>
+
+#define dev_printf(level, fmt, args...) \
+	RTE_LOG(level, PMD, "rte_cxgbe_pmd: " fmt, ## args)
+
+#define dev_err(x, args...) dev_printf(ERR, args)
+#define dev_info(x, args...) dev_printf(INFO, args)
+#define dev_warn(x, args...) dev_printf(WARNING, args)
+
+#ifdef RTE_LIBRTE_CXGBE_DEBUG
+#define dev_debug(x, args...) dev_printf(DEBUG, args)
+#else
+#define dev_debug(x, args...) do { } while (0)
+#endif
+
+#ifdef RTE_LIBRTE_CXGBE_DEBUG_REG
+#define CXGBE_DEBUG_REG(x, args...) dev_printf(DEBUG, "REG:" args)
+#else
+#define CXGBE_DEBUG_REG(x, args...) do { } while (0)
+#endif
+
+#ifdef RTE_LIBRTE_CXGBE_DEBUG_MBOX
+#define CXGBE_DEBUG_MBOX(x, args...) dev_printf(DEBUG, "MBOX:" args)
+#else
+#define CXGBE_DEBUG_MBOX(x, args...) do { } while (0)
+#endif
+
+#ifdef RTE_LIBRTE_CXGBE_DEBUG_TX
+#define CXGBE_DEBUG_TX(x, args...) dev_printf(DEBUG, "TX:" args)
+#else
+#define CXGBE_DEBUG_TX(x, args...) do { } while (0)
+#endif
+
+#ifdef RTE_LIBRTE_CXGBE_DEBUG_RX
+#define CXGBE_DEBUG_RX(x, args...) dev_printf(DEBUG, "RX:" args)
+#else
+#define CXGBE_DEBUG_RX(x, args...) do { } while (0)
+#endif
+
+#ifdef RTE_LIBRTE_CXGBE_DEBUG
+#define CXGBE_FUNC_TRACE() \
+	RTE_LOG(DEBUG, PMD, "CXGBE trace: %s\n", __func__)
+#else
+#define CXGBE_FUNC_TRACE() do { } while (0)
+#endif
+
+#define pr_err(y, args...) dev_err(0, y, ##args)
+#define pr_warn(y, args...) dev_warn(0, y, ##args)
+#define pr_info(y, args...) dev_info(0, y, ##args)
+#define BUG() pr_err("BUG at %s:%d", __func__, __LINE__)
+
+#define ASSERT(x) do {\
+	if (!(x)) \
+		rte_panic("CXGBE: x"); \
+} while (0)
+#define BUG_ON(x) ASSERT(!(x))
+
+#ifndef WARN_ON
+#define WARN_ON(x) do { \
+	int ret = !!(x); \
+	if (unlikely(ret)) \
+		pr_warn("WARN_ON: \"" #x "\" at %s:%d\n", __func__, __LINE__); \
+} while (0)
+#endif
+
+#define __iomem
+
+#ifndef BIT
+#define BIT(n) (1 << (n))
+#endif
+
+#define L1_CACHE_SHIFT  6
+#define L1_CACHE_BYTES  BIT(L1_CACHE_SHIFT)
+
+#define PAGE_SHIFT  12
+#define ALIGN(x, a) (((x) + (a) - 1) & ~((a) - 1))
+#define PTR_ALIGN(p, a) ((typeof(p))ALIGN((unsigned long)(p), (a)))
+
+#define VLAN_HLEN 4
+
+#define rmb()     rte_rmb() /* dpdk rte provided rmb */
+#define wmb()     rte_wmb() /* dpdk rte provided wmb */
+
+typedef uint8_t   u8;
+typedef int8_t    s8;
+typedef uint16_t  u16;
+typedef uint32_t  u32;
+typedef int32_t   s32;
+typedef uint64_t  u64;
+typedef int       bool;
+typedef uint64_t  dma_addr_t;
+
+#ifndef __le16
+#define __le16	uint16_t
+#endif
+#ifndef __le32
+#define __le32	uint32_t
+#endif
+#ifndef __le64
+#define __le64	uint64_t
+#endif
+#ifndef __be16
+#define __be16	uint16_t
+#endif
+#ifndef __be32
+#define __be32	uint32_t
+#endif
+#ifndef __be64
+#define __be64	uint64_t
+#endif
+#ifndef __u8
+#define __u8	uint8_t
+#endif
+#ifndef __u16
+#define __u16	uint16_t
+#endif
+#ifndef __u32
+#define __u32	uint32_t
+#endif
+#ifndef __u64
+#define __u64	uint64_t
+#endif
+
+#define FALSE	0
+#define TRUE	1
+#define false	0
+#define true	1
+
+#define min(a, b) RTE_MIN(a, b)
+#define max(a, b) RTE_MAX(a, b)
+
+/*
+ * round up val _p to a power of 2 size _s
+ */
+#define roundup(_p, _s) (((unsigned long)(_p) + (_s - 1)) & ~(_s - 1))
+
+#undef container_of
+#define container_of(ptr, type, member) ({ \
+		typeof(((type *)0)->member)(*__mptr) = (ptr); \
+		(type *)((char *)__mptr - offsetof(type, member)); })
+
+#define ARRAY_SIZE(arr) (sizeof(arr) / sizeof(arr[0]))
+
+#define cpu_to_be16(o) rte_cpu_to_be_16(o)
+#define cpu_to_be32(o) rte_cpu_to_be_32(o)
+#define cpu_to_be64(o) rte_cpu_to_be_64(o)
+#define cpu_to_le32(o) rte_cpu_to_le_32(o)
+#define be16_to_cpu(o) rte_be_to_cpu_16(o)
+#define be32_to_cpu(o) rte_be_to_cpu_32(o)
+#define be64_to_cpu(o) rte_be_to_cpu_64(o)
+#define le32_to_cpu(o) rte_le_to_cpu_32(o)
+
+#define DIV_ROUND_UP(n, d) (((n) + (d) - 1) / (d))
+#define DELAY(x) rte_delay_us(x)
+#define udelay(x) DELAY(x)
+#define msleep(x) DELAY(1000 * (x))
+#define usleep_range(min, max) msleep(DIV_ROUND_UP(min, 1000))
+
+static inline uint8_t hweight32(uint32_t word32)
+{
+	uint32_t res = word32 - ((word32 >> 1) & 0x55555555);
+
+	res = (res & 0x33333333) + ((res >> 2) & 0x33333333);
+	res = (res + (res >> 4)) & 0x0F0F0F0F;
+	res = res + (res >> 8);
+	return (res + (res >> 16)) & 0x000000FF;
+
+} /* weight32 */
+
+/**
+ * fls - find last (most-significant) bit set
+ * @x: the word to search
+ *
+ * This is defined the same way as ffs.
+ * Note fls(0) = 0, fls(1) = 1, fls(0x80000000) = 32.
+ */
+static inline int fls(int x)
+{
+	int r = 32;
+
+	if (!x)
+		return 0;
+	if (!(x & 0xffff0000u)) {
+		x <<= 16;
+		r -= 16;
+	}
+	if (!(x & 0xff000000u)) {
+		x <<= 8;
+		r -= 8;
+	}
+	if (!(x & 0xf0000000u)) {
+		x <<= 4;
+		r -= 4;
+	}
+	if (!(x & 0xc0000000u)) {
+		x <<= 2;
+		r -= 2;
+	}
+	if (!(x & 0x80000000u)) {
+		x <<= 1;
+		r -= 1;
+	}
+	return r;
+}
+
+static inline unsigned long ilog2(unsigned long n)
+{
+	unsigned int e = 0;
+
+	while (n) {
+		if (n & ~((1 << 8) - 1)) {
+			e += 8;
+			n >>= 8;
+			continue;
+		}
+
+		if (n & ~((1 << 4) - 1)) {
+			e += 4;
+			n >>= 4;
+		}
+
+		for (;;) {
+			n >>= 1;
+			if (n == 0)
+				break;
+			e++;
+		}
+	}
+
+	return e;
+}
+
+static inline void writel(unsigned int val, volatile void __iomem *addr)
+{
+	*(volatile unsigned int *)addr = val;
+}
+
+static inline void writeq(u64 val, volatile void __iomem *addr)
+{
+	writel(val, addr);
+	writel(val >> 32, (void *)((uintptr_t)addr + 4));
+}
+
+#endif /* _CXGBE_COMPAT_H_ */
diff --git a/lib/librte_pmd_cxgbe/cxgbe_ethdev.c b/lib/librte_pmd_cxgbe/cxgbe_ethdev.c
new file mode 100644
index 0000000..aa2c50e
--- /dev/null
+++ b/lib/librte_pmd_cxgbe/cxgbe_ethdev.c
@@ -0,0 +1,796 @@
+/*-
+ *   BSD LICENSE
+ *
+ *   Copyright(c) 2014-2015 Chelsio Communications.
+ *   All rights reserved.
+ *
+ *   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.
+ *     * Neither the name of Chelsio Communications nor the names of its
+ *       contributors may be used to endorse or promote products derived
+ *       from this software without specific prior written permission.
+ *
+ *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#include <sys/queue.h>
+#include <stdio.h>
+#include <errno.h>
+#include <stdint.h>
+#include <string.h>
+#include <unistd.h>
+#include <stdarg.h>
+#include <inttypes.h>
+#include <netinet/in.h>
+
+#include <rte_byteorder.h>
+#include <rte_common.h>
+#include <rte_cycles.h>
+#include <rte_interrupts.h>
+#include <rte_log.h>
+#include <rte_debug.h>
+#include <rte_pci.h>
+#include <rte_atomic.h>
+#include <rte_branch_prediction.h>
+#include <rte_memory.h>
+#include <rte_memzone.h>
+#include <rte_tailq.h>
+#include <rte_eal.h>
+#include <rte_alarm.h>
+#include <rte_ether.h>
+#include <rte_ethdev.h>
+#include <rte_atomic.h>
+#include <rte_malloc.h>
+#include <rte_random.h>
+#include <rte_dev.h>
+
+#include "cxgbe.h"
+
+/*
+ * Macros needed to support the PCI Device ID Table ...
+ */
+#define CH_PCI_DEVICE_ID_TABLE_DEFINE_BEGIN \
+	static struct rte_pci_id cxgb4_pci_tbl[] = {
+#define CH_PCI_DEVICE_ID_FUNCTION 0x4
+
+#define PCI_VENDOR_ID_CHELSIO 0x1425
+
+#define CH_PCI_ID_TABLE_ENTRY(devid) \
+		{ RTE_PCI_DEVICE(PCI_VENDOR_ID_CHELSIO, (devid)) }
+
+#define CH_PCI_DEVICE_ID_TABLE_DEFINE_END \
+		{ .vendor_id = 0, } \
+	}
+
+/*
+ *... and the PCI ID Table itself ...
+ */
+#include "t4_pci_id_tbl.h"
+
+static uint16_t cxgbe_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
+				uint16_t nb_pkts)
+{
+	struct sge_eth_txq *txq = (struct sge_eth_txq *)tx_queue;
+	uint16_t pkts_sent, pkts_remain;
+	uint16_t total_sent = 0;
+	int ret = 0;
+
+	CXGBE_DEBUG_TX(adapter, "%s: txq = %p; tx_pkts = %p; nb_pkts = %d\n",
+		       __func__, txq, tx_pkts, nb_pkts);
+
+	/* free up desc from already completed tx */
+	reclaim_completed_tx(&txq->q);
+	while (total_sent < nb_pkts) {
+		pkts_remain = nb_pkts - total_sent;
+
+		for (pkts_sent = 0; pkts_sent < pkts_remain; pkts_sent++) {
+			ret = t4_eth_xmit(txq, tx_pkts[total_sent + pkts_sent],
+					  pkts_remain - pkts_sent - 1);
+			if (ret < 0)
+				break;
+		}
+		if (!pkts_sent)
+			break;
+		total_sent += pkts_sent;
+		/* reclaim as much as possible */
+		reclaim_completed_tx(&txq->q);
+	}
+	return total_sent;
+}
+
+static uint16_t cxgbe_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
+				uint16_t nb_pkts)
+{
+	struct sge_eth_rxq *rxq = (struct sge_eth_rxq *)rx_queue;
+	unsigned int work_done;
+
+	CXGBE_DEBUG_RX(adapter, "%s: rxq->rspq.cntxt_id = %u; nb_pkts = %d\n",
+		       __func__, rxq->rspq.cntxt_id, nb_pkts);
+
+	if (cxgbe_poll(&rxq->rspq, rx_pkts, (unsigned int)nb_pkts, &work_done))
+		dev_err(adapter, "error in cxgbe poll\n");
+
+	CXGBE_DEBUG_RX(adapter, "%s: work_done = %u\n", __func__, work_done);
+	return work_done;
+}
+
+static void cxgbe_dev_info_get(struct rte_eth_dev *eth_dev,
+			       struct rte_eth_dev_info *device_info)
+{
+	struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
+	struct adapter *adapter = pi->adapter;
+	int max_queues = adapter->sge.max_ethqsets / adapter->params.nports;
+
+	device_info->min_rx_bufsize = 68; /* XXX: Smallest pkt size */
+	device_info->max_rx_pktlen = 1500; /* XXX: For now we support mtu */
+	device_info->max_rx_queues = max_queues;
+	device_info->max_tx_queues = max_queues;
+	device_info->max_mac_addrs = 1;
+	/* XXX: For now we support one MAC/port */
+	device_info->max_vfs = adapter->params.arch.vfcount;
+	device_info->max_vmdq_pools = 0; /* XXX: For now no support for VMDQ */
+
+	device_info->rx_offload_capa = DEV_RX_OFFLOAD_VLAN_STRIP |
+				       DEV_RX_OFFLOAD_IPV4_CKSUM |
+				       DEV_RX_OFFLOAD_UDP_CKSUM |
+				       DEV_RX_OFFLOAD_TCP_CKSUM;
+
+	device_info->tx_offload_capa = DEV_TX_OFFLOAD_VLAN_INSERT |
+				       DEV_TX_OFFLOAD_IPV4_CKSUM |
+				       DEV_TX_OFFLOAD_UDP_CKSUM |
+				       DEV_TX_OFFLOAD_TCP_CKSUM |
+				       DEV_TX_OFFLOAD_TCP_TSO;
+
+	device_info->reta_size = pi->rss_size;
+}
+
+static void cxgbe_dev_promiscuous_enable(struct rte_eth_dev *eth_dev)
+{
+	struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
+	struct adapter *adapter = pi->adapter;
+
+	t4_set_rxmode(adapter, adapter->mbox, pi->viid, -1,
+		      1, -1, 1, -1, false);
+}
+
+static void cxgbe_dev_promiscuous_disable(struct rte_eth_dev *eth_dev)
+{
+	struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
+	struct adapter *adapter = pi->adapter;
+
+	t4_set_rxmode(adapter, adapter->mbox, pi->viid, -1,
+		      0, -1, 1, -1, false);
+}
+
+static void cxgbe_dev_allmulticast_enable(struct rte_eth_dev *eth_dev)
+{
+	struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
+	struct adapter *adapter = pi->adapter;
+
+	/* TODO: address filters ?? */
+
+	t4_set_rxmode(adapter, adapter->mbox, pi->viid, -1,
+		      -1, 1, 1, -1, false);
+}
+
+static void cxgbe_dev_allmulticast_disable(struct rte_eth_dev *eth_dev)
+{
+	struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
+	struct adapter *adapter = pi->adapter;
+
+	/* TODO: address filters ?? */
+
+	t4_set_rxmode(adapter, adapter->mbox, pi->viid, -1,
+		      -1, 0, 1, -1, false);
+}
+
+static int cxgbe_dev_link_update(struct rte_eth_dev *eth_dev,
+				 __rte_unused int wait_to_complete)
+{
+	struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
+	struct adapter *adapter = pi->adapter;
+	struct sge *s = &adapter->sge;
+	unsigned int work_done, budget = 4;
+	int ret;
+
+	cxgbe_poll(&s->fw_evtq, NULL, budget, &work_done);
+	ret = pi->link_cfg.link_ok;
+	eth_dev->data->dev_link.link_status = pi->link_cfg.link_ok;
+	eth_dev->data->dev_link.link_duplex = ETH_LINK_FULL_DUPLEX;
+	eth_dev->data->dev_link.link_speed = pi->link_cfg.speed;
+	return ret;
+}
+
+static int cxgbe_dev_tx_queue_start(struct rte_eth_dev *eth_dev,
+				    uint16_t tx_queue_id);
+static int cxgbe_dev_rx_queue_start(struct rte_eth_dev *eth_dev,
+				    uint16_t tx_queue_id);
+static void cxgbe_dev_tx_queue_release(void *q);
+static void cxgbe_dev_rx_queue_release(void *q);
+
+/*
+ * Stop device.
+ */
+static void cxgbe_dev_close(struct rte_eth_dev *eth_dev)
+{
+	struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
+	struct adapter *adapter = pi->adapter;
+	int i, dev_down = 0;
+
+	CXGBE_FUNC_TRACE();
+
+	if (!(adapter->flags & FULL_INIT_DONE))
+		return;
+
+	cxgbe_down(pi);
+
+	/*
+	 *  We clear queues only if both tx and rx path of the port
+	 *  have been disabled
+	 */
+	t4_sge_eth_clear_queues(pi);
+
+	/*  See if all ports are down */
+	for_each_port(adapter, i) {
+		pi = adap2pinfo(adapter, i);
+		/*
+		 * Skip first port of the adapter since it will be closed
+		 * by DPDK
+		 */
+		if (i == 0)
+			continue;
+		dev_down += (pi->eth_dev->data->dev_started == 0) ? 1 : 0;
+	}
+
+	/* If rest of the ports are stopped, then free up resources */
+	if (dev_down == (adapter->params.nports - 1))
+		cxgbe_close(adapter);
+}
+
+/* Start the device.
+ * It returns 0 on success.
+ */
+static int cxgbe_dev_start(struct rte_eth_dev *eth_dev)
+{
+	struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
+	struct adapter *adapter = pi->adapter;
+	int err = 0, i;
+
+	CXGBE_FUNC_TRACE();
+
+	/*
+	 * If we don't have a connection to the firmware there's nothing we
+	 * can do.
+	 */
+	if (!(adapter->flags & FW_OK)) {
+		err = -ENXIO;
+		goto out;
+	}
+
+	if (!(adapter->flags & FULL_INIT_DONE)) {
+		err = cxgbe_up(adapter);
+		if (err < 0)
+			goto out;
+	}
+
+	err = setup_rss(pi);
+	if (err)
+		goto out;
+
+	for (i = 0; i < pi->n_tx_qsets; i++) {
+		err = cxgbe_dev_tx_queue_start(eth_dev, i);
+		if (err)
+			goto out;
+	}
+
+	for (i = 0; i < pi->n_rx_qsets; i++) {
+		err = cxgbe_dev_rx_queue_start(eth_dev, i);
+		if (err)
+			goto out;
+	}
+
+	err = link_start(pi);
+	if (err)
+		goto out;
+
+out:
+	return err;
+}
+
+/*
+ * Stop device: disable rx and tx functions to allow for reconfiguring.
+ */
+static void cxgbe_dev_stop(struct rte_eth_dev *eth_dev)
+{
+	struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
+	struct adapter *adapter = pi->adapter;
+
+	CXGBE_FUNC_TRACE();
+
+	if (!(adapter->flags & FULL_INIT_DONE))
+		return;
+
+	cxgbe_down(pi);
+
+	/*
+	 *  We clear queues only if both tx and rx path of the port
+	 *  have been disabled
+	 */
+	t4_sge_eth_clear_queues(pi);
+}
+
+static int cxgbe_dev_configure(struct rte_eth_dev *eth_dev)
+{
+	struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
+	struct adapter *adapter = pi->adapter;
+	int err;
+
+	CXGBE_FUNC_TRACE();
+
+	if (!(adapter->flags & FW_QUEUE_BOUND)) {
+		err = setup_sge_fwevtq(adapter);
+		if (err)
+			return err;
+		adapter->flags |= FW_QUEUE_BOUND;
+	}
+
+	err = cfg_queue_count(eth_dev);
+	if (err)
+		return err;
+
+	return 0;
+}
+
+static int cxgbe_dev_tx_queue_start(struct rte_eth_dev *eth_dev,
+				    uint16_t tx_queue_id)
+{
+	struct sge_eth_txq *txq = (struct sge_eth_txq *)
+				  (eth_dev->data->tx_queues[tx_queue_id]);
+
+	dev_debug(NULL, "%s: tx_queue_id = %d\n", __func__, tx_queue_id);
+
+	return t4_sge_eth_txq_start(txq);
+}
+
+static int cxgbe_dev_tx_queue_stop(struct rte_eth_dev *eth_dev,
+				   uint16_t tx_queue_id)
+{
+	struct sge_eth_txq *txq = (struct sge_eth_txq *)
+				  (eth_dev->data->tx_queues[tx_queue_id]);
+
+	dev_debug(NULL, "%s: tx_queue_id = %d\n", __func__, tx_queue_id);
+
+	return t4_sge_eth_txq_stop(txq);
+}
+
+static int cxgbe_dev_tx_queue_setup(struct rte_eth_dev *eth_dev,
+				    uint16_t queue_idx,	uint16_t nb_desc,
+				    unsigned int socket_id,
+				    const struct rte_eth_txconf *tx_conf)
+{
+	struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
+	struct adapter *adapter = pi->adapter;
+	struct sge *s = &adapter->sge;
+	struct sge_eth_txq *txq = &s->ethtxq[pi->first_qset + queue_idx];
+	int err = 0;
+	unsigned int temp_nb_desc;
+
+	RTE_SET_USED(tx_conf);
+
+	dev_debug(adapter, "%s: eth_dev->data->nb_tx_queues = %d; queue_idx = %d; nb_desc = %d; socket_id = %d; pi->first_qset = %u\n",
+		  __func__, eth_dev->data->nb_tx_queues, queue_idx, nb_desc,
+		  socket_id, pi->first_qset);
+
+	/*  Free up the existing queue  */
+	if (eth_dev->data->tx_queues[queue_idx]) {
+		cxgbe_dev_tx_queue_release(eth_dev->data->tx_queues[queue_idx]);
+		eth_dev->data->tx_queues[queue_idx] = NULL;
+	}
+
+	eth_dev->data->tx_queues[queue_idx] = (void *)txq;
+
+	/* Sanity Checking
+	 *
+	 * nb_desc should be > 1023 and <= CXGBE_MAX_RING_DESC_SIZE
+	 */
+	temp_nb_desc = nb_desc;
+	if (nb_desc < CXGBE_MIN_RING_DESC_SIZE) {
+		dev_warn(adapter, "%s: number of descriptors must be >= %d. Using default [%d]\n",
+			 __func__, CXGBE_MIN_RING_DESC_SIZE,
+			 CXGBE_DEFAULT_TX_DESC_SIZE);
+		temp_nb_desc = CXGBE_DEFAULT_TX_DESC_SIZE;
+	} else if (nb_desc > CXGBE_MAX_RING_DESC_SIZE) {
+		dev_err(adapter, "%s: number of descriptors must be between %d and %d inclusive. Default [%d]\n",
+			__func__, CXGBE_MIN_RING_DESC_SIZE,
+			CXGBE_MAX_RING_DESC_SIZE, CXGBE_DEFAULT_TX_DESC_SIZE);
+		return -(EINVAL);
+	}
+
+	txq->q.size = temp_nb_desc;
+
+	err = t4_sge_alloc_eth_txq(adapter, txq, eth_dev, queue_idx,
+				   s->fw_evtq.cntxt_id, socket_id);
+
+	dev_debug(adapter, "%s: txq->q.cntxt_id= %d err = %d\n",
+		  __func__, txq->q.cntxt_id, err);
+
+	return err;
+}
+
+static void cxgbe_dev_tx_queue_release(void *q)
+{
+	struct sge_eth_txq *txq = (struct sge_eth_txq *)q;
+
+	if (txq) {
+		struct port_info *pi = (struct port_info *)
+				       (txq->eth_dev->data->dev_private);
+		struct adapter *adap = pi->adapter;
+
+		dev_debug(adapter, "%s: pi->port_id = %d; tx_queue_id = %d\n",
+			  __func__, pi->port_id, txq->q.cntxt_id);
+
+		t4_sge_eth_txq_release(adap, txq);
+	}
+}
+
+static int cxgbe_dev_rx_queue_start(struct rte_eth_dev *eth_dev,
+				    uint16_t rx_queue_id)
+{
+	struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
+	struct adapter *adap = pi->adapter;
+	struct sge_rspq *q;
+
+	dev_debug(adapter, "%s: pi->port_id = %d; rx_queue_id = %d\n",
+		  __func__, pi->port_id, rx_queue_id);
+
+	q = eth_dev->data->rx_queues[rx_queue_id];
+	return t4_sge_eth_rxq_start(adap, q);
+}
+
+static int cxgbe_dev_rx_queue_stop(struct rte_eth_dev *eth_dev,
+				   uint16_t rx_queue_id)
+{
+	struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
+	struct adapter *adap = pi->adapter;
+	struct sge_rspq *q;
+
+	dev_debug(adapter, "%s: pi->port_id = %d; rx_queue_id = %d\n",
+		  __func__, pi->port_id, rx_queue_id);
+
+	q = eth_dev->data->rx_queues[rx_queue_id];
+	return t4_sge_eth_rxq_stop(adap, q);
+}
+
+static int cxgbe_dev_rx_queue_setup(struct rte_eth_dev *eth_dev,
+				    uint16_t queue_idx,	uint16_t nb_desc,
+				    unsigned int socket_id,
+				    const struct rte_eth_rxconf *rx_conf,
+				    struct rte_mempool *mp)
+{
+	struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
+	struct adapter *adapter = pi->adapter;
+	struct sge *s = &adapter->sge;
+	struct sge_eth_rxq *rxq = &s->ethrxq[pi->first_qset + queue_idx];
+	int err = 0;
+	int msi_idx = 0;
+	unsigned int temp_nb_desc;
+
+	RTE_SET_USED(rx_conf);
+
+	dev_debug(adapter, "%s: eth_dev->data->nb_rx_queues = %d; queue_idx = %d; nb_desc = %d; socket_id = %d; mp = %p\n",
+		  __func__, eth_dev->data->nb_rx_queues, queue_idx, nb_desc,
+		  socket_id, mp);
+
+	/*  Free up the existing queue  */
+	if (eth_dev->data->rx_queues[queue_idx]) {
+		cxgbe_dev_rx_queue_release(eth_dev->data->rx_queues[queue_idx]);
+		eth_dev->data->rx_queues[queue_idx] = NULL;
+	}
+
+	eth_dev->data->rx_queues[queue_idx] = (void *)rxq;
+
+	/* Sanity Checking
+	 *
+	 * nb_desc should be > 0 and <= CXGBE_MAX_RING_DESC_SIZE
+	 */
+	temp_nb_desc = nb_desc;
+	if (nb_desc < CXGBE_MIN_RING_DESC_SIZE) {
+		dev_warn(adapter, "%s: number of descriptors must be >= %d. Using default [%d]\n",
+			 __func__, CXGBE_MIN_RING_DESC_SIZE,
+			 CXGBE_DEFAULT_RX_DESC_SIZE);
+		temp_nb_desc = CXGBE_DEFAULT_RX_DESC_SIZE;
+	} else if (nb_desc > CXGBE_MAX_RING_DESC_SIZE) {
+		dev_err(adapter, "%s: number of descriptors must be between %d and %d inclusive. Default [%d]\n",
+			__func__, CXGBE_MIN_RING_DESC_SIZE,
+			CXGBE_MAX_RING_DESC_SIZE, CXGBE_DEFAULT_RX_DESC_SIZE);
+		return -(EINVAL);
+	}
+
+	rxq->rspq.size = temp_nb_desc;
+	if ((&rxq->fl) != NULL)
+		rxq->fl.size = temp_nb_desc;
+
+	err = t4_sge_alloc_rxq(adapter, &rxq->rspq, false, eth_dev, msi_idx,
+			       &rxq->fl, t4_ethrx_handler,
+			       t4_get_mps_bg_map(adapter, pi->tx_chan), mp,
+			       queue_idx, socket_id);
+
+	dev_debug(adapter, "%s: err = %d; port_id = %d; cntxt_id = %u\n",
+		  __func__, err, pi->port_id, rxq->rspq.cntxt_id);
+	return err;
+}
+
+static void cxgbe_dev_rx_queue_release(void *q)
+{
+	struct sge_eth_rxq *rxq = (struct sge_eth_rxq *)q;
+	struct sge_rspq *rq = &rxq->rspq;
+
+	if (rq) {
+		struct port_info *pi = (struct port_info *)
+				       (rq->eth_dev->data->dev_private);
+		struct adapter *adap = pi->adapter;
+
+		dev_debug(adapter, "%s: pi->port_id = %d; rx_queue_id = %d\n",
+			  __func__, pi->port_id, rxq->rspq.cntxt_id);
+
+		t4_sge_eth_rxq_release(adap, rxq);
+	}
+}
+
+/*
+ * Get port statistics.
+ */
+static void cxgbe_dev_stats_get(struct rte_eth_dev *eth_dev,
+				struct rte_eth_stats *eth_stats)
+{
+	struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
+	struct adapter *adapter = pi->adapter;
+	struct sge *s = &adapter->sge;
+	struct port_stats ps;
+	unsigned int i;
+
+	cxgbe_stats_get(pi, &ps);
+
+	/* RX Stats */
+	eth_stats->ipackets = ps.rx_frames;
+	eth_stats->ibytes   = ps.rx_octets;
+	eth_stats->imcasts  = ps.rx_mcast_frames;
+	eth_stats->imissed  = ps.rx_ovflow0 + ps.rx_ovflow1 +
+			      ps.rx_ovflow2 + ps.rx_ovflow3 +
+			      ps.rx_trunc0 + ps.rx_trunc1 +
+			      ps.rx_trunc2 + ps.rx_trunc3;
+	eth_stats->ibadcrc  = ps.rx_fcs_err;
+	eth_stats->ibadlen  = ps.rx_jabber + ps.rx_too_long + ps.rx_runt;
+	eth_stats->ierrors  = ps.rx_symbol_err + eth_stats->ibadcrc +
+			      eth_stats->ibadlen + ps.rx_len_err +
+			      eth_stats->imissed;
+	eth_stats->rx_pause_xon  = ps.rx_pause;
+
+	/* TX Stats */
+	eth_stats->opackets = ps.tx_frames;
+	eth_stats->obytes   = ps.tx_octets;
+	eth_stats->oerrors  = ps.tx_error_frames;
+	eth_stats->tx_pause_xon  = ps.tx_pause;
+
+	for (i = 0; i < pi->n_rx_qsets; i++) {
+		struct sge_eth_rxq *rxq =
+			&s->ethrxq[pi->first_qset + i];
+
+		eth_stats->q_ipackets[i] = rxq->stats.pkts;
+		eth_stats->q_ibytes[i] = rxq->stats.rx_bytes;
+	}
+
+	for (i = 0; i < pi->n_tx_qsets; i++) {
+		struct sge_eth_txq *txq =
+			&s->ethtxq[pi->first_qset + i];
+
+		eth_stats->q_opackets[i] = txq->stats.pkts;
+		eth_stats->q_obytes[i] = txq->stats.tx_bytes;
+		eth_stats->q_errors[i] = txq->stats.mapping_err;
+	}
+}
+
+/*
+ * Reset port statistics.
+ */
+static void cxgbe_dev_stats_reset(struct rte_eth_dev *eth_dev)
+{
+	struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
+	struct adapter *adapter = pi->adapter;
+	struct sge *s = &adapter->sge;
+	unsigned int i;
+
+	cxgbe_stats_reset(pi);
+	for (i = 0; i < pi->n_rx_qsets; i++) {
+		struct sge_eth_rxq *rxq =
+			&s->ethrxq[pi->first_qset + i];
+
+		rxq->stats.pkts = 0;
+		rxq->stats.rx_bytes = 0;
+	}
+	for (i = 0; i < pi->n_tx_qsets; i++) {
+		struct sge_eth_txq *txq =
+			&s->ethtxq[pi->first_qset + i];
+
+		txq->stats.pkts = 0;
+		txq->stats.tx_bytes = 0;
+		txq->stats.mapping_err = 0;
+	}
+}
+
+static int cxgbe_flow_ctrl_get(struct rte_eth_dev *eth_dev,
+			       struct rte_eth_fc_conf *fc_conf)
+{
+	struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
+	struct link_config *lc = &pi->link_cfg;
+	int rx_pause, tx_pause;
+
+	fc_conf->autoneg = lc->fc & PAUSE_AUTONEG;
+	rx_pause = lc->fc & PAUSE_RX;
+	tx_pause = lc->fc & PAUSE_TX;
+
+	if (rx_pause && tx_pause)
+		fc_conf->mode = RTE_FC_FULL;
+	else if (rx_pause)
+		fc_conf->mode = RTE_FC_RX_PAUSE;
+	else if (tx_pause)
+		fc_conf->mode = RTE_FC_TX_PAUSE;
+	else
+		fc_conf->mode = RTE_FC_NONE;
+	return 0;
+}
+
+static int cxgbe_flow_ctrl_set(struct rte_eth_dev *eth_dev,
+			       struct rte_eth_fc_conf *fc_conf)
+{
+	struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
+	struct adapter *adapter = pi->adapter;
+	struct link_config *lc = &pi->link_cfg;
+
+	if (lc->supported & FW_PORT_CAP_ANEG) {
+		if (fc_conf->autoneg)
+			lc->requested_fc |= PAUSE_AUTONEG;
+		else
+			lc->requested_fc &= ~PAUSE_AUTONEG;
+	}
+
+	if (((fc_conf->mode & RTE_FC_FULL) == RTE_FC_FULL) ||
+	    (fc_conf->mode & RTE_FC_RX_PAUSE))
+		lc->requested_fc |= PAUSE_RX;
+	else
+		lc->requested_fc &= ~PAUSE_RX;
+
+	if (((fc_conf->mode & RTE_FC_FULL) == RTE_FC_FULL) ||
+	    (fc_conf->mode & RTE_FC_TX_PAUSE))
+		lc->requested_fc |= PAUSE_TX;
+	else
+		lc->requested_fc &= ~PAUSE_TX;
+
+	return t4_link_l1cfg(adapter, adapter->mbox, pi->tx_chan,
+			     &pi->link_cfg);
+}
+
+static struct eth_dev_ops cxgbe_eth_dev_ops = {
+	.dev_start		= cxgbe_dev_start,
+	.dev_stop		= cxgbe_dev_stop,
+	.dev_close		= cxgbe_dev_close,
+	.promiscuous_enable	= cxgbe_dev_promiscuous_enable,
+	.promiscuous_disable	= cxgbe_dev_promiscuous_disable,
+	.allmulticast_enable	= cxgbe_dev_allmulticast_enable,
+	.allmulticast_disable	= cxgbe_dev_allmulticast_disable,
+	.dev_configure		= cxgbe_dev_configure,
+	.dev_infos_get		= cxgbe_dev_info_get,
+	.link_update		= cxgbe_dev_link_update,
+	.tx_queue_setup         = cxgbe_dev_tx_queue_setup,
+	.tx_queue_start		= cxgbe_dev_tx_queue_start,
+	.tx_queue_stop		= cxgbe_dev_tx_queue_stop,
+	.tx_queue_release	= cxgbe_dev_tx_queue_release,
+	.rx_queue_setup         = cxgbe_dev_rx_queue_setup,
+	.rx_queue_start		= cxgbe_dev_rx_queue_start,
+	.rx_queue_stop		= cxgbe_dev_rx_queue_stop,
+	.rx_queue_release	= cxgbe_dev_rx_queue_release,
+	.stats_get		= cxgbe_dev_stats_get,
+	.stats_reset		= cxgbe_dev_stats_reset,
+	.flow_ctrl_get		= cxgbe_flow_ctrl_get,
+	.flow_ctrl_set		= cxgbe_flow_ctrl_set,
+};
+
+/*
+ * Initialize driver
+ * It returns 0 on success.
+ */
+static int eth_cxgbe_dev_init(struct rte_eth_dev *eth_dev)
+{
+	struct rte_pci_device *pci_dev;
+	struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
+	struct adapter *adapter = NULL;
+	char name[RTE_ETH_NAME_MAX_LEN];
+	int err = 0;
+
+	CXGBE_FUNC_TRACE();
+
+	eth_dev->dev_ops = &cxgbe_eth_dev_ops;
+	eth_dev->rx_pkt_burst = &cxgbe_recv_pkts;
+	eth_dev->tx_pkt_burst = &cxgbe_xmit_pkts;
+
+	/* for secondary processes, we don't initialise any further as primary
+	 * has already done this work.
+	 */
+	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
+		return 0;
+
+	pci_dev = eth_dev->pci_dev;
+	snprintf(name, sizeof(name), "cxgbeadapter%d", eth_dev->data->port_id);
+	adapter = rte_zmalloc(name, sizeof(*adapter), 0);
+	if (!adapter)
+		return -1;
+
+	adapter->use_unpacked_mode = 1;
+	adapter->regs = (void *)pci_dev->mem_resource[0].addr;
+	if (!adapter->regs) {
+		dev_err(adapter, "%s: cannot map device registers\n", __func__);
+		err = -ENOMEM;
+		goto out_free_adapter;
+	}
+	adapter->pdev = pci_dev;
+	adapter->eth_dev = eth_dev;
+	pi->adapter = adapter;
+
+	err = cxgbe_probe(adapter);
+	if (err)
+		dev_err(adapter, "%s: cxgbe probe failed with err %d\n",
+			__func__, err);
+
+out_free_adapter:
+	return err;
+}
+
+static struct eth_driver rte_cxgbe_pmd = {
+	{
+		.name = "rte_cxgbe_pmd",
+		.id_table = cxgb4_pci_tbl,
+		.drv_flags = RTE_PCI_DRV_NEED_MAPPING | RTE_PCI_DRV_INTR_LSC,
+	},
+	.eth_dev_init = eth_cxgbe_dev_init,
+	.dev_private_size = sizeof(struct port_info),
+};
+
+/*
+ * Driver initialization routine.
+ * Invoked once at EAL init time.
+ * Register itself as the [Poll Mode] Driver of PCI CXGBE devices.
+ */
+static int rte_cxgbe_pmd_init(const char *name __rte_unused,
+			      const char *params __rte_unused)
+{
+	CXGBE_FUNC_TRACE();
+
+	rte_eth_driver_register(&rte_cxgbe_pmd);
+	return 0;
+}
+
+static struct rte_driver rte_cxgbe_driver = {
+	.name = "cxgbe_driver",
+	.type = PMD_PDEV,
+	.init = rte_cxgbe_pmd_init,
+};
+
+PMD_REGISTER_DRIVER(rte_cxgbe_driver);
diff --git a/lib/librte_pmd_cxgbe/cxgbe_main.c b/lib/librte_pmd_cxgbe/cxgbe_main.c
new file mode 100644
index 0000000..b39a798
--- /dev/null
+++ b/lib/librte_pmd_cxgbe/cxgbe_main.c
@@ -0,0 +1,1219 @@
+/*-
+ *   BSD LICENSE
+ *
+ *   Copyright(c) 2014-2015 Chelsio Communications.
+ *   All rights reserved.
+ *
+ *   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.
+ *     * Neither the name of Chelsio Communications nor the names of its
+ *       contributors may be used to endorse or promote products derived
+ *       from this software without specific prior written permission.
+ *
+ *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#include <sys/queue.h>
+#include <stdio.h>
+#include <errno.h>
+#include <stdint.h>
+#include <string.h>
+#include <unistd.h>
+#include <stdarg.h>
+#include <inttypes.h>
+#include <netinet/in.h>
+
+#include <rte_byteorder.h>
+#include <rte_common.h>
+#include <rte_cycles.h>
+#include <rte_interrupts.h>
+#include <rte_log.h>
+#include <rte_debug.h>
+#include <rte_pci.h>
+#include <rte_atomic.h>
+#include <rte_branch_prediction.h>
+#include <rte_memory.h>
+#include <rte_memzone.h>
+#include <rte_tailq.h>
+#include <rte_eal.h>
+#include <rte_alarm.h>
+#include <rte_ether.h>
+#include <rte_ethdev.h>
+#include <rte_atomic.h>
+#include <rte_malloc.h>
+#include <rte_random.h>
+#include <rte_dev.h>
+
+#include "common.h"
+#include "t4_regs.h"
+#include "t4_msg.h"
+#include "cxgbe.h"
+
+/*
+ * Response queue handler for the FW event queue.
+ */
+static int fwevtq_handler(struct sge_rspq *q, const __be64 *rsp,
+			  __rte_unused const struct pkt_gl *gl)
+{
+	u8 opcode = ((const struct rss_header *)rsp)->opcode;
+
+	rsp++;                                          /* skip RSS header */
+
+	/*
+	 * FW can send EGR_UPDATEs encapsulated in a CPL_FW4_MSG.
+	 */
+	if (unlikely(opcode == CPL_FW4_MSG &&
+		     ((const struct cpl_fw4_msg *)rsp)->type ==
+		      FW_TYPE_RSSCPL)) {
+		rsp++;
+		opcode = ((const struct rss_header *)rsp)->opcode;
+		rsp++;
+		if (opcode != CPL_SGE_EGR_UPDATE) {
+			dev_err(q->adapter, "unexpected FW4/CPL %#x on FW event queue\n",
+				opcode);
+			goto out;
+		}
+	}
+
+	if (likely(opcode == CPL_SGE_EGR_UPDATE)) {
+		/* do nothing */
+	} else if (opcode == CPL_FW6_MSG || opcode == CPL_FW4_MSG) {
+		const struct cpl_fw6_msg *msg = (const void *)rsp;
+
+		t4_handle_fw_rpl(q->adapter, msg->data);
+	} else {
+		dev_err(adapter, "unexpected CPL %#x on FW event queue\n",
+			opcode);
+	}
+out:
+	return 0;
+}
+
+int setup_sge_fwevtq(struct adapter *adapter)
+{
+	struct sge *s = &adapter->sge;
+	int err = 0;
+	int msi_idx = 0;
+
+	err = t4_sge_alloc_rxq(adapter, &s->fw_evtq, true, adapter->eth_dev,
+			       msi_idx, NULL, fwevtq_handler, -1, NULL, 0,
+			       rte_socket_id());
+	return err;
+}
+
+static int closest_timer(const struct sge *s, int time)
+{
+	unsigned int i, match = 0;
+	int delta, min_delta = INT_MAX;
+
+	for (i = 0; i < ARRAY_SIZE(s->timer_val); i++) {
+		delta = time - s->timer_val[i];
+		if (delta < 0)
+			delta = -delta;
+		if (delta < min_delta) {
+			min_delta = delta;
+			match = i;
+		}
+	}
+	return match;
+}
+
+static int closest_thres(const struct sge *s, int thres)
+{
+	unsigned int i, match = 0;
+	int delta, min_delta = INT_MAX;
+
+	for (i = 0; i < ARRAY_SIZE(s->counter_val); i++) {
+		delta = thres - s->counter_val[i];
+		if (delta < 0)
+			delta = -delta;
+		if (delta < min_delta) {
+			min_delta = delta;
+			match = i;
+		}
+	}
+	return match;
+}
+
+/**
+ * cxgb4_set_rspq_intr_params - set a queue's interrupt holdoff parameters
+ * @q: the Rx queue
+ * @us: the hold-off time in us, or 0 to disable timer
+ * @cnt: the hold-off packet count, or 0 to disable counter
+ *
+ * Sets an Rx queue's interrupt hold-off time and packet count.  At least
+ * one of the two needs to be enabled for the queue to generate interrupts.
+ */
+int cxgb4_set_rspq_intr_params(struct sge_rspq *q, unsigned int us,
+			       unsigned int cnt)
+{
+	struct adapter *adap = q->adapter;
+	unsigned int timer_val;
+
+	if (cnt) {
+		int err;
+		u32 v, new_idx;
+
+		new_idx = closest_thres(&adap->sge, cnt);
+		if (q->desc && q->pktcnt_idx != new_idx) {
+			/* the queue has already been created, update it */
+			v = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DMAQ) |
+			    V_FW_PARAMS_PARAM_X(
+			    FW_PARAMS_PARAM_DMAQ_IQ_INTCNTTHRESH) |
+			    V_FW_PARAMS_PARAM_YZ(q->cntxt_id);
+			err = t4_set_params(adap, adap->mbox, adap->pf, 0, 1,
+					    &v, &new_idx);
+			if (err)
+				return err;
+		}
+		q->pktcnt_idx = new_idx;
+	}
+
+	timer_val = (us == 0) ? X_TIMERREG_RESTART_COUNTER :
+				closest_timer(&adap->sge, us);
+
+	if ((us | cnt) == 0)
+		q->intr_params = V_QINTR_TIMER_IDX(X_TIMERREG_UPDATE_CIDX);
+	else
+		q->intr_params = V_QINTR_TIMER_IDX(timer_val) |
+				 V_QINTR_CNT_EN(cnt > 0);
+	return 0;
+}
+
+static inline bool is_x_1g_port(const struct link_config *lc)
+{
+	return ((lc->supported & FW_PORT_CAP_SPEED_1G) != 0);
+}
+
+static inline bool is_x_10g_port(const struct link_config *lc)
+{
+	return ((lc->supported & FW_PORT_CAP_SPEED_10G) != 0 ||
+		(lc->supported & FW_PORT_CAP_SPEED_40G) != 0 ||
+		(lc->supported & FW_PORT_CAP_SPEED_100G) != 0);
+}
+
+inline void init_rspq(struct adapter *adap, struct sge_rspq *q,
+		      unsigned int us, unsigned int cnt,
+		      unsigned int size, unsigned int iqe_size)
+{
+	q->adapter = adap;
+	cxgb4_set_rspq_intr_params(q, us, cnt);
+	q->iqe_len = iqe_size;
+	q->size = size;
+}
+
+int cfg_queue_count(struct rte_eth_dev *eth_dev)
+{
+	struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
+	struct adapter *adap = pi->adapter;
+	struct sge *s = &adap->sge;
+	unsigned int max_queues = s->max_ethqsets / adap->params.nports;
+
+	if ((eth_dev->data->nb_rx_queues < 1) ||
+	    (eth_dev->data->nb_tx_queues < 1))
+		return -EINVAL;
+
+	if ((eth_dev->data->nb_rx_queues > max_queues) ||
+	    (eth_dev->data->nb_tx_queues > max_queues))
+		return -EINVAL;
+
+	if (eth_dev->data->nb_rx_queues > pi->rss_size)
+		return -EINVAL;
+
+	/* We must configure RSS, since config has changed*/
+	pi->flags &= ~PORT_RSS_DONE;
+
+	pi->n_rx_qsets = eth_dev->data->nb_rx_queues;
+	pi->n_tx_qsets = eth_dev->data->nb_tx_queues;
+
+	return 0;
+}
+
+void cfg_queues(struct rte_eth_dev *eth_dev)
+{
+	struct rte_config *config = rte_eal_get_configuration();
+	struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
+	struct adapter *adap = pi->adapter;
+	struct sge *s = &adap->sge;
+	unsigned int i, nb_ports = 0, qidx = 0;
+	unsigned int q_per_port = 0;
+
+	if (!(adap->flags & CFG_QUEUES)) {
+		for_each_port(adap, i) {
+			struct port_info *tpi = adap2pinfo(adap, i);
+
+			nb_ports += (is_x_10g_port(&tpi->link_cfg)) ||
+				     is_x_1g_port(&tpi->link_cfg) ? 1 : 0;
+		}
+
+		/*
+		 * We default up to # of cores queues per 1G/10G port.
+		 */
+		if (nb_ports)
+			q_per_port = (MAX_ETH_QSETS -
+				     (adap->params.nports - nb_ports)) /
+				     nb_ports;
+
+		if (q_per_port > config->lcore_count)
+			q_per_port = config->lcore_count;
+
+		for_each_port(adap, i) {
+			struct port_info *pi = adap2pinfo(adap, i);
+
+			pi->first_qset = qidx;
+
+			/* Initially n_rx_qsets == n_tx_qsets */
+			pi->n_rx_qsets = (is_x_10g_port(&pi->link_cfg) ||
+					  is_x_1g_port(&pi->link_cfg)) ?
+					  q_per_port : 1;
+			pi->n_tx_qsets = pi->n_rx_qsets;
+
+			if (pi->n_rx_qsets > pi->rss_size)
+				pi->n_rx_qsets = pi->rss_size;
+
+			qidx += pi->n_rx_qsets;
+		}
+
+		s->max_ethqsets = qidx;
+
+		for (i = 0; i < ARRAY_SIZE(s->ethrxq); i++) {
+			struct sge_eth_rxq *r = &s->ethrxq[i];
+
+			init_rspq(adap, &r->rspq, 0, 0, 1024, 64);
+			r->usembufs = 1;
+			r->fl.size = (r->usembufs ? 1024 : 72);
+		}
+
+		for (i = 0; i < ARRAY_SIZE(s->ethtxq); i++)
+			s->ethtxq[i].q.size = 1024;
+
+		init_rspq(adap, &adap->sge.fw_evtq, 0, 0, 1024, 64);
+		adap->flags |= CFG_QUEUES;
+	}
+}
+
+void cxgbe_stats_get(struct port_info *pi, struct port_stats *stats)
+{
+	t4_get_port_stats_offset(pi->adapter, pi->tx_chan, stats,
+				 &pi->stats_base);
+}
+
+void cxgbe_stats_reset(struct port_info *pi)
+{
+	t4_clr_port_stats(pi->adapter, pi->tx_chan);
+}
+
+static void setup_memwin(struct adapter *adap)
+{
+	u32 mem_win0_base;
+
+	/* For T5, only relative offset inside the PCIe BAR is passed */
+	mem_win0_base = MEMWIN0_BASE;
+
+	/*
+	 * Set up memory window for accessing adapter memory ranges.  (Read
+	 * back MA register to ensure that changes propagate before we attempt
+	 * to use the new values.)
+	 */
+	t4_write_reg(adap,
+		     PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN,
+					 MEMWIN_NIC),
+		     mem_win0_base | V_BIR(0) |
+		     V_WINDOW(ilog2(MEMWIN0_APERTURE) - X_WINDOW_SHIFT));
+	t4_read_reg(adap,
+		    PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN,
+					MEMWIN_NIC));
+}
+
+static int init_rss(struct adapter *adap)
+{
+	unsigned int i;
+	int err;
+
+	err = t4_init_rss_mode(adap, adap->mbox);
+	if (err)
+		return err;
+
+	for_each_port(adap, i) {
+		struct port_info *pi = adap2pinfo(adap, i);
+
+		pi->rss = rte_zmalloc(NULL, pi->rss_size, 0);
+		if (!pi->rss)
+			return -ENOMEM;
+	}
+	return 0;
+}
+
+static void print_port_info(struct adapter *adap)
+{
+	int i;
+	char buf[80];
+	struct rte_pci_addr *loc = &adap->pdev->addr;
+
+	for_each_port(adap, i) {
+		const struct port_info *pi = &adap->port[i];
+		char *bufp = buf;
+
+		if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_100M)
+			bufp += sprintf(bufp, "100/");
+		if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_1G)
+			bufp += sprintf(bufp, "1000/");
+		if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_10G)
+			bufp += sprintf(bufp, "10G/");
+		if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_40G)
+			bufp += sprintf(bufp, "40G/");
+		if (bufp != buf)
+			--bufp;
+		sprintf(bufp, "BASE-%s",
+			t4_get_port_type_description(pi->port_type));
+
+		dev_info(adap,
+			 " " PCI_PRI_FMT " Chelsio rev %d %s %s\n",
+			 loc->domain, loc->bus, loc->devid, loc->function,
+			 CHELSIO_CHIP_RELEASE(adap->params.chip), buf,
+			 (adap->flags & USING_MSIX) ? " MSI-X" :
+			 (adap->flags & USING_MSI) ? " MSI" : "");
+	}
+}
+
+/*
+ * Tweak configuration based on module parameters, etc.  Most of these have
+ * defaults assigned to them by Firmware Configuration Files (if we're using
+ * them) but need to be explicitly set if we're using hard-coded
+ * initialization.  But even in the case of using Firmware Configuration
+ * Files, we'd like to expose the ability to change these via module
+ * parameters so these are essentially common tweaks/settings for
+ * Configuration Files and hard-coded initialization ...
+ */
+static int adap_init0_tweaks(struct adapter *adapter)
+{
+	u8 rx_dma_offset;
+
+	/*
+	 * Fix up various Host-Dependent Parameters like Page Size, Cache
+	 * Line Size, etc.  The firmware default is for a 4KB Page Size and
+	 * 64B Cache Line Size ...
+	 */
+	t4_fixup_host_params_compat(adapter, PAGE_SIZE, L1_CACHE_BYTES,
+				    T5_LAST_REV);
+
+	/*
+	 * Normally we tell the chip to deliver Ingress Packets into our DMA
+	 * buffers offset by 2 bytes in order to have the IP headers line up on
+	 * 4-byte boundaries.  This is a requirement for many architectures
+	 * which will throw a machine check fault if an attempt is made to
+	 * access one of the 4-byte IP header fields on a non-4-byte boundary.
+	 * And it's a major performance issue even on some architectures whichi
+	 * allow it like some implementations of the x86 ISA.  However, some
+	 * architectures don't mind this and for some very edge-case performance
+	 * sensitive applications (like forwarding large volumes of small
+	 * packets), setting this DMA offset to 0 will decrease the number of
+	 * PCI-E Bus transfers enough to measurably affect performance.
+	 */
+	rx_dma_offset = 2;
+	t4_set_reg_field(adapter, A_SGE_CONTROL, V_PKTSHIFT(M_PKTSHIFT),
+			 V_PKTSHIFT(rx_dma_offset));
+
+	/*
+	 * Don't include the "IP Pseudo Header" in CPL_RX_PKT checksums: Linux
+	 * adds the pseudo header itself.
+	 */
+	t4_tp_wr_bits_indirect(adapter, A_TP_INGRESS_CONFIG,
+			       F_CSUM_HAS_PSEUDO_HDR, 0);
+
+	return 0;
+}
+
+/*
+ * Attempt to initialize the adapter via a Firmware Configuration File.
+ */
+static int adap_init0_config(struct adapter *adapter, int reset)
+{
+	struct fw_caps_config_cmd caps_cmd;
+	unsigned long mtype = 0, maddr = 0;
+	u32 finiver, finicsum, cfcsum;
+	int ret;
+	int config_issued = 0;
+	int cfg_addr;
+	char config_name[20];
+
+	/*
+	 * Reset device if necessary.
+	 */
+	if (reset) {
+		ret = t4_fw_reset(adapter, adapter->mbox,
+				  F_PIORSTMODE | F_PIORST);
+		if (ret < 0) {
+			dev_warn(adapter, "Firmware reset failed, error %d\n",
+				 -ret);
+			goto bye;
+		}
+	}
+
+	cfg_addr = t4_flash_cfg_addr(adapter);
+	if (cfg_addr < 0) {
+		ret = cfg_addr;
+		dev_warn(adapter, "Finding address for firmware config file in flash failed, error %d\n",
+			 -ret);
+		goto bye;
+	}
+
+	strcpy(config_name, "On Flash");
+	mtype = FW_MEMTYPE_CF_FLASH;
+	maddr = cfg_addr;
+
+	/*
+	 * Issue a Capability Configuration command to the firmware to get it
+	 * to parse the Configuration File.  We don't use t4_fw_config_file()
+	 * because we want the ability to modify various features after we've
+	 * processed the configuration file ...
+	 */
+	memset(&caps_cmd, 0, sizeof(caps_cmd));
+	caps_cmd.op_to_write = cpu_to_be32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
+					   F_FW_CMD_REQUEST | F_FW_CMD_READ);
+	caps_cmd.cfvalid_to_len16 =
+		cpu_to_be32(F_FW_CAPS_CONFIG_CMD_CFVALID |
+			    V_FW_CAPS_CONFIG_CMD_MEMTYPE_CF(mtype) |
+			    V_FW_CAPS_CONFIG_CMD_MEMADDR64K_CF(maddr >> 16) |
+			    FW_LEN16(caps_cmd));
+	ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd, sizeof(caps_cmd),
+			 &caps_cmd);
+	/*
+	 * If the CAPS_CONFIG failed with an ENOENT (for a Firmware
+	 * Configuration File in FLASH), our last gasp effort is to use the
+	 * Firmware Configuration File which is embedded in the firmware.  A
+	 * very few early versions of the firmware didn't have one embedded
+	 * but we can ignore those.
+	 */
+	if (ret == -ENOENT) {
+		dev_info(adapter, "%s: Going for embedded config in firmware..\n",
+			 __func__);
+
+		memset(&caps_cmd, 0, sizeof(caps_cmd));
+		caps_cmd.op_to_write =
+			cpu_to_be32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
+				    F_FW_CMD_REQUEST | F_FW_CMD_READ);
+		caps_cmd.cfvalid_to_len16 = cpu_to_be32(FW_LEN16(caps_cmd));
+		ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd,
+				 sizeof(caps_cmd), &caps_cmd);
+		strcpy(config_name, "Firmware Default");
+	}
+
+	config_issued = 1;
+	if (ret < 0)
+		goto bye;
+
+	finiver = be32_to_cpu(caps_cmd.finiver);
+	finicsum = be32_to_cpu(caps_cmd.finicsum);
+	cfcsum = be32_to_cpu(caps_cmd.cfcsum);
+	if (finicsum != cfcsum)
+		dev_warn(adapter, "Configuration File checksum mismatch: [fini] csum=%#x, computed csum=%#x\n",
+			 finicsum, cfcsum);
+
+	/*
+	 * If we're a pure NIC driver then disable all offloading facilities.
+	 * This will allow the firmware to optimize aspects of the hardware
+	 * configuration which will result in improved performance.
+	 */
+	caps_cmd.niccaps &= cpu_to_be16(~(FW_CAPS_CONFIG_NIC_HASHFILTER |
+					  FW_CAPS_CONFIG_NIC_ETHOFLD));
+	caps_cmd.toecaps = 0;
+	caps_cmd.iscsicaps = 0;
+	caps_cmd.rdmacaps = 0;
+	caps_cmd.fcoecaps = 0;
+
+	/*
+	 * And now tell the firmware to use the configuration we just loaded.
+	 */
+	caps_cmd.op_to_write = cpu_to_be32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
+					   F_FW_CMD_REQUEST | F_FW_CMD_WRITE);
+	caps_cmd.cfvalid_to_len16 = htonl(FW_LEN16(caps_cmd));
+	ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd, sizeof(caps_cmd),
+			 NULL);
+	if (ret < 0) {
+		dev_warn(adapter, "Unable to finalize Firmware Capabilities %d\n",
+			 -ret);
+		goto bye;
+	}
+
+	/*
+	 * Tweak configuration based on system architecture, module
+	 * parameters, etc.
+	 */
+	ret = adap_init0_tweaks(adapter);
+	if (ret < 0) {
+		dev_warn(adapter, "Unable to do init0-tweaks %d\n", -ret);
+		goto bye;
+	}
+
+	/*
+	 * And finally tell the firmware to initialize itself using the
+	 * parameters from the Configuration File.
+	 */
+	ret = t4_fw_initialize(adapter, adapter->mbox);
+	if (ret < 0) {
+		dev_warn(adapter, "Initializing Firmware failed, error %d\n",
+			 -ret);
+		goto bye;
+	}
+
+	/*
+	 * Return successfully and note that we're operating with parameters
+	 * not supplied by the driver, rather than from hard-wired
+	 * initialization constants burried in the driver.
+	 */
+	dev_info(adapter,
+		 "Successfully configured using Firmware Configuration File \"%s\", version %#x, computed checksum %#x\n",
+		 config_name, finiver, cfcsum);
+
+	return 0;
+
+	/*
+	 * Something bad happened.  Return the error ...  (If the "error"
+	 * is that there's no Configuration File on the adapter we don't
+	 * want to issue a warning since this is fairly common.)
+	 */
+bye:
+	if (config_issued && ret != -ENOENT)
+		dev_warn(adapter, "\"%s\" configuration file error %d\n",
+			 config_name, -ret);
+
+	dev_debug(adapter, "%s: returning ret = %d ..\n", __func__, ret);
+	return ret;
+}
+
+static int adap_init0(struct adapter *adap)
+{
+	int ret = 0;
+	u32 v, port_vec;
+	enum dev_state state;
+	u32 params[7], val[7];
+	int reset = 1;
+	int mbox = adap->mbox;
+
+	/*
+	 * Contact FW, advertising Master capability (and potentially forcing
+	 * ourselves as the Master PF if our module parameter force_init is
+	 * set).
+	 */
+	ret = t4_fw_hello(adap, adap->mbox, adap->mbox, MASTER_MAY, &state);
+	if (ret < 0) {
+		dev_err(adap, "%s: could not connect to FW, error %d\n",
+			__func__, -ret);
+		goto bye;
+	}
+
+	CXGBE_DEBUG_MBOX(adap, "%s: adap->mbox = %d; ret = %d\n", __func__,
+			 adap->mbox, ret);
+
+	if (ret == mbox)
+		adap->flags |= MASTER_PF;
+
+	if (state == DEV_STATE_INIT) {
+		/*
+		 * Force halt and reset FW because a previous instance may have
+		 * exited abnormally without properly shutting down
+		 */
+		ret = t4_fw_halt(adap, adap->mbox, reset);
+		if (ret < 0) {
+			dev_err(adap, "Failed to halt. Exit.\n");
+			goto bye;
+		}
+
+		ret = t4_fw_restart(adap, adap->mbox, reset);
+		if (ret < 0) {
+			dev_err(adap, "Failed to restart. Exit.\n");
+			goto bye;
+		}
+		state &= ~DEV_STATE_INIT;
+	}
+
+	t4_get_fw_version(adap, &adap->params.fw_vers);
+	t4_get_tp_version(adap, &adap->params.tp_vers);
+
+	dev_info(adap, "fw: %u.%u.%u.%u, TP: %u.%u.%u.%u\n",
+		 G_FW_HDR_FW_VER_MAJOR(adap->params.fw_vers),
+		 G_FW_HDR_FW_VER_MINOR(adap->params.fw_vers),
+		 G_FW_HDR_FW_VER_MICRO(adap->params.fw_vers),
+		 G_FW_HDR_FW_VER_BUILD(adap->params.fw_vers),
+		 G_FW_HDR_FW_VER_MAJOR(adap->params.tp_vers),
+		 G_FW_HDR_FW_VER_MINOR(adap->params.tp_vers),
+		 G_FW_HDR_FW_VER_MICRO(adap->params.tp_vers),
+		 G_FW_HDR_FW_VER_BUILD(adap->params.tp_vers));
+
+	ret = t4_get_core_clock(adap, &adap->params.vpd);
+	if (ret < 0) {
+		dev_err(adap, "%s: could not get core clock, error %d\n",
+			__func__, -ret);
+		goto bye;
+	}
+
+	/*
+	 * Find out what ports are available to us.  Note that we need to do
+	 * this before calling adap_init0_no_config() since it needs nports
+	 * and portvec ...
+	 */
+	v = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
+	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_PORTVEC);
+	ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 1, &v, &port_vec);
+	if (ret < 0) {
+		dev_err(adap, "%s: failure in t4_queury_params; error = %d\n",
+			__func__, ret);
+		goto bye;
+	}
+
+	adap->params.nports = hweight32(port_vec);
+	adap->params.portvec = port_vec;
+
+	dev_debug(adap, "%s: adap->params.nports = %u\n", __func__,
+		  adap->params.nports);
+
+	/*
+	 * If the firmware is initialized already (and we're not forcing a
+	 * master initialization), note that we're living with existing
+	 * adapter parameters.  Otherwise, it's time to try initializing the
+	 * adapter ...
+	 */
+	if (state == DEV_STATE_INIT) {
+		dev_info(adap, "Coming up as %s: Adapter already initialized\n",
+			 adap->flags & MASTER_PF ? "MASTER" : "SLAVE");
+	} else {
+		dev_info(adap, "Coming up as MASTER: Initializing adapter\n");
+
+		ret = adap_init0_config(adap, reset);
+		if (ret == -ENOENT) {
+			dev_err(adap,
+				"No Configuration File present on adapter. Using hard-wired configuration parameters.\n");
+			goto bye;
+		}
+	}
+	if (ret < 0) {
+		dev_err(adap, "could not initialize adapter, error %d\n", -ret);
+		goto bye;
+	}
+
+	/*
+	 * Give the SGE code a chance to pull in anything that it needs ...
+	 * Note that this must be called after we retrieve our VPD parameters
+	 * in order to know how to convert core ticks to seconds, etc.
+	 */
+	ret = t4_sge_init(adap);
+	if (ret < 0) {
+		dev_err(adap, "t4_sge_init failed with error %d\n",
+			-ret);
+		goto bye;
+	}
+
+	/*
+	 * Grab some of our basic fundamental operating parameters.
+	 */
+#define FW_PARAM_DEV(param) \
+	(V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) | \
+	 V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_##param))
+
+#define FW_PARAM_PFVF(param) \
+	(V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_PFVF) | \
+	 V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_PFVF_##param) |  \
+	 V_FW_PARAMS_PARAM_Y(0) | \
+	 V_FW_PARAMS_PARAM_Z(0))
+
+	/* If we're running on newer firmware, let it know that we're
+	 * prepared to deal with encapsulated CPL messages.  Older
+	 * firmware won't understand this and we'll just get
+	 * unencapsulated messages ...
+	 */
+	params[0] = FW_PARAM_PFVF(CPLFW4MSG_ENCAP);
+	val[0] = 1;
+	(void)t4_set_params(adap, adap->mbox, adap->pf, 0, 1, params, val);
+
+	/*
+	 * Find out whether we're allowed to use the T5+ ULPTX MEMWRITE DSGL
+	 * capability.  Earlier versions of the firmware didn't have the
+	 * ULPTX_MEMWRITE_DSGL so we'll interpret a query failure as no
+	 * permission to use ULPTX MEMWRITE DSGL.
+	 */
+	if (is_t4(adap->params.chip)) {
+		adap->params.ulptx_memwrite_dsgl = false;
+	} else {
+		params[0] = FW_PARAM_DEV(ULPTX_MEMWRITE_DSGL);
+		ret = t4_query_params(adap, adap->mbox, adap->pf, 0,
+				      1, params, val);
+		adap->params.ulptx_memwrite_dsgl = (ret == 0 && val[0] != 0);
+	}
+
+	/*
+	 * The MTU/MSS Table is initialized by now, so load their values.  If
+	 * we're initializing the adapter, then we'll make any modifications
+	 * we want to the MTU/MSS Table and also initialize the congestion
+	 * parameters.
+	 */
+	t4_read_mtu_tbl(adap, adap->params.mtus, NULL);
+	if (state != DEV_STATE_INIT) {
+		int i;
+
+		/*
+		 * The default MTU Table contains values 1492 and 1500.
+		 * However, for TCP, it's better to have two values which are
+		 * a multiple of 8 +/- 4 bytes apart near this popular MTU.
+		 * This allows us to have a TCP Data Payload which is a
+		 * multiple of 8 regardless of what combination of TCP Options
+		 * are in use (always a multiple of 4 bytes) which is
+		 * important for performance reasons.  For instance, if no
+		 * options are in use, then we have a 20-byte IP header and a
+		 * 20-byte TCP header.  In this case, a 1500-byte MSS would
+		 * result in a TCP Data Payload of 1500 - 40 == 1460 bytes
+		 * which is not a multiple of 8.  So using an MSS of 1488 in
+		 * this case results in a TCP Data Payload of 1448 bytes which
+		 * is a multiple of 8.  On the other hand, if 12-byte TCP Time
+		 * Stamps have been negotiated, then an MTU of 1500 bytes
+		 * results in a TCP Data Payload of 1448 bytes which, as
+		 * above, is a multiple of 8 bytes ...
+		 */
+		for (i = 0; i < NMTUS; i++)
+			if (adap->params.mtus[i] == 1492) {
+				adap->params.mtus[i] = 1488;
+				break;
+			}
+
+		t4_load_mtus(adap, adap->params.mtus, adap->params.a_wnd,
+			     adap->params.b_wnd);
+	}
+	t4_init_sge_params(adap);
+	t4_init_tp_params(adap);
+
+	adap->params.drv_memwin = MEMWIN_NIC;
+	adap->flags |= FW_OK;
+	dev_debug(adap, "%s: returning zero..\n", __func__);
+	return 0;
+
+	/*
+	 * Something bad happened.  If a command timed out or failed with EIO
+	 * FW does not operate within its spec or something catastrophic
+	 * happened to HW/FW, stop issuing commands.
+	 */
+bye:
+	if (ret != -ETIMEDOUT && ret != -EIO)
+		t4_fw_bye(adap, adap->mbox);
+	return ret;
+}
+
+/**
+ * t4_os_portmod_changed - handle port module changes
+ * @adap: the adapter associated with the module change
+ * @port_id: the port index whose module status has changed
+ *
+ * This is the OS-dependent handler for port module changes.  It is
+ * invoked when a port module is removed or inserted for any OS-specific
+ * processing.
+ */
+void t4_os_portmod_changed(const struct adapter *adap, int port_id)
+{
+	static const char * const mod_str[] = {
+		NULL, "LR", "SR", "ER", "passive DA", "active DA", "LRM"
+	};
+
+	const struct port_info *pi = &adap->port[port_id];
+
+	if (pi->mod_type == FW_PORT_MOD_TYPE_NONE)
+		dev_info(adap, "Port%d: port module unplugged\n", pi->port_id);
+	else if (pi->mod_type < ARRAY_SIZE(mod_str))
+		dev_info(adap, "Port%d: %s port module inserted\n", pi->port_id,
+			 mod_str[pi->mod_type]);
+	else if (pi->mod_type == FW_PORT_MOD_TYPE_NOTSUPPORTED)
+		dev_info(adap, "Port%d: unsupported optical port module inserted\n",
+			 pi->port_id);
+	else if (pi->mod_type == FW_PORT_MOD_TYPE_UNKNOWN)
+		dev_info(adap, "Port%d: unknown port module inserted, forcing TWINAX\n",
+			 pi->port_id);
+	else if (pi->mod_type == FW_PORT_MOD_TYPE_ERROR)
+		dev_info(adap, "Port%d: transceiver module error\n",
+			 pi->port_id);
+	else
+		dev_info(adap, "Port%d: unknown module type %d inserted\n",
+			 pi->port_id, pi->mod_type);
+}
+
+/**
+ * link_start - enable a port
+ * @dev: the port to enable
+ *
+ * Performs the MAC and PHY actions needed to enable a port.
+ */
+int link_start(struct port_info *pi)
+{
+	struct adapter *adapter = pi->adapter;
+	int ret;
+
+	/*
+	 * We do not set address filters and promiscuity here, the stack does
+	 * that step explicitly.
+	 */
+	ret = t4_set_rxmode(adapter, adapter->mbox, pi->viid, 1500, -1, -1,
+			    -1, 1, true);
+	if (ret == 0) {
+		ret = t4_change_mac(adapter, adapter->mbox, pi->viid,
+				    pi->xact_addr_filt,
+				    (u8 *)&pi->eth_dev->data->mac_addrs[0],
+				    true, true);
+		if (ret >= 0) {
+			pi->xact_addr_filt = ret;
+			ret = 0;
+		}
+	}
+	if (ret == 0)
+		ret = t4_link_l1cfg(adapter, adapter->mbox, pi->tx_chan,
+				    &pi->link_cfg);
+	if (ret == 0) {
+		/*
+		 * Enabling a Virtual Interface can result in an interrupt
+		 * during the processing of the VI Enable command and, in some
+		 * paths, result in an attempt to issue another command in the
+		 * interrupt context.  Thus, we disable interrupts during the
+		 * course of the VI Enable command ...
+		 */
+		ret = t4_enable_vi_params(adapter, adapter->mbox, pi->viid,
+					  true, true, false);
+	}
+	return ret;
+}
+
+/**
+ * cxgb4_write_rss - write the RSS table for a given port
+ * @pi: the port
+ * @queues: array of queue indices for RSS
+ *
+ * Sets up the portion of the HW RSS table for the port's VI to distribute
+ * packets to the Rx queues in @queues.
+ */
+int cxgb4_write_rss(const struct port_info *pi, const u16 *queues)
+{
+	u16 *rss;
+	int i, err;
+	struct adapter *adapter = pi->adapter;
+	const struct sge_eth_rxq *rxq;
+
+	/*  Should never be called before setting up sge eth rx queues */
+	BUG_ON(!(adapter->flags & FULL_INIT_DONE));
+
+	rxq = &adapter->sge.ethrxq[pi->first_qset];
+	rss = rte_zmalloc(NULL, pi->rss_size * sizeof(u16), 0);
+	if (!rss)
+		return -ENOMEM;
+
+	/* map the queue indices to queue ids */
+	for (i = 0; i < pi->rss_size; i++, queues++)
+		rss[i] = rxq[*queues].rspq.abs_id;
+
+	err = t4_config_rss_range(adapter, adapter->pf, pi->viid, 0,
+				  pi->rss_size, rss, pi->rss_size);
+	/*
+	 * If Tunnel All Lookup isn't specified in the global RSS
+	 * Configuration, then we need to specify a default Ingress
+	 * Queue for any ingress packets which aren't hashed.  We'll
+	 * use our first ingress queue ...
+	 */
+	if (!err)
+		err = t4_config_vi_rss(adapter, adapter->mbox, pi->viid,
+				       F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN |
+				       F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN |
+				       F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN |
+				       F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN |
+				       F_FW_RSS_VI_CONFIG_CMD_UDPEN,
+				       rss[0]);
+	rte_free(rss);
+	return err;
+}
+
+/**
+ * setup_rss - configure RSS
+ * @adapter: the adapter
+ *
+ * Sets up RSS to distribute packets to multiple receive queues.  We
+ * configure the RSS CPU lookup table to distribute to the number of HW
+ * receive queues, and the response queue lookup table to narrow that
+ * down to the response queues actually configured for each port.
+ * We always configure the RSS mapping for all ports since the mapping
+ * table has plenty of entries.
+ */
+int setup_rss(struct port_info *pi)
+{
+	int j, err;
+	struct adapter *adapter = pi->adapter;
+
+	dev_debug(adapter, "%s:  pi->rss_size = %u; pi->n_rx_qsets = %u\n",
+		  __func__, pi->rss_size, pi->n_rx_qsets);
+
+	if (!pi->flags & PORT_RSS_DONE) {
+		if (adapter->flags & FULL_INIT_DONE) {
+			/* Fill default values with equal distribution */
+			for (j = 0; j < pi->rss_size; j++)
+				pi->rss[j] = j % pi->n_rx_qsets;
+
+			err = cxgb4_write_rss(pi, pi->rss);
+			if (err)
+				return err;
+			pi->flags |= PORT_RSS_DONE;
+		}
+	}
+	return 0;
+}
+
+/*
+ * Enable NAPI scheduling and interrupt generation for all Rx queues.
+ */
+static void enable_rx(struct adapter *adap)
+{
+	struct sge *s = &adap->sge;
+	struct sge_rspq *q = &s->fw_evtq;
+	int i, j;
+
+	/* 0-increment GTS to start the timer and enable interrupts */
+	t4_write_reg(adap, MYPF_REG(A_SGE_PF_GTS),
+		     V_SEINTARM(q->intr_params) |
+		     V_INGRESSQID(q->cntxt_id));
+
+	for_each_port(adap, i) {
+		const struct port_info *pi = &adap->port[i];
+		struct rte_eth_dev *eth_dev = pi->eth_dev;
+
+		for (j = 0; j < eth_dev->data->nb_rx_queues; j++) {
+			q = eth_dev->data->rx_queues[j];
+
+			/*
+			 * 0-increment GTS to start the timer and enable
+			 * interrupts
+			 */
+			t4_write_reg(adap, MYPF_REG(A_SGE_PF_GTS),
+				     V_SEINTARM(q->intr_params) |
+				     V_INGRESSQID(q->cntxt_id));
+		}
+	}
+}
+
+/**
+ * cxgb_up - enable the adapter
+ * @adap: adapter being enabled
+ *
+ * Called when the first port is enabled, this function performs the
+ * actions necessary to make an adapter operational, such as completing
+ * the initialization of HW modules, and enabling interrupts.
+ */
+int cxgbe_up(struct adapter *adap)
+{
+	enable_rx(adap);
+	t4_intr_enable(adap);
+	adap->flags |= FULL_INIT_DONE;
+
+	/* TODO: deadman watchdog ?? */
+	return 0;
+}
+
+/*
+ * Close the port
+ */
+int cxgbe_down(struct port_info *pi)
+{
+	struct adapter *adapter = pi->adapter;
+	int err = 0;
+
+	err = t4_enable_vi(adapter, adapter->mbox, pi->viid, false, false);
+	if (err) {
+		dev_err(adapter, "%s: disable_vi failed: %d\n", __func__, err);
+		return err;
+	}
+
+	t4_reset_link_config(adapter, pi->port_id);
+	return 0;
+}
+
+/*
+ * Release resources when all the ports have been stopped.
+ */
+void cxgbe_close(struct adapter *adapter)
+{
+	struct port_info *pi;
+	int i;
+
+	if (adapter->flags & FULL_INIT_DONE) {
+		t4_intr_disable(adapter);
+		t4_free_sge_resources(adapter);
+		for_each_port(adapter, i) {
+			pi = adap2pinfo(adapter, i);
+			if (pi->viid != 0)
+				t4_free_vi(adapter, adapter->mbox,
+					   adapter->pf, 0, pi->viid);
+			rte_free(pi->eth_dev->data->mac_addrs);
+		}
+		adapter->flags &= ~FULL_INIT_DONE;
+	}
+
+	if (adapter->flags & FW_OK)
+		t4_fw_bye(adapter, adapter->mbox);
+}
+
+int cxgbe_probe(struct adapter *adapter)
+{
+	struct port_info *pi;
+	int func, i;
+	int err = 0;
+
+	func = G_SOURCEPF(t4_read_reg(adapter, A_PL_WHOAMI));
+	adapter->mbox = func;
+	adapter->pf = func;
+
+	t4_os_lock_init(&adapter->mbox_lock);
+	TAILQ_INIT(&adapter->mbox_list);
+
+	err = t4_prep_adapter(adapter);
+	if (err)
+		return err;
+
+	setup_memwin(adapter);
+	err = adap_init0(adapter);
+	if (err) {
+		dev_err(adapter, "%s: Adapter initialization failed, error %d\n",
+			__func__, err);
+		goto out_free;
+	}
+
+	if (!is_t4(adapter->params.chip)) {
+		/*
+		 * The userspace doorbell BAR is split evenly into doorbell
+		 * regions, each associated with an egress queue.  If this
+		 * per-queue region is large enough (at least UDBS_SEG_SIZE)
+		 * then it can be used to submit a tx work request with an
+		 * implied doorbell.  Enable write combining on the BAR if
+		 * there is room for such work requests.
+		 */
+		int s_qpp, qpp, num_seg;
+
+		s_qpp = (S_QUEUESPERPAGEPF0 +
+			(S_QUEUESPERPAGEPF1 - S_QUEUESPERPAGEPF0) *
+			adapter->pf);
+		qpp = 1 << ((t4_read_reg(adapter,
+				A_SGE_EGRESS_QUEUES_PER_PAGE_PF) >> s_qpp)
+				& M_QUEUESPERPAGEPF0);
+		num_seg = PAGE_SIZE / UDBS_SEG_SIZE;
+		if (qpp > num_seg)
+			dev_warn(adapter, "Incorrect SGE EGRESS QUEUES_PER_PAGE configuration, continuing in debug mode\n");
+
+		adapter->bar2 = (void *)adapter->pdev->mem_resource[2].addr;
+		if (!adapter->bar2) {
+			dev_err(adapter, "cannot map device bar2 region\n");
+			err = -ENOMEM;
+			goto out_free;
+		}
+		t4_write_reg(adapter, A_SGE_STAT_CFG, V_STATSOURCE_T5(7) |
+			     V_STATMODE(0));
+	}
+
+	for_each_port(adapter, i) {
+		char name[RTE_ETH_NAME_MAX_LEN];
+		struct rte_eth_dev_data *data = NULL;
+		const unsigned int numa_node = rte_socket_id();
+
+		pi = &adapter->port[i];
+		pi->adapter = adapter;
+		pi->xact_addr_filt = -1;
+		pi->port_id = i;
+
+		snprintf(name, sizeof(name), "cxgbe%d",
+			 adapter->eth_dev->data->port_id + i);
+
+		if (i == 0) {
+			/* First port is already allocated by DPDK */
+			pi->eth_dev = adapter->eth_dev;
+			goto allocate_mac;
+		}
+
+		/*
+		 * now do all data allocation - for eth_dev structure,
+		 * and internal (private) data for the remaining ports
+		 */
+
+		/* reserve an ethdev entry */
+		pi->eth_dev = rte_eth_dev_allocate(name, RTE_ETH_DEV_PCI);
+		if (!pi->eth_dev)
+			goto out_free;
+
+		data = rte_zmalloc_socket(name, sizeof(*data), 0, numa_node);
+		if (!data)
+			goto out_free;
+
+		data->port_id = adapter->eth_dev->data->port_id + i;
+
+		pi->eth_dev->data = data;
+
+allocate_mac:
+		pi->eth_dev->pci_dev = adapter->pdev;
+		pi->eth_dev->data->dev_private = pi;
+		pi->eth_dev->driver = adapter->eth_dev->driver;
+		pi->eth_dev->dev_ops = adapter->eth_dev->dev_ops;
+		pi->eth_dev->tx_pkt_burst = adapter->eth_dev->tx_pkt_burst;
+		pi->eth_dev->rx_pkt_burst = adapter->eth_dev->rx_pkt_burst;
+		TAILQ_INIT(&pi->eth_dev->link_intr_cbs);
+
+		pi->eth_dev->data->mac_addrs = rte_zmalloc(name,
+							   ETHER_ADDR_LEN, 0);
+		if (!pi->eth_dev->data->mac_addrs) {
+			dev_err(adapter, "%s: Mem allocation failed for storing mac addr, aborting\n",
+				__func__);
+			err = -1;
+			goto out_free;
+		}
+	}
+
+	if (adapter->flags & FW_OK) {
+		err = t4_port_init(adapter, adapter->mbox, adapter->pf, 0);
+		if (err) {
+			dev_err(adapter, "%s: t4_port_init failed with err %d\n",
+				__func__, err);
+			goto out_free;
+		}
+	}
+
+	cfg_queues(adapter->eth_dev);
+
+	print_port_info(adapter);
+
+	err = init_rss(adapter);
+	if (err)
+		goto out_free;
+
+	return 0;
+
+out_free:
+	for_each_port(adapter, i) {
+		pi = adap2pinfo(adapter, i);
+		if (pi->viid != 0)
+			t4_free_vi(adapter, adapter->mbox, adapter->pf,
+				   0, pi->viid);
+		/* Skip first port since it'll be de-allocated by DPDK */
+		if (i == 0)
+			continue;
+		if (pi->eth_dev->data)
+			rte_free(pi->eth_dev->data);
+	}
+
+	if (adapter->flags & FW_OK)
+		t4_fw_bye(adapter, adapter->mbox);
+	return -err;
+}
diff --git a/lib/librte_pmd_cxgbe/sge.c b/lib/librte_pmd_cxgbe/sge.c
new file mode 100644
index 0000000..6343dff
--- /dev/null
+++ b/lib/librte_pmd_cxgbe/sge.c
@@ -0,0 +1,2250 @@
+/*-
+ *   BSD LICENSE
+ *
+ *   Copyright(c) 2014-2015 Chelsio Communications.
+ *   All rights reserved.
+ *
+ *   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.
+ *     * Neither the name of Chelsio Communications nor the names of its
+ *       contributors may be used to endorse or promote products derived
+ *       from this software without specific prior written permission.
+ *
+ *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#include <linux/if_ether.h>
+#include <sys/queue.h>
+#include <stdio.h>
+#include <errno.h>
+#include <stdint.h>
+#include <string.h>
+#include <unistd.h>
+#include <stdarg.h>
+#include <inttypes.h>
+#include <netinet/in.h>
+
+#include <rte_byteorder.h>
+#include <rte_common.h>
+#include <rte_cycles.h>
+#include <rte_interrupts.h>
+#include <rte_log.h>
+#include <rte_debug.h>
+#include <rte_pci.h>
+#include <rte_atomic.h>
+#include <rte_branch_prediction.h>
+#include <rte_memory.h>
+#include <rte_memzone.h>
+#include <rte_tailq.h>
+#include <rte_eal.h>
+#include <rte_alarm.h>
+#include <rte_ether.h>
+#include <rte_ethdev.h>
+#include <rte_atomic.h>
+#include <rte_malloc.h>
+#include <rte_random.h>
+#include <rte_dev.h>
+
+#include "common.h"
+#include "t4_regs.h"
+#include "t4_msg.h"
+#include "cxgbe.h"
+
+static inline void ship_tx_pkt_coalesce_wr(struct adapter *adap,
+					   struct sge_eth_txq *txq,
+					   uint16_t left);
+
+/*
+ * Rx buffer size for "packed" pages Free List buffers (multiple ingress
+ * packets packed per page buffer).  We use largish buffers if possible but
+ * settle for single pages under memory shortage.
+ */
+#if PAGE_SHIFT >= 16
+#define FL_PG_ORDER 0
+#else
+#define FL_PG_ORDER (16 - PAGE_SHIFT)
+#endif
+
+/* RX_PULL_LEN should be <= RX_COPY_THRES */
+#define RX_COPY_THRES    256
+#define RX_PULL_LEN      128
+
+/*
+ * Max number of Rx buffers we replenish at a time.
+ */
+#define MAX_RX_REFILL 16U
+
+#define NOMEM_TMR_IDX (SGE_NTIMERS - 1)
+
+/*
+ * Max Tx descriptor space we allow for an Ethernet packet to be inlined
+ * into a WR.
+ */
+#define MAX_IMM_TX_PKT_LEN 256
+
+/*
+ * Rx buffer sizes for "usembufs" Free List buffers (one ingress packet
+ * per mbuf buffer).  We currently only support two sizes for 1500- and
+ * 9000-byte MTUs. We could easily support more but there doesn't seem to be
+ * much need for that ...
+ */
+#define FL_MTU_SMALL 1500
+#define FL_MTU_LARGE 9000
+
+static inline unsigned int fl_mtu_bufsize(struct adapter *adapter,
+					  unsigned int mtu)
+{
+	struct sge *s = &adapter->sge;
+
+	return ALIGN(s->pktshift + ETH_HLEN + VLAN_HLEN + mtu, s->fl_align);
+}
+
+#define FL_MTU_SMALL_BUFSIZE(adapter) fl_mtu_bufsize(adapter, FL_MTU_SMALL)
+#define FL_MTU_LARGE_BUFSIZE(adapter) fl_mtu_bufsize(adapter, FL_MTU_LARGE)
+
+/*
+ * Bits 0..3 of rx_sw_desc.dma_addr have special meaning.  The hardware uses
+ * these to specify the buffer size as an index into the SGE Free List Buffer
+ * Size register array.  We also use bit 4, when the buffer has been unmapped
+ * for DMA, but this is of course never sent to the hardware and is only used
+ * to prevent double unmappings.  All of the above requires that the Free List
+ * Buffers which we allocate have the bottom 5 bits free (0) -- i.e. are
+ * 32-byte or or a power of 2 greater in alignment.  Since the SGE's minimal
+ * Free List Buffer alignment is 32 bytes, this works out for us ...
+ */
+enum {
+	RX_BUF_FLAGS     = 0x1f,   /* bottom five bits are special */
+	RX_BUF_SIZE      = 0x0f,   /* bottom three bits are for buf sizes */
+	RX_UNMAPPED_BUF  = 0x10,   /* buffer is not mapped */
+
+	/*
+	 * XXX We shouldn't depend on being able to use these indices.
+	 * XXX Especially when some other Master PF has initialized the
+	 * XXX adapter or we use the Firmware Configuration File.  We
+	 * XXX should really search through the Host Buffer Size register
+	 * XXX array for the appropriately sized buffer indices.
+	 */
+	RX_SMALL_PG_BUF  = 0x0,   /* small (PAGE_SIZE) page buffer */
+	RX_LARGE_PG_BUF  = 0x1,   /* buffer large (FL_PG_ORDER) page buffer */
+
+	RX_SMALL_MTU_BUF = 0x2,   /* small MTU buffer */
+	RX_LARGE_MTU_BUF = 0x3,   /* large MTU buffer */
+};
+
+/**
+ * txq_avail - return the number of available slots in a Tx queue
+ * @q: the Tx queue
+ *
+ * Returns the number of descriptors in a Tx queue available to write new
+ * packets.
+ */
+static inline unsigned int txq_avail(const struct sge_txq *q)
+{
+	return q->size - 1 - q->in_use;
+}
+
+static int map_mbuf(struct rte_mbuf *mbuf, dma_addr_t *addr)
+{
+	struct rte_mbuf *m = mbuf;
+
+	for (; m; m = m->next, addr++) {
+		*addr = m->buf_physaddr + rte_pktmbuf_headroom(m);
+		if (*addr == 0)
+			goto out_err;
+	}
+	return 0;
+
+out_err:
+	return -ENOMEM;
+}
+
+/**
+ * free_tx_desc - reclaims Tx descriptors and their buffers
+ * @q: the Tx queue to reclaim descriptors from
+ * @n: the number of descriptors to reclaim
+ *
+ * Reclaims Tx descriptors from an SGE Tx queue and frees the associated
+ * Tx buffers.  Called with the Tx queue lock held.
+ */
+static void free_tx_desc(struct sge_txq *q, unsigned int n)
+{
+	struct tx_sw_desc *d;
+	unsigned int cidx = 0;
+
+	d = &q->sdesc[cidx];
+	while (n--) {
+		if (d->mbuf) {                       /* an SGL is present */
+			rte_pktmbuf_free(d->mbuf);
+			d->mbuf = NULL;
+		}
+		if (d->coalesce.idx) {
+			int i;
+
+			for (i = 0; i < d->coalesce.idx; i++) {
+				rte_pktmbuf_free(d->coalesce.mbuf[i]);
+				d->coalesce.mbuf[i] = NULL;
+			}
+			d->coalesce.idx = 0;
+		}
+		++d;
+		if (++cidx == q->size) {
+			cidx = 0;
+			d = q->sdesc;
+		}
+		RTE_MBUF_PREFETCH_TO_FREE(&q->sdesc->mbuf->pool);
+	}
+}
+
+static void reclaim_tx_desc(struct sge_txq *q, unsigned int n)
+{
+	unsigned int cidx = q->cidx;
+
+	while (n--) {
+		if (++cidx == q->size)
+			cidx = 0;
+	}
+	q->cidx = cidx;
+}
+
+/**
+ * fl_cap - return the capacity of a free-buffer list
+ * @fl: the FL
+ *
+ * Returns the capacity of a free-buffer list.  The capacity is less than
+ * the size because one descriptor needs to be left unpopulated, otherwise
+ * HW will think the FL is empty.
+ */
+static inline unsigned int fl_cap(const struct sge_fl *fl)
+{
+	return fl->size - 8;   /* 1 descriptor = 8 buffers */
+}
+
+/**
+ * fl_starving - return whether a Free List is starving.
+ * @adapter: pointer to the adapter
+ * @fl: the Free List
+ *
+ * Tests specified Free List to see whether the number of buffers
+ * available to the hardware has falled below our "starvation"
+ * threshold.
+ */
+static inline bool fl_starving(const struct adapter *adapter,
+			       const struct sge_fl *fl)
+{
+	const struct sge *s = &adapter->sge;
+
+	return fl->avail - fl->pend_cred <= s->fl_starve_thres;
+}
+
+static inline unsigned int get_buf_size(struct adapter *adapter,
+					const struct rx_sw_desc *d)
+{
+	struct sge *s = &adapter->sge;
+	unsigned int rx_buf_size_idx = d->dma_addr & RX_BUF_SIZE;
+	unsigned int buf_size;
+
+	switch (rx_buf_size_idx) {
+	case RX_SMALL_PG_BUF:
+		buf_size = PAGE_SIZE;
+		break;
+
+	case RX_LARGE_PG_BUF:
+		buf_size = PAGE_SIZE << s->fl_pg_order;
+		break;
+
+	case RX_SMALL_MTU_BUF:
+		buf_size = FL_MTU_SMALL_BUFSIZE(adapter);
+		break;
+
+	case RX_LARGE_MTU_BUF:
+		buf_size = FL_MTU_LARGE_BUFSIZE(adapter);
+		break;
+
+	default:
+		BUG_ON(1);
+		buf_size = 0; /* deal with bogus compiler warnings */
+		/* NOTREACHED */
+	}
+
+	return buf_size;
+}
+
+/**
+ * free_rx_bufs - free the Rx buffers on an SGE free list
+ * @q: the SGE free list to free buffers from
+ * @n: how many buffers to free
+ *
+ * Release the next @n buffers on an SGE free-buffer Rx queue.   The
+ * buffers must be made inaccessible to HW before calling this function.
+ */
+static void free_rx_bufs(struct sge_fl *q, int n)
+{
+	unsigned int cidx = q->cidx;
+	struct rx_sw_desc *d;
+
+	d = &q->sdesc[cidx];
+	while (n--) {
+		if (d->buf) {
+			rte_pktmbuf_free(d->buf);
+			d->buf = NULL;
+		}
+		++d;
+		if (++cidx == q->size) {
+			cidx = 0;
+			d = q->sdesc;
+		}
+		q->avail--;
+	}
+	q->cidx = cidx;
+}
+
+/**
+ * unmap_rx_buf - unmap the current Rx buffer on an SGE free list
+ * @q: the SGE free list
+ *
+ * Unmap the current buffer on an SGE free-buffer Rx queue.   The
+ * buffer must be made inaccessible to HW before calling this function.
+ *
+ * This is similar to @free_rx_bufs above but does not free the buffer.
+ * Do note that the FL still loses any further access to the buffer.
+ */
+static void unmap_rx_buf(struct sge_fl *q)
+{
+	if (++q->cidx == q->size)
+		q->cidx = 0;
+	q->avail--;
+}
+
+static inline void ring_fl_db(struct adapter *adap, struct sge_fl *q)
+{
+	if (q->pend_cred >= 8) {
+		u32 val = adap->params.arch.sge_fl_db;
+
+		if (is_t4(adap->params.chip))
+			val |= V_PIDX(q->pend_cred / 8);
+		else
+			val |= V_PIDX_T5(q->pend_cred / 8);
+
+		/*
+		 * Make sure all memory writes to the Free List queue are
+		 * committed before we tell the hardware about them.
+		 */
+		wmb();
+
+		/*
+		 * If we don't have access to the new User Doorbell (T5+), use
+		 * the old doorbell mechanism; otherwise use the new BAR2
+		 * mechanism.
+		 */
+		if (unlikely(!q->bar2_addr)) {
+			t4_write_reg(adap, MYPF_REG(A_SGE_PF_KDOORBELL),
+				     val | V_QID(q->cntxt_id));
+		} else {
+			writel(val | V_QID(q->bar2_qid),
+			       (void *)((uintptr_t)q->bar2_addr +
+			       SGE_UDB_KDOORBELL));
+
+			/*
+			 * This Write memory Barrier will force the write to
+			 * the User Doorbell area to be flushed.
+			 */
+			wmb();
+		}
+		q->pend_cred &= 7;
+	}
+}
+
+static inline struct rte_mbuf *cxgbe_rxmbuf_alloc(struct rte_mempool *mp)
+{
+	struct rte_mbuf *m;
+
+	m = __rte_mbuf_raw_alloc(mp);
+	__rte_mbuf_sanity_check_raw(m, 0);
+	return m;
+}
+
+static inline void set_rx_sw_desc(struct rx_sw_desc *sd, void *buf,
+				  dma_addr_t mapping)
+{
+	sd->buf = buf;
+	sd->dma_addr = mapping;      /* includes size low bits */
+}
+
+/**
+ * refill_fl_usembufs - refill an SGE Rx buffer ring with mbufs
+ * @adap: the adapter
+ * @q: the ring to refill
+ * @n: the number of new buffers to allocate
+ *
+ * (Re)populate an SGE free-buffer queue with up to @n new packet buffers,
+ * allocated with the supplied gfp flags.  The caller must assure that
+ * @n does not exceed the queue's capacity.  If afterwards the queue is
+ * found critically low mark it as starving in the bitmap of starving FLs.
+ *
+ * Returns the number of buffers allocated.
+ */
+static unsigned int refill_fl_usembufs(struct adapter *adap, struct sge_fl *q,
+				       int n)
+{
+	struct sge_eth_rxq *rxq = container_of(q, struct sge_eth_rxq, fl);
+	unsigned int cred = q->avail;
+	__be64 *d = &q->desc[q->pidx];
+	struct rx_sw_desc *sd = &q->sdesc[q->pidx];
+	unsigned int buf_size_idx = RX_SMALL_MTU_BUF;
+
+	while (n--) {
+		struct rte_mbuf *mbuf = cxgbe_rxmbuf_alloc(rxq->rspq.mb_pool);
+		dma_addr_t mapping;
+
+		if (!mbuf) {
+			dev_debug(adap, "%s: mbuf alloc failed\n", __func__);
+			q->alloc_failed++;
+			rxq->rspq.eth_dev->data->rx_mbuf_alloc_failed++;
+			goto out;
+		}
+
+		mbuf->data_off = RTE_PKTMBUF_HEADROOM;
+		mbuf->next = NULL;
+
+		mapping = (dma_addr_t)(mbuf->buf_physaddr + mbuf->data_off);
+
+		mapping |= buf_size_idx;
+		*d++ = cpu_to_be64(mapping);
+		set_rx_sw_desc(sd, mbuf, mapping);
+		sd++;
+
+		q->avail++;
+		if (++q->pidx == q->size) {
+			q->pidx = 0;
+			sd = q->sdesc;
+			d = q->desc;
+		}
+	}
+
+out:    cred = q->avail - cred;
+	q->pend_cred += cred;
+	ring_fl_db(adap, q);
+
+	if (unlikely(fl_starving(adap, q))) {
+		/*
+		 * Make sure data has been written to free list
+		 */
+		wmb();
+		q->low++;
+	}
+
+	return cred;
+}
+
+/**
+ * refill_fl - refill an SGE Rx buffer ring with mbufs
+ * @adap: the adapter
+ * @q: the ring to refill
+ * @n: the number of new buffers to allocate
+ *
+ * (Re)populate an SGE free-buffer queue with up to @n new packet buffers,
+ * allocated with the supplied gfp flags.  The caller must assure that
+ * @n does not exceed the queue's capacity.  Returns the number of buffers
+ * allocated.
+ */
+static unsigned int refill_fl(struct adapter *adap, struct sge_fl *q, int n)
+{
+	return refill_fl_usembufs(adap, q, n);
+}
+
+static inline void __refill_fl(struct adapter *adap, struct sge_fl *fl)
+{
+	refill_fl(adap, fl, min(MAX_RX_REFILL, fl_cap(fl) - fl->avail));
+}
+
+/*
+ * Return the number of reclaimable descriptors in a Tx queue.
+ */
+static inline int reclaimable(const struct sge_txq *q)
+{
+	int hw_cidx = ntohs(q->stat->cidx);
+
+	hw_cidx -= q->cidx;
+	if (hw_cidx < 0)
+		return hw_cidx + q->size;
+	return hw_cidx;
+}
+
+/**
+ * reclaim_completed_tx - reclaims completed Tx descriptors
+ * @q: the Tx queue to reclaim completed descriptors from
+ *
+ * Reclaims Tx descriptors that the SGE has indicated it has processed.
+ */
+void reclaim_completed_tx(struct sge_txq *q)
+{
+	unsigned int avail = reclaimable(q);
+
+	do {
+		/* reclaim as much as possible */
+		reclaim_tx_desc(q, avail);
+		q->in_use -= avail;
+		avail = reclaimable(q);
+	} while (avail);
+}
+
+/**
+ * sgl_len - calculates the size of an SGL of the given capacity
+ * @n: the number of SGL entries
+ *
+ * Calculates the number of flits needed for a scatter/gather list that
+ * can hold the given number of entries.
+ */
+static inline unsigned int sgl_len(unsigned int n)
+{
+	/*
+	 * A Direct Scatter Gather List uses 32-bit lengths and 64-bit PCI DMA
+	 * addresses.  The DSGL Work Request starts off with a 32-bit DSGL
+	 * ULPTX header, then Length0, then Address0, then, for 1 <= i <= N,
+	 * repeated sequences of { Length[i], Length[i+1], Address[i],
+	 * Address[i+1] } (this ensures that all addresses are on 64-bit
+	 * boundaries).  If N is even, then Length[N+1] should be set to 0 and
+	 * Address[N+1] is omitted.
+	 *
+	 * The following calculation incorporates all of the above.  It's
+	 * somewhat hard to follow but, briefly: the "+2" accounts for the
+	 * first two flits which include the DSGL header, Length0 and
+	 * Address0; the "(3*(n-1))/2" covers the main body of list entries (3
+	 * flits for every pair of the remaining N) +1 if (n-1) is odd; and
+	 * finally the "+((n-1)&1)" adds the one remaining flit needed if
+	 * (n-1) is odd ...
+	 */
+	n--;
+	return (3 * n) / 2 + (n & 1) + 2;
+}
+
+/**
+ * flits_to_desc - returns the num of Tx descriptors for the given flits
+ * @n: the number of flits
+ *
+ * Returns the number of Tx descriptors needed for the supplied number
+ * of flits.
+ */
+static inline unsigned int flits_to_desc(unsigned int n)
+{
+	return DIV_ROUND_UP(n, 8);
+}
+
+/**
+ * is_eth_imm - can an Ethernet packet be sent as immediate data?
+ * @m: the packet
+ *
+ * Returns whether an Ethernet packet is small enough to fit as
+ * immediate data. Return value corresponds to the headroom required.
+ */
+static inline int is_eth_imm(const struct rte_mbuf *m)
+{
+	unsigned int hdrlen = (m->ol_flags & PKT_TX_TCP_SEG) ?
+			      sizeof(struct cpl_tx_pkt_lso_core) : 0;
+
+	hdrlen += sizeof(struct cpl_tx_pkt);
+	if (m->pkt_len <= MAX_IMM_TX_PKT_LEN - hdrlen)
+		return hdrlen;
+
+	return 0;
+}
+
+/**
+ * calc_tx_flits - calculate the number of flits for a packet Tx WR
+ * @m: the packet
+ *
+ * Returns the number of flits needed for a Tx WR for the given Ethernet
+ * packet, including the needed WR and CPL headers.
+ */
+static inline unsigned int calc_tx_flits(const struct rte_mbuf *m)
+{
+	unsigned int flits;
+	int hdrlen;
+
+	/*
+	 * If the mbuf is small enough, we can pump it out as a work request
+	 * with only immediate data.  In that case we just have to have the
+	 * TX Packet header plus the mbuf data in the Work Request.
+	 */
+
+	hdrlen = is_eth_imm(m);
+	if (hdrlen)
+		return DIV_ROUND_UP(m->pkt_len + hdrlen, sizeof(__be64));
+
+	/*
+	 * Otherwise, we're going to have to construct a Scatter gather list
+	 * of the mbuf body and fragments.  We also include the flits necessary
+	 * for the TX Packet Work Request and CPL.  We always have a firmware
+	 * Write Header (incorporated as part of the cpl_tx_pkt_lso and
+	 * cpl_tx_pkt structures), followed by either a TX Packet Write CPL
+	 * message or, if we're doing a Large Send Offload, an LSO CPL message
+	 * with an embeded TX Packet Write CPL message.
+	 */
+	flits = sgl_len(m->nb_segs);
+	if (m->tso_segsz)
+		flits += (sizeof(struct fw_eth_tx_pkt_wr) +
+			  sizeof(struct cpl_tx_pkt_lso_core) +
+			  sizeof(struct cpl_tx_pkt_core)) / sizeof(__be64);
+	else
+		flits += (sizeof(struct fw_eth_tx_pkt_wr) +
+			  sizeof(struct cpl_tx_pkt_core)) / sizeof(__be64);
+	return flits;
+}
+
+/**
+ * calc_tx_descs - calculate the number of Tx descriptors for a packet
+ * @m: the packet
+ *
+ * Returns the number of Tx descriptors needed for the given Ethernet
+ * packet, including the needed WR and CPL headers.
+ */
+static inline unsigned int calc_tx_descs(const struct rte_mbuf *m)
+{
+	return flits_to_desc(calc_tx_flits(m));
+}
+
+/**
+ * write_sgl - populate a scatter/gather list for a packet
+ * @mbuf: the packet
+ * @q: the Tx queue we are writing into
+ * @sgl: starting location for writing the SGL
+ * @end: points right after the end of the SGL
+ * @start: start offset into mbuf main-body data to include in the SGL
+ * @addr: address of mapped region
+ *
+ * Generates a scatter/gather list for the buffers that make up a packet.
+ * The caller must provide adequate space for the SGL that will be written.
+ * The SGL includes all of the packet's page fragments and the data in its
+ * main body except for the first @start bytes.  @sgl must be 16-byte
+ * aligned and within a Tx descriptor with available space.  @end points
+ * write after the end of the SGL but does not account for any potential
+ * wrap around, i.e., @end > @sgl.
+ */
+static void write_sgl(struct rte_mbuf *mbuf, struct sge_txq *q,
+		      struct ulptx_sgl *sgl, u64 *end, unsigned int start,
+		      const dma_addr_t *addr)
+{
+	unsigned int i, len;
+	struct ulptx_sge_pair *to;
+	struct rte_mbuf *m = mbuf;
+	unsigned int nfrags = m->nb_segs;
+	struct ulptx_sge_pair buf[nfrags / 2];
+
+	len = m->data_len - start;
+	sgl->len0 = htonl(len);
+	sgl->addr0 = rte_cpu_to_be_64(addr[0]);
+
+	sgl->cmd_nsge = htonl(V_ULPTX_CMD(ULP_TX_SC_DSGL) |
+			      V_ULPTX_NSGE(nfrags));
+	if (likely(--nfrags == 0))
+		return;
+	/*
+	 * Most of the complexity below deals with the possibility we hit the
+	 * end of the queue in the middle of writing the SGL.  For this case
+	 * only we create the SGL in a temporary buffer and then copy it.
+	 */
+	to = (u8 *)end > (u8 *)q->stat ? buf : sgl->sge;
+
+	for (i = 0; nfrags >= 2; nfrags -= 2, to++) {
+		m = m->next;
+		to->len[0] = rte_cpu_to_be_32(m->data_len);
+		to->addr[0] = rte_cpu_to_be_64(addr[++i]);
+		m = m->next;
+		to->len[1] = rte_cpu_to_be_32(m->data_len);
+		to->addr[1] = rte_cpu_to_be_64(addr[++i]);
+	}
+	if (nfrags) {
+		m = m->next;
+		to->len[0] = rte_cpu_to_be_32(m->data_len);
+		to->len[1] = rte_cpu_to_be_32(0);
+		to->addr[0] = rte_cpu_to_be_64(addr[i + 1]);
+	}
+	if (unlikely((u8 *)end > (u8 *)q->stat)) {
+		unsigned int part0 = RTE_PTR_DIFF((u8 *)q->stat,
+						  (u8 *)sgl->sge);
+		unsigned int part1;
+
+		if (likely(part0))
+			memcpy(sgl->sge, buf, part0);
+		part1 = RTE_PTR_DIFF((u8 *)end, (u8 *)q->stat);
+		rte_memcpy(q->desc, RTE_PTR_ADD((u8 *)buf, part0), part1);
+		end = RTE_PTR_ADD((void *)q->desc, part1);
+	}
+	if ((uintptr_t)end & 8)           /* 0-pad to multiple of 16 */
+		*(u64 *)end = 0;
+}
+
+#define IDXDIFF(head, tail, wrap) \
+	((head) >= (tail) ? (head) - (tail) : (wrap) - (tail) + (head))
+
+#define Q_IDXDIFF(q, idx) IDXDIFF((q)->pidx, (q)->idx, (q)->size)
+
+/**
+ * ring_tx_db - ring a Tx queue's doorbell
+ * @adap: the adapter
+ * @q: the Tx queue
+ * @n: number of new descriptors to give to HW
+ *
+ * Ring the doorbel for a Tx queue.
+ */
+static inline void ring_tx_db(struct adapter *adap, struct sge_txq *q)
+{
+	int n = Q_IDXDIFF(q, dbidx);
+
+	/*
+	 * Make sure that all writes to the TX Descriptors are committed
+	 * before we tell the hardware about them.
+	 */
+	rte_wmb();
+
+	/*
+	 * If we don't have access to the new User Doorbell (T5+), use the old
+	 * doorbell mechanism; otherwise use the new BAR2 mechanism.
+	 */
+	if (unlikely(!q->bar2_addr)) {
+		u32 val = V_PIDX(n);
+
+		/*
+		 * For T4 we need to participate in the Doorbell Recovery
+		 * mechanism.
+		 */
+		if (!q->db_disabled)
+			t4_write_reg(adap, MYPF_REG(A_SGE_PF_KDOORBELL),
+				     V_QID(q->cntxt_id) | val);
+		else
+			q->db_pidx_inc += n;
+		q->db_pidx = q->pidx;
+	} else {
+		u32 val = V_PIDX_T5(n);
+
+		/*
+		 * T4 and later chips share the same PIDX field offset within
+		 * the doorbell, but T5 and later shrank the field in order to
+		 * gain a bit for Doorbell Priority.  The field was absurdly
+		 * large in the first place (14 bits) so we just use the T5
+		 * and later limits and warn if a Queue ID is too large.
+		 */
+		WARN_ON(val & F_DBPRIO);
+
+		writel(val | V_QID(q->bar2_qid),
+		       (void *)((uintptr_t)q->bar2_addr + SGE_UDB_KDOORBELL));
+
+		/*
+		 * This Write Memory Barrier will force the write to the User
+		 * Doorbell area to be flushed.  This is needed to prevent
+		 * writes on different CPUs for the same queue from hitting
+		 * the adapter out of order.  This is required when some Work
+		 * Requests take the Write Combine Gather Buffer path (user
+		 * doorbell area offset [SGE_UDB_WCDOORBELL..+63]) and some
+		 * take the traditional path where we simply increment the
+		 * PIDX (User Doorbell area SGE_UDB_KDOORBELL) and have the
+		 * hardware DMA read the actual Work Request.
+		 */
+		rte_wmb();
+	}
+	q->dbidx = q->pidx;
+}
+
+/*
+ * Figure out what HW csum a packet wants and return the appropriate control
+ * bits.
+ */
+static u64 hwcsum(enum chip_type chip, const struct rte_mbuf *m)
+{
+	int csum_type;
+
+	if (m->ol_flags & PKT_TX_IP_CKSUM) {
+		switch (m->ol_flags & PKT_TX_L4_MASK) {
+		case PKT_TX_TCP_CKSUM:
+			csum_type = TX_CSUM_TCPIP;
+			break;
+		case PKT_TX_UDP_CKSUM:
+			csum_type = TX_CSUM_UDPIP;
+			break;
+		default:
+			goto nocsum;
+		}
+	} else {
+		goto nocsum;
+	}
+
+	if (likely(csum_type >= TX_CSUM_TCPIP)) {
+		int hdr_len = V_TXPKT_IPHDR_LEN(m->l3_len);
+		int eth_hdr_len = m->l2_len;
+
+		if (CHELSIO_CHIP_VERSION(chip) <= CHELSIO_T5)
+			hdr_len |= V_TXPKT_ETHHDR_LEN(eth_hdr_len);
+		else
+			hdr_len |= V_T6_TXPKT_ETHHDR_LEN(eth_hdr_len);
+		return V_TXPKT_CSUM_TYPE(csum_type) | hdr_len;
+	}
+nocsum:
+	/*
+	 * unknown protocol, disable HW csum
+	 * and hope a bad packet is detected
+	 */
+	return F_TXPKT_L4CSUM_DIS;
+}
+
+static inline void txq_advance(struct sge_txq *q, unsigned int n)
+{
+	q->in_use += n;
+	q->pidx += n;
+	if (q->pidx >= q->size)
+		q->pidx -= q->size;
+}
+
+#define MAX_COALESCE_LEN 64000
+
+static inline int wraps_around(struct sge_txq *q, int ndesc)
+{
+	return (q->pidx + ndesc) > q->size ? 1 : 0;
+}
+
+/**
+ * ship_tx_pkt_coalesce_wr - finalizes and ships a coalesce WR
+ * @ adap: adapter structure
+ * @txq: tx queue
+ * @left: how many more left
+ *
+ * writes the different fields of the pkts WR and sends it.
+ */
+static inline void ship_tx_pkt_coalesce_wr(struct adapter *adap,
+					   struct sge_eth_txq *txq,
+					   uint16_t left)
+{
+	u32 wr_mid;
+	struct sge_txq *q = &txq->q;
+	struct fw_eth_tx_pkts_wr *wr;
+	unsigned int ndesc;
+
+	/* fill the pkts WR header */
+	wr = (void *)&q->desc[q->pidx];
+	wr->op_pkd = htonl(V_FW_WR_OP(FW_ETH_TX_PKTS_WR));
+
+	wr_mid = V_FW_WR_LEN16(DIV_ROUND_UP(q->coalesce.flits, 2));
+	ndesc = flits_to_desc(q->coalesce.flits);
+	wr->equiq_to_len16 = htonl(wr_mid);
+	wr->plen = cpu_to_be16(q->coalesce.len);
+	wr->npkt = q->coalesce.idx;
+	wr->r3 = 0;
+	wr->type = q->coalesce.type;
+
+	/* zero out coalesce structure members */
+	q->coalesce.idx = 0;
+	q->coalesce.flits = 0;
+	q->coalesce.len = 0;
+
+	txq_advance(q, ndesc);
+	txq->stats.coal_wr++;
+	txq->stats.coal_pkts += wr->npkt;
+
+	if (Q_IDXDIFF(q, equeidx) >= q->size / 2) {
+		q->equeidx = q->pidx;
+		wr_mid |= F_FW_WR_EQUEQ;
+		wr->equiq_to_len16 = htonl(wr_mid);
+	}
+
+	if (left == 0) {
+		ring_tx_db(adap, q);
+		return;
+	}
+
+	if (Q_IDXDIFF(q, dbidx) >= SGE_MAX_WR_NDESC * 2)
+		ring_tx_db(adap, q);
+}
+
+/**
+ * should_tx_packet_coalesce - decides wether to coalesce an mbuf or not
+ * @txq: tx queue where the mbuf is sent
+ * @mbuf: mbuf to be sent
+ * @nflits: return value for number of flits needed
+ * @adap: adapter structure
+ * @left: how many more left
+ *
+ * This function decides if a packet should be coalesced or not.
+ */
+static inline int should_tx_packet_coalesce(struct sge_eth_txq *txq,
+					    struct rte_mbuf *mbuf,
+					    unsigned int *nflits,
+					    struct adapter *adap,
+					    uint16_t left)
+{
+	struct sge_txq *q = &txq->q;
+	unsigned int flits, ndesc;
+	unsigned char type = 0;
+	int credits, hw_cidx = ntohs(q->stat->cidx);
+	int in_use = q->pidx - hw_cidx + flits_to_desc(q->coalesce.flits);
+
+	/* use coal WR type 1 when no frags are present */
+	type = (mbuf->nb_segs == 1) ? 1 : 0;
+
+	if (in_use < 0)
+		in_use += q->size;
+
+	if (unlikely(type != q->coalesce.type && q->coalesce.idx))
+		ship_tx_pkt_coalesce_wr(adap, txq, left);
+
+	/* calculate the number of flits required for coalescing this packet
+	 * without the 2 flits of the WR header. These are added further down
+	 * if we are just starting in new PKTS WR. sgl_len doesn't account for
+	 * the possible 16 bytes alignment ULP TX commands so we do it here.
+	 */
+	flits = (sgl_len(mbuf->nb_segs) + 1) & ~1U;
+	if (type == 0)
+		flits += (sizeof(struct ulp_txpkt) +
+			  sizeof(struct ulptx_idata)) / sizeof(__be64);
+	flits += sizeof(struct cpl_tx_pkt_core) / sizeof(__be64);
+	*nflits = flits;
+
+	/* If coalescing is on, the mbuf is added to a pkts WR */
+	if (q->coalesce.idx) {
+		ndesc = DIV_ROUND_UP(q->coalesce.flits + flits, 8);
+		credits = txq_avail(q) - ndesc;
+
+		/* If we are wrapping or this is last mbuf then, send the
+		 * already coalesced mbufs and let the non-coalesce pass
+		 * handle the mbuf.
+		 */
+		if (unlikely(credits < 0 || wraps_around(q, ndesc))) {
+			ship_tx_pkt_coalesce_wr(adap, txq, left);
+			return 0;
+		}
+
+		/* If the max coalesce len or the max WR len is reached
+		 * ship the WR and keep coalescing on.
+		 */
+		if (unlikely((q->coalesce.len + mbuf->pkt_len >
+						MAX_COALESCE_LEN) ||
+			     (q->coalesce.flits + flits >
+			      q->coalesce.max))) {
+			ship_tx_pkt_coalesce_wr(adap, txq, left);
+			goto new;
+		}
+		return 1;
+	}
+
+new:
+	/* start a new pkts WR, the WR header is not filled below */
+	flits += sizeof(struct fw_eth_tx_pkts_wr) / sizeof(__be64);
+	ndesc = flits_to_desc(q->coalesce.flits + flits);
+	credits = txq_avail(q) - ndesc;
+
+	if (unlikely(credits < 0 || wraps_around(q, ndesc)))
+		return 0;
+	q->coalesce.flits += 2;
+	q->coalesce.type = type;
+	q->coalesce.ptr = (unsigned char *)&q->desc[q->pidx] +
+			   2 * sizeof(__be64);
+	return 1;
+}
+
+/**
+ * tx_do_packet_coalesce - add an mbuf to a coalesce WR
+ * @txq: sge_eth_txq used send the mbuf
+ * @mbuf: mbuf to be sent
+ * @flits: flits needed for this mbuf
+ * @adap: adapter structure
+ * @pi: port_info structure
+ * @addr: mapped address of the mbuf
+ * @left: how many more left
+ *
+ * Adds an mbuf to be sent as part of a coalesce WR by filling a
+ * ulp_tx_pkt command, ulp_tx_sc_imm command, cpl message and
+ * ulp_tx_sc_dsgl command.
+ */
+static inline int tx_do_packet_coalesce(struct sge_eth_txq *txq,
+					struct rte_mbuf *mbuf,
+					int flits, struct adapter *adap,
+					const struct port_info *pi,
+					dma_addr_t *addr,
+					uint16_t left)
+{
+	u64 cntrl, *end;
+	struct sge_txq *q = &txq->q;
+	struct ulp_txpkt *mc;
+	struct ulptx_idata *sc_imm;
+	struct cpl_tx_pkt_core *cpl;
+	struct tx_sw_desc *sd;
+	unsigned int idx = q->coalesce.idx, len = mbuf->pkt_len;
+
+	if (q->coalesce.type == 0) {
+		mc = (struct ulp_txpkt *)q->coalesce.ptr;
+		mc->cmd_dest = htonl(V_ULPTX_CMD(4) | V_ULP_TXPKT_DEST(0) |
+				     V_ULP_TXPKT_FID(adap->sge.fw_evtq.cntxt_id) |
+				     F_ULP_TXPKT_RO);
+		mc->len = htonl(DIV_ROUND_UP(flits, 2));
+		sc_imm = (struct ulptx_idata *)(mc + 1);
+		sc_imm->cmd_more = htonl(V_ULPTX_CMD(ULP_TX_SC_IMM) |
+					 F_ULP_TX_SC_MORE);
+		sc_imm->len = htonl(sizeof(*cpl));
+		end = (u64 *)mc + flits;
+		cpl = (struct cpl_tx_pkt_core *)(sc_imm + 1);
+	} else {
+		end = (u64 *)q->coalesce.ptr + flits;
+		cpl = (struct cpl_tx_pkt_core *)q->coalesce.ptr;
+	}
+
+	/* update coalesce structure for this txq */
+	q->coalesce.flits += flits;
+	q->coalesce.ptr += flits * sizeof(__be64);
+	q->coalesce.len += mbuf->pkt_len;
+
+	/* fill the cpl message, same as in t4_eth_xmit, this should be kept
+	 * similar to t4_eth_xmit
+	 */
+	if (mbuf->ol_flags & PKT_TX_IP_CKSUM) {
+		cntrl = hwcsum(adap->params.chip, mbuf) |
+			       F_TXPKT_IPCSUM_DIS;
+		txq->stats.tx_cso++;
+	} else {
+		cntrl = F_TXPKT_L4CSUM_DIS | F_TXPKT_IPCSUM_DIS;
+	}
+
+	if (mbuf->ol_flags & PKT_TX_VLAN_PKT) {
+		txq->stats.vlan_ins++;
+		cntrl |= F_TXPKT_VLAN_VLD | V_TXPKT_VLAN(mbuf->vlan_tci);
+	}
+
+	cpl->ctrl0 = htonl(V_TXPKT_OPCODE(CPL_TX_PKT_XT) |
+			   V_TXPKT_INTF(pi->tx_chan) |
+			   V_TXPKT_PF(adap->pf));
+	cpl->pack = htons(0);
+	cpl->len = htons(len);
+	cpl->ctrl1 = cpu_to_be64(cntrl);
+	write_sgl(mbuf, q, (struct ulptx_sgl *)(cpl + 1), end, 0,  addr);
+	txq->stats.pkts++;
+	txq->stats.tx_bytes += len;
+
+	sd = &q->sdesc[q->pidx + (idx >> 1)];
+	if (!(idx & 1)) {
+		if (sd->coalesce.idx) {
+			int i;
+
+			for (i = 0; i < sd->coalesce.idx; i++) {
+				rte_pktmbuf_free(sd->coalesce.mbuf[i]);
+				sd->coalesce.mbuf[i] = NULL;
+			}
+		}
+	}
+
+	/* store pointers to the mbuf and the sgl used in free_tx_desc.
+	 * each tx desc can hold two pointers corresponding to the value
+	 * of ETH_COALESCE_PKT_PER_DESC
+	 */
+	sd->coalesce.mbuf[idx & 1] = mbuf;
+	sd->coalesce.sgl[idx & 1] = (struct ulptx_sgl *)(cpl + 1);
+	sd->coalesce.idx = (idx & 1) + 1;
+
+	/* send the coaelsced work request if max reached */
+	if (++q->coalesce.idx == ETH_COALESCE_PKT_NUM || left == 0)
+		ship_tx_pkt_coalesce_wr(adap, txq, left);
+	return 0;
+}
+
+/**
+ * t4_eth_xmit - add a packet to an Ethernet Tx queue
+ * @txq: the egress queue
+ * @mbuf: the packet
+ * @left: how many more left
+ *
+ * Add a packet to an SGE Ethernet Tx queue.  Runs with softirqs disabled.
+ */
+int t4_eth_xmit(struct sge_eth_txq *txq, struct rte_mbuf *mbuf, uint16_t left)
+{
+	const struct port_info *pi;
+	struct cpl_tx_pkt_lso_core *lso;
+	struct adapter *adap;
+	struct rte_mbuf *m = mbuf;
+	struct fw_eth_tx_pkt_wr *wr;
+	struct cpl_tx_pkt_core *cpl;
+	struct tx_sw_desc *d;
+	dma_addr_t addr[m->nb_segs];
+	unsigned int flits, ndesc, cflits;
+	int l3hdr_len, l4hdr_len, eth_xtra_len;
+	int len, last_desc;
+	int credits;
+	u32 wr_mid;
+	u64 cntrl, *end;
+	bool v6;
+
+	/* Reject xmit if queue is stopped */
+	if (unlikely(txq->flags & EQ_STOPPED))
+		return -(EBUSY);
+
+	/*
+	 * The chip min packet length is 10 octets but play safe and reject
+	 * anything shorter than an Ethernet header.
+	 */
+	if (unlikely(m->pkt_len < ETHER_HDR_LEN)) {
+out_free:
+		rte_pktmbuf_free(m);
+		return 0;
+	}
+
+	rte_prefetch0(&((&txq->q)->sdesc->mbuf->pool));
+	pi = (struct port_info *)txq->eth_dev->data->dev_private;
+	adap = pi->adapter;
+
+	cntrl = F_TXPKT_L4CSUM_DIS | F_TXPKT_IPCSUM_DIS;
+	/* align the end of coalesce WR to a 512 byte boundary */
+	txq->q.coalesce.max = (8 - (txq->q.pidx & 7)) * 8;
+
+	if (!(m->ol_flags & PKT_TX_TCP_SEG)) {
+		if (should_tx_packet_coalesce(txq, mbuf, &cflits, adap, left)) {
+			if (unlikely(map_mbuf(mbuf, addr) < 0)) {
+				dev_warn(adap, "%s: mapping err for coalesce\n",
+					 __func__);
+				txq->stats.mapping_err++;
+				goto out_free;
+			}
+			return tx_do_packet_coalesce(txq, mbuf, cflits, adap,
+						     pi, addr, left);
+		} else {
+			return -EBUSY;
+		}
+	}
+
+	if (txq->q.coalesce.idx)
+		ship_tx_pkt_coalesce_wr(adap, txq, left);
+
+	flits = calc_tx_flits(m);
+	ndesc = flits_to_desc(flits);
+	credits = txq_avail(&txq->q) - ndesc;
+
+	if (unlikely(credits < 0)) {
+		dev_debug(adap, "%s: Tx ring %u full; credits = %d\n",
+			  __func__, txq->q.cntxt_id, credits);
+		return -EBUSY;
+	}
+
+	if (unlikely(map_mbuf(m, addr) < 0)) {
+		txq->stats.mapping_err++;
+		goto out_free;
+	}
+
+	wr_mid = V_FW_WR_LEN16(DIV_ROUND_UP(flits, 2));
+	if (Q_IDXDIFF(&txq->q, equeidx)  >= 64 || left == 0) {
+		txq->q.equeidx = txq->q.pidx;
+		wr_mid |= F_FW_WR_EQUEQ;
+	}
+
+	wr = (void *)&txq->q.desc[txq->q.pidx];
+	wr->equiq_to_len16 = htonl(wr_mid);
+	wr->r3 = rte_cpu_to_be_64(0);
+	end = (u64 *)wr + flits;
+
+	len = 0;
+	len += sizeof(*cpl);
+	lso = (void *)(wr + 1);
+	v6 = (m->ol_flags & PKT_TX_IPV6) != 0;
+	l3hdr_len = m->l3_len;
+	l4hdr_len = m->l4_len;
+	eth_xtra_len = m->l2_len - ETHER_HDR_LEN;
+	len += sizeof(*lso);
+	wr->op_immdlen = htonl(V_FW_WR_OP(FW_ETH_TX_PKT_WR) |
+			       V_FW_WR_IMMDLEN(len));
+	lso->lso_ctrl = htonl(V_LSO_OPCODE(CPL_TX_PKT_LSO) |
+			      F_LSO_FIRST_SLICE | F_LSO_LAST_SLICE |
+			      V_LSO_IPV6(v6) |
+			      V_LSO_ETHHDR_LEN(eth_xtra_len / 4) |
+			      V_LSO_IPHDR_LEN(l3hdr_len / 4) |
+			      V_LSO_TCPHDR_LEN(l4hdr_len / 4));
+	lso->ipid_ofst = htons(0);
+	lso->mss = htons(m->tso_segsz);
+	lso->seqno_offset = htonl(0);
+	if (is_t4(adap->params.chip))
+		lso->len = htonl(m->pkt_len);
+	else
+		lso->len = htonl(V_LSO_T5_XFER_SIZE(m->pkt_len));
+	cpl = (void *)(lso + 1);
+	cntrl = V_TXPKT_CSUM_TYPE(v6 ? TX_CSUM_TCPIP6 : TX_CSUM_TCPIP) |
+				  V_TXPKT_IPHDR_LEN(l3hdr_len) |
+				  V_TXPKT_ETHHDR_LEN(eth_xtra_len);
+	txq->stats.tso++;
+	txq->stats.tx_cso += m->tso_segsz;
+
+	if (m->ol_flags & PKT_TX_VLAN_PKT) {
+		txq->stats.vlan_ins++;
+		cntrl |= F_TXPKT_VLAN_VLD | V_TXPKT_VLAN(m->vlan_tci);
+	}
+
+	cpl->ctrl0 = htonl(V_TXPKT_OPCODE(CPL_TX_PKT_XT) |
+			   V_TXPKT_INTF(pi->tx_chan) |
+			   V_TXPKT_PF(adap->pf));
+	cpl->pack = htons(0);
+	cpl->len = htons(m->pkt_len);
+	cpl->ctrl1 = cpu_to_be64(cntrl);
+
+	txq->stats.pkts++;
+	txq->stats.tx_bytes += m->pkt_len;
+	last_desc = txq->q.pidx + ndesc - 1;
+	if (last_desc >= (int)txq->q.size)
+		last_desc -= txq->q.size;
+
+	d = &txq->q.sdesc[last_desc];
+	if (d->mbuf) {
+		rte_pktmbuf_free(d->mbuf);
+		d->mbuf = NULL;
+	}
+	write_sgl(m, &txq->q, (struct ulptx_sgl *)(cpl + 1), end, 0,
+		  addr);
+	txq->q.sdesc[last_desc].mbuf = m;
+	txq->q.sdesc[last_desc].sgl = (struct ulptx_sgl *)(cpl + 1);
+	txq_advance(&txq->q, ndesc);
+	ring_tx_db(adap, &txq->q);
+	return 0;
+}
+
+/**
+ * alloc_ring - allocate resources for an SGE descriptor ring
+ * @dev: the PCI device's core device
+ * @nelem: the number of descriptors
+ * @elem_size: the size of each descriptor
+ * @sw_size: the size of the SW state associated with each ring element
+ * @phys: the physical address of the allocated ring
+ * @metadata: address of the array holding the SW state for the ring
+ * @stat_size: extra space in HW ring for status information
+ * @node: preferred node for memory allocations
+ *
+ * Allocates resources for an SGE descriptor ring, such as Tx queues,
+ * free buffer lists, or response queues.  Each SGE ring requires
+ * space for its HW descriptors plus, optionally, space for the SW state
+ * associated with each HW entry (the metadata).  The function returns
+ * three values: the virtual address for the HW ring (the return value
+ * of the function), the bus address of the HW ring, and the address
+ * of the SW ring.
+ */
+static void *alloc_ring(size_t nelem, size_t elem_size,
+			size_t sw_size, dma_addr_t *phys, void *metadata,
+			size_t stat_size, __rte_unused uint16_t queue_id,
+			int socket_id, const char *z_name,
+			const char *z_name_sw)
+{
+	size_t len = CXGBE_MAX_RING_DESC_SIZE * elem_size + stat_size;
+	const struct rte_memzone *tz;
+	void *s = NULL;
+
+	dev_debug(adapter, "%s: nelem = %lu; elem_size = %lu; sw_size = %lu; "
+		  "stat_size = %lu; queue_id = %u; socket_id = %d; z_name = %s;"
+		  " z_name_sw = %s\n", __func__, nelem, elem_size, sw_size,
+		  stat_size, queue_id, socket_id, z_name, z_name_sw);
+
+	tz = rte_memzone_lookup(z_name);
+	if (tz) {
+		dev_debug(adapter, "%s: tz exists...returning existing..\n",
+			  __func__);
+		goto alloc_sw_ring;
+	}
+
+	/*
+	 * Allocate TX/RX ring hardware descriptors. A memzone large enough to
+	 * handle the maximum ring size is allocated in order to allow for
+	 * resizing in later calls to the queue setup function.
+	 */
+	tz = rte_memzone_reserve_aligned(z_name, len, socket_id, 0, 4096);
+	if (!tz)
+		return NULL;
+
+alloc_sw_ring:
+	memset(tz->addr, 0, len);
+	if (sw_size) {
+		s = rte_zmalloc_socket(z_name_sw, nelem * sw_size,
+				       RTE_CACHE_LINE_SIZE, socket_id);
+
+		if (!s) {
+			dev_err(adapter, "%s: failed to get sw_ring memory\n",
+				__func__);
+			return NULL;
+		}
+	}
+	if (metadata)
+		*(void **)metadata = s;
+
+	*phys = (uint64_t)tz->phys_addr;
+	return tz->addr;
+}
+
+/**
+ * t4_pktgl_to_mbuf_usembufs - build an mbuf from a packet gather list
+ * @gl: the gather list
+ *
+ * Builds an mbuf from the given packet gather list.  Returns the mbuf or
+ * %NULL if mbuf allocation failed.
+ */
+static struct rte_mbuf *t4_pktgl_to_mbuf_usembufs(const struct pkt_gl *gl)
+{
+	/*
+	 * If there's only one mbuf fragment, just return that.
+	 */
+	if (likely(gl->nfrags == 1))
+		return gl->mbufs[0];
+
+	return NULL;
+}
+
+/**
+ * t4_pktgl_to_mbuf - build an mbuf from a packet gather list
+ * @gl: the gather list
+ *
+ * Builds an mbuf from the given packet gather list.  Returns the mbuf or
+ * %NULL if mbuf allocation failed.
+ */
+static struct rte_mbuf *t4_pktgl_to_mbuf(const struct pkt_gl *gl)
+{
+	return t4_pktgl_to_mbuf_usembufs(gl);
+}
+
+#define RTE_MBUF_DATA_DMA_ADDR_DEFAULT(mb) \
+	((dma_addr_t) ((mb)->buf_physaddr + (mb)->data_off))
+
+/**
+ * t4_ethrx_handler - process an ingress ethernet packet
+ * @q: the response queue that received the packet
+ * @rsp: the response queue descriptor holding the RX_PKT message
+ * @si: the gather list of packet fragments
+ *
+ * Process an ingress ethernet packet and deliver it to the stack.
+ */
+int t4_ethrx_handler(struct sge_rspq *q, const __be64 *rsp,
+		     const struct pkt_gl *si)
+{
+	struct rte_mbuf *mbuf;
+	const struct cpl_rx_pkt *pkt;
+	const struct rss_header *rss_hdr;
+	bool csum_ok;
+	struct sge_eth_rxq *rxq = container_of(q, struct sge_eth_rxq, rspq);
+	struct adapter *adapter = q->adapter;
+	struct sge *s = &adapter->sge;
+
+	rss_hdr = (const void *)rsp;
+	pkt = (const void *)&rsp[1];
+	csum_ok = pkt->csum_calc && !pkt->err_vec;
+
+	mbuf = t4_pktgl_to_mbuf(si);
+	if (unlikely(!mbuf)) {
+		rxq->stats.rx_drops++;
+		return 0;
+	}
+
+	rte_pktmbuf_adj(mbuf, s->pktshift);
+	mbuf->port = pkt->iff;
+	if (pkt->l2info & htonl(F_RXF_IP)) {
+		mbuf->ol_flags |= PKT_RX_IPV4_HDR;
+		if (unlikely(!csum_ok))
+			mbuf->ol_flags |= PKT_RX_IP_CKSUM_BAD;
+
+		if ((pkt->l2info & htonl(F_RXF_UDP | F_RXF_TCP)) && !csum_ok)
+			mbuf->ol_flags |= PKT_RX_L4_CKSUM_BAD;
+	} else if (pkt->l2info & htonl(F_RXF_IP6)) {
+		mbuf->ol_flags |= PKT_RX_IPV6_HDR;
+	}
+
+	mbuf->port = pkt->iff;
+
+	if (!rss_hdr->filter_tid && rss_hdr->hash_type) {
+		mbuf->ol_flags |= PKT_RX_RSS_HASH;
+		mbuf->hash.rss = ntohl(rss_hdr->hash_val);
+	}
+
+	if (pkt->vlan_ex) {
+		mbuf->ol_flags |= PKT_RX_VLAN_PKT;
+		mbuf->vlan_tci = ntohs(pkt->vlan);
+	}
+	rxq->stats.pkts++;
+	rxq->stats.rx_bytes += mbuf->pkt_len;
+
+	return 0;
+}
+
+/**
+ * restore_rx_bufs - put back a packet's Rx buffers
+ * @q: the SGE free list
+ * @frags: number of FL buffers to restore
+ *
+ * Puts back on an FL the Rx buffers.  The buffers have already been
+ * unmapped and are left unmapped, we mark them so to prevent further
+ * unmapping attempts.
+ *
+ * This function undoes a series of @unmap_rx_buf calls when we find out
+ * that the current packet can't be processed right away afterall and we
+ * need to come back to it later.  This is a very rare event and there's
+ * no effort to make this particularly efficient.
+ */
+static void restore_rx_bufs(struct sge_fl *q, int frags)
+{
+	while (frags--) {
+		if (q->cidx == 0)
+			q->cidx = q->size - 1;
+		else
+			q->cidx--;
+		q->avail++;
+	}
+}
+
+/**
+ * is_new_response - check if a response is newly written
+ * @r: the response descriptor
+ * @q: the response queue
+ *
+ * Returns true if a response descriptor contains a yet unprocessed
+ * response.
+ */
+static inline bool is_new_response(const struct rsp_ctrl *r,
+				   const struct sge_rspq *q)
+{
+	return (r->u.type_gen >> S_RSPD_GEN) == q->gen;
+}
+
+#define CXGB4_MSG_AN ((void *)1)
+
+/**
+ * rspq_next - advance to the next entry in a response queue
+ * @q: the queue
+ *
+ * Updates the state of a response queue to advance it to the next entry.
+ */
+static inline void rspq_next(struct sge_rspq *q)
+{
+	q->cur_desc = (const __be64 *)((const char *)q->cur_desc + q->iqe_len);
+	if (unlikely(++q->cidx == q->size)) {
+		q->cidx = 0;
+		q->gen ^= 1;
+		q->cur_desc = q->desc;
+	}
+}
+
+/**
+ * process_responses - process responses from an SGE response queue
+ * @q: the ingress queue to process
+ * @budget: how many responses can be processed in this round
+ * @rx_pkts: mbuf to put the pkts
+ *
+ * Process responses from an SGE response queue up to the supplied budget.
+ * Responses include received packets as well as control messages from FW
+ * or HW.
+ *
+ * Additionally choose the interrupt holdoff time for the next interrupt
+ * on this queue.  If the system is under memory shortage use a fairly
+ * long delay to help recovery.
+ */
+static int process_responses(struct sge_rspq *q, int budget,
+			     struct rte_mbuf **rx_pkts)
+{
+	int ret = 0, rsp_type;
+	int budget_left = budget;
+	const struct rsp_ctrl *rc;
+	struct sge_eth_rxq *rxq = container_of(q, struct sge_eth_rxq, rspq);
+	struct adapter *adapter = q->adapter;
+
+	while (likely(budget_left)) {
+		rc = (const struct rsp_ctrl *)
+		     ((const char *)q->cur_desc + (q->iqe_len - sizeof(*rc)));
+
+		if (!is_new_response(rc, q))
+			break;
+
+		/*
+		 * Ensure response has been read
+		 */
+		rmb();
+		rsp_type = G_RSPD_TYPE(rc->u.type_gen);
+
+		if (likely(rsp_type == X_RSPD_TYPE_FLBUF)) {
+			struct pkt_gl si;
+			const struct rx_sw_desc *rsd;
+			struct rte_mbuf *pkt = NULL;
+			u32 len = ntohl(rc->pldbuflen_qid), bufsz, frags;
+
+			si.usembufs = rxq->usembufs;
+			/*
+			 * In "use mbufs" mode, we don't pack multiple
+			 * ingress packets per buffer (mbuf) so we
+			 * should _always_ get a "New Buffer" flags
+			 * from the SGE.  Also, since we hand the
+			 * mbuf's up to the host stack for it to
+			 * eventually free, we don't release the mbuf's
+			 * in the driver (in contrast to the "packed
+			 * page" mode where the driver needs to
+			 * release its reference on the page buffers).
+			 */
+			BUG_ON(!(len & F_RSPD_NEWBUF));
+			len = G_RSPD_LEN(len);
+			si.tot_len = len;
+
+			/* gather packet fragments */
+			for (frags = 0; len; frags++) {
+				rsd = &rxq->fl.sdesc[rxq->fl.cidx];
+				bufsz = min(get_buf_size(adapter, rsd),	len);
+				pkt = rsd->buf;
+				pkt->data_len = bufsz;
+				pkt->pkt_len = bufsz;
+				si.mbufs[frags] = pkt;
+				len -= bufsz;
+				unmap_rx_buf(&rxq->fl);
+			}
+
+			si.va = RTE_PTR_ADD(si.mbufs[0]->buf_addr,
+					    si.mbufs[0]->data_off);
+			rte_prefetch1(si.va);
+
+			/*
+			 * For the "use mbuf" case here, we can end up
+			 * chewing through our Free List very rapidly
+			 * with one entry per Ingress packet getting
+			 * consumed.  So if the handler() successfully
+			 * consumed the mbuf, check to see if we can
+			 * refill the Free List incrementally in the
+			 * loop ...
+			 */
+			si.nfrags = frags;
+			ret = q->handler(q, q->cur_desc, &si);
+
+			if (unlikely(ret != 0)) {
+				restore_rx_bufs(&rxq->fl, frags);
+			} else {
+				rx_pkts[budget - budget_left] = pkt;
+				if (fl_cap(&rxq->fl) - rxq->fl.avail >= 8)
+					__refill_fl(q->adapter, &rxq->fl);
+			}
+
+		} else if (likely(rsp_type == X_RSPD_TYPE_CPL)) {
+			ret = q->handler(q, q->cur_desc, NULL);
+		} else {
+			ret = q->handler(q, (const __be64 *)rc, CXGB4_MSG_AN);
+		}
+
+		if (unlikely(ret)) {
+			/* couldn't process descriptor, back off for recovery */
+			q->next_intr_params = V_QINTR_TIMER_IDX(NOMEM_TMR_IDX);
+			break;
+		}
+
+		rspq_next(q);
+		budget_left--;
+	}
+
+	/*
+	 * If this is a Response Queue with an associated Free List and
+	 * there's room for another chunk of new Free List buffer pointers,
+	 * refill the Free List.
+	 */
+
+	if (q->offset >= 0 && fl_cap(&rxq->fl) - rxq->fl.avail >= 8)
+		__refill_fl(q->adapter, &rxq->fl);
+
+	return budget - budget_left;
+}
+
+int cxgbe_poll(struct sge_rspq *q, struct rte_mbuf **rx_pkts,
+	       unsigned int budget, unsigned int *work_done)
+{
+	unsigned int params;
+	u32 val;
+	int err = 0;
+
+	*work_done = process_responses(q, budget, rx_pkts);
+	params = V_QINTR_TIMER_IDX(X_TIMERREG_UPDATE_CIDX);
+	q->next_intr_params = params;
+	val = V_CIDXINC(*work_done) | V_SEINTARM(params);
+
+	if (*work_done) {
+		/*
+		 * If we don't have access to the new User GTS (T5+),
+		 * use the old doorbell mechanism; otherwise use the new
+		 * BAR2 mechanism.
+		 */
+		if (unlikely(!q->bar2_addr))
+			t4_write_reg(q->adapter, MYPF_REG(A_SGE_PF_GTS),
+				     val | V_INGRESSQID((u32)q->cntxt_id));
+		else {
+			writel(val | V_INGRESSQID(q->bar2_qid),
+			       (void *)((uintptr_t)q->bar2_addr +
+			       SGE_UDB_GTS));
+			/*
+			 * This Write memory Barrier will force the write to
+			 * the User Doorbell area to be flushed.
+			 */
+			wmb();
+		}
+	}
+
+	return err;
+}
+
+/**
+ * bar2_address - return the BAR2 address for an SGE Queue's Registers
+ * @adapter: the adapter
+ * @qid: the SGE Queue ID
+ * @qtype: the SGE Queue Type (Egress or Ingress)
+ * @pbar2_qid: BAR2 Queue ID or 0 for Queue ID inferred SGE Queues
+ *
+ * Returns the BAR2 address for the SGE Queue Registers associated with
+ * @qid.  If BAR2 SGE Registers aren't available, returns NULL.  Also
+ * returns the BAR2 Queue ID to be used with writes to the BAR2 SGE
+ * Queue Registers.  If the BAR2 Queue ID is 0, then "Inferred Queue ID"
+ * Registers are supported (e.g. the Write Combining Doorbell Buffer).
+ */
+static void __iomem *bar2_address(struct adapter *adapter, unsigned int qid,
+				  enum t4_bar2_qtype qtype,
+				  unsigned int *pbar2_qid)
+{
+	u64 bar2_qoffset;
+	int ret;
+
+	ret = t4_bar2_sge_qregs(adapter, qid, qtype, &bar2_qoffset, pbar2_qid);
+	if (ret)
+		return NULL;
+
+	return adapter->bar2 + bar2_qoffset;
+}
+
+int t4_sge_eth_rxq_start(struct adapter *adap, struct sge_rspq *rq)
+{
+	struct sge_eth_rxq *rxq = container_of(rq, struct sge_eth_rxq, rspq);
+	unsigned int fl_id = rxq->fl.size ? rxq->fl.cntxt_id : 0xffff;
+
+	return t4_iq_start_stop(adap, adap->mbox, true, adap->pf, 0,
+				rq->cntxt_id, fl_id, 0xffff);
+}
+
+int t4_sge_eth_rxq_stop(struct adapter *adap, struct sge_rspq *rq)
+{
+	struct sge_eth_rxq *rxq = container_of(rq, struct sge_eth_rxq, rspq);
+	unsigned int fl_id = rxq->fl.size ? rxq->fl.cntxt_id : 0xffff;
+
+	return t4_iq_start_stop(adap, adap->mbox, false, adap->pf, 0,
+				rq->cntxt_id, fl_id, 0xffff);
+}
+
+/*
+ * @intr_idx: MSI/MSI-X vector if >=0, -(absolute qid + 1) if < 0
+ * @cong: < 0 -> no congestion feedback, >= 0 -> congestion channel map
+ */
+int t4_sge_alloc_rxq(struct adapter *adap, struct sge_rspq *iq, bool fwevtq,
+		     struct rte_eth_dev *eth_dev, int intr_idx,
+		     struct sge_fl *fl, rspq_handler_t hnd, int cong,
+		     struct rte_mempool *mp, int queue_id, int socket_id)
+{
+	int ret, flsz = 0;
+	struct fw_iq_cmd c;
+	struct sge *s = &adap->sge;
+	struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
+	char z_name[RTE_MEMZONE_NAMESIZE];
+	char z_name_sw[RTE_MEMZONE_NAMESIZE];
+	unsigned int nb_refill;
+
+	/* Size needs to be multiple of 16, including status entry. */
+	iq->size = roundup(iq->size, 16);
+
+	snprintf(z_name, sizeof(z_name), "%s_%s_%d_%d",
+		 eth_dev->driver->pci_drv.name, fwevtq ? "fwq_ring" : "rx_ring",
+		 eth_dev->data->port_id, queue_id);
+	snprintf(z_name_sw, sizeof(z_name_sw), "%s_sw_ring", z_name);
+
+	iq->desc = alloc_ring(iq->size, iq->iqe_len, 0, &iq->phys_addr, NULL, 0,
+			      queue_id, socket_id, z_name, z_name_sw);
+	if (!iq->desc)
+		return -ENOMEM;
+
+	memset(&c, 0, sizeof(c));
+	c.op_to_vfn = htonl(V_FW_CMD_OP(FW_IQ_CMD) | F_FW_CMD_REQUEST |
+			    F_FW_CMD_WRITE | F_FW_CMD_EXEC |
+			    V_FW_IQ_CMD_PFN(adap->pf) | V_FW_IQ_CMD_VFN(0));
+	c.alloc_to_len16 = htonl(F_FW_IQ_CMD_ALLOC | F_FW_IQ_CMD_IQSTART |
+				 (sizeof(c) / 16));
+	c.type_to_iqandstindex =
+		htonl(V_FW_IQ_CMD_TYPE(FW_IQ_TYPE_FL_INT_CAP) |
+		      V_FW_IQ_CMD_IQASYNCH(fwevtq) |
+		      V_FW_IQ_CMD_VIID(pi->viid) |
+		      V_FW_IQ_CMD_IQANDST(intr_idx < 0) |
+		      V_FW_IQ_CMD_IQANUD(X_UPDATEDELIVERY_INTERRUPT) |
+		      V_FW_IQ_CMD_IQANDSTINDEX(intr_idx >= 0 ? intr_idx :
+							       -intr_idx - 1));
+	c.iqdroprss_to_iqesize =
+		htons(V_FW_IQ_CMD_IQPCIECH(pi->tx_chan) |
+		      F_FW_IQ_CMD_IQGTSMODE |
+		      V_FW_IQ_CMD_IQINTCNTTHRESH(iq->pktcnt_idx) |
+		      V_FW_IQ_CMD_IQESIZE(ilog2(iq->iqe_len) - 4));
+	c.iqsize = htons(iq->size);
+	c.iqaddr = cpu_to_be64(iq->phys_addr);
+	if (cong >= 0)
+		c.iqns_to_fl0congen = htonl(F_FW_IQ_CMD_IQFLINTCONGEN);
+
+	if (fl) {
+		struct sge_eth_rxq *rxq = container_of(fl, struct sge_eth_rxq,
+						       fl);
+		enum chip_type chip = CHELSIO_CHIP_VERSION(adap->params.chip);
+
+		/*
+		 * Allocate the ring for the hardware free list (with space
+		 * for its status page) along with the associated software
+		 * descriptor ring.  The free list size needs to be a multiple
+		 * of the Egress Queue Unit and at least 2 Egress Units larger
+		 * than the SGE's Egress Congrestion Threshold
+		 * (fl_starve_thres - 1).
+		 */
+		if (fl->size < s->fl_starve_thres - 1 + 2 * 8)
+			fl->size = s->fl_starve_thres - 1 + 2 * 8;
+		fl->size = roundup(fl->size, 8);
+
+		snprintf(z_name, sizeof(z_name), "%s_%s_%d_%d",
+			 eth_dev->driver->pci_drv.name,
+			 fwevtq ? "fwq_ring" : "fl_ring",
+			 eth_dev->data->port_id, queue_id);
+		snprintf(z_name_sw, sizeof(z_name_sw), "%s_sw_ring", z_name);
+
+		fl->desc = alloc_ring(fl->size, sizeof(__be64),
+				      sizeof(struct rx_sw_desc),
+				      &fl->addr, &fl->sdesc, s->stat_len,
+				      queue_id, socket_id, z_name, z_name_sw);
+
+		if (!fl->desc)
+			goto fl_nomem;
+
+		flsz = fl->size / 8 + s->stat_len / sizeof(struct tx_desc);
+		c.iqns_to_fl0congen |=
+			htonl(V_FW_IQ_CMD_FL0HOSTFCMODE(X_HOSTFCMODE_NONE) |
+			      (unlikely(rxq->usembufs) ?
+			       0 : F_FW_IQ_CMD_FL0PACKEN) |
+			      F_FW_IQ_CMD_FL0FETCHRO | F_FW_IQ_CMD_FL0DATARO |
+			      F_FW_IQ_CMD_FL0PADEN);
+		if (cong >= 0)
+			c.iqns_to_fl0congen |=
+				htonl(V_FW_IQ_CMD_FL0CNGCHMAP(cong) |
+				      F_FW_IQ_CMD_FL0CONGCIF |
+				      F_FW_IQ_CMD_FL0CONGEN);
+
+		/* In T6, for egress queue type FL there is internal overhead
+		 * of 16B for header going into FLM module.
+		 * Hence maximum allowed burst size will be 448 bytes.
+		 */
+		c.fl0dcaen_to_fl0cidxfthresh =
+			htons(V_FW_IQ_CMD_FL0FBMIN(X_FETCHBURSTMIN_64B) |
+			      V_FW_IQ_CMD_FL0FBMAX((chip <= CHELSIO_T5) ?
+			      X_FETCHBURSTMAX_512B : X_FETCHBURSTMAX_256B));
+		c.fl0size = htons(flsz);
+		c.fl0addr = cpu_to_be64(fl->addr);
+	}
+
+	ret = t4_wr_mbox(adap, adap->mbox, &c, sizeof(c), &c);
+	if (ret)
+		goto err;
+
+	iq->cur_desc = iq->desc;
+	iq->cidx = 0;
+	iq->gen = 1;
+	iq->next_intr_params = iq->intr_params;
+	iq->cntxt_id = ntohs(c.iqid);
+	iq->abs_id = ntohs(c.physiqid);
+	iq->bar2_addr = bar2_address(adap, iq->cntxt_id, T4_BAR2_QTYPE_INGRESS,
+				     &iq->bar2_qid);
+	iq->size--;                           /* subtract status entry */
+	iq->eth_dev = eth_dev;
+	iq->handler = hnd;
+	iq->mb_pool = mp;
+
+	/* set offset to -1 to distinguish ingress queues without FL */
+	iq->offset = fl ? 0 : -1;
+
+	if (fl) {
+		fl->cntxt_id = ntohs(c.fl0id);
+		fl->avail = 0;
+		fl->pend_cred = 0;
+		fl->pidx = 0;
+		fl->cidx = 0;
+		fl->alloc_failed = 0;
+
+		/*
+		 * Note, we must initialize the BAR2 Free List User Doorbell
+		 * information before refilling the Free List!
+		 */
+		fl->bar2_addr = bar2_address(adap, fl->cntxt_id,
+					     T4_BAR2_QTYPE_EGRESS,
+					     &fl->bar2_qid);
+
+		nb_refill = refill_fl(adap, fl, fl_cap(fl));
+		if (nb_refill != fl_cap(fl)) {
+			ret = -ENOMEM;
+			dev_err(adap, "%s: mbuf alloc failed with error: %d\n",
+				__func__, ret);
+			goto refill_fl_err;
+		}
+	}
+
+	/*
+	 * For T5 and later we attempt to set up the Congestion Manager values
+	 * of the new RX Ethernet Queue.  This should really be handled by
+	 * firmware because it's more complex than any host driver wants to
+	 * get involved with and it's different per chip and this is almost
+	 * certainly wrong.  Formware would be wrong as well, but it would be
+	 * a lot easier to fix in one place ...  For now we do something very
+	 * simple (and hopefully less wrong).
+	 */
+	if (!is_t4(adap->params.chip) && cong >= 0) {
+		u32 param, val;
+		int i;
+
+		param = (V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DMAQ) |
+			 V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DMAQ_CONM_CTXT) |
+			 V_FW_PARAMS_PARAM_YZ(iq->cntxt_id));
+		if (cong == 0) {
+			val = V_CONMCTXT_CNGTPMODE(X_CONMCTXT_CNGTPMODE_QUEUE);
+		} else {
+			val = V_CONMCTXT_CNGTPMODE(
+					X_CONMCTXT_CNGTPMODE_CHANNEL);
+			for (i = 0; i < 4; i++) {
+				if (cong & (1 << i))
+					val |= V_CONMCTXT_CNGCHMAP(1 <<
+								   (i << 2));
+			}
+		}
+		ret = t4_set_params(adap, adap->mbox, adap->pf, 0, 1,
+				    &param, &val);
+		if (ret)
+			dev_warn(adap->pdev_dev, "Failed to set Congestion Manager Context for Ingress Queue %d: %d\n",
+				 iq->cntxt_id, -ret);
+	}
+
+	return 0;
+
+refill_fl_err:
+	t4_iq_free(adap, adap->mbox, adap->pf, 0, FW_IQ_TYPE_FL_INT_CAP,
+		   iq->cntxt_id, fl ? fl->cntxt_id : 0xffff, 0xffff);
+fl_nomem:
+	ret = -ENOMEM;
+err:
+	iq->cntxt_id = 0;
+	iq->abs_id = 0;
+	if (iq->desc)
+		iq->desc = NULL;
+
+	if (fl && fl->desc) {
+		rte_free(fl->sdesc);
+		fl->cntxt_id = 0;
+		fl->sdesc = NULL;
+		fl->desc = NULL;
+	}
+	return ret;
+}
+
+static void init_txq(struct adapter *adap, struct sge_txq *q, unsigned int id)
+{
+	q->cntxt_id = id;
+	q->bar2_addr = bar2_address(adap, q->cntxt_id, T4_BAR2_QTYPE_EGRESS,
+				    &q->bar2_qid);
+	q->cidx = 0;
+	q->pidx = 0;
+	q->dbidx = 0;
+	q->in_use = 0;
+	q->equeidx = 0;
+	q->coalesce.idx = 0;
+	q->coalesce.len = 0;
+	q->coalesce.flits = 0;
+	q->stat = (void *)&q->desc[q->size];
+}
+
+int t4_sge_eth_txq_start(struct sge_eth_txq *txq)
+{
+	/*
+	 *  TODO: For flow-control, queue may be stopped waiting to reclaim
+	 *  credits.
+	 *  Ensure queue is in EQ_STOPPED state before starting it.
+	 */
+	if (!(txq->flags & EQ_STOPPED))
+		return -(EBUSY);
+
+	txq->flags &= ~EQ_STOPPED;
+
+	return 0;
+}
+
+int t4_sge_eth_txq_stop(struct sge_eth_txq *txq)
+{
+	txq->flags |= EQ_STOPPED;
+
+	return 0;
+}
+
+int t4_sge_alloc_eth_txq(struct adapter *adap, struct sge_eth_txq *txq,
+			 struct rte_eth_dev *eth_dev, uint16_t queue_id,
+			 unsigned int iqid, int socket_id)
+{
+	int ret, nentries;
+	struct fw_eq_eth_cmd c;
+	struct sge *s = &adap->sge;
+	struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
+	char z_name[RTE_MEMZONE_NAMESIZE];
+	char z_name_sw[RTE_MEMZONE_NAMESIZE];
+
+	/* Add status entries */
+	nentries = txq->q.size + s->stat_len / sizeof(struct tx_desc);
+
+	snprintf(z_name, sizeof(z_name), "%s_%s_%d_%d",
+		 eth_dev->driver->pci_drv.name, "tx_ring",
+		 eth_dev->data->port_id, queue_id);
+	snprintf(z_name_sw, sizeof(z_name_sw), "%s_sw_ring", z_name);
+
+	txq->q.desc = alloc_ring(txq->q.size, sizeof(struct tx_desc),
+				 sizeof(struct tx_sw_desc), &txq->q.phys_addr,
+				 &txq->q.sdesc, s->stat_len, queue_id,
+				 socket_id, z_name, z_name_sw);
+	if (!txq->q.desc)
+		return -ENOMEM;
+
+	memset(&c, 0, sizeof(c));
+	c.op_to_vfn = htonl(V_FW_CMD_OP(FW_EQ_ETH_CMD) | F_FW_CMD_REQUEST |
+			    F_FW_CMD_WRITE | F_FW_CMD_EXEC |
+			    V_FW_EQ_ETH_CMD_PFN(adap->pf) |
+			    V_FW_EQ_ETH_CMD_VFN(0));
+	c.alloc_to_len16 = htonl(F_FW_EQ_ETH_CMD_ALLOC |
+				 F_FW_EQ_ETH_CMD_EQSTART | (sizeof(c) / 16));
+	c.autoequiqe_to_viid = htonl(F_FW_EQ_ETH_CMD_AUTOEQUEQE |
+				     V_FW_EQ_ETH_CMD_VIID(pi->viid));
+	c.fetchszm_to_iqid =
+		htonl(V_FW_EQ_ETH_CMD_HOSTFCMODE(X_HOSTFCMODE_NONE) |
+		      V_FW_EQ_ETH_CMD_PCIECHN(pi->tx_chan) |
+		      F_FW_EQ_ETH_CMD_FETCHRO | V_FW_EQ_ETH_CMD_IQID(iqid));
+	c.dcaen_to_eqsize =
+		htonl(V_FW_EQ_ETH_CMD_FBMIN(X_FETCHBURSTMIN_64B) |
+		      V_FW_EQ_ETH_CMD_FBMAX(X_FETCHBURSTMAX_512B) |
+		      V_FW_EQ_ETH_CMD_EQSIZE(nentries));
+	c.eqaddr = rte_cpu_to_be_64(txq->q.phys_addr);
+
+	ret = t4_wr_mbox(adap, adap->mbox, &c, sizeof(c), &c);
+	if (ret) {
+		rte_free(txq->q.sdesc);
+		txq->q.sdesc = NULL;
+		txq->q.desc = NULL;
+		return ret;
+	}
+
+	init_txq(adap, &txq->q, G_FW_EQ_ETH_CMD_EQID(ntohl(c.eqid_pkd)));
+	txq->stats.tso = 0;
+	txq->stats.pkts = 0;
+	txq->stats.tx_cso = 0;
+	txq->stats.coal_wr = 0;
+	txq->stats.vlan_ins = 0;
+	txq->stats.tx_bytes = 0;
+	txq->stats.coal_pkts = 0;
+	txq->stats.mapping_err = 0;
+	txq->flags |= EQ_STOPPED;
+	txq->eth_dev = eth_dev;
+	return 0;
+}
+
+static void free_txq(struct sge_txq *q)
+{
+	q->cntxt_id = 0;
+	q->sdesc = NULL;
+	q->desc = NULL;
+}
+
+static void free_rspq_fl(struct adapter *adap, struct sge_rspq *rq,
+			 struct sge_fl *fl)
+{
+	unsigned int fl_id = fl ? fl->cntxt_id : 0xffff;
+
+	t4_iq_free(adap, adap->mbox, adap->pf, 0, FW_IQ_TYPE_FL_INT_CAP,
+		   rq->cntxt_id, fl_id, 0xffff);
+	rq->cntxt_id = 0;
+	rq->abs_id = 0;
+	rq->desc = NULL;
+
+	if (fl) {
+		free_rx_bufs(fl, fl->avail);
+		rte_free(fl->sdesc);
+		fl->sdesc = NULL;
+		fl->cntxt_id = 0;
+		fl->desc = NULL;
+	}
+}
+
+/*
+ * Clear all queues of the port
+ *
+ * Note:  This function must only be called after rx and tx path
+ * of the port have been disabled.
+ */
+void t4_sge_eth_clear_queues(struct port_info *pi)
+{
+	int i;
+	struct adapter *adap = pi->adapter;
+	struct sge_eth_rxq *rxq = &adap->sge.ethrxq[pi->first_qset];
+	struct sge_eth_txq *txq = &adap->sge.ethtxq[pi->first_qset];
+
+	for (i = 0; i < pi->n_rx_qsets; i++, rxq++) {
+		if (rxq->rspq.desc)
+			t4_sge_eth_rxq_stop(adap, &rxq->rspq);
+	}
+	for (i = 0; i < pi->n_tx_qsets; i++, txq++) {
+		if (txq->q.desc) {
+			struct sge_txq *q = &txq->q;
+
+			t4_sge_eth_txq_stop(txq);
+			reclaim_completed_tx(q);
+			free_tx_desc(q, q->size);
+			q->equeidx = q->pidx;
+		}
+	}
+}
+
+void t4_sge_eth_rxq_release(struct adapter *adap, struct sge_eth_rxq *rxq)
+{
+	if (rxq->rspq.desc) {
+		t4_sge_eth_rxq_stop(adap, &rxq->rspq);
+		free_rspq_fl(adap, &rxq->rspq, rxq->fl.size ? &rxq->fl : NULL);
+	}
+}
+
+void t4_sge_eth_txq_release(struct adapter *adap, struct sge_eth_txq *txq)
+{
+	if (txq->q.desc) {
+		t4_sge_eth_txq_stop(txq);
+		reclaim_completed_tx(&txq->q);
+		t4_eth_eq_free(adap, adap->mbox, adap->pf, 0, txq->q.cntxt_id);
+		free_tx_desc(&txq->q, txq->q.size);
+		rte_free(txq->q.sdesc);
+		free_txq(&txq->q);
+	}
+}
+
+/**
+ * t4_free_sge_resources - free SGE resources
+ * @adap: the adapter
+ *
+ * Frees resources used by the SGE queue sets.
+ */
+void t4_free_sge_resources(struct adapter *adap)
+{
+	int i;
+	struct sge_eth_rxq *rxq = &adap->sge.ethrxq[0];
+	struct sge_eth_txq *txq = &adap->sge.ethtxq[0];
+
+	/* clean up Ethernet Tx/Rx queues */
+	for (i = 0; i < adap->sge.max_ethqsets; i++, rxq++, txq++) {
+		/* Free only the queues allocated */
+		if (rxq->rspq.desc) {
+			t4_sge_eth_rxq_release(adap, rxq);
+			rxq->rspq.eth_dev = NULL;
+		}
+		if (txq->q.desc) {
+			t4_sge_eth_txq_release(adap, txq);
+			txq->eth_dev = NULL;
+		}
+	}
+
+	if (adap->sge.fw_evtq.desc)
+		free_rspq_fl(adap, &adap->sge.fw_evtq, NULL);
+}
+
+/**
+ * t4_sge_init - initialize SGE
+ * @adap: the adapter
+ *
+ * Performs SGE initialization needed every time after a chip reset.
+ * We do not initialize any of the queues here, instead the driver
+ * top-level must request those individually.
+ *
+ * Called in two different modes:
+ *
+ *  1. Perform actual hardware initialization and record hard-coded
+ *     parameters which were used.  This gets used when we're the
+ *     Master PF and the Firmware Configuration File support didn't
+ *     work for some reason.
+ *
+ *  2. We're not the Master PF or initialization was performed with
+ *     a Firmware Configuration File.  In this case we need to grab
+ *     any of the SGE operating parameters that we need to have in
+ *     order to do our job and make sure we can live with them ...
+ */
+static int t4_sge_init_soft(struct adapter *adap)
+{
+	struct sge *s = &adap->sge;
+	u32 fl_small_pg, fl_large_pg, fl_small_mtu, fl_large_mtu;
+	u32 timer_value_0_and_1, timer_value_2_and_3, timer_value_4_and_5;
+	u32 ingress_rx_threshold;
+
+	/*
+	 * Verify that CPL messages are going to the Ingress Queue for
+	 * process_responses() and that only packet data is going to the
+	 * Free Lists.
+	 */
+	if ((t4_read_reg(adap, A_SGE_CONTROL) & F_RXPKTCPLMODE) !=
+	    V_RXPKTCPLMODE(X_RXPKTCPLMODE_SPLIT)) {
+		dev_err(adap, "bad SGE CPL MODE\n");
+		return -EINVAL;
+	}
+
+	/*
+	 * Validate the Host Buffer Register Array indices that we want to
+	 * use ...
+	 *
+	 * XXX Note that we should really read through the Host Buffer Size
+	 * XXX register array and find the indices of the Buffer Sizes which
+	 * XXX meet our needs!
+	 */
+#define READ_FL_BUF(x) \
+	t4_read_reg(adap, A_SGE_FL_BUFFER_SIZE0 + (x) * sizeof(u32))
+
+	fl_small_pg = READ_FL_BUF(RX_SMALL_PG_BUF);
+	fl_large_pg = READ_FL_BUF(RX_LARGE_PG_BUF);
+	fl_small_mtu = READ_FL_BUF(RX_SMALL_MTU_BUF);
+	fl_large_mtu = READ_FL_BUF(RX_LARGE_MTU_BUF);
+
+	/*
+	 * We only bother using the Large Page logic if the Large Page Buffer
+	 * is larger than our Page Size Buffer.
+	 */
+	if (fl_large_pg <= fl_small_pg)
+		fl_large_pg = 0;
+
+#undef READ_FL_BUF
+
+	/*
+	 * The Page Size Buffer must be exactly equal to our Page Size and the
+	 * Large Page Size Buffer should be 0 (per above) or a power of 2.
+	 */
+	if (fl_small_pg != PAGE_SIZE ||
+	    (fl_large_pg & (fl_large_pg - 1)) != 0) {
+		dev_err(adap, "bad SGE FL page buffer sizes [%d, %d]\n",
+			fl_small_pg, fl_large_pg);
+		return -EINVAL;
+	}
+	if (fl_large_pg)
+		s->fl_pg_order = ilog2(fl_large_pg) - PAGE_SHIFT;
+
+	if (adap->use_unpacked_mode) {
+		int err = 0;
+
+		if (fl_small_mtu < FL_MTU_SMALL_BUFSIZE(adap)) {
+			dev_err(adap, "bad SGE FL small MTU %d\n",
+				fl_small_mtu);
+			err = -EINVAL;
+		}
+		if (fl_large_mtu < FL_MTU_LARGE_BUFSIZE(adap)) {
+			dev_err(adap, "bad SGE FL large MTU %d\n",
+				fl_large_mtu);
+			err = -EINVAL;
+		}
+		if (err)
+			return err;
+	}
+
+	/*
+	 * Retrieve our RX interrupt holdoff timer values and counter
+	 * threshold values from the SGE parameters.
+	 */
+	timer_value_0_and_1 = t4_read_reg(adap, A_SGE_TIMER_VALUE_0_AND_1);
+	timer_value_2_and_3 = t4_read_reg(adap, A_SGE_TIMER_VALUE_2_AND_3);
+	timer_value_4_and_5 = t4_read_reg(adap, A_SGE_TIMER_VALUE_4_AND_5);
+	s->timer_val[0] = core_ticks_to_us(adap,
+					   G_TIMERVALUE0(timer_value_0_and_1));
+	s->timer_val[1] = core_ticks_to_us(adap,
+					   G_TIMERVALUE1(timer_value_0_and_1));
+	s->timer_val[2] = core_ticks_to_us(adap,
+					   G_TIMERVALUE2(timer_value_2_and_3));
+	s->timer_val[3] = core_ticks_to_us(adap,
+					   G_TIMERVALUE3(timer_value_2_and_3));
+	s->timer_val[4] = core_ticks_to_us(adap,
+					   G_TIMERVALUE4(timer_value_4_and_5));
+	s->timer_val[5] = core_ticks_to_us(adap,
+					   G_TIMERVALUE5(timer_value_4_and_5));
+
+	ingress_rx_threshold = t4_read_reg(adap, A_SGE_INGRESS_RX_THRESHOLD);
+	s->counter_val[0] = G_THRESHOLD_0(ingress_rx_threshold);
+	s->counter_val[1] = G_THRESHOLD_1(ingress_rx_threshold);
+	s->counter_val[2] = G_THRESHOLD_2(ingress_rx_threshold);
+	s->counter_val[3] = G_THRESHOLD_3(ingress_rx_threshold);
+
+	return 0;
+}
+
+int t4_sge_init(struct adapter *adap)
+{
+	struct sge *s = &adap->sge;
+	u32 sge_control, sge_control2, sge_conm_ctrl;
+	unsigned int ingpadboundary, ingpackboundary;
+	int ret, egress_threshold;
+
+	/*
+	 * Ingress Padding Boundary and Egress Status Page Size are set up by
+	 * t4_fixup_host_params().
+	 */
+	sge_control = t4_read_reg(adap, A_SGE_CONTROL);
+	s->pktshift = G_PKTSHIFT(sge_control);
+	s->stat_len = (sge_control & F_EGRSTATUSPAGESIZE) ? 128 : 64;
+
+	/*
+	 * T4 uses a single control field to specify both the PCIe Padding and
+	 * Packing Boundary.  T5 introduced the ability to specify these
+	 * separately.  The actual Ingress Packet Data alignment boundary
+	 * within Packed Buffer Mode is the maximum of these two
+	 * specifications.
+	 */
+	ingpadboundary = 1 << (G_INGPADBOUNDARY(sge_control) +
+			 X_INGPADBOUNDARY_SHIFT);
+	s->fl_align = ingpadboundary;
+
+	if (!is_t4(adap->params.chip) && !adap->use_unpacked_mode) {
+		/*
+		 * T5 has a weird interpretation of one of the PCIe Packing
+		 * Boundary values.  No idea why ...
+		 */
+		sge_control2 = t4_read_reg(adap, A_SGE_CONTROL2);
+		ingpackboundary = G_INGPACKBOUNDARY(sge_control2);
+		if (ingpackboundary == X_INGPACKBOUNDARY_16B)
+			ingpackboundary = 16;
+		else
+			ingpackboundary = 1 << (ingpackboundary +
+					  X_INGPACKBOUNDARY_SHIFT);
+
+		s->fl_align = max(ingpadboundary, ingpackboundary);
+	}
+
+	ret = t4_sge_init_soft(adap);
+	if (ret < 0) {
+		dev_err(adap, "%s: t4_sge_init_soft failed, error %d\n",
+			__func__, -ret);
+		return ret;
+	}
+
+	/*
+	 * A FL with <= fl_starve_thres buffers is starving and a periodic
+	 * timer will attempt to refill it.  This needs to be larger than the
+	 * SGE's Egress Congestion Threshold.  If it isn't, then we can get
+	 * stuck waiting for new packets while the SGE is waiting for us to
+	 * give it more Free List entries.  (Note that the SGE's Egress
+	 * Congestion Threshold is in units of 2 Free List pointers.)  For T4,
+	 * there was only a single field to control this.  For T5 there's the
+	 * original field which now only applies to Unpacked Mode Free List
+	 * buffers and a new field which only applies to Packed Mode Free List
+	 * buffers.
+	 */
+	sge_conm_ctrl = t4_read_reg(adap, A_SGE_CONM_CTRL);
+	if (is_t4(adap->params.chip) || adap->use_unpacked_mode)
+		egress_threshold = G_EGRTHRESHOLD(sge_conm_ctrl);
+	else
+		egress_threshold = G_EGRTHRESHOLDPACKING(sge_conm_ctrl);
+	s->fl_starve_thres = 2 * egress_threshold + 1;
+
+	return 0;
+}
-- 
1.7.1

  parent reply	other threads:[~2015-05-22 13:25 UTC|newest]

Thread overview: 61+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2015-05-22 13:24 [dpdk-dev] [PATCH 0/5] Chelsio Terminator 5 (T5) 10G/40G Poll Mode Driver Rahul Lakkireddy
2015-05-22 13:24 ` [dpdk-dev] [PATCH 1/5] cxgbe: add hardware specific api for all supported Chelsio T5 series adapters Rahul Lakkireddy
2015-05-22 13:24 ` Rahul Lakkireddy [this message]
2015-05-22 16:42   ` [dpdk-dev] [PATCH 2/5] cxgbe: add cxgbe poll mode driver Stephen Hemminger
2015-05-23  5:57     ` Rahul Lakkireddy
2015-05-26 17:02       ` Rahul Lakkireddy
2015-05-26 17:24         ` Stephen Hemminger
2015-05-26 18:13           ` Rahul Lakkireddy
2015-05-22 16:43   ` Stephen Hemminger
2015-05-23  5:56     ` Rahul Lakkireddy
2015-05-22 16:46   ` Stephen Hemminger
2015-05-23  5:53     ` Rahul Lakkireddy
2015-05-27  5:49       ` Thomas Monjalon
2015-05-27 11:26         ` Rahul Lakkireddy
2015-05-22 13:24 ` [dpdk-dev] [PATCH 3/5] doc: add cxgbe PMD documentation under doc/guides/nics/cxgbe.rst Rahul Lakkireddy
2015-05-27  5:38   ` Thomas Monjalon
2015-05-27 11:24     ` Rahul Lakkireddy
2015-05-22 13:24 ` [dpdk-dev] [PATCH 4/5] config: enable cxgbe PMD for compilation and linking Rahul Lakkireddy
2015-05-22 13:24 ` [dpdk-dev] [PATCH 5/5] maintainers: claim responsibility for cxgbe PMD Rahul Lakkireddy
2015-06-01 17:30 ` [dpdk-dev] [PATCH v2 00/11] Chelsio Terminator 5 (T5) 10G/40G Poll Mode Driver Rahul Lakkireddy
2015-06-01 17:30   ` [dpdk-dev] [PATCH v2 01/11] cxgbe: add hardware specific api for all supported Chelsio T5 series adapters Rahul Lakkireddy
2015-06-01 17:30   ` [dpdk-dev] [PATCH v2 02/11] cxgbe: add cxgbe poll mode driver Rahul Lakkireddy
2015-06-17 12:30     ` Thomas Monjalon
2015-06-18  7:06       ` Rahul Lakkireddy
2015-06-01 17:30   ` [dpdk-dev] [PATCH v2 03/11] cxgbe: add device configuration and RX support for cxgbe PMD Rahul Lakkireddy
2015-06-01 17:30   ` [dpdk-dev] [PATCH v2 04/11] cxgbe: add TX " Rahul Lakkireddy
2015-06-01 17:30   ` [dpdk-dev] [PATCH v2 05/11] cxgbe: add device related operations " Rahul Lakkireddy
2015-06-01 17:30   ` [dpdk-dev] [PATCH v2 06/11] cxgbe: add port statistics " Rahul Lakkireddy
2015-06-01 17:30   ` [dpdk-dev] [PATCH v2 07/11] cxgbe: add link related functions " Rahul Lakkireddy
2015-06-01 17:30   ` [dpdk-dev] [PATCH v2 08/11] cxgbe: add flow control " Rahul Lakkireddy
2015-06-01 17:30   ` [dpdk-dev] [PATCH v2 09/11] doc: add cxgbe PMD documentation under doc/guides/nics/cxgbe.rst Rahul Lakkireddy
2015-06-01 17:30   ` [dpdk-dev] [PATCH v2 10/11] config: enable cxgbe PMD for compilation and linking Rahul Lakkireddy
2015-06-01 17:30   ` [dpdk-dev] [PATCH v2 11/11] maintainers: claim responsibility for cxgbe PMD Rahul Lakkireddy
2015-06-18 12:17   ` [dpdk-dev] [PATCH v3 0/9] Chelsio Terminator 5 (T5) 10G/40G Poll Mode Driver Rahul Lakkireddy
2015-06-18 12:17     ` [dpdk-dev] [PATCH v3 1/9] cxgbe: add hardware specific api for all supported Chelsio T5 series adapters Rahul Lakkireddy
2015-06-18 12:17     ` [dpdk-dev] [PATCH v3 2/9] cxgbe: add cxgbe poll mode driver Rahul Lakkireddy
2015-06-28 19:32       ` Thomas Monjalon
2015-06-29 23:23         ` Rahul Lakkireddy
2015-06-18 12:17     ` [dpdk-dev] [PATCH v3 3/9] cxgbe: add device configuration and RX support for cxgbe PMD Rahul Lakkireddy
2015-06-28 19:34       ` Thomas Monjalon
2015-06-29 23:18         ` Rahul Lakkireddy
2015-06-18 12:17     ` [dpdk-dev] [PATCH v3 4/9] cxgbe: add TX " Rahul Lakkireddy
2015-06-18 12:17     ` [dpdk-dev] [PATCH v3 5/9] cxgbe: add device related operations " Rahul Lakkireddy
2015-06-18 12:17     ` [dpdk-dev] [PATCH v3 6/9] cxgbe: add port statistics " Rahul Lakkireddy
2015-06-18 12:17     ` [dpdk-dev] [PATCH v3 7/9] cxgbe: add link related functions " Rahul Lakkireddy
2015-06-18 12:17     ` [dpdk-dev] [PATCH v3 8/9] cxgbe: add flow control " Rahul Lakkireddy
2015-06-18 12:17     ` [dpdk-dev] [PATCH v3 9/9] doc: add cxgbe PMD documentation under doc/guides/nics/cxgbe.rst Rahul Lakkireddy
2015-06-18 13:47       ` Mcnamara, John
2015-06-18 13:44     ` [dpdk-dev] [PATCH v3 0/9] Chelsio Terminator 5 (T5) 10G/40G Poll Mode Driver Mcnamara, John
2015-06-29 23:28     ` [dpdk-dev] [PATCH v4 " Rahul Lakkireddy
2015-06-29 23:28       ` [dpdk-dev] [PATCH v4 1/9] cxgbe: add hardware specific api for all supported Chelsio T5 series adapters Rahul Lakkireddy
2015-06-29 23:28       ` [dpdk-dev] [PATCH v4 2/9] cxgbe: add cxgbe poll mode driver Rahul Lakkireddy
2015-06-29 23:28       ` [dpdk-dev] [PATCH v4 3/9] cxgbe: add device configuration and RX support for cxgbe PMD Rahul Lakkireddy
2015-06-29 23:28       ` [dpdk-dev] [PATCH v4 4/9] cxgbe: add TX " Rahul Lakkireddy
2015-06-29 23:28       ` [dpdk-dev] [PATCH v4 5/9] cxgbe: add device related operations " Rahul Lakkireddy
2015-06-29 23:28       ` [dpdk-dev] [PATCH v4 6/9] cxgbe: add port statistics " Rahul Lakkireddy
2015-06-29 23:28       ` [dpdk-dev] [PATCH v4 7/9] cxgbe: add link related functions " Rahul Lakkireddy
2015-06-29 23:28       ` [dpdk-dev] [PATCH v4 8/9] cxgbe: add flow control " Rahul Lakkireddy
2015-06-29 23:28       ` [dpdk-dev] [PATCH v4 9/9] doc: add cxgbe PMD documentation under doc/guides/nics/cxgbe.rst Rahul Lakkireddy
2015-06-30 21:01       ` [dpdk-dev] [PATCH v4 0/9] Chelsio Terminator 5 (T5) 10G/40G Poll Mode Driver Thomas Monjalon
2015-07-01  6:35         ` Rahul Lakkireddy

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=25f06823ef87d05bf57d996df387478e7fbc966f.1432300701.git.rahul.lakkireddy@chelsio.com \
    --to=rahul.lakkireddy@chelsio.com \
    --cc=dev@dpdk.org \
    --cc=felix@chelsio.com \
    --cc=kumaras@chelsio.com \
    --cc=nirranjan@chelsio.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).