DPDK patches and discussions
 help / color / mirror / Atom feed
* [dpdk-dev] [PATCH 00/13] Mellanox ConnectX-4 PMD (mlx5)
@ 2015-10-05 17:52 Adrien Mazarguil
  2015-10-05 17:52 ` [dpdk-dev] [PATCH 01/13] mlx5: new poll-mode driver for Mellanox ConnectX-4 adapters Adrien Mazarguil
                   ` (13 more replies)
  0 siblings, 14 replies; 32+ messages in thread
From: Adrien Mazarguil @ 2015-10-05 17:52 UTC (permalink / raw)
  To: dev

This PMD adds basic support for Mellanox ConnectX-4 (mlx5) families of
10/25/40/50/100 Gb/s adapters through the Verbs framework.

Its design is very similar to that of mlx4 from which most of its code is
borrowed without the mistake of putting it all in a single huge file.

It is disabled by default due to its dependency on libibverbs.

Adrien Mazarguil (13):
  mlx5: new poll-mode driver for Mellanox ConnectX-4 adapters
  mlx5: add non-scattered TX and RX support
  mlx5: add MAC handling
  mlx5: add device configure/start/stop
  mlx5: add support for scattered RX and TX buffers
  mlx5: add MTU configuration support
  mlx5: add software counters and related callbacks
  mlx5: add promiscuous and allmulticast RX modes
  mlx5: add link update device operation
  mlx5: add flow control device operations
  mlx5: add VLAN filtering
  mlx5: add checksum offloading support
  doc: add mlx5 documentation and release notes for version 2.2

 MAINTAINERS                               |    4 +
 config/common_bsdapp                      |    9 +
 config/common_linuxapp                    |    9 +
 doc/guides/nics/mlx5.rst                  |  308 +++++++++
 doc/guides/rel_notes/release_2_2.rst      |    8 +
 drivers/net/Makefile                      |    1 +
 drivers/net/mlx5/Makefile                 |  128 ++++
 drivers/net/mlx5/mlx5.c                   |  553 +++++++++++++++
 drivers/net/mlx5/mlx5.h                   |  215 ++++++
 drivers/net/mlx5/mlx5_defs.h              |   85 +++
 drivers/net/mlx5/mlx5_ethdev.c            |  843 +++++++++++++++++++++++
 drivers/net/mlx5/mlx5_mac.c               |  497 ++++++++++++++
 drivers/net/mlx5/mlx5_rxmode.c            |  327 +++++++++
 drivers/net/mlx5/mlx5_rxq.c               | 1067 +++++++++++++++++++++++++++++
 drivers/net/mlx5/mlx5_rxtx.c              | 1008 +++++++++++++++++++++++++++
 drivers/net/mlx5/mlx5_rxtx.h              |  197 ++++++
 drivers/net/mlx5/mlx5_stats.c             |  144 ++++
 drivers/net/mlx5/mlx5_trigger.c           |  153 +++++
 drivers/net/mlx5/mlx5_txq.c               |  513 ++++++++++++++
 drivers/net/mlx5/mlx5_utils.h             |  166 +++++
 drivers/net/mlx5/mlx5_vlan.c              |  166 +++++
 drivers/net/mlx5/rte_pmd_mlx5_version.map |    3 +
 mk/rte.app.mk                             |    5 +
 23 files changed, 6409 insertions(+)
 create mode 100644 doc/guides/nics/mlx5.rst
 create mode 100644 drivers/net/mlx5/Makefile
 create mode 100644 drivers/net/mlx5/mlx5.c
 create mode 100644 drivers/net/mlx5/mlx5.h
 create mode 100644 drivers/net/mlx5/mlx5_defs.h
 create mode 100644 drivers/net/mlx5/mlx5_ethdev.c
 create mode 100644 drivers/net/mlx5/mlx5_mac.c
 create mode 100644 drivers/net/mlx5/mlx5_rxmode.c
 create mode 100644 drivers/net/mlx5/mlx5_rxq.c
 create mode 100644 drivers/net/mlx5/mlx5_rxtx.c
 create mode 100644 drivers/net/mlx5/mlx5_rxtx.h
 create mode 100644 drivers/net/mlx5/mlx5_stats.c
 create mode 100644 drivers/net/mlx5/mlx5_trigger.c
 create mode 100644 drivers/net/mlx5/mlx5_txq.c
 create mode 100644 drivers/net/mlx5/mlx5_utils.h
 create mode 100644 drivers/net/mlx5/mlx5_vlan.c
 create mode 100644 drivers/net/mlx5/rte_pmd_mlx5_version.map

-- 
2.1.0

^ permalink raw reply	[flat|nested] 32+ messages in thread

* [dpdk-dev] [PATCH 01/13] mlx5: new poll-mode driver for Mellanox ConnectX-4 adapters
  2015-10-05 17:52 [dpdk-dev] [PATCH 00/13] Mellanox ConnectX-4 PMD (mlx5) Adrien Mazarguil
@ 2015-10-05 17:52 ` Adrien Mazarguil
  2015-10-05 17:52 ` [dpdk-dev] [PATCH 02/13] mlx5: add non-scattered TX and RX support Adrien Mazarguil
                   ` (12 subsequent siblings)
  13 siblings, 0 replies; 32+ messages in thread
From: Adrien Mazarguil @ 2015-10-05 17:52 UTC (permalink / raw)
  To: dev

In its current state, this driver implements the bare minimum to initialize
itself and Mellanox ConnectX-4 adapters without doing anything else
(no RX/TX for instance). It is disabled by default since it is based on the
mlx4 driver and also depends on libibverbs.

Signed-off-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>
Signed-off-by: Nelio Laranjeiro <nelio.laranjeiro@6wind.com>
Signed-off-by: Or Ami <ora@mellanox.com>
---
 MAINTAINERS                               |   4 +
 config/common_bsdapp                      |   6 +
 config/common_linuxapp                    |   6 +
 drivers/net/Makefile                      |   1 +
 drivers/net/mlx5/Makefile                 | 109 +++++++
 drivers/net/mlx5/mlx5.c                   | 496 ++++++++++++++++++++++++++++++
 drivers/net/mlx5/mlx5.h                   | 154 ++++++++++
 drivers/net/mlx5/mlx5_defs.h              |  53 ++++
 drivers/net/mlx5/mlx5_ethdev.c            | 420 +++++++++++++++++++++++++
 drivers/net/mlx5/mlx5_mac.c               | 150 +++++++++
 drivers/net/mlx5/mlx5_utils.h             | 149 +++++++++
 drivers/net/mlx5/rte_pmd_mlx5_version.map |   3 +
 mk/rte.app.mk                             |   5 +
 13 files changed, 1556 insertions(+)
 create mode 100644 drivers/net/mlx5/Makefile
 create mode 100644 drivers/net/mlx5/mlx5.c
 create mode 100644 drivers/net/mlx5/mlx5.h
 create mode 100644 drivers/net/mlx5/mlx5_defs.h
 create mode 100644 drivers/net/mlx5/mlx5_ethdev.c
 create mode 100644 drivers/net/mlx5/mlx5_mac.c
 create mode 100644 drivers/net/mlx5/mlx5_utils.h
 create mode 100644 drivers/net/mlx5/rte_pmd_mlx5_version.map

diff --git a/MAINTAINERS b/MAINTAINERS
index 080a8e8..9d11055 100644
--- a/MAINTAINERS
+++ b/MAINTAINERS
@@ -255,6 +255,10 @@ M: Adrien Mazarguil <adrien.mazarguil@6wind.com>
 F: drivers/net/mlx4/
 F: doc/guides/nics/mlx4.rst
 
+Mellanox mlx5
+M: Adrien Mazarguil <adrien.mazarguil@6wind.com>
+F: drivers/net/mlx5/
+
 RedHat virtio
 M: Huawei Xie <huawei.xie@intel.com>
 M: Changchun Ouyang <changchun.ouyang@intel.com>
diff --git a/config/common_bsdapp b/config/common_bsdapp
index b37dcf4..1e6885f 100644
--- a/config/common_bsdapp
+++ b/config/common_bsdapp
@@ -214,6 +214,12 @@ CONFIG_RTE_LIBRTE_MLX4_TX_MP_CACHE=8
 CONFIG_RTE_LIBRTE_MLX4_SOFT_COUNTERS=1
 
 #
+# Compile burst-oriented Mellanox ConnectX-4 (MLX5) PMD
+#
+CONFIG_RTE_LIBRTE_MLX5_PMD=n
+CONFIG_RTE_LIBRTE_MLX5_DEBUG=n
+
+#
 # Compile burst-oriented Broadcom PMD driver
 #
 CONFIG_RTE_LIBRTE_BNX2X_PMD=n
diff --git a/config/common_linuxapp b/config/common_linuxapp
index 0de43d5..7da7ba7 100644
--- a/config/common_linuxapp
+++ b/config/common_linuxapp
@@ -212,6 +212,12 @@ CONFIG_RTE_LIBRTE_MLX4_TX_MP_CACHE=8
 CONFIG_RTE_LIBRTE_MLX4_SOFT_COUNTERS=1
 
 #
+# Compile burst-oriented Mellanox ConnectX-4 (MLX5) PMD
+#
+CONFIG_RTE_LIBRTE_MLX5_PMD=n
+CONFIG_RTE_LIBRTE_MLX5_DEBUG=n
+
+#
 # Compile burst-oriented Broadcom PMD driver
 #
 CONFIG_RTE_LIBRTE_BNX2X_PMD=n
diff --git a/drivers/net/Makefile b/drivers/net/Makefile
index 5ebf963..6da1ce2 100644
--- a/drivers/net/Makefile
+++ b/drivers/net/Makefile
@@ -41,6 +41,7 @@ DIRS-$(CONFIG_RTE_LIBRTE_FM10K_PMD) += fm10k
 DIRS-$(CONFIG_RTE_LIBRTE_I40E_PMD) += i40e
 DIRS-$(CONFIG_RTE_LIBRTE_IXGBE_PMD) += ixgbe
 DIRS-$(CONFIG_RTE_LIBRTE_MLX4_PMD) += mlx4
+DIRS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5
 DIRS-$(CONFIG_RTE_LIBRTE_MPIPE_PMD) += mpipe
 DIRS-$(CONFIG_RTE_LIBRTE_PMD_NULL) += null
 DIRS-$(CONFIG_RTE_LIBRTE_PMD_PCAP) += pcap
diff --git a/drivers/net/mlx5/Makefile b/drivers/net/mlx5/Makefile
new file mode 100644
index 0000000..6e63073
--- /dev/null
+++ b/drivers/net/mlx5/Makefile
@@ -0,0 +1,109 @@
+#   BSD LICENSE
+#
+#   Copyright 2015 6WIND S.A.
+#   Copyright 2015 Mellanox.
+#
+#   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 6WIND S.A. 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 $(RTE_SDK)/mk/rte.vars.mk
+
+ifeq ($(CONFIG_RTE_BUILD_COMBINE_LIBS)$(CONFIG_RTE_BUILD_SHARED_LIB),yy)
+all:
+	@echo 'MLX5: Not supported in a combined shared library'
+	@false
+endif
+
+# Library name.
+LIB = librte_pmd_mlx5.a
+
+# Sources.
+SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5.c
+SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5_ethdev.c
+SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5_mac.c
+
+# Dependencies.
+DEPDIRS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += lib/librte_ether
+DEPDIRS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += lib/librte_mbuf
+DEPDIRS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += lib/librte_eal
+DEPDIRS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += lib/librte_mempool
+
+# Basic CFLAGS.
+CFLAGS += -O3
+CFLAGS += -std=gnu99 -Wall -Wextra
+CFLAGS += -g
+CFLAGS += -I.
+CFLAGS += -D_XOPEN_SOURCE=600
+CFLAGS += $(WERROR_FLAGS)
+LDLIBS += -libverbs
+
+# A few warnings cannot be avoided in external headers.
+CFLAGS += -Wno-error=cast-qual
+
+EXPORT_MAP := rte_pmd_mlx5_version.map
+LIBABIVER := 1
+
+# DEBUG which is usually provided on the command-line may enable
+# CONFIG_RTE_LIBRTE_MLX5_DEBUG.
+ifeq ($(DEBUG),1)
+CONFIG_RTE_LIBRTE_MLX5_DEBUG := y
+endif
+
+# User-defined CFLAGS.
+ifeq ($(CONFIG_RTE_LIBRTE_MLX5_DEBUG),y)
+CFLAGS += -pedantic -UNDEBUG -DPEDANTIC
+else
+CFLAGS += -DNDEBUG -UPEDANTIC
+endif
+
+include $(RTE_SDK)/mk/rte.lib.mk
+
+# Generate and clean-up mlx5_autoconf.h.
+
+export CC CFLAGS CPPFLAGS EXTRA_CFLAGS EXTRA_CPPFLAGS
+export AUTO_CONFIG_CFLAGS = -Wno-error
+
+ifndef V
+AUTOCONF_OUTPUT := >/dev/null
+endif
+
+mlx5_autoconf.h: $(RTE_SDK)/scripts/auto-config-h.sh
+	$Q $(RM) -f -- '$@'
+	$Q sh -- '$<' '$@' \
+		RSS_SUPPORT \
+		infiniband/verbs.h \
+		enum IBV_EXP_DEVICE_UD_RSS $(AUTOCONF_OUTPUT)
+	$Q sh -- '$<' '$@' \
+		HAVE_EXP_QUERY_DEVICE \
+		infiniband/verbs.h \
+		type 'struct ibv_exp_device_attr' $(AUTOCONF_OUTPUT)
+
+mlx5.o: mlx5_autoconf.h
+
+clean_mlx5: FORCE
+	$Q rm -f -- mlx5_autoconf.h
+
+clean: clean_mlx5
diff --git a/drivers/net/mlx5/mlx5.c b/drivers/net/mlx5/mlx5.c
new file mode 100644
index 0000000..f6fe8a6
--- /dev/null
+++ b/drivers/net/mlx5/mlx5.c
@@ -0,0 +1,496 @@
+/*-
+ *   BSD LICENSE
+ *
+ *   Copyright 2015 6WIND S.A.
+ *   Copyright 2015 Mellanox.
+ *
+ *   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 6WIND S.A. 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 <stddef.h>
+#include <unistd.h>
+#include <string.h>
+#include <assert.h>
+#include <stdint.h>
+#include <stdlib.h>
+#include <net/if.h>
+
+/* Verbs header. */
+/* ISO C doesn't support unnamed structs/unions, disabling -pedantic. */
+#ifdef PEDANTIC
+#pragma GCC diagnostic ignored "-pedantic"
+#endif
+#include <infiniband/verbs.h>
+#ifdef PEDANTIC
+#pragma GCC diagnostic error "-pedantic"
+#endif
+
+/* DPDK headers don't like -pedantic. */
+#ifdef PEDANTIC
+#pragma GCC diagnostic ignored "-pedantic"
+#endif
+#include <rte_malloc.h>
+#include <rte_ethdev.h>
+#include <rte_pci.h>
+#include <rte_common.h>
+#ifdef PEDANTIC
+#pragma GCC diagnostic error "-pedantic"
+#endif
+
+#include "mlx5.h"
+#include "mlx5_utils.h"
+#include "mlx5_autoconf.h"
+
+/**
+ * DPDK callback to close the device.
+ *
+ * Destroy all queues and objects, free memory.
+ *
+ * @param dev
+ *   Pointer to Ethernet device structure.
+ */
+static void
+mlx5_dev_close(struct rte_eth_dev *dev)
+{
+	struct priv *priv = dev->data->dev_private;
+
+	priv_lock(priv);
+	DEBUG("%p: closing device \"%s\"",
+	      (void *)dev,
+	      ((priv->ctx != NULL) ? priv->ctx->device->name : ""));
+	if (priv->pd != NULL) {
+		assert(priv->ctx != NULL);
+		claim_zero(ibv_dealloc_pd(priv->pd));
+		claim_zero(ibv_close_device(priv->ctx));
+	} else
+		assert(priv->ctx == NULL);
+	priv_unlock(priv);
+	memset(priv, 0, sizeof(*priv));
+}
+
+static const struct eth_dev_ops mlx5_dev_ops = {
+	.dev_close = mlx5_dev_close,
+};
+
+static struct {
+	struct rte_pci_addr pci_addr; /* associated PCI address */
+	uint32_t ports; /* physical ports bitfield. */
+} mlx5_dev[32];
+
+/**
+ * Get device index in mlx5_dev[] from PCI bus address.
+ *
+ * @param[in] pci_addr
+ *   PCI bus address to look for.
+ *
+ * @return
+ *   mlx5_dev[] index on success, -1 on failure.
+ */
+static int
+mlx5_dev_idx(struct rte_pci_addr *pci_addr)
+{
+	unsigned int i;
+	int ret = -1;
+
+	assert(pci_addr != NULL);
+	for (i = 0; (i != RTE_DIM(mlx5_dev)); ++i) {
+		if ((mlx5_dev[i].pci_addr.domain == pci_addr->domain) &&
+		    (mlx5_dev[i].pci_addr.bus == pci_addr->bus) &&
+		    (mlx5_dev[i].pci_addr.devid == pci_addr->devid) &&
+		    (mlx5_dev[i].pci_addr.function == pci_addr->function))
+			return i;
+		if ((mlx5_dev[i].ports == 0) && (ret == -1))
+			ret = i;
+	}
+	return ret;
+}
+
+static struct eth_driver mlx5_driver;
+
+/**
+ * DPDK callback to register a PCI device.
+ *
+ * This function creates an Ethernet device for each port of a given
+ * PCI device.
+ *
+ * @param[in] pci_drv
+ *   PCI driver structure (mlx5_driver).
+ * @param[in] pci_dev
+ *   PCI device information.
+ *
+ * @return
+ *   0 on success, negative errno value on failure.
+ */
+static int
+mlx5_pci_devinit(struct rte_pci_driver *pci_drv, struct rte_pci_device *pci_dev)
+{
+	struct ibv_device **list;
+	struct ibv_device *ibv_dev;
+	int err = 0;
+	struct ibv_context *attr_ctx = NULL;
+	struct ibv_device_attr device_attr;
+	unsigned int vf;
+	int idx;
+	int i;
+
+	(void)pci_drv;
+	assert(pci_drv == &mlx5_driver.pci_drv);
+	/* Get mlx5_dev[] index. */
+	idx = mlx5_dev_idx(&pci_dev->addr);
+	if (idx == -1) {
+		ERROR("this driver cannot support any more adapters");
+		return -ENOMEM;
+	}
+	DEBUG("using driver device index %d", idx);
+
+	/* Save PCI address. */
+	mlx5_dev[idx].pci_addr = pci_dev->addr;
+	list = ibv_get_device_list(&i);
+	if (list == NULL) {
+		assert(errno);
+		if (errno == ENOSYS) {
+			WARN("cannot list devices, is ib_uverbs loaded?");
+			return 0;
+		}
+		return -errno;
+	}
+	assert(i >= 0);
+	/*
+	 * For each listed device, check related sysfs entry against
+	 * the provided PCI ID.
+	 */
+	while (i != 0) {
+		struct rte_pci_addr pci_addr;
+
+		--i;
+		DEBUG("checking device \"%s\"", list[i]->name);
+		if (mlx5_ibv_device_to_pci_addr(list[i], &pci_addr))
+			continue;
+		if ((pci_dev->addr.domain != pci_addr.domain) ||
+		    (pci_dev->addr.bus != pci_addr.bus) ||
+		    (pci_dev->addr.devid != pci_addr.devid) ||
+		    (pci_dev->addr.function != pci_addr.function))
+			continue;
+		vf = ((pci_dev->id.device_id ==
+		       PCI_DEVICE_ID_MELLANOX_CONNECTX4VF) ||
+		      (pci_dev->id.device_id ==
+		       PCI_DEVICE_ID_MELLANOX_CONNECTX4LXVF));
+		INFO("PCI information matches, using device \"%s\" (VF: %s)",
+		     list[i]->name, (vf ? "true" : "false"));
+		attr_ctx = ibv_open_device(list[i]);
+		err = errno;
+		break;
+	}
+	if (attr_ctx == NULL) {
+		ibv_free_device_list(list);
+		switch (err) {
+		case 0:
+			WARN("cannot access device, is mlx5_ib loaded?");
+			return 0;
+		case EINVAL:
+			WARN("cannot use device, are drivers up to date?");
+			return 0;
+		}
+		assert(err > 0);
+		return -err;
+	}
+	ibv_dev = list[i];
+
+	DEBUG("device opened");
+	if (ibv_query_device(attr_ctx, &device_attr))
+		goto error;
+	INFO("%u port(s) detected", device_attr.phys_port_cnt);
+
+	for (i = 0; i < device_attr.phys_port_cnt; i++) {
+		uint32_t port = i + 1; /* ports are indexed from one */
+		uint32_t test = (1 << i);
+		struct ibv_context *ctx = NULL;
+		struct ibv_port_attr port_attr;
+		struct ibv_pd *pd = NULL;
+		struct priv *priv = NULL;
+		struct rte_eth_dev *eth_dev;
+#ifdef HAVE_EXP_QUERY_DEVICE
+		struct ibv_exp_device_attr exp_device_attr;
+#endif /* HAVE_EXP_QUERY_DEVICE */
+		struct ether_addr mac;
+
+#ifdef HAVE_EXP_QUERY_DEVICE
+		exp_device_attr.comp_mask = IBV_EXP_DEVICE_ATTR_EXP_CAP_FLAGS;
+#ifdef RSS_SUPPORT
+		exp_device_attr.comp_mask |= IBV_EXP_DEVICE_ATTR_RSS_TBL_SZ;
+#endif /* RSS_SUPPORT */
+#endif /* HAVE_EXP_QUERY_DEVICE */
+
+		DEBUG("using port %u (%08" PRIx32 ")", port, test);
+
+		ctx = ibv_open_device(ibv_dev);
+		if (ctx == NULL)
+			goto port_error;
+
+		/* Check port status. */
+		err = ibv_query_port(ctx, port, &port_attr);
+		if (err) {
+			ERROR("port query failed: %s", strerror(err));
+			goto port_error;
+		}
+		if (port_attr.state != IBV_PORT_ACTIVE)
+			WARN("bad state for port %d: \"%s\" (%d)",
+			     port, ibv_port_state_str(port_attr.state),
+			     port_attr.state);
+
+		/* Allocate protection domain. */
+		pd = ibv_alloc_pd(ctx);
+		if (pd == NULL) {
+			ERROR("PD allocation failure");
+			err = ENOMEM;
+			goto port_error;
+		}
+
+		mlx5_dev[idx].ports |= test;
+
+		/* from rte_ethdev.c */
+		priv = rte_zmalloc("ethdev private structure",
+				   sizeof(*priv),
+				   RTE_CACHE_LINE_SIZE);
+		if (priv == NULL) {
+			ERROR("priv allocation failure");
+			err = ENOMEM;
+			goto port_error;
+		}
+
+		priv->ctx = ctx;
+		priv->device_attr = device_attr;
+		priv->port = port;
+		priv->pd = pd;
+		priv->mtu = ETHER_MTU;
+#ifdef HAVE_EXP_QUERY_DEVICE
+		if (ibv_exp_query_device(ctx, &exp_device_attr)) {
+			ERROR("ibv_exp_query_device() failed");
+			goto port_error;
+		}
+#ifdef RSS_SUPPORT
+		if ((exp_device_attr.exp_device_cap_flags &
+		     IBV_EXP_DEVICE_QPG) &&
+		    (exp_device_attr.exp_device_cap_flags &
+		     IBV_EXP_DEVICE_UD_RSS) &&
+		    (exp_device_attr.comp_mask &
+		     IBV_EXP_DEVICE_ATTR_RSS_TBL_SZ) &&
+		    (exp_device_attr.max_rss_tbl_sz > 0)) {
+			priv->hw_qpg = 1;
+			priv->hw_rss = 1;
+			priv->max_rss_tbl_sz = exp_device_attr.max_rss_tbl_sz;
+		} else {
+			priv->hw_qpg = 0;
+			priv->hw_rss = 0;
+			priv->max_rss_tbl_sz = 0;
+		}
+		priv->hw_tss = !!(exp_device_attr.exp_device_cap_flags &
+				  IBV_EXP_DEVICE_UD_TSS);
+		DEBUG("device flags: %s%s%s",
+		      (priv->hw_qpg ? "IBV_DEVICE_QPG " : ""),
+		      (priv->hw_tss ? "IBV_DEVICE_TSS " : ""),
+		      (priv->hw_rss ? "IBV_DEVICE_RSS " : ""));
+		if (priv->hw_rss)
+			DEBUG("maximum RSS indirection table size: %u",
+			      exp_device_attr.max_rss_tbl_sz);
+#endif /* RSS_SUPPORT */
+
+		priv->hw_csum =
+			((exp_device_attr.exp_device_cap_flags &
+			  IBV_EXP_DEVICE_RX_CSUM_TCP_UDP_PKT) &&
+			 (exp_device_attr.exp_device_cap_flags &
+			  IBV_EXP_DEVICE_RX_CSUM_IP_PKT));
+		DEBUG("checksum offloading is %ssupported",
+		      (priv->hw_csum ? "" : "not "));
+
+		priv->hw_csum_l2tun = !!(exp_device_attr.exp_device_cap_flags &
+					 IBV_EXP_DEVICE_VXLAN_SUPPORT);
+		DEBUG("L2 tunnel checksum offloads are %ssupported",
+		      (priv->hw_csum_l2tun ? "" : "not "));
+
+#endif /* HAVE_EXP_QUERY_DEVICE */
+
+		priv->vf = vf;
+		/* Configure the first MAC address by default. */
+		if (priv_get_mac(priv, &mac.addr_bytes)) {
+			ERROR("cannot get MAC address, is mlx5_en loaded?"
+			      " (errno: %s)", strerror(errno));
+			goto port_error;
+		}
+		INFO("port %u MAC address is %02x:%02x:%02x:%02x:%02x:%02x",
+		     priv->port,
+		     mac.addr_bytes[0], mac.addr_bytes[1],
+		     mac.addr_bytes[2], mac.addr_bytes[3],
+		     mac.addr_bytes[4], mac.addr_bytes[5]);
+		/* Register MAC and broadcast addresses. */
+		claim_zero(priv_mac_addr_add(priv, 0,
+					     (const uint8_t (*)[ETHER_ADDR_LEN])
+					     mac.addr_bytes));
+		claim_zero(priv_mac_addr_add(priv, 1,
+					     &(const uint8_t [ETHER_ADDR_LEN])
+					     { "\xff\xff\xff\xff\xff\xff" }));
+#ifndef NDEBUG
+		{
+			char ifname[IF_NAMESIZE];
+
+			if (priv_get_ifname(priv, &ifname) == 0)
+				DEBUG("port %u ifname is \"%s\"",
+				      priv->port, ifname);
+			else
+				DEBUG("port %u ifname is unknown", priv->port);
+		}
+#endif
+		/* Get actual MTU if possible. */
+		priv_get_mtu(priv, &priv->mtu);
+		DEBUG("port %u MTU is %u", priv->port, priv->mtu);
+
+		/* from rte_ethdev.c */
+		{
+			char name[RTE_ETH_NAME_MAX_LEN];
+
+			snprintf(name, sizeof(name), "%s port %u",
+				 ibv_get_device_name(ibv_dev), port);
+			eth_dev = rte_eth_dev_allocate(name, RTE_ETH_DEV_PCI);
+		}
+		if (eth_dev == NULL) {
+			ERROR("can not allocate rte ethdev");
+			err = ENOMEM;
+			goto port_error;
+		}
+
+		eth_dev->data->dev_private = priv;
+		eth_dev->pci_dev = pci_dev;
+		eth_dev->driver = &mlx5_driver;
+		eth_dev->data->rx_mbuf_alloc_failed = 0;
+		eth_dev->data->mtu = ETHER_MTU;
+
+		priv->dev = eth_dev;
+		eth_dev->dev_ops = &mlx5_dev_ops;
+		eth_dev->data->mac_addrs = priv->mac;
+
+		/* Bring Ethernet device up. */
+		DEBUG("forcing Ethernet interface up");
+		priv_set_flags(priv, ~IFF_UP, IFF_UP);
+		continue;
+
+port_error:
+		rte_free(priv);
+		if (pd)
+			claim_zero(ibv_dealloc_pd(pd));
+		if (ctx)
+			claim_zero(ibv_close_device(ctx));
+		break;
+	}
+
+	/*
+	 * XXX if something went wrong in the loop above, there is a resource
+	 * leak (ctx, pd, priv, dpdk ethdev) but we can do nothing about it as
+	 * long as the dpdk does not provide a way to deallocate a ethdev and a
+	 * way to enumerate the registered ethdevs to free the previous ones.
+	 */
+
+	/* no port found, complain */
+	if (!mlx5_dev[idx].ports) {
+		err = ENODEV;
+		goto error;
+	}
+
+error:
+	if (attr_ctx)
+		claim_zero(ibv_close_device(attr_ctx));
+	if (list)
+		ibv_free_device_list(list);
+	assert(err >= 0);
+	return -err;
+}
+
+static const struct rte_pci_id mlx5_pci_id_map[] = {
+	{
+		.vendor_id = PCI_VENDOR_ID_MELLANOX,
+		.device_id = PCI_DEVICE_ID_MELLANOX_CONNECTX4,
+		.subsystem_vendor_id = PCI_ANY_ID,
+		.subsystem_device_id = PCI_ANY_ID
+	},
+	{
+		.vendor_id = PCI_VENDOR_ID_MELLANOX,
+		.device_id = PCI_DEVICE_ID_MELLANOX_CONNECTX4VF,
+		.subsystem_vendor_id = PCI_ANY_ID,
+		.subsystem_device_id = PCI_ANY_ID
+	},
+	{
+		.vendor_id = PCI_VENDOR_ID_MELLANOX,
+		.device_id = PCI_DEVICE_ID_MELLANOX_CONNECTX4LX,
+		.subsystem_vendor_id = PCI_ANY_ID,
+		.subsystem_device_id = PCI_ANY_ID
+	},
+	{
+		.vendor_id = PCI_VENDOR_ID_MELLANOX,
+		.device_id = PCI_DEVICE_ID_MELLANOX_CONNECTX4LXVF,
+		.subsystem_vendor_id = PCI_ANY_ID,
+		.subsystem_device_id = PCI_ANY_ID
+	},
+	{
+		.vendor_id = 0
+	}
+};
+
+static struct eth_driver mlx5_driver = {
+	.pci_drv = {
+		.name = MLX5_DRIVER_NAME,
+		.id_table = mlx5_pci_id_map,
+		.devinit = mlx5_pci_devinit,
+	},
+	.dev_private_size = sizeof(struct priv)
+};
+
+/**
+ * Driver initialization routine.
+ */
+static int
+rte_mlx5_pmd_init(const char *name, const char *args)
+{
+	(void)name;
+	(void)args;
+	/*
+	 * RDMAV_HUGEPAGES_SAFE tells ibv_fork_init() we intend to use
+	 * huge pages. Calling ibv_fork_init() during init allows
+	 * applications to use fork() safely for purposes other than
+	 * using this PMD, which is not supported in forked processes.
+	 */
+	setenv("RDMAV_HUGEPAGES_SAFE", "1", 1);
+	ibv_fork_init();
+	rte_eal_pci_register(&mlx5_driver.pci_drv);
+	return 0;
+}
+
+static struct rte_driver rte_mlx5_driver = {
+	.type = PMD_PDEV,
+	.name = MLX5_DRIVER_NAME,
+	.init = rte_mlx5_pmd_init,
+};
+
+PMD_REGISTER_DRIVER(rte_mlx5_driver)
diff --git a/drivers/net/mlx5/mlx5.h b/drivers/net/mlx5/mlx5.h
new file mode 100644
index 0000000..ef7975d
--- /dev/null
+++ b/drivers/net/mlx5/mlx5.h
@@ -0,0 +1,154 @@
+/*-
+ *   BSD LICENSE
+ *
+ *   Copyright 2015 6WIND S.A.
+ *   Copyright 2015 Mellanox.
+ *
+ *   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 6WIND S.A. 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 RTE_PMD_MLX5_H_
+#define RTE_PMD_MLX5_H_
+
+#include <stddef.h>
+#include <stdint.h>
+#include <limits.h>
+#include <net/if.h>
+#include <netinet/in.h>
+#include <linux/if.h>
+
+/* Verbs header. */
+/* ISO C doesn't support unnamed structs/unions, disabling -pedantic. */
+#ifdef PEDANTIC
+#pragma GCC diagnostic ignored "-pedantic"
+#endif
+#include <infiniband/verbs.h>
+#ifdef PEDANTIC
+#pragma GCC diagnostic error "-pedantic"
+#endif
+
+/* DPDK headers don't like -pedantic. */
+#ifdef PEDANTIC
+#pragma GCC diagnostic ignored "-pedantic"
+#endif
+#include <rte_ether.h>
+#include <rte_ethdev.h>
+#include <rte_spinlock.h>
+#ifdef PEDANTIC
+#pragma GCC diagnostic error "-pedantic"
+#endif
+
+#include "mlx5_utils.h"
+#include "mlx5_autoconf.h"
+#include "mlx5_defs.h"
+
+enum {
+	PCI_VENDOR_ID_MELLANOX = 0x15b3,
+};
+
+enum {
+	PCI_DEVICE_ID_MELLANOX_CONNECTX4 = 0x1013,
+	PCI_DEVICE_ID_MELLANOX_CONNECTX4VF = 0x1014,
+	PCI_DEVICE_ID_MELLANOX_CONNECTX4LX = 0x1015,
+	PCI_DEVICE_ID_MELLANOX_CONNECTX4LXVF = 0x1016,
+};
+
+struct priv {
+	struct rte_eth_dev *dev; /* Ethernet device. */
+	struct ibv_context *ctx; /* Verbs context. */
+	struct ibv_device_attr device_attr; /* Device properties. */
+	struct ibv_pd *pd; /* Protection Domain. */
+	/*
+	 * MAC addresses array and configuration bit-field.
+	 * An extra entry that cannot be modified by the DPDK is reserved
+	 * for broadcast frames (destination MAC address ff:ff:ff:ff:ff:ff).
+	 */
+	struct ether_addr mac[MLX5_MAX_MAC_ADDRESSES];
+	BITFIELD_DECLARE(mac_configured, uint32_t, MLX5_MAX_MAC_ADDRESSES);
+	/* VLAN filters. */
+	struct {
+		unsigned int enabled:1; /* If enabled. */
+		unsigned int id:12; /* VLAN ID (0-4095). */
+	} vlan_filter[MLX5_MAX_VLAN_IDS]; /* VLAN filters table. */
+	/* Device properties. */
+	uint16_t mtu; /* Configured MTU. */
+	uint8_t port; /* Physical port number. */
+	unsigned int started:1; /* Device started, flows enabled. */
+	unsigned int promisc:1; /* Device in promiscuous mode. */
+	unsigned int allmulti:1; /* Device receives all multicast packets. */
+	unsigned int hw_qpg:1; /* QP groups are supported. */
+	unsigned int hw_tss:1; /* TSS is supported. */
+	unsigned int hw_rss:1; /* RSS is supported. */
+	unsigned int hw_csum:1; /* Checksum offload is supported. */
+	unsigned int hw_csum_l2tun:1; /* Same for L2 tunnels. */
+	unsigned int rss:1; /* RSS is enabled. */
+	unsigned int vf:1; /* This is a VF device. */
+	unsigned int max_rss_tbl_sz; /* Maximum number of RSS queues. */
+	rte_spinlock_t lock; /* Lock for control functions. */
+};
+
+/**
+ * Lock private structure to protect it from concurrent access in the
+ * control path.
+ *
+ * @param priv
+ *   Pointer to private structure.
+ */
+static inline void
+priv_lock(struct priv *priv)
+{
+	rte_spinlock_lock(&priv->lock);
+}
+
+/**
+ * Unlock private structure.
+ *
+ * @param priv
+ *   Pointer to private structure.
+ */
+static inline void
+priv_unlock(struct priv *priv)
+{
+	rte_spinlock_unlock(&priv->lock);
+}
+
+/* mlx5_ethdev.c */
+
+int priv_get_ifname(const struct priv *, char (*)[IF_NAMESIZE]);
+int priv_ifreq(const struct priv *, int req, struct ifreq *);
+int priv_get_mtu(struct priv *, uint16_t *);
+int priv_set_flags(struct priv *, unsigned int, unsigned int);
+int mlx5_ibv_device_to_pci_addr(const struct ibv_device *,
+				struct rte_pci_addr *);
+
+/* mlx5_mac.c */
+
+int priv_get_mac(struct priv *, uint8_t (*)[ETHER_ADDR_LEN]);
+int priv_mac_addr_add(struct priv *, unsigned int,
+		      const uint8_t (*)[ETHER_ADDR_LEN]);
+
+#endif /* RTE_PMD_MLX5_H_ */
diff --git a/drivers/net/mlx5/mlx5_defs.h b/drivers/net/mlx5/mlx5_defs.h
new file mode 100644
index 0000000..987ddcf
--- /dev/null
+++ b/drivers/net/mlx5/mlx5_defs.h
@@ -0,0 +1,53 @@
+/*-
+ *   BSD LICENSE
+ *
+ *   Copyright 2015 6WIND S.A.
+ *   Copyright 2015 Mellanox.
+ *
+ *   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 6WIND S.A. 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 RTE_PMD_MLX5_DEFS_H_
+#define RTE_PMD_MLX5_DEFS_H_
+
+/* Reported driver name. */
+#define MLX5_DRIVER_NAME "librte_pmd_mlx5"
+
+/*
+ * Maximum number of simultaneous MAC addresses supported.
+ *
+ * According to ConnectX's Programmer Reference Manual:
+ *   The L2 Address Match is implemented by comparing a MAC/VLAN combination
+ *   of 128 MAC addresses and 127 VLAN values, comprising 128x127 possible
+ *   L2 addresses.
+ */
+#define MLX5_MAX_MAC_ADDRESSES 128
+
+/* Maximum number of simultaneous VLAN filters supported. See above. */
+#define MLX5_MAX_VLAN_IDS 127
+
+#endif /* RTE_PMD_MLX5_DEFS_H_ */
diff --git a/drivers/net/mlx5/mlx5_ethdev.c b/drivers/net/mlx5/mlx5_ethdev.c
new file mode 100644
index 0000000..b6c7d7a
--- /dev/null
+++ b/drivers/net/mlx5/mlx5_ethdev.c
@@ -0,0 +1,420 @@
+/*-
+ *   BSD LICENSE
+ *
+ *   Copyright 2015 6WIND S.A.
+ *   Copyright 2015 Mellanox.
+ *
+ *   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 6WIND S.A. 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 <stddef.h>
+#include <unistd.h>
+#include <stdint.h>
+#include <stdio.h>
+#include <string.h>
+#include <stdlib.h>
+#include <errno.h>
+#include <dirent.h>
+#include <net/if.h>
+#include <sys/ioctl.h>
+#include <sys/socket.h>
+#include <netinet/in.h>
+#include <linux/if.h>
+
+/* DPDK headers don't like -pedantic. */
+#ifdef PEDANTIC
+#pragma GCC diagnostic ignored "-pedantic"
+#endif
+#include <rte_atomic.h>
+#include <rte_ethdev.h>
+#include <rte_mbuf.h>
+#include <rte_common.h>
+#ifdef PEDANTIC
+#pragma GCC diagnostic error "-pedantic"
+#endif
+
+#include "mlx5.h"
+#include "mlx5_utils.h"
+
+/**
+ * Get interface name from private structure.
+ *
+ * @param[in] priv
+ *   Pointer to private structure.
+ * @param[out] ifname
+ *   Interface name output buffer.
+ *
+ * @return
+ *   0 on success, -1 on failure and errno is set.
+ */
+int
+priv_get_ifname(const struct priv *priv, char (*ifname)[IF_NAMESIZE])
+{
+	DIR *dir;
+	struct dirent *dent;
+	unsigned int dev_type = 0;
+	unsigned int dev_port_prev = ~0u;
+	char match[IF_NAMESIZE] = "";
+
+	{
+		MKSTR(path, "%s/device/net", priv->ctx->device->ibdev_path);
+
+		dir = opendir(path);
+		if (dir == NULL)
+			return -1;
+	}
+	while ((dent = readdir(dir)) != NULL) {
+		char *name = dent->d_name;
+		FILE *file;
+		unsigned int dev_port;
+		int r;
+
+		if ((name[0] == '.') &&
+		    ((name[1] == '\0') ||
+		     ((name[1] == '.') && (name[2] == '\0'))))
+			continue;
+
+		MKSTR(path, "%s/device/net/%s/%s",
+		      priv->ctx->device->ibdev_path, name,
+		      (dev_type ? "dev_id" : "dev_port"));
+
+		file = fopen(path, "rb");
+		if (file == NULL) {
+			if (errno != ENOENT)
+				continue;
+			/*
+			 * Switch to dev_id when dev_port does not exist as
+			 * is the case with Linux kernel versions < 3.15.
+			 */
+try_dev_id:
+			match[0] = '\0';
+			if (dev_type)
+				break;
+			dev_type = 1;
+			dev_port_prev = ~0u;
+			rewinddir(dir);
+			continue;
+		}
+		r = fscanf(file, (dev_type ? "%x" : "%u"), &dev_port);
+		fclose(file);
+		if (r != 1)
+			continue;
+		/*
+		 * Switch to dev_id when dev_port returns the same value for
+		 * all ports. May happen when using a MOFED release older than
+		 * 3.0 with a Linux kernel >= 3.15.
+		 */
+		if (dev_port == dev_port_prev)
+			goto try_dev_id;
+		dev_port_prev = dev_port;
+		if (dev_port == (priv->port - 1u))
+			snprintf(match, sizeof(match), "%s", name);
+	}
+	closedir(dir);
+	if (match[0] == '\0')
+		return -1;
+	strncpy(*ifname, match, sizeof(*ifname));
+	return 0;
+}
+
+/**
+ * Read from sysfs entry.
+ *
+ * @param[in] priv
+ *   Pointer to private structure.
+ * @param[in] entry
+ *   Entry name relative to sysfs path.
+ * @param[out] buf
+ *   Data output buffer.
+ * @param size
+ *   Buffer size.
+ *
+ * @return
+ *   0 on success, -1 on failure and errno is set.
+ */
+static int
+priv_sysfs_read(const struct priv *priv, const char *entry,
+		char *buf, size_t size)
+{
+	char ifname[IF_NAMESIZE];
+	FILE *file;
+	int ret;
+	int err;
+
+	if (priv_get_ifname(priv, &ifname))
+		return -1;
+
+	MKSTR(path, "%s/device/net/%s/%s", priv->ctx->device->ibdev_path,
+	      ifname, entry);
+
+	file = fopen(path, "rb");
+	if (file == NULL)
+		return -1;
+	ret = fread(buf, 1, size, file);
+	err = errno;
+	if (((size_t)ret < size) && (ferror(file)))
+		ret = -1;
+	else
+		ret = size;
+	fclose(file);
+	errno = err;
+	return ret;
+}
+
+/**
+ * Write to sysfs entry.
+ *
+ * @param[in] priv
+ *   Pointer to private structure.
+ * @param[in] entry
+ *   Entry name relative to sysfs path.
+ * @param[in] buf
+ *   Data buffer.
+ * @param size
+ *   Buffer size.
+ *
+ * @return
+ *   0 on success, -1 on failure and errno is set.
+ */
+static int
+priv_sysfs_write(const struct priv *priv, const char *entry,
+		 char *buf, size_t size)
+{
+	char ifname[IF_NAMESIZE];
+	FILE *file;
+	int ret;
+	int err;
+
+	if (priv_get_ifname(priv, &ifname))
+		return -1;
+
+	MKSTR(path, "%s/device/net/%s/%s", priv->ctx->device->ibdev_path,
+	      ifname, entry);
+
+	file = fopen(path, "wb");
+	if (file == NULL)
+		return -1;
+	ret = fwrite(buf, 1, size, file);
+	err = errno;
+	if (((size_t)ret < size) || (ferror(file)))
+		ret = -1;
+	else
+		ret = size;
+	fclose(file);
+	errno = err;
+	return ret;
+}
+
+/**
+ * Get unsigned long sysfs property.
+ *
+ * @param priv
+ *   Pointer to private structure.
+ * @param[in] name
+ *   Entry name relative to sysfs path.
+ * @param[out] value
+ *   Value output buffer.
+ *
+ * @return
+ *   0 on success, -1 on failure and errno is set.
+ */
+static int
+priv_get_sysfs_ulong(struct priv *priv, const char *name, unsigned long *value)
+{
+	int ret;
+	unsigned long value_ret;
+	char value_str[32];
+
+	ret = priv_sysfs_read(priv, name, value_str, (sizeof(value_str) - 1));
+	if (ret == -1) {
+		DEBUG("cannot read %s value from sysfs: %s",
+		      name, strerror(errno));
+		return -1;
+	}
+	value_str[ret] = '\0';
+	errno = 0;
+	value_ret = strtoul(value_str, NULL, 0);
+	if (errno) {
+		DEBUG("invalid %s value `%s': %s", name, value_str,
+		      strerror(errno));
+		return -1;
+	}
+	*value = value_ret;
+	return 0;
+}
+
+/**
+ * Set unsigned long sysfs property.
+ *
+ * @param priv
+ *   Pointer to private structure.
+ * @param[in] name
+ *   Entry name relative to sysfs path.
+ * @param value
+ *   Value to set.
+ *
+ * @return
+ *   0 on success, -1 on failure and errno is set.
+ */
+static int
+priv_set_sysfs_ulong(struct priv *priv, const char *name, unsigned long value)
+{
+	int ret;
+	MKSTR(value_str, "%lu", value);
+
+	ret = priv_sysfs_write(priv, name, value_str, (sizeof(value_str) - 1));
+	if (ret == -1) {
+		DEBUG("cannot write %s `%s' (%lu) to sysfs: %s",
+		      name, value_str, value, strerror(errno));
+		return -1;
+	}
+	return 0;
+}
+
+/**
+ * Perform ifreq ioctl() on associated Ethernet device.
+ *
+ * @param[in] priv
+ *   Pointer to private structure.
+ * @param req
+ *   Request number to pass to ioctl().
+ * @param[out] ifr
+ *   Interface request structure output buffer.
+ *
+ * @return
+ *   0 on success, -1 on failure and errno is set.
+ */
+int
+priv_ifreq(const struct priv *priv, int req, struct ifreq *ifr)
+{
+	int sock = socket(PF_INET, SOCK_DGRAM, IPPROTO_IP);
+	int ret = -1;
+
+	if (sock == -1)
+		return ret;
+	if (priv_get_ifname(priv, &ifr->ifr_name) == 0)
+		ret = ioctl(sock, req, ifr);
+	close(sock);
+	return ret;
+}
+
+/**
+ * Get device MTU.
+ *
+ * @param priv
+ *   Pointer to private structure.
+ * @param[out] mtu
+ *   MTU value output buffer.
+ *
+ * @return
+ *   0 on success, -1 on failure and errno is set.
+ */
+int
+priv_get_mtu(struct priv *priv, uint16_t *mtu)
+{
+	unsigned long ulong_mtu;
+
+	if (priv_get_sysfs_ulong(priv, "mtu", &ulong_mtu) == -1)
+		return -1;
+	*mtu = ulong_mtu;
+	return 0;
+}
+
+/**
+ * Set device flags.
+ *
+ * @param priv
+ *   Pointer to private structure.
+ * @param keep
+ *   Bitmask for flags that must remain untouched.
+ * @param flags
+ *   Bitmask for flags to modify.
+ *
+ * @return
+ *   0 on success, -1 on failure and errno is set.
+ */
+int
+priv_set_flags(struct priv *priv, unsigned int keep, unsigned int flags)
+{
+	unsigned long tmp;
+
+	if (priv_get_sysfs_ulong(priv, "flags", &tmp) == -1)
+		return -1;
+	tmp &= keep;
+	tmp |= flags;
+	return priv_set_sysfs_ulong(priv, "flags", tmp);
+}
+
+/**
+ * Get PCI information from struct ibv_device.
+ *
+ * @param device
+ *   Pointer to Ethernet device structure.
+ * @param[out] pci_addr
+ *   PCI bus address output buffer.
+ *
+ * @return
+ *   0 on success, -1 on failure and errno is set.
+ */
+int
+mlx5_ibv_device_to_pci_addr(const struct ibv_device *device,
+			    struct rte_pci_addr *pci_addr)
+{
+	FILE *file;
+	char line[32];
+	MKSTR(path, "%s/device/uevent", device->ibdev_path);
+
+	file = fopen(path, "rb");
+	if (file == NULL)
+		return -1;
+	while (fgets(line, sizeof(line), file) == line) {
+		size_t len = strlen(line);
+		int ret;
+
+		/* Truncate long lines. */
+		if (len == (sizeof(line) - 1))
+			while (line[(len - 1)] != '\n') {
+				ret = fgetc(file);
+				if (ret == EOF)
+					break;
+				line[(len - 1)] = ret;
+			}
+		/* Extract information. */
+		if (sscanf(line,
+			   "PCI_SLOT_NAME="
+			   "%" SCNx16 ":%" SCNx8 ":%" SCNx8 ".%" SCNx8 "\n",
+			   &pci_addr->domain,
+			   &pci_addr->bus,
+			   &pci_addr->devid,
+			   &pci_addr->function) == 4) {
+			ret = 0;
+			break;
+		}
+	}
+	fclose(file);
+	return 0;
+}
diff --git a/drivers/net/mlx5/mlx5_mac.c b/drivers/net/mlx5/mlx5_mac.c
new file mode 100644
index 0000000..f7e1cf6
--- /dev/null
+++ b/drivers/net/mlx5/mlx5_mac.c
@@ -0,0 +1,150 @@
+/*-
+ *   BSD LICENSE
+ *
+ *   Copyright 2015 6WIND S.A.
+ *   Copyright 2015 Mellanox.
+ *
+ *   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 6WIND S.A. 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 <stddef.h>
+#include <assert.h>
+#include <stdint.h>
+#include <string.h>
+#include <inttypes.h>
+#include <errno.h>
+#include <netinet/in.h>
+#include <linux/if.h>
+#include <sys/ioctl.h>
+#include <arpa/inet.h>
+
+/* Verbs header. */
+/* ISO C doesn't support unnamed structs/unions, disabling -pedantic. */
+#ifdef PEDANTIC
+#pragma GCC diagnostic ignored "-pedantic"
+#endif
+#include <infiniband/verbs.h>
+#ifdef PEDANTIC
+#pragma GCC diagnostic error "-pedantic"
+#endif
+
+/* DPDK headers don't like -pedantic. */
+#ifdef PEDANTIC
+#pragma GCC diagnostic ignored "-pedantic"
+#endif
+#include <rte_ether.h>
+#include <rte_ethdev.h>
+#include <rte_common.h>
+#ifdef PEDANTIC
+#pragma GCC diagnostic error "-pedantic"
+#endif
+
+#include "mlx5.h"
+#include "mlx5_utils.h"
+
+/**
+ * Get MAC address by querying netdevice.
+ *
+ * @param[in] priv
+ *   struct priv for the requested device.
+ * @param[out] mac
+ *   MAC address output buffer.
+ *
+ * @return
+ *   0 on success, -1 on failure and errno is set.
+ */
+int
+priv_get_mac(struct priv *priv, uint8_t (*mac)[ETHER_ADDR_LEN])
+{
+	struct ifreq request;
+
+	if (priv_ifreq(priv, SIOCGIFHWADDR, &request))
+		return -1;
+	memcpy(mac, request.ifr_hwaddr.sa_data, ETHER_ADDR_LEN);
+	return 0;
+}
+
+/**
+ * Unregister a MAC address.
+ *
+ * @param priv
+ *   Pointer to private structure.
+ * @param mac_index
+ *   MAC address index.
+ */
+static void
+priv_mac_addr_del(struct priv *priv, unsigned int mac_index)
+{
+	assert(mac_index < RTE_DIM(priv->mac));
+	if (!BITFIELD_ISSET(priv->mac_configured, mac_index))
+		return;
+	BITFIELD_RESET(priv->mac_configured, mac_index);
+}
+
+/**
+ * Register a MAC address.
+ *
+ * @param priv
+ *   Pointer to private structure.
+ * @param mac_index
+ *   MAC address index to use.
+ * @param mac
+ *   MAC address to register.
+ *
+ * @return
+ *   0 on success, errno value on failure.
+ */
+int
+priv_mac_addr_add(struct priv *priv, unsigned int mac_index,
+		  const uint8_t (*mac)[ETHER_ADDR_LEN])
+{
+	unsigned int i;
+
+	assert(mac_index < RTE_DIM(priv->mac));
+	/* First, make sure this address isn't already configured. */
+	for (i = 0; (i != RTE_DIM(priv->mac)); ++i) {
+		/* Skip this index, it's going to be reconfigured. */
+		if (i == mac_index)
+			continue;
+		if (!BITFIELD_ISSET(priv->mac_configured, i))
+			continue;
+		if (memcmp(priv->mac[i].addr_bytes, *mac, sizeof(*mac)))
+			continue;
+		/* Address already configured elsewhere, return with error. */
+		return EADDRINUSE;
+	}
+	if (BITFIELD_ISSET(priv->mac_configured, mac_index))
+		priv_mac_addr_del(priv, mac_index);
+	priv->mac[mac_index] = (struct ether_addr){
+		{
+			(*mac)[0], (*mac)[1], (*mac)[2],
+			(*mac)[3], (*mac)[4], (*mac)[5]
+		}
+	};
+	BITFIELD_SET(priv->mac_configured, mac_index);
+	return 0;
+}
diff --git a/drivers/net/mlx5/mlx5_utils.h b/drivers/net/mlx5/mlx5_utils.h
new file mode 100644
index 0000000..cc6aab6
--- /dev/null
+++ b/drivers/net/mlx5/mlx5_utils.h
@@ -0,0 +1,149 @@
+/*-
+ *   BSD LICENSE
+ *
+ *   Copyright 2015 6WIND S.A.
+ *   Copyright 2015 Mellanox.
+ *
+ *   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 6WIND S.A. 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 RTE_PMD_MLX5_UTILS_H_
+#define RTE_PMD_MLX5_UTILS_H_
+
+#include <stddef.h>
+#include <stdio.h>
+#include <limits.h>
+#include <assert.h>
+#include <errno.h>
+
+#include "mlx5_defs.h"
+
+/* Bit-field manipulation. */
+#define BITFIELD_DECLARE(bf, type, size) \
+	type bf[(((size_t)(size) / (sizeof(type) * CHAR_BIT)) + \
+		 !!((size_t)(size) % (sizeof(type) * CHAR_BIT)))]
+#define BITFIELD_DEFINE(bf, type, size) \
+	BITFIELD_DECLARE((bf), type, (size)) = { 0 }
+#define BITFIELD_SET(bf, b) \
+	(assert((size_t)(b) < (sizeof(bf) * CHAR_BIT)), \
+	 (void)((bf)[((b) / (sizeof((bf)[0]) * CHAR_BIT))] |= \
+		((size_t)1 << ((b) % (sizeof((bf)[0]) * CHAR_BIT)))))
+#define BITFIELD_RESET(bf, b) \
+	(assert((size_t)(b) < (sizeof(bf) * CHAR_BIT)), \
+	 (void)((bf)[((b) / (sizeof((bf)[0]) * CHAR_BIT))] &= \
+		~((size_t)1 << ((b) % (sizeof((bf)[0]) * CHAR_BIT)))))
+#define BITFIELD_ISSET(bf, b) \
+	(assert((size_t)(b) < (sizeof(bf) * CHAR_BIT)), \
+	 !!(((bf)[((b) / (sizeof((bf)[0]) * CHAR_BIT))] & \
+	     ((size_t)1 << ((b) % (sizeof((bf)[0]) * CHAR_BIT))))))
+
+/* Save and restore errno around argument evaluation. */
+#define ERRNO_SAFE(x) ((errno = (int []){ errno, ((x), 0) }[0]))
+
+/*
+ * Helper macros to work around __VA_ARGS__ limitations in a C99 compliant
+ * manner.
+ */
+#define PMD_DRV_LOG_STRIP(a, b) a
+#define PMD_DRV_LOG_OPAREN (
+#define PMD_DRV_LOG_CPAREN )
+#define PMD_DRV_LOG_COMMA ,
+
+/* Return the file name part of a path. */
+static inline const char *
+pmd_drv_log_basename(const char *s)
+{
+	const char *n = s;
+
+	while (*n)
+		if (*(n++) == '/')
+			s = n;
+	return s;
+}
+
+/*
+ * When debugging is enabled (NDEBUG not defined), file, line and function
+ * information replace the driver name (MLX5_DRIVER_NAME) in log messages.
+ */
+#ifndef NDEBUG
+
+#define PMD_DRV_LOG___(level, ...) \
+	ERRNO_SAFE(RTE_LOG(level, PMD, __VA_ARGS__))
+#define PMD_DRV_LOG__(level, ...) \
+	PMD_DRV_LOG___(level, "%s:%u: %s(): " __VA_ARGS__)
+#define PMD_DRV_LOG_(level, s, ...) \
+	PMD_DRV_LOG__(level, \
+		s "\n" PMD_DRV_LOG_COMMA \
+		pmd_drv_log_basename(__FILE__) PMD_DRV_LOG_COMMA \
+		__LINE__ PMD_DRV_LOG_COMMA \
+		__func__, \
+		__VA_ARGS__)
+
+#else /* NDEBUG */
+
+#define PMD_DRV_LOG___(level, ...) \
+	ERRNO_SAFE(RTE_LOG(level, PMD, MLX5_DRIVER_NAME ": " __VA_ARGS__))
+#define PMD_DRV_LOG__(level, ...) \
+	PMD_DRV_LOG___(level, __VA_ARGS__)
+#define PMD_DRV_LOG_(level, s, ...) \
+	PMD_DRV_LOG__(level, s "\n", __VA_ARGS__)
+
+#endif /* NDEBUG */
+
+/* Generic printf()-like logging macro with automatic line feed. */
+#define PMD_DRV_LOG(level, ...) \
+	PMD_DRV_LOG_(level, \
+		__VA_ARGS__ PMD_DRV_LOG_STRIP PMD_DRV_LOG_OPAREN, \
+		PMD_DRV_LOG_CPAREN)
+
+/*
+ * Like assert(), DEBUG() becomes a no-op and claim_zero() does not perform
+ * any check when debugging is disabled.
+ */
+#ifndef NDEBUG
+
+#define DEBUG(...) PMD_DRV_LOG(DEBUG, __VA_ARGS__)
+#define claim_zero(...) assert((__VA_ARGS__) == 0)
+
+#else /* NDEBUG */
+
+#define DEBUG(...) (void)0
+#define claim_zero(...) (__VA_ARGS__)
+
+#endif /* NDEBUG */
+
+#define INFO(...) PMD_DRV_LOG(INFO, __VA_ARGS__)
+#define WARN(...) PMD_DRV_LOG(WARNING, __VA_ARGS__)
+#define ERROR(...) PMD_DRV_LOG(ERR, __VA_ARGS__)
+
+/* Allocate a buffer on the stack and fill it with a printf format string. */
+#define MKSTR(name, ...) \
+	char name[snprintf(NULL, 0, __VA_ARGS__) + 1]; \
+	\
+	snprintf(name, sizeof(name), __VA_ARGS__)
+
+#endif /* RTE_PMD_MLX5_UTILS_H_ */
diff --git a/drivers/net/mlx5/rte_pmd_mlx5_version.map b/drivers/net/mlx5/rte_pmd_mlx5_version.map
new file mode 100644
index 0000000..ad607bb
--- /dev/null
+++ b/drivers/net/mlx5/rte_pmd_mlx5_version.map
@@ -0,0 +1,3 @@
+DPDK_2.2 {
+	local: *;
+};
diff --git a/mk/rte.app.mk b/mk/rte.app.mk
index 9e1909e..724efa7 100644
--- a/mk/rte.app.mk
+++ b/mk/rte.app.mk
@@ -104,6 +104,10 @@ ifeq ($(CONFIG_RTE_BUILD_SHARED_LIB),n)
 _LDLIBS-$(CONFIG_RTE_LIBRTE_MLX4_PMD)       += -libverbs
 endif # ! CONFIG_RTE_BUILD_SHARED_LIBS
 
+ifeq ($(CONFIG_RTE_BUILD_SHARED_LIB),n)
+_LDLIBS-$(CONFIG_RTE_LIBRTE_MLX5_PMD)       += -libverbs
+endif # ! CONFIG_RTE_BUILD_SHARED_LIBS
+
 _LDLIBS-$(CONFIG_RTE_LIBRTE_BNX2X_PMD)      += -lz
 
 _LDLIBS-y += --start-group
@@ -137,6 +141,7 @@ _LDLIBS-$(CONFIG_RTE_LIBRTE_FM10K_PMD)      += -lrte_pmd_fm10k
 _LDLIBS-$(CONFIG_RTE_LIBRTE_IXGBE_PMD)      += -lrte_pmd_ixgbe
 _LDLIBS-$(CONFIG_RTE_LIBRTE_E1000_PMD)      += -lrte_pmd_e1000
 _LDLIBS-$(CONFIG_RTE_LIBRTE_MLX4_PMD)       += -lrte_pmd_mlx4
+_LDLIBS-$(CONFIG_RTE_LIBRTE_MLX5_PMD)       += -lrte_pmd_mlx5
 _LDLIBS-$(CONFIG_RTE_LIBRTE_MPIPE_PMD)      += -lrte_pmd_mpipe -lgxio
 _LDLIBS-$(CONFIG_RTE_LIBRTE_PMD_RING)       += -lrte_pmd_ring
 _LDLIBS-$(CONFIG_RTE_LIBRTE_PMD_PCAP)       += -lrte_pmd_pcap
-- 
2.1.0

^ permalink raw reply	[flat|nested] 32+ messages in thread

* [dpdk-dev] [PATCH 02/13] mlx5: add non-scattered TX and RX support
  2015-10-05 17:52 [dpdk-dev] [PATCH 00/13] Mellanox ConnectX-4 PMD (mlx5) Adrien Mazarguil
  2015-10-05 17:52 ` [dpdk-dev] [PATCH 01/13] mlx5: new poll-mode driver for Mellanox ConnectX-4 adapters Adrien Mazarguil
@ 2015-10-05 17:52 ` Adrien Mazarguil
  2015-10-05 17:52 ` [dpdk-dev] [PATCH 03/13] mlx5: add MAC handling Adrien Mazarguil
                   ` (11 subsequent siblings)
  13 siblings, 0 replies; 32+ messages in thread
From: Adrien Mazarguil @ 2015-10-05 17:52 UTC (permalink / raw)
  To: dev

RSS implementation with parent/child QPs comes from mlx4 and is temporary.

Signed-off-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>
Signed-off-by: Nelio Laranjeiro <nelio.laranjeiro@6wind.com>
---
 config/common_bsdapp          |   3 +
 config/common_linuxapp        |   3 +
 drivers/net/mlx5/Makefile     |  15 +
 drivers/net/mlx5/mlx5.c       |  40 +++
 drivers/net/mlx5/mlx5.h       |  25 ++
 drivers/net/mlx5/mlx5_defs.h  |  24 ++
 drivers/net/mlx5/mlx5_rxq.c   | 682 ++++++++++++++++++++++++++++++++++++++++++
 drivers/net/mlx5/mlx5_rxtx.c  | 495 ++++++++++++++++++++++++++++++
 drivers/net/mlx5/mlx5_rxtx.h  | 161 ++++++++++
 drivers/net/mlx5/mlx5_txq.c   | 512 +++++++++++++++++++++++++++++++
 drivers/net/mlx5/mlx5_utils.h |  11 +
 11 files changed, 1971 insertions(+)
 create mode 100644 drivers/net/mlx5/mlx5_rxq.c
 create mode 100644 drivers/net/mlx5/mlx5_rxtx.c
 create mode 100644 drivers/net/mlx5/mlx5_rxtx.h
 create mode 100644 drivers/net/mlx5/mlx5_txq.c

diff --git a/config/common_bsdapp b/config/common_bsdapp
index 1e6885f..3b50ff9 100644
--- a/config/common_bsdapp
+++ b/config/common_bsdapp
@@ -218,6 +218,9 @@ CONFIG_RTE_LIBRTE_MLX4_SOFT_COUNTERS=1
 #
 CONFIG_RTE_LIBRTE_MLX5_PMD=n
 CONFIG_RTE_LIBRTE_MLX5_DEBUG=n
+CONFIG_RTE_LIBRTE_MLX5_SGE_WR_N=4
+CONFIG_RTE_LIBRTE_MLX5_MAX_INLINE=0
+CONFIG_RTE_LIBRTE_MLX5_TX_MP_CACHE=8
 
 #
 # Compile burst-oriented Broadcom PMD driver
diff --git a/config/common_linuxapp b/config/common_linuxapp
index 7da7ba7..eed8fc0 100644
--- a/config/common_linuxapp
+++ b/config/common_linuxapp
@@ -216,6 +216,9 @@ CONFIG_RTE_LIBRTE_MLX4_SOFT_COUNTERS=1
 #
 CONFIG_RTE_LIBRTE_MLX5_PMD=n
 CONFIG_RTE_LIBRTE_MLX5_DEBUG=n
+CONFIG_RTE_LIBRTE_MLX5_SGE_WR_N=4
+CONFIG_RTE_LIBRTE_MLX5_MAX_INLINE=0
+CONFIG_RTE_LIBRTE_MLX5_TX_MP_CACHE=8
 
 #
 # Compile burst-oriented Broadcom PMD driver
diff --git a/drivers/net/mlx5/Makefile b/drivers/net/mlx5/Makefile
index 6e63073..7b9c57b 100644
--- a/drivers/net/mlx5/Makefile
+++ b/drivers/net/mlx5/Makefile
@@ -42,6 +42,9 @@ LIB = librte_pmd_mlx5.a
 
 # Sources.
 SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5.c
+SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5_rxq.c
+SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5_txq.c
+SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5_rxtx.c
 SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5_ethdev.c
 SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5_mac.c
 
@@ -79,6 +82,18 @@ else
 CFLAGS += -DNDEBUG -UPEDANTIC
 endif
 
+ifdef CONFIG_RTE_LIBRTE_MLX5_SGE_WR_N
+CFLAGS += -DMLX5_PMD_SGE_WR_N=$(CONFIG_RTE_LIBRTE_MLX5_SGE_WR_N)
+endif
+
+ifdef CONFIG_RTE_LIBRTE_MLX5_MAX_INLINE
+CFLAGS += -DMLX5_PMD_MAX_INLINE=$(CONFIG_RTE_LIBRTE_MLX5_MAX_INLINE)
+endif
+
+ifdef CONFIG_RTE_LIBRTE_MLX5_TX_MP_CACHE
+CFLAGS += -DMLX5_PMD_TX_MP_CACHE=$(CONFIG_RTE_LIBRTE_MLX5_TX_MP_CACHE)
+endif
+
 include $(RTE_SDK)/mk/rte.lib.mk
 
 # Generate and clean-up mlx5_autoconf.h.
diff --git a/drivers/net/mlx5/mlx5.c b/drivers/net/mlx5/mlx5.c
index f6fe8a6..31ce5ec 100644
--- a/drivers/net/mlx5/mlx5.c
+++ b/drivers/net/mlx5/mlx5.c
@@ -63,6 +63,7 @@
 
 #include "mlx5.h"
 #include "mlx5_utils.h"
+#include "mlx5_rxtx.h"
 #include "mlx5_autoconf.h"
 
 /**
@@ -77,11 +78,46 @@ static void
 mlx5_dev_close(struct rte_eth_dev *dev)
 {
 	struct priv *priv = dev->data->dev_private;
+	void *tmp;
+	unsigned int i;
 
 	priv_lock(priv);
 	DEBUG("%p: closing device \"%s\"",
 	      (void *)dev,
 	      ((priv->ctx != NULL) ? priv->ctx->device->name : ""));
+	/* Prevent crashes when queues are still in use. */
+	dev->rx_pkt_burst = removed_rx_burst;
+	dev->tx_pkt_burst = removed_tx_burst;
+	if (priv->rxqs != NULL) {
+		/* XXX race condition if mlx5_rx_burst() is still running. */
+		usleep(1000);
+		for (i = 0; (i != priv->rxqs_n); ++i) {
+			tmp = (*priv->rxqs)[i];
+			if (tmp == NULL)
+				continue;
+			(*priv->rxqs)[i] = NULL;
+			rxq_cleanup(tmp);
+			rte_free(tmp);
+		}
+		priv->rxqs_n = 0;
+		priv->rxqs = NULL;
+	}
+	if (priv->txqs != NULL) {
+		/* XXX race condition if mlx5_tx_burst() is still running. */
+		usleep(1000);
+		for (i = 0; (i != priv->txqs_n); ++i) {
+			tmp = (*priv->txqs)[i];
+			if (tmp == NULL)
+				continue;
+			(*priv->txqs)[i] = NULL;
+			txq_cleanup(tmp);
+			rte_free(tmp);
+		}
+		priv->txqs_n = 0;
+		priv->txqs = NULL;
+	}
+	if (priv->rss)
+		rxq_cleanup(&priv->rxq_parent);
 	if (priv->pd != NULL) {
 		assert(priv->ctx != NULL);
 		claim_zero(ibv_dealloc_pd(priv->pd));
@@ -94,6 +130,10 @@ mlx5_dev_close(struct rte_eth_dev *dev)
 
 static const struct eth_dev_ops mlx5_dev_ops = {
 	.dev_close = mlx5_dev_close,
+	.rx_queue_setup = mlx5_rx_queue_setup,
+	.tx_queue_setup = mlx5_tx_queue_setup,
+	.rx_queue_release = mlx5_rx_queue_release,
+	.tx_queue_release = mlx5_tx_queue_release,
 };
 
 static struct {
diff --git a/drivers/net/mlx5/mlx5.h b/drivers/net/mlx5/mlx5.h
index ef7975d..7e60e70 100644
--- a/drivers/net/mlx5/mlx5.h
+++ b/drivers/net/mlx5/mlx5.h
@@ -63,6 +63,7 @@
 #endif
 
 #include "mlx5_utils.h"
+#include "mlx5_rxtx.h"
 #include "mlx5_autoconf.h"
 #include "mlx5_defs.h"
 
@@ -108,9 +109,33 @@ struct priv {
 	unsigned int rss:1; /* RSS is enabled. */
 	unsigned int vf:1; /* This is a VF device. */
 	unsigned int max_rss_tbl_sz; /* Maximum number of RSS queues. */
+	/* RX/TX queues. */
+	struct rxq rxq_parent; /* Parent queue when RSS is enabled. */
+	unsigned int rxqs_n; /* RX queues array size. */
+	unsigned int txqs_n; /* TX queues array size. */
+	struct rxq *(*rxqs)[]; /* RX queues. */
+	struct txq *(*txqs)[]; /* TX queues. */
 	rte_spinlock_t lock; /* Lock for control functions. */
 };
 
+/* Work Request ID data type (64 bit). */
+typedef union {
+	struct {
+		uint32_t id;
+		uint16_t offset;
+	} data;
+	uint64_t raw;
+} wr_id_t;
+
+/* Compile-time check. */
+static inline void wr_id_t_check(void)
+{
+	wr_id_t check[1 + (2 * -!(sizeof(wr_id_t) == sizeof(uint64_t)))];
+
+	(void)check;
+	(void)wr_id_t_check;
+}
+
 /**
  * Lock private structure to protect it from concurrent access in the
  * control path.
diff --git a/drivers/net/mlx5/mlx5_defs.h b/drivers/net/mlx5/mlx5_defs.h
index 987ddcf..4f13a4e 100644
--- a/drivers/net/mlx5/mlx5_defs.h
+++ b/drivers/net/mlx5/mlx5_defs.h
@@ -50,4 +50,28 @@
 /* Maximum number of simultaneous VLAN filters supported. See above. */
 #define MLX5_MAX_VLAN_IDS 127
 
+/* Request send completion once in every 64 sends, might be less. */
+#define MLX5_PMD_TX_PER_COMP_REQ 64
+
+/* Maximum number of Scatter/Gather Elements per Work Request. */
+#ifndef MLX5_PMD_SGE_WR_N
+#define MLX5_PMD_SGE_WR_N 4
+#endif
+
+/* Maximum size for inline data. */
+#ifndef MLX5_PMD_MAX_INLINE
+#define MLX5_PMD_MAX_INLINE 0
+#endif
+
+/*
+ * Maximum number of cached Memory Pools (MPs) per TX queue. Each RTE MP
+ * from which buffers are to be transmitted will have to be mapped by this
+ * driver to their own Memory Region (MR). This is a slow operation.
+ *
+ * This value is always 1 for RX queues.
+ */
+#ifndef MLX5_PMD_TX_MP_CACHE
+#define MLX5_PMD_TX_MP_CACHE 8
+#endif
+
 #endif /* RTE_PMD_MLX5_DEFS_H_ */
diff --git a/drivers/net/mlx5/mlx5_rxq.c b/drivers/net/mlx5/mlx5_rxq.c
new file mode 100644
index 0000000..01cc649
--- /dev/null
+++ b/drivers/net/mlx5/mlx5_rxq.c
@@ -0,0 +1,682 @@
+/*-
+ *   BSD LICENSE
+ *
+ *   Copyright 2015 6WIND S.A.
+ *   Copyright 2015 Mellanox.
+ *
+ *   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 6WIND S.A. 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 <stddef.h>
+#include <assert.h>
+#include <errno.h>
+#include <string.h>
+#include <stdint.h>
+
+/* Verbs header. */
+/* ISO C doesn't support unnamed structs/unions, disabling -pedantic. */
+#ifdef PEDANTIC
+#pragma GCC diagnostic ignored "-pedantic"
+#endif
+#include <infiniband/verbs.h>
+#ifdef PEDANTIC
+#pragma GCC diagnostic error "-pedantic"
+#endif
+
+/* DPDK headers don't like -pedantic. */
+#ifdef PEDANTIC
+#pragma GCC diagnostic ignored "-pedantic"
+#endif
+#include <rte_mbuf.h>
+#include <rte_malloc.h>
+#include <rte_ethdev.h>
+#include <rte_common.h>
+#ifdef PEDANTIC
+#pragma GCC diagnostic error "-pedantic"
+#endif
+
+#include "mlx5.h"
+#include "mlx5_rxtx.h"
+#include "mlx5_utils.h"
+#include "mlx5_defs.h"
+
+/**
+ * Allocate RX queue elements.
+ *
+ * @param rxq
+ *   Pointer to RX queue structure.
+ * @param elts_n
+ *   Number of elements to allocate.
+ * @param[in] pool
+ *   If not NULL, fetch buffers from this array instead of allocating them
+ *   with rte_pktmbuf_alloc().
+ *
+ * @return
+ *   0 on success, errno value on failure.
+ */
+static int
+rxq_alloc_elts(struct rxq *rxq, unsigned int elts_n, struct rte_mbuf **pool)
+{
+	unsigned int i;
+	struct rxq_elt (*elts)[elts_n] =
+		rte_calloc_socket("RXQ elements", 1, sizeof(*elts), 0,
+				  rxq->socket);
+	int ret = 0;
+
+	if (elts == NULL) {
+		ERROR("%p: can't allocate packets array", (void *)rxq);
+		ret = ENOMEM;
+		goto error;
+	}
+	/* For each WR (packet). */
+	for (i = 0; (i != elts_n); ++i) {
+		struct rxq_elt *elt = &(*elts)[i];
+		struct ibv_recv_wr *wr = &elt->wr;
+		struct ibv_sge *sge = &(*elts)[i].sge;
+		struct rte_mbuf *buf;
+
+		if (pool != NULL) {
+			buf = *(pool++);
+			assert(buf != NULL);
+			rte_pktmbuf_reset(buf);
+		} else
+			buf = rte_pktmbuf_alloc(rxq->mp);
+		if (buf == NULL) {
+			assert(pool == NULL);
+			ERROR("%p: empty mbuf pool", (void *)rxq);
+			ret = ENOMEM;
+			goto error;
+		}
+		/* Configure WR. Work request ID contains its own index in
+		 * the elts array and the offset between SGE buffer header and
+		 * its data. */
+		WR_ID(wr->wr_id).id = i;
+		WR_ID(wr->wr_id).offset =
+			(((uintptr_t)buf->buf_addr + RTE_PKTMBUF_HEADROOM) -
+			 (uintptr_t)buf);
+		wr->next = &(*elts)[(i + 1)].wr;
+		wr->sg_list = sge;
+		wr->num_sge = 1;
+		/* Headroom is reserved by rte_pktmbuf_alloc(). */
+		assert(DATA_OFF(buf) == RTE_PKTMBUF_HEADROOM);
+		/* Buffer is supposed to be empty. */
+		assert(rte_pktmbuf_data_len(buf) == 0);
+		assert(rte_pktmbuf_pkt_len(buf) == 0);
+		/* sge->addr must be able to store a pointer. */
+		assert(sizeof(sge->addr) >= sizeof(uintptr_t));
+		/* SGE keeps its headroom. */
+		sge->addr = (uintptr_t)
+			((uint8_t *)buf->buf_addr + RTE_PKTMBUF_HEADROOM);
+		sge->length = (buf->buf_len - RTE_PKTMBUF_HEADROOM);
+		sge->lkey = rxq->mr->lkey;
+		/* Redundant check for tailroom. */
+		assert(sge->length == rte_pktmbuf_tailroom(buf));
+		/* Make sure elts index and SGE mbuf pointer can be deduced
+		 * from WR ID. */
+		if ((WR_ID(wr->wr_id).id != i) ||
+		    ((void *)((uintptr_t)sge->addr -
+			WR_ID(wr->wr_id).offset) != buf)) {
+			ERROR("%p: cannot store index and offset in WR ID",
+			      (void *)rxq);
+			sge->addr = 0;
+			rte_pktmbuf_free(buf);
+			ret = EOVERFLOW;
+			goto error;
+		}
+	}
+	/* The last WR pointer must be NULL. */
+	(*elts)[(i - 1)].wr.next = NULL;
+	DEBUG("%p: allocated and configured %u single-segment WRs",
+	      (void *)rxq, elts_n);
+	rxq->elts_n = elts_n;
+	rxq->elts_head = 0;
+	rxq->elts.no_sp = elts;
+	assert(ret == 0);
+	return 0;
+error:
+	if (elts != NULL) {
+		assert(pool == NULL);
+		for (i = 0; (i != RTE_DIM(*elts)); ++i) {
+			struct rxq_elt *elt = &(*elts)[i];
+			struct rte_mbuf *buf;
+
+			if (elt->sge.addr == 0)
+				continue;
+			assert(WR_ID(elt->wr.wr_id).id == i);
+			buf = (void *)((uintptr_t)elt->sge.addr -
+				WR_ID(elt->wr.wr_id).offset);
+			rte_pktmbuf_free_seg(buf);
+		}
+		rte_free(elts);
+	}
+	DEBUG("%p: failed, freed everything", (void *)rxq);
+	assert(ret > 0);
+	return ret;
+}
+
+/**
+ * Free RX queue elements.
+ *
+ * @param rxq
+ *   Pointer to RX queue structure.
+ */
+static void
+rxq_free_elts(struct rxq *rxq)
+{
+	unsigned int i;
+	unsigned int elts_n = rxq->elts_n;
+	struct rxq_elt (*elts)[elts_n] = rxq->elts.no_sp;
+
+	DEBUG("%p: freeing WRs", (void *)rxq);
+	rxq->elts_n = 0;
+	rxq->elts.no_sp = NULL;
+	if (elts == NULL)
+		return;
+	for (i = 0; (i != RTE_DIM(*elts)); ++i) {
+		struct rxq_elt *elt = &(*elts)[i];
+		struct rte_mbuf *buf;
+
+		if (elt->sge.addr == 0)
+			continue;
+		assert(WR_ID(elt->wr.wr_id).id == i);
+		buf = (void *)((uintptr_t)elt->sge.addr -
+			WR_ID(elt->wr.wr_id).offset);
+		rte_pktmbuf_free_seg(buf);
+	}
+	rte_free(elts);
+}
+
+/**
+ * Clean up a RX queue.
+ *
+ * Destroy objects, free allocated memory and reset the structure for reuse.
+ *
+ * @param rxq
+ *   Pointer to RX queue structure.
+ */
+void
+rxq_cleanup(struct rxq *rxq)
+{
+	struct ibv_exp_release_intf_params params;
+
+	DEBUG("cleaning up %p", (void *)rxq);
+	rxq_free_elts(rxq);
+	if (rxq->if_qp != NULL) {
+		assert(rxq->priv != NULL);
+		assert(rxq->priv->ctx != NULL);
+		assert(rxq->qp != NULL);
+		params = (struct ibv_exp_release_intf_params){
+			.comp_mask = 0,
+		};
+		claim_zero(ibv_exp_release_intf(rxq->priv->ctx,
+						rxq->if_qp,
+						&params));
+	}
+	if (rxq->if_cq != NULL) {
+		assert(rxq->priv != NULL);
+		assert(rxq->priv->ctx != NULL);
+		assert(rxq->cq != NULL);
+		params = (struct ibv_exp_release_intf_params){
+			.comp_mask = 0,
+		};
+		claim_zero(ibv_exp_release_intf(rxq->priv->ctx,
+						rxq->if_cq,
+						&params));
+	}
+	if (rxq->qp != NULL) {
+		claim_zero(ibv_destroy_qp(rxq->qp));
+	}
+	if (rxq->cq != NULL)
+		claim_zero(ibv_destroy_cq(rxq->cq));
+	if (rxq->rd != NULL) {
+		struct ibv_exp_destroy_res_domain_attr attr = {
+			.comp_mask = 0,
+		};
+
+		assert(rxq->priv != NULL);
+		assert(rxq->priv->ctx != NULL);
+		claim_zero(ibv_exp_destroy_res_domain(rxq->priv->ctx,
+						      rxq->rd,
+						      &attr));
+	}
+	if (rxq->mr != NULL)
+		claim_zero(ibv_dereg_mr(rxq->mr));
+	memset(rxq, 0, sizeof(*rxq));
+}
+
+/**
+ * Allocate a Queue Pair.
+ * Optionally setup inline receive if supported.
+ *
+ * @param priv
+ *   Pointer to private structure.
+ * @param cq
+ *   Completion queue to associate with QP.
+ * @param desc
+ *   Number of descriptors in QP (hint only).
+ *
+ * @return
+ *   QP pointer or NULL in case of error.
+ */
+static struct ibv_qp *
+rxq_setup_qp(struct priv *priv, struct ibv_cq *cq, uint16_t desc,
+	     struct ibv_exp_res_domain *rd)
+{
+	struct ibv_exp_qp_init_attr attr = {
+		/* CQ to be associated with the send queue. */
+		.send_cq = cq,
+		/* CQ to be associated with the receive queue. */
+		.recv_cq = cq,
+		.cap = {
+			/* Max number of outstanding WRs. */
+			.max_recv_wr = ((priv->device_attr.max_qp_wr < desc) ?
+					priv->device_attr.max_qp_wr :
+					desc),
+			/* Max number of scatter/gather elements in a WR. */
+			.max_recv_sge = ((priv->device_attr.max_sge <
+					  MLX5_PMD_SGE_WR_N) ?
+					 priv->device_attr.max_sge :
+					 MLX5_PMD_SGE_WR_N),
+		},
+		.qp_type = IBV_QPT_RAW_PACKET,
+		.comp_mask = (IBV_EXP_QP_INIT_ATTR_PD |
+			      IBV_EXP_QP_INIT_ATTR_RES_DOMAIN),
+		.pd = priv->pd,
+		.res_domain = rd,
+	};
+
+	return ibv_exp_create_qp(priv->ctx, &attr);
+}
+
+#ifdef RSS_SUPPORT
+
+/**
+ * Allocate a RSS Queue Pair.
+ * Optionally setup inline receive if supported.
+ *
+ * @param priv
+ *   Pointer to private structure.
+ * @param cq
+ *   Completion queue to associate with QP.
+ * @param desc
+ *   Number of descriptors in QP (hint only).
+ * @param parent
+ *   If nonzero, create a parent QP, otherwise a child.
+ *
+ * @return
+ *   QP pointer or NULL in case of error.
+ */
+static struct ibv_qp *
+rxq_setup_qp_rss(struct priv *priv, struct ibv_cq *cq, uint16_t desc,
+		 int parent, struct ibv_exp_res_domain *rd)
+{
+	struct ibv_exp_qp_init_attr attr = {
+		/* CQ to be associated with the send queue. */
+		.send_cq = cq,
+		/* CQ to be associated with the receive queue. */
+		.recv_cq = cq,
+		.cap = {
+			/* Max number of outstanding WRs. */
+			.max_recv_wr = ((priv->device_attr.max_qp_wr < desc) ?
+					priv->device_attr.max_qp_wr :
+					desc),
+			/* Max number of scatter/gather elements in a WR. */
+			.max_recv_sge = ((priv->device_attr.max_sge <
+					  MLX5_PMD_SGE_WR_N) ?
+					 priv->device_attr.max_sge :
+					 MLX5_PMD_SGE_WR_N),
+		},
+		.qp_type = IBV_QPT_RAW_PACKET,
+		.comp_mask = (IBV_EXP_QP_INIT_ATTR_PD |
+			      IBV_EXP_QP_INIT_ATTR_RES_DOMAIN |
+			      IBV_EXP_QP_INIT_ATTR_QPG),
+		.pd = priv->pd,
+		.res_domain = rd,
+	};
+
+	if (parent) {
+		attr.qpg.qpg_type = IBV_EXP_QPG_PARENT;
+		/* TSS isn't necessary. */
+		attr.qpg.parent_attrib.tss_child_count = 0;
+		attr.qpg.parent_attrib.rss_child_count = priv->rxqs_n;
+		DEBUG("initializing parent RSS queue");
+	} else {
+		attr.qpg.qpg_type = IBV_EXP_QPG_CHILD_RX;
+		attr.qpg.qpg_parent = priv->rxq_parent.qp;
+		DEBUG("initializing child RSS queue");
+	}
+	return ibv_exp_create_qp(priv->ctx, &attr);
+}
+
+#endif /* RSS_SUPPORT */
+
+/**
+ * Configure a RX queue.
+ *
+ * @param dev
+ *   Pointer to Ethernet device structure.
+ * @param rxq
+ *   Pointer to RX queue structure.
+ * @param desc
+ *   Number of descriptors to configure in queue.
+ * @param socket
+ *   NUMA socket on which memory must be allocated.
+ * @param[in] conf
+ *   Thresholds parameters.
+ * @param mp
+ *   Memory pool for buffer allocations.
+ *
+ * @return
+ *   0 on success, errno value on failure.
+ */
+int
+rxq_setup(struct rte_eth_dev *dev, struct rxq *rxq, uint16_t desc,
+	  unsigned int socket, const struct rte_eth_rxconf *conf,
+	  struct rte_mempool *mp)
+{
+	struct priv *priv = dev->data->dev_private;
+	struct rxq tmpl = {
+		.priv = priv,
+		.mp = mp,
+		.socket = socket
+	};
+	struct ibv_exp_qp_attr mod;
+	union {
+		struct ibv_exp_query_intf_params params;
+		struct ibv_exp_cq_init_attr cq;
+		struct ibv_exp_res_domain_init_attr rd;
+	} attr;
+	enum ibv_exp_query_intf_status status;
+	struct ibv_recv_wr *bad_wr;
+	struct rte_mbuf *buf;
+	int ret = 0;
+	int parent = (rxq == &priv->rxq_parent);
+
+	(void)conf; /* Thresholds configuration (ignored). */
+	/*
+	 * If this is a parent queue, hardware must support RSS and
+	 * RSS must be enabled.
+	 */
+	assert((!parent) || ((priv->hw_rss) && (priv->rss)));
+	if (parent) {
+		/* Even if unused, ibv_create_cq() requires at least one
+		 * descriptor. */
+		desc = 1;
+		goto skip_mr;
+	}
+	if ((desc == 0) || (desc % MLX5_PMD_SGE_WR_N)) {
+		ERROR("%p: invalid number of RX descriptors (must be a"
+		      " multiple of %d)", (void *)dev, MLX5_PMD_SGE_WR_N);
+		return EINVAL;
+	}
+	/* Get mbuf length. */
+	buf = rte_pktmbuf_alloc(mp);
+	if (buf == NULL) {
+		ERROR("%p: unable to allocate mbuf", (void *)dev);
+		return ENOMEM;
+	}
+	tmpl.mb_len = buf->buf_len;
+	assert((rte_pktmbuf_headroom(buf) +
+		rte_pktmbuf_tailroom(buf)) == tmpl.mb_len);
+	assert(rte_pktmbuf_headroom(buf) == RTE_PKTMBUF_HEADROOM);
+	rte_pktmbuf_free(buf);
+	/* Use the entire RX mempool as the memory region. */
+	tmpl.mr = ibv_reg_mr(priv->pd,
+			     (void *)mp->elt_va_start,
+			     (mp->elt_va_end - mp->elt_va_start),
+			     (IBV_ACCESS_LOCAL_WRITE |
+			      IBV_ACCESS_REMOTE_WRITE));
+	if (tmpl.mr == NULL) {
+		ret = EINVAL;
+		ERROR("%p: MR creation failure: %s",
+		      (void *)dev, strerror(ret));
+		goto error;
+	}
+skip_mr:
+	attr.rd = (struct ibv_exp_res_domain_init_attr){
+		.comp_mask = (IBV_EXP_RES_DOMAIN_THREAD_MODEL |
+			      IBV_EXP_RES_DOMAIN_MSG_MODEL),
+		.thread_model = IBV_EXP_THREAD_SINGLE,
+		.msg_model = IBV_EXP_MSG_HIGH_BW,
+	};
+	tmpl.rd = ibv_exp_create_res_domain(priv->ctx, &attr.rd);
+	if (tmpl.rd == NULL) {
+		ret = ENOMEM;
+		ERROR("%p: RD creation failure: %s",
+		      (void *)dev, strerror(ret));
+		goto error;
+	}
+	attr.cq = (struct ibv_exp_cq_init_attr){
+		.comp_mask = IBV_EXP_CQ_INIT_ATTR_RES_DOMAIN,
+		.res_domain = tmpl.rd,
+	};
+	tmpl.cq = ibv_exp_create_cq(priv->ctx, desc, NULL, NULL, 0, &attr.cq);
+	if (tmpl.cq == NULL) {
+		ret = ENOMEM;
+		ERROR("%p: CQ creation failure: %s",
+		      (void *)dev, strerror(ret));
+		goto error;
+	}
+	DEBUG("priv->device_attr.max_qp_wr is %d",
+	      priv->device_attr.max_qp_wr);
+	DEBUG("priv->device_attr.max_sge is %d",
+	      priv->device_attr.max_sge);
+#ifdef RSS_SUPPORT
+	if (priv->rss)
+		tmpl.qp = rxq_setup_qp_rss(priv, tmpl.cq, desc, parent,
+					   tmpl.rd);
+	else
+#endif /* RSS_SUPPORT */
+		tmpl.qp = rxq_setup_qp(priv, tmpl.cq, desc, tmpl.rd);
+	if (tmpl.qp == NULL) {
+		ret = (errno ? errno : EINVAL);
+		ERROR("%p: QP creation failure: %s",
+		      (void *)dev, strerror(ret));
+		goto error;
+	}
+	mod = (struct ibv_exp_qp_attr){
+		/* Move the QP to this state. */
+		.qp_state = IBV_QPS_INIT,
+		/* Primary port number. */
+		.port_num = priv->port
+	};
+	ret = ibv_exp_modify_qp(tmpl.qp, &mod,
+				(IBV_EXP_QP_STATE |
+#ifdef RSS_SUPPORT
+				 (parent ? IBV_EXP_QP_GROUP_RSS : 0) |
+#endif /* RSS_SUPPORT */
+				 IBV_EXP_QP_PORT));
+	if (ret) {
+		ERROR("%p: QP state to IBV_QPS_INIT failed: %s",
+		      (void *)dev, strerror(ret));
+		goto error;
+	}
+	/* Allocate descriptors for RX queues, except for the RSS parent. */
+	if (parent)
+		goto skip_alloc;
+	ret = rxq_alloc_elts(&tmpl, desc, NULL);
+	if (ret) {
+		ERROR("%p: RXQ allocation failed: %s",
+		      (void *)dev, strerror(ret));
+		goto error;
+	}
+	ret = ibv_post_recv(tmpl.qp,
+			    &(*tmpl.elts.no_sp)[0].wr,
+			    &bad_wr);
+	if (ret) {
+		ERROR("%p: ibv_post_recv() failed for WR %p: %s",
+		      (void *)dev,
+		      (void *)bad_wr,
+		      strerror(ret));
+		goto error;
+	}
+skip_alloc:
+	mod = (struct ibv_exp_qp_attr){
+		.qp_state = IBV_QPS_RTR
+	};
+	ret = ibv_exp_modify_qp(tmpl.qp, &mod, IBV_EXP_QP_STATE);
+	if (ret) {
+		ERROR("%p: QP state to IBV_QPS_RTR failed: %s",
+		      (void *)dev, strerror(ret));
+		goto error;
+	}
+	/* Save port ID. */
+	tmpl.port_id = dev->data->port_id;
+	DEBUG("%p: RTE port ID: %u", (void *)rxq, tmpl.port_id);
+	attr.params = (struct ibv_exp_query_intf_params){
+		.intf_scope = IBV_EXP_INTF_GLOBAL,
+		.intf = IBV_EXP_INTF_CQ,
+		.obj = tmpl.cq,
+	};
+	tmpl.if_cq = ibv_exp_query_intf(priv->ctx, &attr.params, &status);
+	if (tmpl.if_cq == NULL) {
+		ERROR("%p: CQ interface family query failed with status %d",
+		      (void *)dev, status);
+		goto error;
+	}
+	attr.params = (struct ibv_exp_query_intf_params){
+		.intf_scope = IBV_EXP_INTF_GLOBAL,
+		.intf = IBV_EXP_INTF_QP_BURST,
+		.obj = tmpl.qp,
+	};
+	tmpl.if_qp = ibv_exp_query_intf(priv->ctx, &attr.params, &status);
+	if (tmpl.if_qp == NULL) {
+		ERROR("%p: QP interface family query failed with status %d",
+		      (void *)dev, status);
+		goto error;
+	}
+	/* Clean up rxq in case we're reinitializing it. */
+	DEBUG("%p: cleaning-up old rxq just in case", (void *)rxq);
+	rxq_cleanup(rxq);
+	*rxq = tmpl;
+	DEBUG("%p: rxq updated with %p", (void *)rxq, (void *)&tmpl);
+	assert(ret == 0);
+	return 0;
+error:
+	rxq_cleanup(&tmpl);
+	assert(ret > 0);
+	return ret;
+}
+
+/**
+ * DPDK callback to configure a RX queue.
+ *
+ * @param dev
+ *   Pointer to Ethernet device structure.
+ * @param idx
+ *   RX queue index.
+ * @param desc
+ *   Number of descriptors to configure in queue.
+ * @param socket
+ *   NUMA socket on which memory must be allocated.
+ * @param[in] conf
+ *   Thresholds parameters.
+ * @param mp
+ *   Memory pool for buffer allocations.
+ *
+ * @return
+ *   0 on success, negative errno value on failure.
+ */
+int
+mlx5_rx_queue_setup(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc,
+		    unsigned int socket, const struct rte_eth_rxconf *conf,
+		    struct rte_mempool *mp)
+{
+	struct priv *priv = dev->data->dev_private;
+	struct rxq *rxq = (*priv->rxqs)[idx];
+	int ret;
+
+	priv_lock(priv);
+	DEBUG("%p: configuring queue %u for %u descriptors",
+	      (void *)dev, idx, desc);
+	if (idx >= priv->rxqs_n) {
+		ERROR("%p: queue index out of range (%u >= %u)",
+		      (void *)dev, idx, priv->rxqs_n);
+		priv_unlock(priv);
+		return -EOVERFLOW;
+	}
+	if (rxq != NULL) {
+		DEBUG("%p: reusing already allocated queue index %u (%p)",
+		      (void *)dev, idx, (void *)rxq);
+		if (priv->started) {
+			priv_unlock(priv);
+			return -EEXIST;
+		}
+		(*priv->rxqs)[idx] = NULL;
+		rxq_cleanup(rxq);
+	} else {
+		rxq = rte_calloc_socket("RXQ", 1, sizeof(*rxq), 0, socket);
+		if (rxq == NULL) {
+			ERROR("%p: unable to allocate queue index %u",
+			      (void *)dev, idx);
+			priv_unlock(priv);
+			return -ENOMEM;
+		}
+	}
+	ret = rxq_setup(dev, rxq, desc, socket, conf, mp);
+	if (ret)
+		rte_free(rxq);
+	else {
+		DEBUG("%p: adding RX queue %p to list",
+		      (void *)dev, (void *)rxq);
+		(*priv->rxqs)[idx] = rxq;
+		/* Update receive callback. */
+		dev->rx_pkt_burst = mlx5_rx_burst;
+	}
+	priv_unlock(priv);
+	return -ret;
+}
+
+/**
+ * DPDK callback to release a RX queue.
+ *
+ * @param dpdk_rxq
+ *   Generic RX queue pointer.
+ */
+void
+mlx5_rx_queue_release(void *dpdk_rxq)
+{
+	struct rxq *rxq = (struct rxq *)dpdk_rxq;
+	struct priv *priv;
+	unsigned int i;
+
+	if (rxq == NULL)
+		return;
+	priv = rxq->priv;
+	priv_lock(priv);
+	assert(rxq != &priv->rxq_parent);
+	for (i = 0; (i != priv->rxqs_n); ++i)
+		if ((*priv->rxqs)[i] == rxq) {
+			DEBUG("%p: removing RX queue %p from list",
+			      (void *)priv->dev, (void *)rxq);
+			(*priv->rxqs)[i] = NULL;
+			break;
+		}
+	rxq_cleanup(rxq);
+	rte_free(rxq);
+	priv_unlock(priv);
+}
diff --git a/drivers/net/mlx5/mlx5_rxtx.c b/drivers/net/mlx5/mlx5_rxtx.c
new file mode 100644
index 0000000..40bddf0
--- /dev/null
+++ b/drivers/net/mlx5/mlx5_rxtx.c
@@ -0,0 +1,495 @@
+/*-
+ *   BSD LICENSE
+ *
+ *   Copyright 2015 6WIND S.A.
+ *   Copyright 2015 Mellanox.
+ *
+ *   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 6WIND S.A. 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 <assert.h>
+#include <stdint.h>
+#include <string.h>
+#include <stdlib.h>
+
+/* Verbs header. */
+/* ISO C doesn't support unnamed structs/unions, disabling -pedantic. */
+#ifdef PEDANTIC
+#pragma GCC diagnostic ignored "-pedantic"
+#endif
+#include <infiniband/verbs.h>
+#ifdef PEDANTIC
+#pragma GCC diagnostic error "-pedantic"
+#endif
+
+/* DPDK headers don't like -pedantic. */
+#ifdef PEDANTIC
+#pragma GCC diagnostic ignored "-pedantic"
+#endif
+#include <rte_mbuf.h>
+#include <rte_mempool.h>
+#include <rte_prefetch.h>
+#include <rte_common.h>
+#include <rte_branch_prediction.h>
+#ifdef PEDANTIC
+#pragma GCC diagnostic error "-pedantic"
+#endif
+
+#include "mlx5.h"
+#include "mlx5_utils.h"
+#include "mlx5_rxtx.h"
+#include "mlx5_defs.h"
+
+/**
+ * Manage TX completions.
+ *
+ * When sending a burst, mlx5_tx_burst() posts several WRs.
+ * To improve performance, a completion event is only required once every
+ * MLX5_PMD_TX_PER_COMP_REQ sends. Doing so discards completion information
+ * for other WRs, but this information would not be used anyway.
+ *
+ * @param txq
+ *   Pointer to TX queue structure.
+ *
+ * @return
+ *   0 on success, -1 on failure.
+ */
+static int
+txq_complete(struct txq *txq)
+{
+	unsigned int elts_comp = txq->elts_comp;
+	unsigned int elts_tail = txq->elts_tail;
+	const unsigned int elts_n = txq->elts_n;
+	int wcs_n;
+
+	if (unlikely(elts_comp == 0))
+		return 0;
+#ifdef DEBUG_SEND
+	DEBUG("%p: processing %u work requests completions",
+	      (void *)txq, elts_comp);
+#endif
+	wcs_n = txq->if_cq->poll_cnt(txq->cq, elts_comp);
+	if (unlikely(wcs_n == 0))
+		return 0;
+	if (unlikely(wcs_n < 0)) {
+		DEBUG("%p: ibv_poll_cq() failed (wcs_n=%d)",
+		      (void *)txq, wcs_n);
+		return -1;
+	}
+	elts_comp -= wcs_n;
+	assert(elts_comp <= txq->elts_comp);
+	/*
+	 * Assume WC status is successful as nothing can be done about it
+	 * anyway.
+	 */
+	elts_tail += wcs_n * txq->elts_comp_cd_init;
+	if (elts_tail >= elts_n)
+		elts_tail -= elts_n;
+	txq->elts_tail = elts_tail;
+	txq->elts_comp = elts_comp;
+	return 0;
+}
+
+/**
+ * Get Memory Region (MR) <-> Memory Pool (MP) association from txq->mp2mr[].
+ * Add MP to txq->mp2mr[] if it's not registered yet. If mp2mr[] is full,
+ * remove an entry first.
+ *
+ * @param txq
+ *   Pointer to TX queue structure.
+ * @param[in] mp
+ *   Memory Pool for which a Memory Region lkey must be returned.
+ *
+ * @return
+ *   mr->lkey on success, (uint32_t)-1 on failure.
+ */
+static uint32_t
+txq_mp2mr(struct txq *txq, struct rte_mempool *mp)
+{
+	unsigned int i;
+	struct ibv_mr *mr;
+
+	for (i = 0; (i != RTE_DIM(txq->mp2mr)); ++i) {
+		if (unlikely(txq->mp2mr[i].mp == NULL)) {
+			/* Unknown MP, add a new MR for it. */
+			break;
+		}
+		if (txq->mp2mr[i].mp == mp) {
+			assert(txq->mp2mr[i].lkey != (uint32_t)-1);
+			assert(txq->mp2mr[i].mr->lkey == txq->mp2mr[i].lkey);
+			return txq->mp2mr[i].lkey;
+		}
+	}
+	/* Add a new entry, register MR first. */
+	DEBUG("%p: discovered new memory pool %p", (void *)txq, (void *)mp);
+	mr = ibv_reg_mr(txq->priv->pd,
+			(void *)mp->elt_va_start,
+			(mp->elt_va_end - mp->elt_va_start),
+			(IBV_ACCESS_LOCAL_WRITE | IBV_ACCESS_REMOTE_WRITE));
+	if (unlikely(mr == NULL)) {
+		DEBUG("%p: unable to configure MR, ibv_reg_mr() failed.",
+		      (void *)txq);
+		return (uint32_t)-1;
+	}
+	if (unlikely(i == RTE_DIM(txq->mp2mr))) {
+		/* Table is full, remove oldest entry. */
+		DEBUG("%p: MR <-> MP table full, dropping oldest entry.",
+		      (void *)txq);
+		--i;
+		claim_zero(ibv_dereg_mr(txq->mp2mr[i].mr));
+		memmove(&txq->mp2mr[0], &txq->mp2mr[1],
+			(sizeof(txq->mp2mr) - sizeof(txq->mp2mr[0])));
+	}
+	/* Store the new entry. */
+	txq->mp2mr[i].mp = mp;
+	txq->mp2mr[i].mr = mr;
+	txq->mp2mr[i].lkey = mr->lkey;
+	DEBUG("%p: new MR lkey for MP %p: 0x%08" PRIu32,
+	      (void *)txq, (void *)mp, txq->mp2mr[i].lkey);
+	return txq->mp2mr[i].lkey;
+}
+
+/**
+ * DPDK callback for TX.
+ *
+ * @param dpdk_txq
+ *   Generic pointer to TX queue structure.
+ * @param[in] pkts
+ *   Packets to transmit.
+ * @param pkts_n
+ *   Number of packets in array.
+ *
+ * @return
+ *   Number of packets successfully transmitted (<= pkts_n).
+ */
+uint16_t
+mlx5_tx_burst(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n)
+{
+	struct txq *txq = (struct txq *)dpdk_txq;
+	unsigned int elts_head = txq->elts_head;
+	const unsigned int elts_tail = txq->elts_tail;
+	const unsigned int elts_n = txq->elts_n;
+	unsigned int elts_comp_cd = txq->elts_comp_cd;
+	unsigned int elts_comp = 0;
+	unsigned int i;
+	unsigned int max;
+	int err;
+
+	assert(elts_comp_cd != 0);
+	txq_complete(txq);
+	max = (elts_n - (elts_head - elts_tail));
+	if (max > elts_n)
+		max -= elts_n;
+	assert(max >= 1);
+	assert(max <= elts_n);
+	/* Always leave one free entry in the ring. */
+	--max;
+	if (max == 0)
+		return 0;
+	if (max > pkts_n)
+		max = pkts_n;
+	for (i = 0; (i != max); ++i) {
+		struct rte_mbuf *buf = pkts[i];
+		unsigned int elts_head_next =
+			(((elts_head + 1) == elts_n) ? 0 : elts_head + 1);
+		struct txq_elt *elt_next = &(*txq->elts)[elts_head_next];
+		struct txq_elt *elt = &(*txq->elts)[elts_head];
+		unsigned int segs = NB_SEGS(buf);
+		uint32_t send_flags = 0;
+
+		/* Clean up old buffer. */
+		if (likely(elt->buf != NULL)) {
+			struct rte_mbuf *tmp = elt->buf;
+
+			/* Faster than rte_pktmbuf_free(). */
+			do {
+				struct rte_mbuf *next = NEXT(tmp);
+
+				rte_pktmbuf_free_seg(tmp);
+				tmp = next;
+			} while (tmp != NULL);
+		}
+		/* Request TX completion. */
+		if (unlikely(--elts_comp_cd == 0)) {
+			elts_comp_cd = txq->elts_comp_cd_init;
+			++elts_comp;
+			send_flags |= IBV_EXP_QP_BURST_SIGNALED;
+		}
+		if (likely(segs == 1)) {
+			uintptr_t addr;
+			uint32_t length;
+			uint32_t lkey;
+
+			/* Retrieve buffer information. */
+			addr = rte_pktmbuf_mtod(buf, uintptr_t);
+			length = DATA_LEN(buf);
+			/* Retrieve Memory Region key for this memory pool. */
+			lkey = txq_mp2mr(txq, buf->pool);
+			if (unlikely(lkey == (uint32_t)-1)) {
+				/* MR does not exist. */
+				DEBUG("%p: unable to get MP <-> MR"
+				      " association", (void *)txq);
+				/* Clean up TX element. */
+				elt->buf = NULL;
+				goto stop;
+			}
+			/* Update element. */
+			elt->buf = buf;
+			if (txq->priv->vf)
+				rte_prefetch0((volatile void *)
+					      (uintptr_t)addr);
+			RTE_MBUF_PREFETCH_TO_FREE(elt_next->buf);
+			/* Put packet into send queue. */
+#if MLX5_PMD_MAX_INLINE > 0
+			if (length <= txq->max_inline)
+				err = txq->if_qp->send_pending_inline
+					(txq->qp,
+					 (void *)addr,
+					 length,
+					 send_flags);
+			else
+#endif
+				err = txq->if_qp->send_pending
+					(txq->qp,
+					 addr,
+					 length,
+					 lkey,
+					 send_flags);
+			if (unlikely(err))
+				goto stop;
+		} else {
+			DEBUG("%p: TX scattered buffers support not"
+			      " compiled in", (void *)txq);
+			goto stop;
+		}
+		elts_head = elts_head_next;
+	}
+stop:
+	/* Take a shortcut if nothing must be sent. */
+	if (unlikely(i == 0))
+		return 0;
+	/* Ring QP doorbell. */
+	err = txq->if_qp->send_flush(txq->qp);
+	if (unlikely(err)) {
+		/* A nonzero value is not supposed to be returned.
+		 * Nothing can be done about it. */
+		DEBUG("%p: send_flush() failed with error %d",
+		      (void *)txq, err);
+	}
+	txq->elts_head = elts_head;
+	txq->elts_comp += elts_comp;
+	txq->elts_comp_cd = elts_comp_cd;
+	return i;
+}
+
+/**
+ * DPDK callback for RX.
+ *
+ * @param dpdk_rxq
+ *   Generic pointer to RX queue structure.
+ * @param[out] pkts
+ *   Array to store received packets.
+ * @param pkts_n
+ *   Maximum number of packets in array.
+ *
+ * @return
+ *   Number of packets successfully received (<= pkts_n).
+ */
+uint16_t
+mlx5_rx_burst(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n)
+{
+	struct rxq *rxq = (struct rxq *)dpdk_rxq;
+	struct rxq_elt (*elts)[rxq->elts_n] = rxq->elts.no_sp;
+	const unsigned int elts_n = rxq->elts_n;
+	unsigned int elts_head = rxq->elts_head;
+	struct ibv_sge sges[pkts_n];
+	unsigned int i;
+	unsigned int pkts_ret = 0;
+	int ret;
+
+	for (i = 0; (i != pkts_n); ++i) {
+		struct rxq_elt *elt = &(*elts)[elts_head];
+		struct ibv_recv_wr *wr = &elt->wr;
+		uint64_t wr_id = wr->wr_id;
+		unsigned int len;
+		struct rte_mbuf *seg = (void *)((uintptr_t)elt->sge.addr -
+			WR_ID(wr_id).offset);
+		struct rte_mbuf *rep;
+		uint32_t flags;
+
+		/* Sanity checks. */
+		assert(WR_ID(wr_id).id < rxq->elts_n);
+		assert(wr->sg_list == &elt->sge);
+		assert(wr->num_sge == 1);
+		assert(elts_head < rxq->elts_n);
+		assert(rxq->elts_head < rxq->elts_n);
+		ret = rxq->if_cq->poll_length_flags(rxq->cq, NULL, NULL,
+						    &flags);
+		if (unlikely(ret < 0)) {
+			struct ibv_wc wc;
+			int wcs_n;
+
+			DEBUG("rxq=%p, poll_length() failed (ret=%d)",
+			      (void *)rxq, ret);
+			/* ibv_poll_cq() must be used in case of failure. */
+			wcs_n = ibv_poll_cq(rxq->cq, 1, &wc);
+			if (unlikely(wcs_n == 0))
+				break;
+			if (unlikely(wcs_n < 0)) {
+				DEBUG("rxq=%p, ibv_poll_cq() failed (wcs_n=%d)",
+				      (void *)rxq, wcs_n);
+				break;
+			}
+			assert(wcs_n == 1);
+			if (unlikely(wc.status != IBV_WC_SUCCESS)) {
+				/* Whatever, just repost the offending WR. */
+				DEBUG("rxq=%p, wr_id=%" PRIu64 ": bad work"
+				      " completion status (%d): %s",
+				      (void *)rxq, wc.wr_id, wc.status,
+				      ibv_wc_status_str(wc.status));
+				/* Add SGE to array for repost. */
+				sges[i] = elt->sge;
+				goto repost;
+			}
+			ret = wc.byte_len;
+		}
+		if (ret == 0)
+			break;
+		len = ret;
+		/*
+		 * Fetch initial bytes of packet descriptor into a
+		 * cacheline while allocating rep.
+		 */
+		rte_prefetch0(seg);
+		rep = __rte_mbuf_raw_alloc(rxq->mp);
+		if (unlikely(rep == NULL)) {
+			/*
+			 * Unable to allocate a replacement mbuf,
+			 * repost WR.
+			 */
+			DEBUG("rxq=%p, wr_id=%" PRIu32 ":"
+			      " can't allocate a new mbuf",
+			      (void *)rxq, WR_ID(wr_id).id);
+			/* Increment out of memory counters. */
+			++rxq->priv->dev->data->rx_mbuf_alloc_failed;
+			goto repost;
+		}
+
+		/* Reconfigure sge to use rep instead of seg. */
+		elt->sge.addr = (uintptr_t)rep->buf_addr + RTE_PKTMBUF_HEADROOM;
+		assert(elt->sge.lkey == rxq->mr->lkey);
+		WR_ID(wr->wr_id).offset =
+			(((uintptr_t)rep->buf_addr + RTE_PKTMBUF_HEADROOM) -
+			 (uintptr_t)rep);
+		assert(WR_ID(wr->wr_id).id == WR_ID(wr_id).id);
+
+		/* Add SGE to array for repost. */
+		sges[i] = elt->sge;
+
+		/* Update seg information. */
+		SET_DATA_OFF(seg, RTE_PKTMBUF_HEADROOM);
+		NB_SEGS(seg) = 1;
+		PORT(seg) = rxq->port_id;
+		NEXT(seg) = NULL;
+		PKT_LEN(seg) = len;
+		DATA_LEN(seg) = len;
+
+		/* Return packet. */
+		*(pkts++) = seg;
+		++pkts_ret;
+repost:
+		if (++elts_head >= elts_n)
+			elts_head = 0;
+		continue;
+	}
+	if (unlikely(i == 0))
+		return 0;
+	/* Repost WRs. */
+#ifdef DEBUG_RECV
+	DEBUG("%p: reposting %u WRs", (void *)rxq, i);
+#endif
+	ret = rxq->if_qp->recv_burst(rxq->qp, sges, i);
+	if (unlikely(ret)) {
+		/* Inability to repost WRs is fatal. */
+		DEBUG("%p: recv_burst(): failed (ret=%d)",
+		      (void *)rxq->priv,
+		      ret);
+		abort();
+	}
+	rxq->elts_head = elts_head;
+	return pkts_ret;
+}
+
+/**
+ * Dummy DPDK callback for TX.
+ *
+ * This function is used to temporarily replace the real callback during
+ * unsafe control operations on the queue, or in case of error.
+ *
+ * @param dpdk_txq
+ *   Generic pointer to TX queue structure.
+ * @param[in] pkts
+ *   Packets to transmit.
+ * @param pkts_n
+ *   Number of packets in array.
+ *
+ * @return
+ *   Number of packets successfully transmitted (<= pkts_n).
+ */
+uint16_t
+removed_tx_burst(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n)
+{
+	(void)dpdk_txq;
+	(void)pkts;
+	(void)pkts_n;
+	return 0;
+}
+
+/**
+ * Dummy DPDK callback for RX.
+ *
+ * This function is used to temporarily replace the real callback during
+ * unsafe control operations on the queue, or in case of error.
+ *
+ * @param dpdk_rxq
+ *   Generic pointer to RX queue structure.
+ * @param[out] pkts
+ *   Array to store received packets.
+ * @param pkts_n
+ *   Maximum number of packets in array.
+ *
+ * @return
+ *   Number of packets successfully received (<= pkts_n).
+ */
+uint16_t
+removed_rx_burst(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n)
+{
+	(void)dpdk_rxq;
+	(void)pkts;
+	(void)pkts_n;
+	return 0;
+}
diff --git a/drivers/net/mlx5/mlx5_rxtx.h b/drivers/net/mlx5/mlx5_rxtx.h
new file mode 100644
index 0000000..08a4911
--- /dev/null
+++ b/drivers/net/mlx5/mlx5_rxtx.h
@@ -0,0 +1,161 @@
+/*-
+ *   BSD LICENSE
+ *
+ *   Copyright 2015 6WIND S.A.
+ *   Copyright 2015 Mellanox.
+ *
+ *   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 6WIND S.A. 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 RTE_PMD_MLX5_RXTX_H_
+#define RTE_PMD_MLX5_RXTX_H_
+
+#include <stdint.h>
+
+/* Verbs header. */
+/* ISO C doesn't support unnamed structs/unions, disabling -pedantic. */
+#ifdef PEDANTIC
+#pragma GCC diagnostic ignored "-pedantic"
+#endif
+#include <infiniband/verbs.h>
+#ifdef PEDANTIC
+#pragma GCC diagnostic error "-pedantic"
+#endif
+
+/* DPDK headers don't like -pedantic. */
+#ifdef PEDANTIC
+#pragma GCC diagnostic ignored "-pedantic"
+#endif
+#include <rte_mbuf.h>
+#include <rte_mempool.h>
+#ifdef PEDANTIC
+#pragma GCC diagnostic error "-pedantic"
+#endif
+
+#include "mlx5_utils.h"
+#include "mlx5.h"
+#include "mlx5_defs.h"
+
+/* RX element. */
+struct rxq_elt {
+	struct ibv_recv_wr wr; /* Work Request. */
+	struct ibv_sge sge; /* Scatter/Gather Element. */
+	/* mbuf pointer is derived from WR_ID(wr.wr_id).offset. */
+};
+
+struct priv;
+
+/* RX queue descriptor. */
+struct rxq {
+	struct priv *priv; /* Back pointer to private data. */
+	struct rte_mempool *mp; /* Memory Pool for allocations. */
+	struct ibv_mr *mr; /* Memory Region (for mp). */
+	struct ibv_cq *cq; /* Completion Queue. */
+	struct ibv_qp *qp; /* Queue Pair. */
+	struct ibv_exp_qp_burst_family *if_qp; /* QP burst interface. */
+	struct ibv_exp_cq_family *if_cq; /* CQ interface. */
+	/*
+	 * Each VLAN ID requires a separate flow steering rule.
+	 */
+	BITFIELD_DECLARE(mac_configured, uint32_t, MLX5_MAX_MAC_ADDRESSES);
+	struct ibv_flow *mac_flow[MLX5_MAX_MAC_ADDRESSES][MLX5_MAX_VLAN_IDS];
+	unsigned int port_id; /* Port ID for incoming packets. */
+	unsigned int elts_n; /* (*elts)[] length. */
+	unsigned int elts_head; /* Current index in (*elts)[]. */
+	union {
+		struct rxq_elt (*no_sp)[]; /* RX elements. */
+	} elts;
+	uint32_t mb_len; /* Length of a mp-issued mbuf. */
+	unsigned int socket; /* CPU socket ID for allocations. */
+	struct ibv_exp_res_domain *rd; /* Resource Domain. */
+};
+
+/* TX element. */
+struct txq_elt {
+	struct rte_mbuf *buf;
+};
+
+/* Linear buffer type. It is used when transmitting buffers with too many
+ * segments that do not fit the hardware queue (see max_send_sge).
+ * Extra segments are copied (linearized) in such buffers, replacing the
+ * last SGE during TX.
+ * The size is arbitrary but large enough to hold a jumbo frame with
+ * 8 segments considering mbuf.buf_len is about 2048 bytes. */
+typedef uint8_t linear_t[16384];
+
+/* TX queue descriptor. */
+struct txq {
+	struct priv *priv; /* Back pointer to private data. */
+	struct {
+		struct rte_mempool *mp; /* Cached Memory Pool. */
+		struct ibv_mr *mr; /* Memory Region (for mp). */
+		uint32_t lkey; /* mr->lkey */
+	} mp2mr[MLX5_PMD_TX_MP_CACHE]; /* MP to MR translation table. */
+	struct ibv_cq *cq; /* Completion Queue. */
+	struct ibv_qp *qp; /* Queue Pair. */
+	struct ibv_exp_qp_burst_family *if_qp; /* QP burst interface. */
+	struct ibv_exp_cq_family *if_cq; /* CQ interface. */
+#if MLX5_PMD_MAX_INLINE > 0
+	uint32_t max_inline; /* Max inline send size <= MLX5_PMD_MAX_INLINE. */
+#endif
+	unsigned int elts_n; /* (*elts)[] length. */
+	struct txq_elt (*elts)[]; /* TX elements. */
+	unsigned int elts_head; /* Current index in (*elts)[]. */
+	unsigned int elts_tail; /* First element awaiting completion. */
+	unsigned int elts_comp; /* Number of completion requests. */
+	unsigned int elts_comp_cd; /* Countdown for next completion request. */
+	unsigned int elts_comp_cd_init; /* Initial value for countdown. */
+	linear_t (*elts_linear)[]; /* Linearized buffers. */
+	struct ibv_mr *mr_linear; /* Memory Region for linearized buffers. */
+	unsigned int socket; /* CPU socket ID for allocations. */
+	struct ibv_exp_res_domain *rd; /* Resource Domain. */
+};
+
+/* mlx5_rxq.c */
+
+void rxq_cleanup(struct rxq *);
+int rxq_setup(struct rte_eth_dev *, struct rxq *, uint16_t, unsigned int,
+	      const struct rte_eth_rxconf *, struct rte_mempool *);
+int mlx5_rx_queue_setup(struct rte_eth_dev *, uint16_t, uint16_t, unsigned int,
+			const struct rte_eth_rxconf *, struct rte_mempool *);
+void mlx5_rx_queue_release(void *);
+
+/* mlx5_txq.c */
+
+void txq_cleanup(struct txq *);
+int mlx5_tx_queue_setup(struct rte_eth_dev *, uint16_t, uint16_t, unsigned int,
+			const struct rte_eth_txconf *);
+void mlx5_tx_queue_release(void *);
+
+/* mlx5_rxtx.c */
+
+uint16_t mlx5_tx_burst(void *, struct rte_mbuf **, uint16_t);
+uint16_t mlx5_rx_burst(void *, struct rte_mbuf **, uint16_t);
+uint16_t removed_tx_burst(void *, struct rte_mbuf **, uint16_t);
+uint16_t removed_rx_burst(void *, struct rte_mbuf **, uint16_t);
+
+#endif /* RTE_PMD_MLX5_RXTX_H_ */
diff --git a/drivers/net/mlx5/mlx5_txq.c b/drivers/net/mlx5/mlx5_txq.c
new file mode 100644
index 0000000..2bae61f
--- /dev/null
+++ b/drivers/net/mlx5/mlx5_txq.c
@@ -0,0 +1,512 @@
+/*-
+ *   BSD LICENSE
+ *
+ *   Copyright 2015 6WIND S.A.
+ *   Copyright 2015 Mellanox.
+ *
+ *   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 6WIND S.A. 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 <stddef.h>
+#include <assert.h>
+#include <errno.h>
+#include <string.h>
+#include <stdint.h>
+
+/* Verbs header. */
+/* ISO C doesn't support unnamed structs/unions, disabling -pedantic. */
+#ifdef PEDANTIC
+#pragma GCC diagnostic ignored "-pedantic"
+#endif
+#include <infiniband/verbs.h>
+#ifdef PEDANTIC
+#pragma GCC diagnostic error "-pedantic"
+#endif
+
+/* DPDK headers don't like -pedantic. */
+#ifdef PEDANTIC
+#pragma GCC diagnostic ignored "-pedantic"
+#endif
+#include <rte_mbuf.h>
+#include <rte_malloc.h>
+#include <rte_ethdev.h>
+#include <rte_common.h>
+#ifdef PEDANTIC
+#pragma GCC diagnostic error "-pedantic"
+#endif
+
+#include "mlx5_utils.h"
+#include "mlx5.h"
+#include "mlx5_rxtx.h"
+#include "mlx5_autoconf.h"
+#include "mlx5_defs.h"
+
+/**
+ * Allocate TX queue elements.
+ *
+ * @param txq
+ *   Pointer to TX queue structure.
+ * @param elts_n
+ *   Number of elements to allocate.
+ *
+ * @return
+ *   0 on success, errno value on failure.
+ */
+static int
+txq_alloc_elts(struct txq *txq, unsigned int elts_n)
+{
+	unsigned int i;
+	struct txq_elt (*elts)[elts_n] =
+		rte_calloc_socket("TXQ", 1, sizeof(*elts), 0, txq->socket);
+	linear_t (*elts_linear)[elts_n] =
+		rte_calloc_socket("TXQ", 1, sizeof(*elts_linear), 0,
+				  txq->socket);
+	struct ibv_mr *mr_linear = NULL;
+	int ret = 0;
+
+	if ((elts == NULL) || (elts_linear == NULL)) {
+		ERROR("%p: can't allocate packets array", (void *)txq);
+		ret = ENOMEM;
+		goto error;
+	}
+	mr_linear =
+		ibv_reg_mr(txq->priv->pd, elts_linear, sizeof(*elts_linear),
+			   (IBV_ACCESS_LOCAL_WRITE | IBV_ACCESS_REMOTE_WRITE));
+	if (mr_linear == NULL) {
+		ERROR("%p: unable to configure MR, ibv_reg_mr() failed",
+		      (void *)txq);
+		ret = EINVAL;
+		goto error;
+	}
+	for (i = 0; (i != elts_n); ++i) {
+		struct txq_elt *elt = &(*elts)[i];
+
+		elt->buf = NULL;
+	}
+	DEBUG("%p: allocated and configured %u WRs", (void *)txq, elts_n);
+	txq->elts_n = elts_n;
+	txq->elts = elts;
+	txq->elts_head = 0;
+	txq->elts_tail = 0;
+	txq->elts_comp = 0;
+	/* Request send completion every MLX5_PMD_TX_PER_COMP_REQ packets or
+	 * at least 4 times per ring. */
+	txq->elts_comp_cd_init =
+		((MLX5_PMD_TX_PER_COMP_REQ < (elts_n / 4)) ?
+		 MLX5_PMD_TX_PER_COMP_REQ : (elts_n / 4));
+	txq->elts_comp_cd = txq->elts_comp_cd_init;
+	txq->elts_linear = elts_linear;
+	txq->mr_linear = mr_linear;
+	assert(ret == 0);
+	return 0;
+error:
+	if (mr_linear != NULL)
+		claim_zero(ibv_dereg_mr(mr_linear));
+
+	rte_free(elts_linear);
+	rte_free(elts);
+
+	DEBUG("%p: failed, freed everything", (void *)txq);
+	assert(ret > 0);
+	return ret;
+}
+
+/**
+ * Free TX queue elements.
+ *
+ * @param txq
+ *   Pointer to TX queue structure.
+ */
+static void
+txq_free_elts(struct txq *txq)
+{
+	unsigned int i;
+	unsigned int elts_n = txq->elts_n;
+	struct txq_elt (*elts)[elts_n] = txq->elts;
+	linear_t (*elts_linear)[elts_n] = txq->elts_linear;
+	struct ibv_mr *mr_linear = txq->mr_linear;
+
+	DEBUG("%p: freeing WRs", (void *)txq);
+	txq->elts_n = 0;
+	txq->elts = NULL;
+	txq->elts_linear = NULL;
+	txq->mr_linear = NULL;
+	if (mr_linear != NULL)
+		claim_zero(ibv_dereg_mr(mr_linear));
+
+	rte_free(elts_linear);
+	if (elts == NULL)
+		return;
+	for (i = 0; (i != RTE_DIM(*elts)); ++i) {
+		struct txq_elt *elt = &(*elts)[i];
+
+		if (elt->buf == NULL)
+			continue;
+		rte_pktmbuf_free(elt->buf);
+	}
+	rte_free(elts);
+}
+
+/**
+ * Clean up a TX queue.
+ *
+ * Destroy objects, free allocated memory and reset the structure for reuse.
+ *
+ * @param txq
+ *   Pointer to TX queue structure.
+ */
+void
+txq_cleanup(struct txq *txq)
+{
+	struct ibv_exp_release_intf_params params;
+	size_t i;
+
+	DEBUG("cleaning up %p", (void *)txq);
+	txq_free_elts(txq);
+	if (txq->if_qp != NULL) {
+		assert(txq->priv != NULL);
+		assert(txq->priv->ctx != NULL);
+		assert(txq->qp != NULL);
+		params = (struct ibv_exp_release_intf_params){
+			.comp_mask = 0,
+		};
+		claim_zero(ibv_exp_release_intf(txq->priv->ctx,
+						txq->if_qp,
+						&params));
+	}
+	if (txq->if_cq != NULL) {
+		assert(txq->priv != NULL);
+		assert(txq->priv->ctx != NULL);
+		assert(txq->cq != NULL);
+		params = (struct ibv_exp_release_intf_params){
+			.comp_mask = 0,
+		};
+		claim_zero(ibv_exp_release_intf(txq->priv->ctx,
+						txq->if_cq,
+						&params));
+	}
+	if (txq->qp != NULL)
+		claim_zero(ibv_destroy_qp(txq->qp));
+	if (txq->cq != NULL)
+		claim_zero(ibv_destroy_cq(txq->cq));
+	if (txq->rd != NULL) {
+		struct ibv_exp_destroy_res_domain_attr attr = {
+			.comp_mask = 0,
+		};
+
+		assert(txq->priv != NULL);
+		assert(txq->priv->ctx != NULL);
+		claim_zero(ibv_exp_destroy_res_domain(txq->priv->ctx,
+						      txq->rd,
+						      &attr));
+	}
+	for (i = 0; (i != RTE_DIM(txq->mp2mr)); ++i) {
+		if (txq->mp2mr[i].mp == NULL)
+			break;
+		assert(txq->mp2mr[i].mr != NULL);
+		claim_zero(ibv_dereg_mr(txq->mp2mr[i].mr));
+	}
+	memset(txq, 0, sizeof(*txq));
+}
+
+/**
+ * Configure a TX queue.
+ *
+ * @param dev
+ *   Pointer to Ethernet device structure.
+ * @param txq
+ *   Pointer to TX queue structure.
+ * @param desc
+ *   Number of descriptors to configure in queue.
+ * @param socket
+ *   NUMA socket on which memory must be allocated.
+ * @param[in] conf
+ *   Thresholds parameters.
+ *
+ * @return
+ *   0 on success, errno value on failure.
+ */
+static int
+txq_setup(struct rte_eth_dev *dev, struct txq *txq, uint16_t desc,
+	  unsigned int socket, const struct rte_eth_txconf *conf)
+{
+	struct priv *priv = dev->data->dev_private;
+	struct txq tmpl = {
+		.priv = priv,
+		.socket = socket
+	};
+	union {
+		struct ibv_exp_query_intf_params params;
+		struct ibv_exp_qp_init_attr init;
+		struct ibv_exp_res_domain_init_attr rd;
+		struct ibv_exp_cq_init_attr cq;
+		struct ibv_exp_qp_attr mod;
+	} attr;
+	enum ibv_exp_query_intf_status status;
+	int ret = 0;
+
+	(void)conf; /* Thresholds configuration (ignored). */
+	if ((desc == 0) || (desc % MLX5_PMD_SGE_WR_N)) {
+		ERROR("%p: invalid number of TX descriptors (must be a"
+		      " multiple of %d)", (void *)dev, MLX5_PMD_SGE_WR_N);
+		return EINVAL;
+	}
+	desc /= MLX5_PMD_SGE_WR_N;
+	/* MRs will be registered in mp2mr[] later. */
+	attr.rd = (struct ibv_exp_res_domain_init_attr){
+		.comp_mask = (IBV_EXP_RES_DOMAIN_THREAD_MODEL |
+			      IBV_EXP_RES_DOMAIN_MSG_MODEL),
+		.thread_model = IBV_EXP_THREAD_SINGLE,
+		.msg_model = IBV_EXP_MSG_HIGH_BW,
+	};
+	tmpl.rd = ibv_exp_create_res_domain(priv->ctx, &attr.rd);
+	if (tmpl.rd == NULL) {
+		ret = ENOMEM;
+		ERROR("%p: RD creation failure: %s",
+		      (void *)dev, strerror(ret));
+		goto error;
+	}
+	attr.cq = (struct ibv_exp_cq_init_attr){
+		.comp_mask = IBV_EXP_CQ_INIT_ATTR_RES_DOMAIN,
+		.res_domain = tmpl.rd,
+	};
+	tmpl.cq = ibv_exp_create_cq(priv->ctx, desc, NULL, NULL, 0, &attr.cq);
+	if (tmpl.cq == NULL) {
+		ret = ENOMEM;
+		ERROR("%p: CQ creation failure: %s",
+		      (void *)dev, strerror(ret));
+		goto error;
+	}
+	DEBUG("priv->device_attr.max_qp_wr is %d",
+	      priv->device_attr.max_qp_wr);
+	DEBUG("priv->device_attr.max_sge is %d",
+	      priv->device_attr.max_sge);
+	attr.init = (struct ibv_exp_qp_init_attr){
+		/* CQ to be associated with the send queue. */
+		.send_cq = tmpl.cq,
+		/* CQ to be associated with the receive queue. */
+		.recv_cq = tmpl.cq,
+		.cap = {
+			/* Max number of outstanding WRs. */
+			.max_send_wr = ((priv->device_attr.max_qp_wr < desc) ?
+					priv->device_attr.max_qp_wr :
+					desc),
+			/* Max number of scatter/gather elements in a WR. */
+			.max_send_sge = ((priv->device_attr.max_sge <
+					  MLX5_PMD_SGE_WR_N) ?
+					 priv->device_attr.max_sge :
+					 MLX5_PMD_SGE_WR_N),
+#if MLX5_PMD_MAX_INLINE > 0
+			.max_inline_data = MLX5_PMD_MAX_INLINE,
+#endif
+		},
+		.qp_type = IBV_QPT_RAW_PACKET,
+		/* Do *NOT* enable this, completions events are managed per
+		 * TX burst. */
+		.sq_sig_all = 0,
+		.pd = priv->pd,
+		.res_domain = tmpl.rd,
+		.comp_mask = (IBV_EXP_QP_INIT_ATTR_PD |
+			      IBV_EXP_QP_INIT_ATTR_RES_DOMAIN),
+	};
+	tmpl.qp = ibv_exp_create_qp(priv->ctx, &attr.init);
+	if (tmpl.qp == NULL) {
+		ret = (errno ? errno : EINVAL);
+		ERROR("%p: QP creation failure: %s",
+		      (void *)dev, strerror(ret));
+		goto error;
+	}
+#if MLX5_PMD_MAX_INLINE > 0
+	/* ibv_create_qp() updates this value. */
+	tmpl.max_inline = attr.init.cap.max_inline_data;
+#endif
+	attr.mod = (struct ibv_exp_qp_attr){
+		/* Move the QP to this state. */
+		.qp_state = IBV_QPS_INIT,
+		/* Primary port number. */
+		.port_num = priv->port
+	};
+	ret = ibv_exp_modify_qp(tmpl.qp, &attr.mod,
+				(IBV_EXP_QP_STATE | IBV_EXP_QP_PORT));
+	if (ret) {
+		ERROR("%p: QP state to IBV_QPS_INIT failed: %s",
+		      (void *)dev, strerror(ret));
+		goto error;
+	}
+	ret = txq_alloc_elts(&tmpl, desc);
+	if (ret) {
+		ERROR("%p: TXQ allocation failed: %s",
+		      (void *)dev, strerror(ret));
+		goto error;
+	}
+	attr.mod = (struct ibv_exp_qp_attr){
+		.qp_state = IBV_QPS_RTR
+	};
+	ret = ibv_exp_modify_qp(tmpl.qp, &attr.mod, IBV_EXP_QP_STATE);
+	if (ret) {
+		ERROR("%p: QP state to IBV_QPS_RTR failed: %s",
+		      (void *)dev, strerror(ret));
+		goto error;
+	}
+	attr.mod.qp_state = IBV_QPS_RTS;
+	ret = ibv_exp_modify_qp(tmpl.qp, &attr.mod, IBV_EXP_QP_STATE);
+	if (ret) {
+		ERROR("%p: QP state to IBV_QPS_RTS failed: %s",
+		      (void *)dev, strerror(ret));
+		goto error;
+	}
+	attr.params = (struct ibv_exp_query_intf_params){
+		.intf_scope = IBV_EXP_INTF_GLOBAL,
+		.intf = IBV_EXP_INTF_CQ,
+		.obj = tmpl.cq,
+	};
+	tmpl.if_cq = ibv_exp_query_intf(priv->ctx, &attr.params, &status);
+	if (tmpl.if_cq == NULL) {
+		ret = EINVAL;
+		ERROR("%p: CQ interface family query failed with status %d",
+		      (void *)dev, status);
+		goto error;
+	}
+	attr.params = (struct ibv_exp_query_intf_params){
+		.intf_scope = IBV_EXP_INTF_GLOBAL,
+		.intf = IBV_EXP_INTF_QP_BURST,
+		.obj = tmpl.qp,
+	};
+	tmpl.if_qp = ibv_exp_query_intf(priv->ctx, &attr.params, &status);
+	if (tmpl.if_qp == NULL) {
+		ret = EINVAL;
+		ERROR("%p: QP interface family query failed with status %d",
+		      (void *)dev, status);
+		goto error;
+	}
+	/* Clean up txq in case we're reinitializing it. */
+	DEBUG("%p: cleaning-up old txq just in case", (void *)txq);
+	txq_cleanup(txq);
+	*txq = tmpl;
+	DEBUG("%p: txq updated with %p", (void *)txq, (void *)&tmpl);
+	assert(ret == 0);
+	return 0;
+error:
+	txq_cleanup(&tmpl);
+	assert(ret > 0);
+	return ret;
+}
+
+/**
+ * DPDK callback to configure a TX queue.
+ *
+ * @param dev
+ *   Pointer to Ethernet device structure.
+ * @param idx
+ *   TX queue index.
+ * @param desc
+ *   Number of descriptors to configure in queue.
+ * @param socket
+ *   NUMA socket on which memory must be allocated.
+ * @param[in] conf
+ *   Thresholds parameters.
+ *
+ * @return
+ *   0 on success, negative errno value on failure.
+ */
+int
+mlx5_tx_queue_setup(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc,
+		    unsigned int socket, const struct rte_eth_txconf *conf)
+{
+	struct priv *priv = dev->data->dev_private;
+	struct txq *txq = (*priv->txqs)[idx];
+	int ret;
+
+	priv_lock(priv);
+	DEBUG("%p: configuring queue %u for %u descriptors",
+	      (void *)dev, idx, desc);
+	if (idx >= priv->txqs_n) {
+		ERROR("%p: queue index out of range (%u >= %u)",
+		      (void *)dev, idx, priv->txqs_n);
+		priv_unlock(priv);
+		return -EOVERFLOW;
+	}
+	if (txq != NULL) {
+		DEBUG("%p: reusing already allocated queue index %u (%p)",
+		      (void *)dev, idx, (void *)txq);
+		if (priv->started) {
+			priv_unlock(priv);
+			return -EEXIST;
+		}
+		(*priv->txqs)[idx] = NULL;
+		txq_cleanup(txq);
+	} else {
+		txq = rte_calloc_socket("TXQ", 1, sizeof(*txq), 0, socket);
+		if (txq == NULL) {
+			ERROR("%p: unable to allocate queue index %u",
+			      (void *)dev, idx);
+			priv_unlock(priv);
+			return -ENOMEM;
+		}
+	}
+	ret = txq_setup(dev, txq, desc, socket, conf);
+	if (ret)
+		rte_free(txq);
+	else {
+		DEBUG("%p: adding TX queue %p to list",
+		      (void *)dev, (void *)txq);
+		(*priv->txqs)[idx] = txq;
+		/* Update send callback. */
+		dev->tx_pkt_burst = mlx5_tx_burst;
+	}
+	priv_unlock(priv);
+	return -ret;
+}
+
+/**
+ * DPDK callback to release a TX queue.
+ *
+ * @param dpdk_txq
+ *   Generic TX queue pointer.
+ */
+void
+mlx5_tx_queue_release(void *dpdk_txq)
+{
+	struct txq *txq = (struct txq *)dpdk_txq;
+	struct priv *priv;
+	unsigned int i;
+
+	if (txq == NULL)
+		return;
+	priv = txq->priv;
+	priv_lock(priv);
+	for (i = 0; (i != priv->txqs_n); ++i)
+		if ((*priv->txqs)[i] == txq) {
+			DEBUG("%p: removing TX queue %p from list",
+			      (void *)priv->dev, (void *)txq);
+			(*priv->txqs)[i] = NULL;
+			break;
+		}
+	txq_cleanup(txq);
+	rte_free(txq);
+	priv_unlock(priv);
+}
diff --git a/drivers/net/mlx5/mlx5_utils.h b/drivers/net/mlx5/mlx5_utils.h
index cc6aab6..e48e6b6 100644
--- a/drivers/net/mlx5/mlx5_utils.h
+++ b/drivers/net/mlx5/mlx5_utils.h
@@ -140,10 +140,21 @@ pmd_drv_log_basename(const char *s)
 #define WARN(...) PMD_DRV_LOG(WARNING, __VA_ARGS__)
 #define ERROR(...) PMD_DRV_LOG(ERR, __VA_ARGS__)
 
+/* Convenience macros for accessing mbuf fields. */
+#define NEXT(m) ((m)->next)
+#define DATA_LEN(m) ((m)->data_len)
+#define PKT_LEN(m) ((m)->pkt_len)
+#define DATA_OFF(m) ((m)->data_off)
+#define SET_DATA_OFF(m, o) ((m)->data_off = (o))
+#define NB_SEGS(m) ((m)->nb_segs)
+#define PORT(m) ((m)->port)
+
 /* Allocate a buffer on the stack and fill it with a printf format string. */
 #define MKSTR(name, ...) \
 	char name[snprintf(NULL, 0, __VA_ARGS__) + 1]; \
 	\
 	snprintf(name, sizeof(name), __VA_ARGS__)
 
+#define WR_ID(o) (((wr_id_t *)&(o))->data)
+
 #endif /* RTE_PMD_MLX5_UTILS_H_ */
-- 
2.1.0

^ permalink raw reply	[flat|nested] 32+ messages in thread

* [dpdk-dev] [PATCH 03/13] mlx5: add MAC handling
  2015-10-05 17:52 [dpdk-dev] [PATCH 00/13] Mellanox ConnectX-4 PMD (mlx5) Adrien Mazarguil
  2015-10-05 17:52 ` [dpdk-dev] [PATCH 01/13] mlx5: new poll-mode driver for Mellanox ConnectX-4 adapters Adrien Mazarguil
  2015-10-05 17:52 ` [dpdk-dev] [PATCH 02/13] mlx5: add non-scattered TX and RX support Adrien Mazarguil
@ 2015-10-05 17:52 ` Adrien Mazarguil
  2015-10-05 17:53 ` [dpdk-dev] [PATCH 04/13] mlx5: add device configure/start/stop Adrien Mazarguil
                   ` (10 subsequent siblings)
  13 siblings, 0 replies; 32+ messages in thread
From: Adrien Mazarguil @ 2015-10-05 17:52 UTC (permalink / raw)
  To: dev

This commit adds support for MAC flow steering rules mandatory for the RX
path as well as the related callbacks to add/remove MAC addresses.

Signed-off-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>
Signed-off-by: Nelio Laranjeiro <nelio.laranjeiro@6wind.com>
Signed-off-by: Didier Pallard <didier.pallard@6wind.com>
---
 drivers/net/mlx5/mlx5.c     |   4 +-
 drivers/net/mlx5/mlx5.h     |   5 +
 drivers/net/mlx5/mlx5_mac.c | 347 ++++++++++++++++++++++++++++++++++++++++++++
 drivers/net/mlx5/mlx5_rxq.c |  10 ++
 4 files changed, 365 insertions(+), 1 deletion(-)

diff --git a/drivers/net/mlx5/mlx5.c b/drivers/net/mlx5/mlx5.c
index 31ce5ec..fb306d4 100644
--- a/drivers/net/mlx5/mlx5.c
+++ b/drivers/net/mlx5/mlx5.c
@@ -134,6 +134,8 @@ static const struct eth_dev_ops mlx5_dev_ops = {
 	.tx_queue_setup = mlx5_tx_queue_setup,
 	.rx_queue_release = mlx5_rx_queue_release,
 	.tx_queue_release = mlx5_tx_queue_release,
+	.mac_addr_remove = mlx5_mac_addr_remove,
+	.mac_addr_add = mlx5_mac_addr_add,
 };
 
 static struct {
@@ -390,7 +392,7 @@ mlx5_pci_devinit(struct rte_pci_driver *pci_drv, struct rte_pci_device *pci_dev)
 		claim_zero(priv_mac_addr_add(priv, 0,
 					     (const uint8_t (*)[ETHER_ADDR_LEN])
 					     mac.addr_bytes));
-		claim_zero(priv_mac_addr_add(priv, 1,
+		claim_zero(priv_mac_addr_add(priv, (RTE_DIM(priv->mac) - 1),
 					     &(const uint8_t [ETHER_ADDR_LEN])
 					     { "\xff\xff\xff\xff\xff\xff" }));
 #ifndef NDEBUG
diff --git a/drivers/net/mlx5/mlx5.h b/drivers/net/mlx5/mlx5.h
index 7e60e70..3e0c11e 100644
--- a/drivers/net/mlx5/mlx5.h
+++ b/drivers/net/mlx5/mlx5.h
@@ -173,7 +173,12 @@ int mlx5_ibv_device_to_pci_addr(const struct ibv_device *,
 /* mlx5_mac.c */
 
 int priv_get_mac(struct priv *, uint8_t (*)[ETHER_ADDR_LEN]);
+void rxq_mac_addrs_del(struct rxq *);
+void mlx5_mac_addr_remove(struct rte_eth_dev *, uint32_t);
+int rxq_mac_addrs_add(struct rxq *);
 int priv_mac_addr_add(struct priv *, unsigned int,
 		      const uint8_t (*)[ETHER_ADDR_LEN]);
+void mlx5_mac_addr_add(struct rte_eth_dev *, struct ether_addr *, uint32_t,
+		       uint32_t);
 
 #endif /* RTE_PMD_MLX5_H_ */
diff --git a/drivers/net/mlx5/mlx5_mac.c b/drivers/net/mlx5/mlx5_mac.c
index f7e1cf6..f01faf0 100644
--- a/drivers/net/mlx5/mlx5_mac.c
+++ b/drivers/net/mlx5/mlx5_mac.c
@@ -65,6 +65,8 @@
 
 #include "mlx5.h"
 #include "mlx5_utils.h"
+#include "mlx5_rxtx.h"
+#include "mlx5_defs.h"
 
 /**
  * Get MAC address by querying netdevice.
@@ -89,8 +91,86 @@ priv_get_mac(struct priv *priv, uint8_t (*mac)[ETHER_ADDR_LEN])
 }
 
 /**
+ * Delete flow steering rule.
+ *
+ * @param rxq
+ *   Pointer to RX queue structure.
+ * @param mac_index
+ *   MAC address index.
+ * @param vlan_index
+ *   VLAN index.
+ */
+static void
+rxq_del_flow(struct rxq *rxq, unsigned int mac_index, unsigned int vlan_index)
+{
+#ifndef NDEBUG
+	struct priv *priv = rxq->priv;
+	const uint8_t (*mac)[ETHER_ADDR_LEN] =
+		(const uint8_t (*)[ETHER_ADDR_LEN])
+		priv->mac[mac_index].addr_bytes;
+#endif
+	assert(rxq->mac_flow[mac_index][vlan_index] != NULL);
+	DEBUG("%p: removing MAC address %02x:%02x:%02x:%02x:%02x:%02x index %u"
+	      " (VLAN ID %" PRIu16 ")",
+	      (void *)rxq,
+	      (*mac)[0], (*mac)[1], (*mac)[2], (*mac)[3], (*mac)[4], (*mac)[5],
+	      mac_index, priv->vlan_filter[vlan_index].id);
+	claim_zero(ibv_destroy_flow(rxq->mac_flow[mac_index][vlan_index]));
+	rxq->mac_flow[mac_index][vlan_index] = NULL;
+}
+
+/**
+ * Unregister a MAC address from a RX queue.
+ *
+ * @param rxq
+ *   Pointer to RX queue structure.
+ * @param mac_index
+ *   MAC address index.
+ */
+static void
+rxq_mac_addr_del(struct rxq *rxq, unsigned int mac_index)
+{
+	struct priv *priv = rxq->priv;
+	unsigned int i;
+	unsigned int vlans = 0;
+
+	assert(mac_index < RTE_DIM(priv->mac));
+	if (!BITFIELD_ISSET(rxq->mac_configured, mac_index))
+		return;
+	for (i = 0; (i != RTE_DIM(priv->vlan_filter)); ++i) {
+		if (!priv->vlan_filter[i].enabled)
+			continue;
+		rxq_del_flow(rxq, mac_index, i);
+		vlans++;
+	}
+	if (!vlans) {
+		rxq_del_flow(rxq, mac_index, 0);
+	}
+	BITFIELD_RESET(rxq->mac_configured, mac_index);
+}
+
+/**
+ * Unregister all MAC addresses from a RX queue.
+ *
+ * @param rxq
+ *   Pointer to RX queue structure.
+ */
+void
+rxq_mac_addrs_del(struct rxq *rxq)
+{
+	struct priv *priv = rxq->priv;
+	unsigned int i;
+
+	for (i = 0; (i != RTE_DIM(priv->mac)); ++i)
+		rxq_mac_addr_del(rxq, i);
+}
+
+/**
  * Unregister a MAC address.
  *
+ * In RSS mode, the MAC address is unregistered from the parent queue,
+ * otherwise it is unregistered from each queue directly.
+ *
  * @param priv
  *   Pointer to private structure.
  * @param mac_index
@@ -99,15 +179,217 @@ priv_get_mac(struct priv *priv, uint8_t (*mac)[ETHER_ADDR_LEN])
 static void
 priv_mac_addr_del(struct priv *priv, unsigned int mac_index)
 {
+	unsigned int i;
+
 	assert(mac_index < RTE_DIM(priv->mac));
 	if (!BITFIELD_ISSET(priv->mac_configured, mac_index))
 		return;
+	if (priv->rss) {
+		rxq_mac_addr_del(&priv->rxq_parent, mac_index);
+		goto end;
+	}
+	for (i = 0; (i != priv->dev->data->nb_rx_queues); ++i)
+		rxq_mac_addr_del((*priv->rxqs)[i], mac_index);
+end:
 	BITFIELD_RESET(priv->mac_configured, mac_index);
 }
 
 /**
+ * DPDK callback to remove a MAC address.
+ *
+ * @param dev
+ *   Pointer to Ethernet device structure.
+ * @param index
+ *   MAC address index.
+ */
+void
+mlx5_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
+{
+	struct priv *priv = dev->data->dev_private;
+
+	priv_lock(priv);
+	DEBUG("%p: removing MAC address from index %" PRIu32,
+	      (void *)dev, index);
+	/* Last array entry is reserved for broadcast. */
+	if (index >= (RTE_DIM(priv->mac) - 1))
+		goto end;
+	priv_mac_addr_del(priv, index);
+end:
+	priv_unlock(priv);
+}
+
+/**
+ * Add single flow steering rule.
+ *
+ * @param rxq
+ *   Pointer to RX queue structure.
+ * @param mac_index
+ *   MAC address index to register.
+ * @param vlan_index
+ *   VLAN index. Use -1 for a flow without VLAN.
+ *
+ * @return
+ *   0 on success, errno value on failure.
+ */
+static int
+rxq_add_flow(struct rxq *rxq, unsigned int mac_index, unsigned int vlan_index)
+{
+	struct ibv_flow *flow;
+	struct priv *priv = rxq->priv;
+	const uint8_t (*mac)[ETHER_ADDR_LEN] =
+			(const uint8_t (*)[ETHER_ADDR_LEN])
+			priv->mac[mac_index].addr_bytes;
+
+	/* Allocate flow specification on the stack. */
+	struct __attribute__((packed)) {
+		struct ibv_flow_attr attr;
+		struct ibv_flow_spec_eth spec;
+	} data;
+	struct ibv_flow_attr *attr = &data.attr;
+	struct ibv_flow_spec_eth *spec = &data.spec;
+
+	assert(mac_index < RTE_DIM(priv->mac));
+	assert((vlan_index < RTE_DIM(priv->vlan_filter)) || (vlan_index == -1u));
+	/*
+	 * No padding must be inserted by the compiler between attr and spec.
+	 * This layout is expected by libibverbs.
+	 */
+	assert(((uint8_t *)attr + sizeof(*attr)) == (uint8_t *)spec);
+	*attr = (struct ibv_flow_attr){
+		.type = IBV_FLOW_ATTR_NORMAL,
+		.num_of_specs = 1,
+		.port = priv->port,
+		.flags = 0
+	};
+	*spec = (struct ibv_flow_spec_eth){
+		.type = IBV_FLOW_SPEC_ETH,
+		.size = sizeof(*spec),
+		.val = {
+			.dst_mac = {
+				(*mac)[0], (*mac)[1], (*mac)[2],
+				(*mac)[3], (*mac)[4], (*mac)[5]
+			},
+			.vlan_tag = ((vlan_index != -1u) ?
+				     htons(priv->vlan_filter[vlan_index].id) :
+				     0),
+		},
+		.mask = {
+			.dst_mac = "\xff\xff\xff\xff\xff\xff",
+			.vlan_tag = ((vlan_index != -1u) ? htons(0xfff) : 0),
+		}
+	};
+	DEBUG("%p: adding MAC address %02x:%02x:%02x:%02x:%02x:%02x index %u"
+	      " (VLAN %s %" PRIu16 ")",
+	      (void *)rxq,
+	      (*mac)[0], (*mac)[1], (*mac)[2], (*mac)[3], (*mac)[4], (*mac)[5],
+	      mac_index,
+	      ((vlan_index != -1u) ? "ID" : "index"),
+	      ((vlan_index != -1u) ? priv->vlan_filter[vlan_index].id : -1u));
+	/* Create related flow. */
+	errno = 0;
+	flow = ibv_create_flow(rxq->qp, attr);
+	if (flow == NULL) {
+		/* It's not clear whether errno is always set in this case. */
+		ERROR("%p: flow configuration failed, errno=%d: %s",
+		      (void *)rxq, errno,
+		      (errno ? strerror(errno) : "Unknown error"));
+		if (errno)
+			return errno;
+		return EINVAL;
+	}
+	if (vlan_index == -1u)
+		vlan_index = 0;
+	assert(rxq->mac_flow[mac_index][vlan_index] == NULL);
+	rxq->mac_flow[mac_index][vlan_index] = flow;
+	return 0;
+}
+
+/**
+ * Register a MAC address in a RX queue.
+ *
+ * @param rxq
+ *   Pointer to RX queue structure.
+ * @param mac_index
+ *   MAC address index to register.
+ *
+ * @return
+ *   0 on success, errno value on failure.
+ */
+static int
+rxq_mac_addr_add(struct rxq *rxq, unsigned int mac_index)
+{
+	struct priv *priv = rxq->priv;
+	unsigned int i;
+	unsigned int vlans = 0;
+	int ret;
+
+	assert(mac_index < RTE_DIM(priv->mac));
+	if (BITFIELD_ISSET(rxq->mac_configured, mac_index))
+		rxq_mac_addr_del(rxq, mac_index);
+	/* Fill VLAN specifications. */
+	for (i = 0; (i != RTE_DIM(priv->vlan_filter)); ++i) {
+		if (!priv->vlan_filter[i].enabled)
+			continue;
+		/* Create related flow. */
+		ret = rxq_add_flow(rxq, mac_index, i);
+		if (!ret) {
+			vlans++;
+			continue;
+		}
+		/* Failure, rollback. */
+		while (i != 0)
+			if (priv->vlan_filter[--i].enabled)
+				rxq_del_flow(rxq, mac_index, i);
+		assert(ret > 0);
+		return ret;
+	}
+	/* In case there is no VLAN filter. */
+	if (!vlans) {
+		ret = rxq_add_flow(rxq, mac_index, -1);
+		if (ret)
+			return ret;
+	}
+	BITFIELD_SET(rxq->mac_configured, mac_index);
+	return 0;
+}
+
+/**
+ * Register all MAC addresses in a RX queue.
+ *
+ * @param rxq
+ *   Pointer to RX queue structure.
+ *
+ * @return
+ *   0 on success, errno value on failure.
+ */
+int
+rxq_mac_addrs_add(struct rxq *rxq)
+{
+	struct priv *priv = rxq->priv;
+	unsigned int i;
+	int ret;
+
+	for (i = 0; (i != RTE_DIM(priv->mac)); ++i) {
+		if (!BITFIELD_ISSET(priv->mac_configured, i))
+			continue;
+		ret = rxq_mac_addr_add(rxq, i);
+		if (!ret)
+			continue;
+		/* Failure, rollback. */
+		while (i != 0)
+			rxq_mac_addr_del(rxq, --i);
+		assert(ret > 0);
+		return ret;
+	}
+	return 0;
+}
+
+/**
  * Register a MAC address.
  *
+ * In RSS mode, the MAC address is registered in the parent queue,
+ * otherwise it is registered in each queue directly.
+ *
  * @param priv
  *   Pointer to private structure.
  * @param mac_index
@@ -123,6 +405,7 @@ priv_mac_addr_add(struct priv *priv, unsigned int mac_index,
 		  const uint8_t (*mac)[ETHER_ADDR_LEN])
 {
 	unsigned int i;
+	int ret;
 
 	assert(mac_index < RTE_DIM(priv->mac));
 	/* First, make sure this address isn't already configured. */
@@ -145,6 +428,70 @@ priv_mac_addr_add(struct priv *priv, unsigned int mac_index,
 			(*mac)[3], (*mac)[4], (*mac)[5]
 		}
 	};
+	/* If device isn't started, this is all we need to do. */
+	if (!priv->started) {
+#ifndef NDEBUG
+		/* Verify that all queues have this index disabled. */
+		for (i = 0; (i != priv->rxqs_n); ++i) {
+			if ((*priv->rxqs)[i] == NULL)
+				continue;
+			assert(!BITFIELD_ISSET
+			       ((*priv->rxqs)[i]->mac_configured, mac_index));
+		}
+#endif
+		goto end;
+	}
+	if (priv->rss) {
+		ret = rxq_mac_addr_add(&priv->rxq_parent, mac_index);
+		if (ret)
+			return ret;
+		goto end;
+	}
+	for (i = 0; (i != priv->rxqs_n); ++i) {
+		if ((*priv->rxqs)[i] == NULL)
+			continue;
+		ret = rxq_mac_addr_add((*priv->rxqs)[i], mac_index);
+		if (!ret)
+			continue;
+		/* Failure, rollback. */
+		while (i != 0)
+			if ((*priv->rxqs)[(--i)] != NULL)
+				rxq_mac_addr_del((*priv->rxqs)[i], mac_index);
+		return ret;
+	}
+end:
 	BITFIELD_SET(priv->mac_configured, mac_index);
 	return 0;
 }
+
+/**
+ * DPDK callback to add a MAC address.
+ *
+ * @param dev
+ *   Pointer to Ethernet device structure.
+ * @param mac_addr
+ *   MAC address to register.
+ * @param index
+ *   MAC address index.
+ * @param vmdq
+ *   VMDq pool index to associate address with (ignored).
+ */
+void
+mlx5_mac_addr_add(struct rte_eth_dev *dev, struct ether_addr *mac_addr,
+		  uint32_t index, uint32_t vmdq)
+{
+	struct priv *priv = dev->data->dev_private;
+
+	(void)vmdq;
+	priv_lock(priv);
+	DEBUG("%p: adding MAC address at index %" PRIu32,
+	      (void *)dev, index);
+	/* Last array entry is reserved for broadcast. */
+	if (index >= (RTE_DIM(priv->mac) - 1))
+		goto end;
+	priv_mac_addr_add(priv, index,
+			  (const uint8_t (*)[ETHER_ADDR_LEN])
+			  mac_addr->addr_bytes);
+end:
+	priv_unlock(priv);
+}
diff --git a/drivers/net/mlx5/mlx5_rxq.c b/drivers/net/mlx5/mlx5_rxq.c
index 01cc649..8450fe3 100644
--- a/drivers/net/mlx5/mlx5_rxq.c
+++ b/drivers/net/mlx5/mlx5_rxq.c
@@ -248,6 +248,7 @@ rxq_cleanup(struct rxq *rxq)
 						&params));
 	}
 	if (rxq->qp != NULL) {
+		rxq_mac_addrs_del(rxq);
 		claim_zero(ibv_destroy_qp(rxq->qp));
 	}
 	if (rxq->cq != NULL)
@@ -515,6 +516,15 @@ skip_mr:
 		      (void *)dev, strerror(ret));
 		goto error;
 	}
+	if ((parent) || (!priv->rss))  {
+		/* Configure MAC and broadcast addresses. */
+		ret = rxq_mac_addrs_add(&tmpl);
+		if (ret) {
+			ERROR("%p: QP flow attachment failed: %s",
+			      (void *)dev, strerror(ret));
+			goto error;
+		}
+	}
 	/* Allocate descriptors for RX queues, except for the RSS parent. */
 	if (parent)
 		goto skip_alloc;
-- 
2.1.0

^ permalink raw reply	[flat|nested] 32+ messages in thread

* [dpdk-dev] [PATCH 04/13] mlx5: add device configure/start/stop
  2015-10-05 17:52 [dpdk-dev] [PATCH 00/13] Mellanox ConnectX-4 PMD (mlx5) Adrien Mazarguil
                   ` (2 preceding siblings ...)
  2015-10-05 17:52 ` [dpdk-dev] [PATCH 03/13] mlx5: add MAC handling Adrien Mazarguil
@ 2015-10-05 17:53 ` Adrien Mazarguil
  2015-10-05 17:53 ` [dpdk-dev] [PATCH 05/13] mlx5: add support for scattered RX and TX buffers Adrien Mazarguil
                   ` (9 subsequent siblings)
  13 siblings, 0 replies; 32+ messages in thread
From: Adrien Mazarguil @ 2015-10-05 17:53 UTC (permalink / raw)
  To: dev; +Cc: Francesco Santoro

This commit adds the remaining missing callbacks to make mlx5 usable.
Like mlx4, device start and stop are implemented on top of MAC RX flows.

Signed-off-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>
Signed-off-by: Nelio Laranjeiro <nelio.laranjeiro@6wind.com>
Signed-off-by: Francesco Santoro <francesco.santoro@6wind.com>
Signed-off-by: Didier Pallard <didier.pallard@6wind.com>
---
 drivers/net/mlx5/Makefile       |   1 +
 drivers/net/mlx5/mlx5.c         |   4 ++
 drivers/net/mlx5/mlx5.h         |   7 ++
 drivers/net/mlx5/mlx5_ethdev.c  | 148 ++++++++++++++++++++++++++++++++++++++++
 drivers/net/mlx5/mlx5_trigger.c | 145 +++++++++++++++++++++++++++++++++++++++
 5 files changed, 305 insertions(+)
 create mode 100644 drivers/net/mlx5/mlx5_trigger.c

diff --git a/drivers/net/mlx5/Makefile b/drivers/net/mlx5/Makefile
index 7b9c57b..028c22c 100644
--- a/drivers/net/mlx5/Makefile
+++ b/drivers/net/mlx5/Makefile
@@ -45,6 +45,7 @@ SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5.c
 SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5_rxq.c
 SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5_txq.c
 SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5_rxtx.c
+SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5_trigger.c
 SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5_ethdev.c
 SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5_mac.c
 
diff --git a/drivers/net/mlx5/mlx5.c b/drivers/net/mlx5/mlx5.c
index fb306d4..56cea7c 100644
--- a/drivers/net/mlx5/mlx5.c
+++ b/drivers/net/mlx5/mlx5.c
@@ -129,7 +129,11 @@ mlx5_dev_close(struct rte_eth_dev *dev)
 }
 
 static const struct eth_dev_ops mlx5_dev_ops = {
+	.dev_configure = mlx5_dev_configure,
+	.dev_start = mlx5_dev_start,
+	.dev_stop = mlx5_dev_stop,
 	.dev_close = mlx5_dev_close,
+	.dev_infos_get = mlx5_dev_infos_get,
 	.rx_queue_setup = mlx5_rx_queue_setup,
 	.tx_queue_setup = mlx5_tx_queue_setup,
 	.rx_queue_release = mlx5_rx_queue_release,
diff --git a/drivers/net/mlx5/mlx5.h b/drivers/net/mlx5/mlx5.h
index 3e0c11e..3f4f8fd 100644
--- a/drivers/net/mlx5/mlx5.h
+++ b/drivers/net/mlx5/mlx5.h
@@ -167,6 +167,8 @@ int priv_get_ifname(const struct priv *, char (*)[IF_NAMESIZE]);
 int priv_ifreq(const struct priv *, int req, struct ifreq *);
 int priv_get_mtu(struct priv *, uint16_t *);
 int priv_set_flags(struct priv *, unsigned int, unsigned int);
+int mlx5_dev_configure(struct rte_eth_dev *);
+void mlx5_dev_infos_get(struct rte_eth_dev *, struct rte_eth_dev_info *);
 int mlx5_ibv_device_to_pci_addr(const struct ibv_device *,
 				struct rte_pci_addr *);
 
@@ -181,4 +183,9 @@ int priv_mac_addr_add(struct priv *, unsigned int,
 void mlx5_mac_addr_add(struct rte_eth_dev *, struct ether_addr *, uint32_t,
 		       uint32_t);
 
+/* mlx5_trigger.c */
+
+int mlx5_dev_start(struct rte_eth_dev *);
+void mlx5_dev_stop(struct rte_eth_dev *);
+
 #endif /* RTE_PMD_MLX5_H_ */
diff --git a/drivers/net/mlx5/mlx5_ethdev.c b/drivers/net/mlx5/mlx5_ethdev.c
index b6c7d7a..6b13cec 100644
--- a/drivers/net/mlx5/mlx5_ethdev.c
+++ b/drivers/net/mlx5/mlx5_ethdev.c
@@ -32,6 +32,7 @@
  */
 
 #include <stddef.h>
+#include <assert.h>
 #include <unistd.h>
 #include <stdint.h>
 #include <stdio.h>
@@ -58,6 +59,7 @@
 #endif
 
 #include "mlx5.h"
+#include "mlx5_rxtx.h"
 #include "mlx5_utils.h"
 
 /**
@@ -370,6 +372,152 @@ priv_set_flags(struct priv *priv, unsigned int keep, unsigned int flags)
 }
 
 /**
+ * Ethernet device configuration.
+ *
+ * Prepare the driver for a given number of TX and RX queues.
+ * Allocate parent RSS queue when several RX queues are requested.
+ *
+ * @param dev
+ *   Pointer to Ethernet device structure.
+ *
+ * @return
+ *   0 on success, errno value on failure.
+ */
+static int
+dev_configure(struct rte_eth_dev *dev)
+{
+	struct priv *priv = dev->data->dev_private;
+	unsigned int rxqs_n = dev->data->nb_rx_queues;
+	unsigned int txqs_n = dev->data->nb_tx_queues;
+	unsigned int tmp;
+	int ret;
+
+	priv->rxqs = (void *)dev->data->rx_queues;
+	priv->txqs = (void *)dev->data->tx_queues;
+	if (txqs_n != priv->txqs_n) {
+		INFO("%p: TX queues number update: %u -> %u",
+		     (void *)dev, priv->txqs_n, txqs_n);
+		priv->txqs_n = txqs_n;
+	}
+	if (rxqs_n == priv->rxqs_n)
+		return 0;
+	INFO("%p: RX queues number update: %u -> %u",
+	     (void *)dev, priv->rxqs_n, rxqs_n);
+	/* If RSS is enabled, disable it first. */
+	if (priv->rss) {
+		unsigned int i;
+
+		/* Only if there are no remaining child RX queues. */
+		for (i = 0; (i != priv->rxqs_n); ++i)
+			if ((*priv->rxqs)[i] != NULL)
+				return EINVAL;
+		rxq_cleanup(&priv->rxq_parent);
+		priv->rss = 0;
+		priv->rxqs_n = 0;
+	}
+	if (rxqs_n <= 1) {
+		/* Nothing else to do. */
+		priv->rxqs_n = rxqs_n;
+		return 0;
+	}
+	/* Allocate a new RSS parent queue if supported by hardware. */
+	if (!priv->hw_rss) {
+		ERROR("%p: only a single RX queue can be configured when"
+		      " hardware doesn't support RSS",
+		      (void *)dev);
+		return EINVAL;
+	}
+	/* Fail if hardware doesn't support that many RSS queues. */
+	if (rxqs_n >= priv->max_rss_tbl_sz) {
+		ERROR("%p: only %u RX queues can be configured for RSS",
+		      (void *)dev, priv->max_rss_tbl_sz);
+		return EINVAL;
+	}
+	priv->rss = 1;
+	tmp = priv->rxqs_n;
+	priv->rxqs_n = rxqs_n;
+	ret = rxq_setup(dev, &priv->rxq_parent, 0, 0, NULL, NULL);
+	if (!ret)
+		return 0;
+	/* Failure, rollback. */
+	priv->rss = 0;
+	priv->rxqs_n = tmp;
+	assert(ret > 0);
+	return ret;
+}
+
+/**
+ * DPDK callback for Ethernet device configuration.
+ *
+ * @param dev
+ *   Pointer to Ethernet device structure.
+ *
+ * @return
+ *   0 on success, negative errno value on failure.
+ */
+int
+mlx5_dev_configure(struct rte_eth_dev *dev)
+{
+	struct priv *priv = dev->data->dev_private;
+	int ret;
+
+	priv_lock(priv);
+	ret = dev_configure(dev);
+	assert(ret >= 0);
+	priv_unlock(priv);
+	return -ret;
+}
+
+/**
+ * DPDK callback to get information about the device.
+ *
+ * @param dev
+ *   Pointer to Ethernet device structure.
+ * @param[out] info
+ *   Info structure output buffer.
+ */
+void
+mlx5_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *info)
+{
+	struct priv *priv = dev->data->dev_private;
+	unsigned int max;
+	char ifname[IF_NAMESIZE];
+
+	priv_lock(priv);
+	/* FIXME: we should ask the device for these values. */
+	info->min_rx_bufsize = 32;
+	info->max_rx_pktlen = 65536;
+	/*
+	 * Since we need one CQ per QP, the limit is the minimum number
+	 * between the two values.
+	 */
+	max = ((priv->device_attr.max_cq > priv->device_attr.max_qp) ?
+	       priv->device_attr.max_qp : priv->device_attr.max_cq);
+	/* If max >= 65535 then max = 0, max_rx_queues is uint16_t. */
+	if (max >= 65535)
+		max = 65535;
+	info->max_rx_queues = max;
+	info->max_tx_queues = max;
+	/* Last array entry is reserved for broadcast. */
+	info->max_mac_addrs = (RTE_DIM(priv->mac) - 1);
+	info->rx_offload_capa =
+		(priv->hw_csum ?
+		 (DEV_RX_OFFLOAD_IPV4_CKSUM |
+		  DEV_RX_OFFLOAD_UDP_CKSUM |
+		  DEV_RX_OFFLOAD_TCP_CKSUM) :
+		 0);
+	info->tx_offload_capa =
+		(priv->hw_csum ?
+		 (DEV_TX_OFFLOAD_IPV4_CKSUM |
+		  DEV_TX_OFFLOAD_UDP_CKSUM |
+		  DEV_TX_OFFLOAD_TCP_CKSUM) :
+		 0);
+	if (priv_get_ifname(priv, &ifname) == 0)
+		info->if_index = if_nametoindex(ifname);
+	priv_unlock(priv);
+}
+
+/**
  * Get PCI information from struct ibv_device.
  *
  * @param device
diff --git a/drivers/net/mlx5/mlx5_trigger.c b/drivers/net/mlx5/mlx5_trigger.c
new file mode 100644
index 0000000..bcec957
--- /dev/null
+++ b/drivers/net/mlx5/mlx5_trigger.c
@@ -0,0 +1,145 @@
+/*-
+ *   BSD LICENSE
+ *
+ *   Copyright 2015 6WIND S.A.
+ *   Copyright 2015 Mellanox.
+ *
+ *   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 6WIND S.A. 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.
+ */
+
+/* DPDK headers don't like -pedantic. */
+#ifdef PEDANTIC
+#pragma GCC diagnostic ignored "-pedantic"
+#endif
+#include <rte_ether.h>
+#include <rte_ethdev.h>
+#ifdef PEDANTIC
+#pragma GCC diagnostic error "-pedantic"
+#endif
+
+#include "mlx5.h"
+#include "mlx5_rxtx.h"
+#include "mlx5_utils.h"
+
+/**
+ * DPDK callback to start the device.
+ *
+ * Simulate device start by attaching all configured flows.
+ *
+ * @param dev
+ *   Pointer to Ethernet device structure.
+ *
+ * @return
+ *   0 on success, negative errno value on failure.
+ */
+int
+mlx5_dev_start(struct rte_eth_dev *dev)
+{
+	struct priv *priv = dev->data->dev_private;
+	unsigned int i = 0;
+	unsigned int r;
+	struct rxq *rxq;
+
+	priv_lock(priv);
+	if (priv->started) {
+		priv_unlock(priv);
+		return 0;
+	}
+	DEBUG("%p: attaching configured flows to all RX queues", (void *)dev);
+	priv->started = 1;
+	if (priv->rss) {
+		rxq = &priv->rxq_parent;
+		r = 1;
+	} else {
+		rxq = (*priv->rxqs)[0];
+		r = priv->rxqs_n;
+	}
+	/* Iterate only once when RSS is enabled. */
+	do {
+		int ret;
+
+		/* Ignore nonexistent RX queues. */
+		if (rxq == NULL)
+			continue;
+		ret = rxq_mac_addrs_add(rxq);
+		if (!ret)
+			continue;
+		WARN("%p: QP flow attachment failed: %s",
+		     (void *)dev, strerror(ret));
+		/* Rollback. */
+		while (i != 0) {
+			rxq = (*priv->rxqs)[--i];
+			if (rxq != NULL) {
+				rxq_mac_addrs_del(rxq);
+			}
+		}
+		priv->started = 0;
+		return -ret;
+	} while ((--r) && ((rxq = (*priv->rxqs)[++i]), i));
+	priv_unlock(priv);
+	return 0;
+}
+
+/**
+ * DPDK callback to stop the device.
+ *
+ * Simulate device stop by detaching all configured flows.
+ *
+ * @param dev
+ *   Pointer to Ethernet device structure.
+ */
+void
+mlx5_dev_stop(struct rte_eth_dev *dev)
+{
+	struct priv *priv = dev->data->dev_private;
+	unsigned int i = 0;
+	unsigned int r;
+	struct rxq *rxq;
+
+	priv_lock(priv);
+	if (!priv->started) {
+		priv_unlock(priv);
+		return;
+	}
+	DEBUG("%p: detaching flows from all RX queues", (void *)dev);
+	priv->started = 0;
+	if (priv->rss) {
+		rxq = &priv->rxq_parent;
+		r = 1;
+	} else {
+		rxq = (*priv->rxqs)[0];
+		r = priv->rxqs_n;
+	}
+	/* Iterate only once when RSS is enabled. */
+	do {
+		/* Ignore nonexistent RX queues. */
+		if (rxq == NULL)
+			continue;
+		rxq_mac_addrs_del(rxq);
+	} while ((--r) && ((rxq = (*priv->rxqs)[++i]), i));
+	priv_unlock(priv);
+}
-- 
2.1.0

^ permalink raw reply	[flat|nested] 32+ messages in thread

* [dpdk-dev] [PATCH 05/13] mlx5: add support for scattered RX and TX buffers
  2015-10-05 17:52 [dpdk-dev] [PATCH 00/13] Mellanox ConnectX-4 PMD (mlx5) Adrien Mazarguil
                   ` (3 preceding siblings ...)
  2015-10-05 17:53 ` [dpdk-dev] [PATCH 04/13] mlx5: add device configure/start/stop Adrien Mazarguil
@ 2015-10-05 17:53 ` Adrien Mazarguil
  2015-10-05 17:53 ` [dpdk-dev] [PATCH 06/13] mlx5: add MTU configuration support Adrien Mazarguil
                   ` (8 subsequent siblings)
  13 siblings, 0 replies; 32+ messages in thread
From: Adrien Mazarguil @ 2015-10-05 17:53 UTC (permalink / raw)
  To: dev

A dedicated RX callback is added to handle scattered buffers. For better
performance, it is only used when jumbo frames are enabled and MTU is larger
than a single mbuf.

On the TX path, scattered buffers are also handled in a separate function.
When there are more than MLX5_PMD_SGE_WR_N segments in a given mbuf, the
remaining segments are linearized in the last SGE entry.

Signed-off-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>
Signed-off-by: Nelio Laranjeiro <nelio.laranjeiro@6wind.com>
---
 drivers/net/mlx5/mlx5_rxq.c  | 175 +++++++++++++++++++-
 drivers/net/mlx5/mlx5_rxtx.c | 376 +++++++++++++++++++++++++++++++++++++++++++
 drivers/net/mlx5/mlx5_rxtx.h |  10 ++
 3 files changed, 557 insertions(+), 4 deletions(-)

diff --git a/drivers/net/mlx5/mlx5_rxq.c b/drivers/net/mlx5/mlx5_rxq.c
index 8450fe3..1eddfc7 100644
--- a/drivers/net/mlx5/mlx5_rxq.c
+++ b/drivers/net/mlx5/mlx5_rxq.c
@@ -65,6 +65,153 @@
 #include "mlx5_defs.h"
 
 /**
+ * Allocate RX queue elements with scattered packets support.
+ *
+ * @param rxq
+ *   Pointer to RX queue structure.
+ * @param elts_n
+ *   Number of elements to allocate.
+ * @param[in] pool
+ *   If not NULL, fetch buffers from this array instead of allocating them
+ *   with rte_pktmbuf_alloc().
+ *
+ * @return
+ *   0 on success, errno value on failure.
+ */
+static int
+rxq_alloc_elts_sp(struct rxq *rxq, unsigned int elts_n,
+		  struct rte_mbuf **pool)
+{
+	unsigned int i;
+	struct rxq_elt_sp (*elts)[elts_n] =
+		rte_calloc_socket("RXQ elements", 1, sizeof(*elts), 0,
+				  rxq->socket);
+	int ret = 0;
+
+	if (elts == NULL) {
+		ERROR("%p: can't allocate packets array", (void *)rxq);
+		ret = ENOMEM;
+		goto error;
+	}
+	/* For each WR (packet). */
+	for (i = 0; (i != elts_n); ++i) {
+		unsigned int j;
+		struct rxq_elt_sp *elt = &(*elts)[i];
+		struct ibv_recv_wr *wr = &elt->wr;
+		struct ibv_sge (*sges)[RTE_DIM(elt->sges)] = &elt->sges;
+
+		/* These two arrays must have the same size. */
+		assert(RTE_DIM(elt->sges) == RTE_DIM(elt->bufs));
+		/* Configure WR. */
+		wr->wr_id = i;
+		wr->next = &(*elts)[(i + 1)].wr;
+		wr->sg_list = &(*sges)[0];
+		wr->num_sge = RTE_DIM(*sges);
+		/* For each SGE (segment). */
+		for (j = 0; (j != RTE_DIM(elt->bufs)); ++j) {
+			struct ibv_sge *sge = &(*sges)[j];
+			struct rte_mbuf *buf;
+
+			if (pool != NULL) {
+				buf = *(pool++);
+				assert(buf != NULL);
+				rte_pktmbuf_reset(buf);
+			} else
+				buf = rte_pktmbuf_alloc(rxq->mp);
+			if (buf == NULL) {
+				assert(pool == NULL);
+				ERROR("%p: empty mbuf pool", (void *)rxq);
+				ret = ENOMEM;
+				goto error;
+			}
+			elt->bufs[j] = buf;
+			/* Headroom is reserved by rte_pktmbuf_alloc(). */
+			assert(DATA_OFF(buf) == RTE_PKTMBUF_HEADROOM);
+			/* Buffer is supposed to be empty. */
+			assert(rte_pktmbuf_data_len(buf) == 0);
+			assert(rte_pktmbuf_pkt_len(buf) == 0);
+			/* sge->addr must be able to store a pointer. */
+			assert(sizeof(sge->addr) >= sizeof(uintptr_t));
+			if (j == 0) {
+				/* The first SGE keeps its headroom. */
+				sge->addr = rte_pktmbuf_mtod(buf, uintptr_t);
+				sge->length = (buf->buf_len -
+					       RTE_PKTMBUF_HEADROOM);
+			} else {
+				/* Subsequent SGEs lose theirs. */
+				assert(DATA_OFF(buf) == RTE_PKTMBUF_HEADROOM);
+				SET_DATA_OFF(buf, 0);
+				sge->addr = (uintptr_t)buf->buf_addr;
+				sge->length = buf->buf_len;
+			}
+			sge->lkey = rxq->mr->lkey;
+			/* Redundant check for tailroom. */
+			assert(sge->length == rte_pktmbuf_tailroom(buf));
+		}
+	}
+	/* The last WR pointer must be NULL. */
+	(*elts)[(i - 1)].wr.next = NULL;
+	DEBUG("%p: allocated and configured %u WRs (%zu segments)",
+	      (void *)rxq, elts_n, (elts_n * RTE_DIM((*elts)[0].sges)));
+	rxq->elts_n = elts_n;
+	rxq->elts_head = 0;
+	rxq->elts.sp = elts;
+	assert(ret == 0);
+	return 0;
+error:
+	if (elts != NULL) {
+		assert(pool == NULL);
+		for (i = 0; (i != RTE_DIM(*elts)); ++i) {
+			unsigned int j;
+			struct rxq_elt_sp *elt = &(*elts)[i];
+
+			for (j = 0; (j != RTE_DIM(elt->bufs)); ++j) {
+				struct rte_mbuf *buf = elt->bufs[j];
+
+				if (buf != NULL)
+					rte_pktmbuf_free_seg(buf);
+			}
+		}
+		rte_free(elts);
+	}
+	DEBUG("%p: failed, freed everything", (void *)rxq);
+	assert(ret > 0);
+	return ret;
+}
+
+/**
+ * Free RX queue elements with scattered packets support.
+ *
+ * @param rxq
+ *   Pointer to RX queue structure.
+ */
+static void
+rxq_free_elts_sp(struct rxq *rxq)
+{
+	unsigned int i;
+	unsigned int elts_n = rxq->elts_n;
+	struct rxq_elt_sp (*elts)[elts_n] = rxq->elts.sp;
+
+	DEBUG("%p: freeing WRs", (void *)rxq);
+	rxq->elts_n = 0;
+	rxq->elts.sp = NULL;
+	if (elts == NULL)
+		return;
+	for (i = 0; (i != RTE_DIM(*elts)); ++i) {
+		unsigned int j;
+		struct rxq_elt_sp *elt = &(*elts)[i];
+
+		for (j = 0; (j != RTE_DIM(elt->bufs)); ++j) {
+			struct rte_mbuf *buf = elt->bufs[j];
+
+			if (buf != NULL)
+				rte_pktmbuf_free_seg(buf);
+		}
+	}
+	rte_free(elts);
+}
+
+/**
  * Allocate RX queue elements.
  *
  * @param rxq
@@ -224,7 +371,10 @@ rxq_cleanup(struct rxq *rxq)
 	struct ibv_exp_release_intf_params params;
 
 	DEBUG("cleaning up %p", (void *)rxq);
-	rxq_free_elts(rxq);
+	if (rxq->sp)
+		rxq_free_elts_sp(rxq);
+	else
+		rxq_free_elts(rxq);
 	if (rxq->if_qp != NULL) {
 		assert(rxq->priv != NULL);
 		assert(rxq->priv->ctx != NULL);
@@ -445,6 +595,15 @@ rxq_setup(struct rte_eth_dev *dev, struct rxq *rxq, uint16_t desc,
 		rte_pktmbuf_tailroom(buf)) == tmpl.mb_len);
 	assert(rte_pktmbuf_headroom(buf) == RTE_PKTMBUF_HEADROOM);
 	rte_pktmbuf_free(buf);
+	/* Enable scattered packets support for this queue if necessary. */
+	if ((dev->data->dev_conf.rxmode.jumbo_frame) &&
+	    (dev->data->dev_conf.rxmode.max_rx_pkt_len >
+	     (tmpl.mb_len - RTE_PKTMBUF_HEADROOM))) {
+		tmpl.sp = 1;
+		desc /= MLX5_PMD_SGE_WR_N;
+	}
+	DEBUG("%p: %s scattered packets support (%u WRs)",
+	      (void *)dev, (tmpl.sp ? "enabling" : "disabling"), desc);
 	/* Use the entire RX mempool as the memory region. */
 	tmpl.mr = ibv_reg_mr(priv->pd,
 			     (void *)mp->elt_va_start,
@@ -528,14 +687,19 @@ skip_mr:
 	/* Allocate descriptors for RX queues, except for the RSS parent. */
 	if (parent)
 		goto skip_alloc;
-	ret = rxq_alloc_elts(&tmpl, desc, NULL);
+	if (tmpl.sp)
+		ret = rxq_alloc_elts_sp(&tmpl, desc, NULL);
+	else
+		ret = rxq_alloc_elts(&tmpl, desc, NULL);
 	if (ret) {
 		ERROR("%p: RXQ allocation failed: %s",
 		      (void *)dev, strerror(ret));
 		goto error;
 	}
 	ret = ibv_post_recv(tmpl.qp,
-			    &(*tmpl.elts.no_sp)[0].wr,
+			    (tmpl.sp ?
+			     &(*tmpl.elts.sp)[0].wr :
+			     &(*tmpl.elts.no_sp)[0].wr),
 			    &bad_wr);
 	if (ret) {
 		ERROR("%p: ibv_post_recv() failed for WR %p: %s",
@@ -655,7 +819,10 @@ mlx5_rx_queue_setup(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc,
 		      (void *)dev, (void *)rxq);
 		(*priv->rxqs)[idx] = rxq;
 		/* Update receive callback. */
-		dev->rx_pkt_burst = mlx5_rx_burst;
+		if (rxq->sp)
+			dev->rx_pkt_burst = mlx5_rx_burst_sp;
+		else
+			dev->rx_pkt_burst = mlx5_rx_burst;
 	}
 	priv_unlock(priv);
 	return -ret;
diff --git a/drivers/net/mlx5/mlx5_rxtx.c b/drivers/net/mlx5/mlx5_rxtx.c
index 40bddf0..5042011 100644
--- a/drivers/net/mlx5/mlx5_rxtx.c
+++ b/drivers/net/mlx5/mlx5_rxtx.c
@@ -173,6 +173,154 @@ txq_mp2mr(struct txq *txq, struct rte_mempool *mp)
 	return txq->mp2mr[i].lkey;
 }
 
+#if MLX5_PMD_SGE_WR_N > 1
+
+/**
+ * Copy scattered mbuf contents to a single linear buffer.
+ *
+ * @param[out] linear
+ *   Linear output buffer.
+ * @param[in] buf
+ *   Scattered input buffer.
+ *
+ * @return
+ *   Number of bytes copied to the output buffer or 0 if not large enough.
+ */
+static unsigned int
+linearize_mbuf(linear_t *linear, struct rte_mbuf *buf)
+{
+	unsigned int size = 0;
+	unsigned int offset;
+
+	do {
+		unsigned int len = DATA_LEN(buf);
+
+		offset = size;
+		size += len;
+		if (unlikely(size > sizeof(*linear)))
+			return 0;
+		memcpy(&(*linear)[offset],
+		       rte_pktmbuf_mtod(buf, uint8_t *),
+		       len);
+		buf = NEXT(buf);
+	} while (buf != NULL);
+	return size;
+}
+
+/**
+ * Handle scattered buffers for mlx5_tx_burst().
+ *
+ * @param txq
+ *   TX queue structure.
+ * @param segs
+ *   Number of segments in buf.
+ * @param elt
+ *   TX queue element to fill.
+ * @param[in] buf
+ *   Buffer to process.
+ * @param elts_head
+ *   Index of the linear buffer to use if necessary (normally txq->elts_head).
+ * @param[out] sges
+ *   Array filled with SGEs on success.
+ *
+ * @return
+ *   A structure containing the processed packet size in bytes and the
+ *   number of SGEs. Both fields are set to (unsigned int)-1 in case of
+ *   failure.
+ */
+static struct tx_burst_sg_ret {
+	unsigned int length;
+	unsigned int num;
+}
+tx_burst_sg(struct txq *txq, unsigned int segs, struct txq_elt *elt,
+	    struct rte_mbuf *buf, unsigned int elts_head,
+	    struct ibv_sge (*sges)[MLX5_PMD_SGE_WR_N])
+{
+	unsigned int sent_size = 0;
+	unsigned int j;
+	int linearize = 0;
+
+	/* When there are too many segments, extra segments are
+	 * linearized in the last SGE. */
+	if (unlikely(segs > RTE_DIM(*sges))) {
+		segs = (RTE_DIM(*sges) - 1);
+		linearize = 1;
+	}
+	/* Update element. */
+	elt->buf = buf;
+	/* Register segments as SGEs. */
+	for (j = 0; (j != segs); ++j) {
+		struct ibv_sge *sge = &(*sges)[j];
+		uint32_t lkey;
+
+		/* Retrieve Memory Region key for this memory pool. */
+		lkey = txq_mp2mr(txq, buf->pool);
+		if (unlikely(lkey == (uint32_t)-1)) {
+			/* MR does not exist. */
+			DEBUG("%p: unable to get MP <-> MR association",
+			      (void *)txq);
+			/* Clean up TX element. */
+			elt->buf = NULL;
+			goto stop;
+		}
+		/* Update SGE. */
+		sge->addr = rte_pktmbuf_mtod(buf, uintptr_t);
+		if (txq->priv->vf)
+			rte_prefetch0((volatile void *)
+				      (uintptr_t)sge->addr);
+		sge->length = DATA_LEN(buf);
+		sge->lkey = lkey;
+		sent_size += sge->length;
+		buf = NEXT(buf);
+	}
+	/* If buf is not NULL here and is not going to be linearized,
+	 * nb_segs is not valid. */
+	assert(j == segs);
+	assert((buf == NULL) || (linearize));
+	/* Linearize extra segments. */
+	if (linearize) {
+		struct ibv_sge *sge = &(*sges)[segs];
+		linear_t *linear = &(*txq->elts_linear)[elts_head];
+		unsigned int size = linearize_mbuf(linear, buf);
+
+		assert(segs == (RTE_DIM(*sges) - 1));
+		if (size == 0) {
+			/* Invalid packet. */
+			DEBUG("%p: packet too large to be linearized.",
+			      (void *)txq);
+			/* Clean up TX element. */
+			elt->buf = NULL;
+			goto stop;
+		}
+		/* If MLX5_PMD_SGE_WR_N is 1, free mbuf immediately. */
+		if (RTE_DIM(*sges) == 1) {
+			do {
+				struct rte_mbuf *next = NEXT(buf);
+
+				rte_pktmbuf_free_seg(buf);
+				buf = next;
+			} while (buf != NULL);
+			elt->buf = NULL;
+		}
+		/* Update SGE. */
+		sge->addr = (uintptr_t)&(*linear)[0];
+		sge->length = size;
+		sge->lkey = txq->mr_linear->lkey;
+		sent_size += size;
+	}
+	return (struct tx_burst_sg_ret){
+		.length = sent_size,
+		.num = segs,
+	};
+stop:
+	return (struct tx_burst_sg_ret){
+		.length = -1,
+		.num = -1,
+	};
+}
+
+#endif /* MLX5_PMD_SGE_WR_N > 1 */
+
 /**
  * DPDK callback for TX.
  *
@@ -282,9 +430,28 @@ mlx5_tx_burst(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n)
 			if (unlikely(err))
 				goto stop;
 		} else {
+#if MLX5_PMD_SGE_WR_N > 1
+			struct ibv_sge sges[MLX5_PMD_SGE_WR_N];
+			struct tx_burst_sg_ret ret;
+
+			ret = tx_burst_sg(txq, segs, elt, buf, elts_head,
+					  &sges);
+			if (ret.length == (unsigned int)-1)
+				goto stop;
+			RTE_MBUF_PREFETCH_TO_FREE(elt_next->buf);
+			/* Put SG list into send queue. */
+			err = txq->if_qp->send_pending_sg_list
+				(txq->qp,
+				 sges,
+				 ret.num,
+				 send_flags);
+			if (unlikely(err))
+				goto stop;
+#else /* MLX5_PMD_SGE_WR_N > 1 */
 			DEBUG("%p: TX scattered buffers support not"
 			      " compiled in", (void *)txq);
 			goto stop;
+#endif /* MLX5_PMD_SGE_WR_N > 1 */
 		}
 		elts_head = elts_head_next;
 	}
@@ -307,8 +474,215 @@ stop:
 }
 
 /**
+ * DPDK callback for RX with scattered packets support.
+ *
+ * @param dpdk_rxq
+ *   Generic pointer to RX queue structure.
+ * @param[out] pkts
+ *   Array to store received packets.
+ * @param pkts_n
+ *   Maximum number of packets in array.
+ *
+ * @return
+ *   Number of packets successfully received (<= pkts_n).
+ */
+uint16_t
+mlx5_rx_burst_sp(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n)
+{
+	struct rxq *rxq = (struct rxq *)dpdk_rxq;
+	struct rxq_elt_sp (*elts)[rxq->elts_n] = rxq->elts.sp;
+	const unsigned int elts_n = rxq->elts_n;
+	unsigned int elts_head = rxq->elts_head;
+	struct ibv_recv_wr head;
+	struct ibv_recv_wr **next = &head.next;
+	struct ibv_recv_wr *bad_wr;
+	unsigned int i;
+	unsigned int pkts_ret = 0;
+	int ret;
+
+	if (unlikely(!rxq->sp))
+		return mlx5_rx_burst(dpdk_rxq, pkts, pkts_n);
+	if (unlikely(elts == NULL)) /* See RTE_DEV_CMD_SET_MTU. */
+		return 0;
+	for (i = 0; (i != pkts_n); ++i) {
+		struct rxq_elt_sp *elt = &(*elts)[elts_head];
+		struct ibv_recv_wr *wr = &elt->wr;
+		uint64_t wr_id = wr->wr_id;
+		unsigned int len;
+		unsigned int pkt_buf_len;
+		struct rte_mbuf *pkt_buf = NULL; /* Buffer returned in pkts. */
+		struct rte_mbuf **pkt_buf_next = &pkt_buf;
+		unsigned int seg_headroom = RTE_PKTMBUF_HEADROOM;
+		unsigned int j = 0;
+		uint32_t flags;
+
+		/* Sanity checks. */
+#ifdef NDEBUG
+		(void)wr_id;
+#endif
+		assert(wr_id < rxq->elts_n);
+		assert(wr->sg_list == elt->sges);
+		assert(wr->num_sge == RTE_DIM(elt->sges));
+		assert(elts_head < rxq->elts_n);
+		assert(rxq->elts_head < rxq->elts_n);
+		ret = rxq->if_cq->poll_length_flags(rxq->cq, NULL, NULL,
+						    &flags);
+		if (unlikely(ret < 0)) {
+			struct ibv_wc wc;
+			int wcs_n;
+
+			DEBUG("rxq=%p, poll_length() failed (ret=%d)",
+			      (void *)rxq, ret);
+			/* ibv_poll_cq() must be used in case of failure. */
+			wcs_n = ibv_poll_cq(rxq->cq, 1, &wc);
+			if (unlikely(wcs_n == 0))
+				break;
+			if (unlikely(wcs_n < 0)) {
+				DEBUG("rxq=%p, ibv_poll_cq() failed (wcs_n=%d)",
+				      (void *)rxq, wcs_n);
+				break;
+			}
+			assert(wcs_n == 1);
+			if (unlikely(wc.status != IBV_WC_SUCCESS)) {
+				/* Whatever, just repost the offending WR. */
+				DEBUG("rxq=%p, wr_id=%" PRIu64 ": bad work"
+				      " completion status (%d): %s",
+				      (void *)rxq, wc.wr_id, wc.status,
+				      ibv_wc_status_str(wc.status));
+				/* Link completed WRs together for repost. */
+				*next = wr;
+				next = &wr->next;
+				goto repost;
+			}
+			ret = wc.byte_len;
+		}
+		if (ret == 0)
+			break;
+		len = ret;
+		pkt_buf_len = len;
+		/* Link completed WRs together for repost. */
+		*next = wr;
+		next = &wr->next;
+		/*
+		 * Replace spent segments with new ones, concatenate and
+		 * return them as pkt_buf.
+		 */
+		while (1) {
+			struct ibv_sge *sge = &elt->sges[j];
+			struct rte_mbuf *seg = elt->bufs[j];
+			struct rte_mbuf *rep;
+			unsigned int seg_tailroom;
+
+			/*
+			 * Fetch initial bytes of packet descriptor into a
+			 * cacheline while allocating rep.
+			 */
+			rte_prefetch0(seg);
+			rep = __rte_mbuf_raw_alloc(rxq->mp);
+			if (unlikely(rep == NULL)) {
+				/*
+				 * Unable to allocate a replacement mbuf,
+				 * repost WR.
+				 */
+				DEBUG("rxq=%p, wr_id=%" PRIu64 ":"
+				      " can't allocate a new mbuf",
+				      (void *)rxq, wr_id);
+				if (pkt_buf != NULL) {
+					*pkt_buf_next = NULL;
+					rte_pktmbuf_free(pkt_buf);
+				}
+				/* Increment out of memory counters. */
+				++rxq->priv->dev->data->rx_mbuf_alloc_failed;
+				goto repost;
+			}
+#ifndef NDEBUG
+			/* Poison user-modifiable fields in rep. */
+			NEXT(rep) = (void *)((uintptr_t)-1);
+			SET_DATA_OFF(rep, 0xdead);
+			DATA_LEN(rep) = 0xd00d;
+			PKT_LEN(rep) = 0xdeadd00d;
+			NB_SEGS(rep) = 0x2a;
+			PORT(rep) = 0x2a;
+			rep->ol_flags = -1;
+#endif
+			assert(rep->buf_len == seg->buf_len);
+			assert(rep->buf_len == rxq->mb_len);
+			/* Reconfigure sge to use rep instead of seg. */
+			assert(sge->lkey == rxq->mr->lkey);
+			sge->addr = ((uintptr_t)rep->buf_addr + seg_headroom);
+			elt->bufs[j] = rep;
+			++j;
+			/* Update pkt_buf if it's the first segment, or link
+			 * seg to the previous one and update pkt_buf_next. */
+			*pkt_buf_next = seg;
+			pkt_buf_next = &NEXT(seg);
+			/* Update seg information. */
+			seg_tailroom = (seg->buf_len - seg_headroom);
+			assert(sge->length == seg_tailroom);
+			SET_DATA_OFF(seg, seg_headroom);
+			if (likely(len <= seg_tailroom)) {
+				/* Last segment. */
+				DATA_LEN(seg) = len;
+				PKT_LEN(seg) = len;
+				/* Sanity check. */
+				assert(rte_pktmbuf_headroom(seg) ==
+				       seg_headroom);
+				assert(rte_pktmbuf_tailroom(seg) ==
+				       (seg_tailroom - len));
+				break;
+			}
+			DATA_LEN(seg) = seg_tailroom;
+			PKT_LEN(seg) = seg_tailroom;
+			/* Sanity check. */
+			assert(rte_pktmbuf_headroom(seg) == seg_headroom);
+			assert(rte_pktmbuf_tailroom(seg) == 0);
+			/* Fix len and clear headroom for next segments. */
+			len -= seg_tailroom;
+			seg_headroom = 0;
+		}
+		/* Update head and tail segments. */
+		*pkt_buf_next = NULL;
+		assert(pkt_buf != NULL);
+		assert(j != 0);
+		NB_SEGS(pkt_buf) = j;
+		PORT(pkt_buf) = rxq->port_id;
+		PKT_LEN(pkt_buf) = pkt_buf_len;
+
+		/* Return packet. */
+		*(pkts++) = pkt_buf;
+		++pkts_ret;
+repost:
+		if (++elts_head >= elts_n)
+			elts_head = 0;
+		continue;
+	}
+	if (unlikely(i == 0))
+		return 0;
+	*next = NULL;
+	/* Repost WRs. */
+#ifdef DEBUG_RECV
+	DEBUG("%p: reposting %d WRs", (void *)rxq, i);
+#endif
+	ret = ibv_post_recv(rxq->qp, head.next, &bad_wr);
+	if (unlikely(ret)) {
+		/* Inability to repost WRs is fatal. */
+		DEBUG("%p: ibv_post_recv(): failed for WR %p: %s",
+		      (void *)rxq->priv,
+		      (void *)bad_wr,
+		      strerror(ret));
+		abort();
+	}
+	rxq->elts_head = elts_head;
+	return pkts_ret;
+}
+
+/**
  * DPDK callback for RX.
  *
+ * The following function is the same as mlx5_rx_burst_sp(), except it doesn't
+ * manage scattered packets. Improves performance when MRU is lower than the
+ * size of the first segment.
+ *
  * @param dpdk_rxq
  *   Generic pointer to RX queue structure.
  * @param[out] pkts
@@ -331,6 +705,8 @@ mlx5_rx_burst(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n)
 	unsigned int pkts_ret = 0;
 	int ret;
 
+	if (unlikely(rxq->sp))
+		return mlx5_rx_burst_sp(dpdk_rxq, pkts, pkts_n);
 	for (i = 0; (i != pkts_n); ++i) {
 		struct rxq_elt *elt = &(*elts)[elts_head];
 		struct ibv_recv_wr *wr = &elt->wr;
diff --git a/drivers/net/mlx5/mlx5_rxtx.h b/drivers/net/mlx5/mlx5_rxtx.h
index 08a4911..44170d7 100644
--- a/drivers/net/mlx5/mlx5_rxtx.h
+++ b/drivers/net/mlx5/mlx5_rxtx.h
@@ -60,6 +60,13 @@
 #include "mlx5.h"
 #include "mlx5_defs.h"
 
+/* RX element (scattered packets). */
+struct rxq_elt_sp {
+	struct ibv_recv_wr wr; /* Work Request. */
+	struct ibv_sge sges[MLX5_PMD_SGE_WR_N]; /* Scatter/Gather Elements. */
+	struct rte_mbuf *bufs[MLX5_PMD_SGE_WR_N]; /* SGEs buffers. */
+};
+
 /* RX element. */
 struct rxq_elt {
 	struct ibv_recv_wr wr; /* Work Request. */
@@ -87,8 +94,10 @@ struct rxq {
 	unsigned int elts_n; /* (*elts)[] length. */
 	unsigned int elts_head; /* Current index in (*elts)[]. */
 	union {
+		struct rxq_elt_sp (*sp)[]; /* Scattered RX elements. */
 		struct rxq_elt (*no_sp)[]; /* RX elements. */
 	} elts;
+	unsigned int sp:1; /* Use scattered RX elements. */
 	uint32_t mb_len; /* Length of a mp-issued mbuf. */
 	unsigned int socket; /* CPU socket ID for allocations. */
 	struct ibv_exp_res_domain *rd; /* Resource Domain. */
@@ -154,6 +163,7 @@ void mlx5_tx_queue_release(void *);
 /* mlx5_rxtx.c */
 
 uint16_t mlx5_tx_burst(void *, struct rte_mbuf **, uint16_t);
+uint16_t mlx5_rx_burst_sp(void *, struct rte_mbuf **, uint16_t);
 uint16_t mlx5_rx_burst(void *, struct rte_mbuf **, uint16_t);
 uint16_t removed_tx_burst(void *, struct rte_mbuf **, uint16_t);
 uint16_t removed_rx_burst(void *, struct rte_mbuf **, uint16_t);
-- 
2.1.0

^ permalink raw reply	[flat|nested] 32+ messages in thread

* [dpdk-dev] [PATCH 06/13] mlx5: add MTU configuration support
  2015-10-05 17:52 [dpdk-dev] [PATCH 00/13] Mellanox ConnectX-4 PMD (mlx5) Adrien Mazarguil
                   ` (4 preceding siblings ...)
  2015-10-05 17:53 ` [dpdk-dev] [PATCH 05/13] mlx5: add support for scattered RX and TX buffers Adrien Mazarguil
@ 2015-10-05 17:53 ` Adrien Mazarguil
  2015-10-05 17:53 ` [dpdk-dev] [PATCH 07/13] mlx5: add software counters and related callbacks Adrien Mazarguil
                   ` (7 subsequent siblings)
  13 siblings, 0 replies; 32+ messages in thread
From: Adrien Mazarguil @ 2015-10-05 17:53 UTC (permalink / raw)
  To: dev

Depending on the MTU and whether jumbo frames are enabled, RX queues may
switch between SG and non-SG modes for better performance.

Signed-off-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>
Signed-off-by: Nelio Laranjeiro <nelio.laranjeiro@6wind.com>
---
 drivers/net/mlx5/mlx5.c        |   1 +
 drivers/net/mlx5/mlx5.h        |   1 +
 drivers/net/mlx5/mlx5_ethdev.c | 101 +++++++++++++++++++++++
 drivers/net/mlx5/mlx5_rxq.c    | 181 +++++++++++++++++++++++++++++++++++++++++
 drivers/net/mlx5/mlx5_rxtx.h   |   1 +
 5 files changed, 285 insertions(+)

diff --git a/drivers/net/mlx5/mlx5.c b/drivers/net/mlx5/mlx5.c
index 56cea7c..5bef742 100644
--- a/drivers/net/mlx5/mlx5.c
+++ b/drivers/net/mlx5/mlx5.c
@@ -140,6 +140,7 @@ static const struct eth_dev_ops mlx5_dev_ops = {
 	.tx_queue_release = mlx5_tx_queue_release,
 	.mac_addr_remove = mlx5_mac_addr_remove,
 	.mac_addr_add = mlx5_mac_addr_add,
+	.mtu_set = mlx5_dev_set_mtu,
 };
 
 static struct {
diff --git a/drivers/net/mlx5/mlx5.h b/drivers/net/mlx5/mlx5.h
index 3f4f8fd..14f55ba 100644
--- a/drivers/net/mlx5/mlx5.h
+++ b/drivers/net/mlx5/mlx5.h
@@ -169,6 +169,7 @@ int priv_get_mtu(struct priv *, uint16_t *);
 int priv_set_flags(struct priv *, unsigned int, unsigned int);
 int mlx5_dev_configure(struct rte_eth_dev *);
 void mlx5_dev_infos_get(struct rte_eth_dev *, struct rte_eth_dev_info *);
+int mlx5_dev_set_mtu(struct rte_eth_dev *, uint16_t);
 int mlx5_ibv_device_to_pci_addr(const struct ibv_device *,
 				struct rte_pci_addr *);
 
diff --git a/drivers/net/mlx5/mlx5_ethdev.c b/drivers/net/mlx5/mlx5_ethdev.c
index 6b13cec..4621282 100644
--- a/drivers/net/mlx5/mlx5_ethdev.c
+++ b/drivers/net/mlx5/mlx5_ethdev.c
@@ -347,6 +347,23 @@ priv_get_mtu(struct priv *priv, uint16_t *mtu)
 }
 
 /**
+ * Set device MTU.
+ *
+ * @param priv
+ *   Pointer to private structure.
+ * @param mtu
+ *   MTU value to set.
+ *
+ * @return
+ *   0 on success, -1 on failure and errno is set.
+ */
+static int
+priv_set_mtu(struct priv *priv, uint16_t mtu)
+{
+	return priv_set_sysfs_ulong(priv, "mtu", mtu);
+}
+
+/**
  * Set device flags.
  *
  * @param priv
@@ -518,6 +535,90 @@ mlx5_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *info)
 }
 
 /**
+ * DPDK callback to change the MTU.
+ *
+ * Setting the MTU affects hardware MRU (packets larger than the MTU cannot be
+ * received). Use this as a hint to enable/disable scattered packets support
+ * and improve performance when not needed.
+ * Since failure is not an option, reconfiguring queues on the fly is not
+ * recommended.
+ *
+ * @param dev
+ *   Pointer to Ethernet device structure.
+ * @param in_mtu
+ *   New MTU.
+ *
+ * @return
+ *   0 on success, negative errno value on failure.
+ */
+int
+mlx5_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
+{
+	struct priv *priv = dev->data->dev_private;
+	int ret = 0;
+	unsigned int i;
+	uint16_t (*rx_func)(void *, struct rte_mbuf **, uint16_t) =
+		mlx5_rx_burst;
+
+	priv_lock(priv);
+	/* Set kernel interface MTU first. */
+	if (priv_set_mtu(priv, mtu)) {
+		ret = errno;
+		WARN("cannot set port %u MTU to %u: %s", priv->port, mtu,
+		     strerror(ret));
+		goto out;
+	} else
+		DEBUG("adapter port %u MTU set to %u", priv->port, mtu);
+	priv->mtu = mtu;
+	/* Temporarily replace RX handler with a fake one, assuming it has not
+	 * been copied elsewhere. */
+	dev->rx_pkt_burst = removed_rx_burst;
+	/* Make sure everyone has left mlx5_rx_burst() and uses
+	 * removed_rx_burst() instead. */
+	rte_wmb();
+	usleep(1000);
+	/* Reconfigure each RX queue. */
+	for (i = 0; (i != priv->rxqs_n); ++i) {
+		struct rxq *rxq = (*priv->rxqs)[i];
+		unsigned int max_frame_len;
+		int sp;
+
+		if (rxq == NULL)
+			continue;
+		/* Calculate new maximum frame length according to MTU and
+		 * toggle scattered support (sp) if necessary. */
+		max_frame_len = (priv->mtu + ETHER_HDR_LEN +
+				 (ETHER_MAX_VLAN_FRAME_LEN - ETHER_MAX_LEN));
+		sp = (max_frame_len > (rxq->mb_len - RTE_PKTMBUF_HEADROOM));
+		/* Provide new values to rxq_setup(). */
+		dev->data->dev_conf.rxmode.jumbo_frame = sp;
+		dev->data->dev_conf.rxmode.max_rx_pkt_len = max_frame_len;
+		ret = rxq_rehash(dev, rxq);
+		if (ret) {
+			/* Force SP RX if that queue requires it and abort. */
+			if (rxq->sp)
+				rx_func = mlx5_rx_burst_sp;
+			break;
+		}
+		/* Reenable non-RSS queue attributes. No need to check
+		 * for errors at this stage. */
+		if (!priv->rss) {
+			rxq_mac_addrs_add(rxq);
+		}
+		/* Scattered burst function takes priority. */
+		if (rxq->sp)
+			rx_func = mlx5_rx_burst_sp;
+	}
+	/* Burst functions can now be called again. */
+	rte_wmb();
+	dev->rx_pkt_burst = rx_func;
+out:
+	priv_unlock(priv);
+	assert(ret >= 0);
+	return -ret;
+}
+
+/**
  * Get PCI information from struct ibv_device.
  *
  * @param device
diff --git a/drivers/net/mlx5/mlx5_rxq.c b/drivers/net/mlx5/mlx5_rxq.c
index 1eddfc7..401b575 100644
--- a/drivers/net/mlx5/mlx5_rxq.c
+++ b/drivers/net/mlx5/mlx5_rxq.c
@@ -526,6 +526,187 @@ rxq_setup_qp_rss(struct priv *priv, struct ibv_cq *cq, uint16_t desc,
 #endif /* RSS_SUPPORT */
 
 /**
+ * Reconfigure a RX queue with new parameters.
+ *
+ * rxq_rehash() does not allocate mbufs, which, if not done from the right
+ * thread (such as a control thread), may corrupt the pool.
+ * In case of failure, the queue is left untouched.
+ *
+ * @param dev
+ *   Pointer to Ethernet device structure.
+ * @param rxq
+ *   RX queue pointer.
+ *
+ * @return
+ *   0 on success, errno value on failure.
+ */
+int
+rxq_rehash(struct rte_eth_dev *dev, struct rxq *rxq)
+{
+	struct priv *priv = rxq->priv;
+	struct rxq tmpl = *rxq;
+	unsigned int mbuf_n;
+	unsigned int desc_n;
+	struct rte_mbuf **pool;
+	unsigned int i, k;
+	struct ibv_exp_qp_attr mod;
+	struct ibv_recv_wr *bad_wr;
+	int err;
+	int parent = (rxq == &priv->rxq_parent);
+
+	if (parent) {
+		ERROR("%p: cannot rehash parent queue %p",
+		      (void *)dev, (void *)rxq);
+		return EINVAL;
+	}
+	DEBUG("%p: rehashing queue %p", (void *)dev, (void *)rxq);
+	/* Number of descriptors and mbufs currently allocated. */
+	desc_n = (tmpl.elts_n * (tmpl.sp ? MLX5_PMD_SGE_WR_N : 1));
+	mbuf_n = desc_n;
+	/* Enable scattered packets support for this queue if necessary. */
+	if ((dev->data->dev_conf.rxmode.jumbo_frame) &&
+	    (dev->data->dev_conf.rxmode.max_rx_pkt_len >
+	     (tmpl.mb_len - RTE_PKTMBUF_HEADROOM))) {
+		tmpl.sp = 1;
+		desc_n /= MLX5_PMD_SGE_WR_N;
+	} else
+		tmpl.sp = 0;
+	DEBUG("%p: %s scattered packets support (%u WRs)",
+	      (void *)dev, (tmpl.sp ? "enabling" : "disabling"), desc_n);
+	/* If scatter mode is the same as before, nothing to do. */
+	if (tmpl.sp == rxq->sp) {
+		DEBUG("%p: nothing to do", (void *)dev);
+		return 0;
+	}
+	/* Remove attached flows if RSS is disabled (no parent queue). */
+	if (!priv->rss) {
+		rxq_mac_addrs_del(&tmpl);
+		/* Update original queue in case of failure. */
+		memcpy(rxq->mac_configured, tmpl.mac_configured,
+		       sizeof(rxq->mac_configured));
+		memcpy(rxq->mac_flow, tmpl.mac_flow, sizeof(rxq->mac_flow));
+	}
+	/* From now on, any failure will render the queue unusable.
+	 * Reinitialize QP. */
+	mod = (struct ibv_exp_qp_attr){ .qp_state = IBV_QPS_RESET };
+	err = ibv_exp_modify_qp(tmpl.qp, &mod, IBV_EXP_QP_STATE);
+	if (err) {
+		ERROR("%p: cannot reset QP: %s", (void *)dev, strerror(err));
+		assert(err > 0);
+		return err;
+	}
+	err = ibv_resize_cq(tmpl.cq, desc_n);
+	if (err) {
+		ERROR("%p: cannot resize CQ: %s", (void *)dev, strerror(err));
+		assert(err > 0);
+		return err;
+	}
+	mod = (struct ibv_exp_qp_attr){
+		/* Move the QP to this state. */
+		.qp_state = IBV_QPS_INIT,
+		/* Primary port number. */
+		.port_num = priv->port
+	};
+	err = ibv_exp_modify_qp(tmpl.qp, &mod,
+				(IBV_EXP_QP_STATE |
+#ifdef RSS_SUPPORT
+				 (parent ? IBV_EXP_QP_GROUP_RSS : 0) |
+#endif /* RSS_SUPPORT */
+				 IBV_EXP_QP_PORT));
+	if (err) {
+		ERROR("%p: QP state to IBV_QPS_INIT failed: %s",
+		      (void *)dev, strerror(err));
+		assert(err > 0);
+		return err;
+	};
+	/* Reconfigure flows. Do not care for errors. */
+	if (!priv->rss) {
+		rxq_mac_addrs_add(&tmpl);
+		/* Update original queue in case of failure. */
+		memcpy(rxq->mac_configured, tmpl.mac_configured,
+		       sizeof(rxq->mac_configured));
+		memcpy(rxq->mac_flow, tmpl.mac_flow, sizeof(rxq->mac_flow));
+	}
+	/* Allocate pool. */
+	pool = rte_malloc(__func__, (mbuf_n * sizeof(*pool)), 0);
+	if (pool == NULL) {
+		ERROR("%p: cannot allocate memory", (void *)dev);
+		return ENOBUFS;
+	}
+	/* Snatch mbufs from original queue. */
+	k = 0;
+	if (rxq->sp) {
+		struct rxq_elt_sp (*elts)[rxq->elts_n] = rxq->elts.sp;
+
+		for (i = 0; (i != RTE_DIM(*elts)); ++i) {
+			struct rxq_elt_sp *elt = &(*elts)[i];
+			unsigned int j;
+
+			for (j = 0; (j != RTE_DIM(elt->bufs)); ++j) {
+				assert(elt->bufs[j] != NULL);
+				pool[k++] = elt->bufs[j];
+			}
+		}
+	} else {
+		struct rxq_elt (*elts)[rxq->elts_n] = rxq->elts.no_sp;
+
+		for (i = 0; (i != RTE_DIM(*elts)); ++i) {
+			struct rxq_elt *elt = &(*elts)[i];
+			struct rte_mbuf *buf = (void *)
+				((uintptr_t)elt->sge.addr -
+				 WR_ID(elt->wr.wr_id).offset);
+
+			assert(WR_ID(elt->wr.wr_id).id == i);
+			pool[k++] = buf;
+		}
+	}
+	assert(k == mbuf_n);
+	tmpl.elts_n = 0;
+	tmpl.elts.sp = NULL;
+	assert((void *)&tmpl.elts.sp == (void *)&tmpl.elts.no_sp);
+	err = ((tmpl.sp) ?
+	       rxq_alloc_elts_sp(&tmpl, desc_n, pool) :
+	       rxq_alloc_elts(&tmpl, desc_n, pool));
+	if (err) {
+		ERROR("%p: cannot reallocate WRs, aborting", (void *)dev);
+		rte_free(pool);
+		assert(err > 0);
+		return err;
+	}
+	assert(tmpl.elts_n == desc_n);
+	assert(tmpl.elts.sp != NULL);
+	rte_free(pool);
+	/* Clean up original data. */
+	rxq->elts_n = 0;
+	rte_free(rxq->elts.sp);
+	rxq->elts.sp = NULL;
+	/* Post WRs. */
+	err = ibv_post_recv(tmpl.qp,
+			    (tmpl.sp ?
+			     &(*tmpl.elts.sp)[0].wr :
+			     &(*tmpl.elts.no_sp)[0].wr),
+			    &bad_wr);
+	if (err) {
+		ERROR("%p: ibv_post_recv() failed for WR %p: %s",
+		      (void *)dev,
+		      (void *)bad_wr,
+		      strerror(err));
+		goto skip_rtr;
+	}
+	mod = (struct ibv_exp_qp_attr){
+		.qp_state = IBV_QPS_RTR
+	};
+	err = ibv_exp_modify_qp(tmpl.qp, &mod, IBV_EXP_QP_STATE);
+	if (err)
+		ERROR("%p: QP state to IBV_QPS_RTR failed: %s",
+		      (void *)dev, strerror(err));
+skip_rtr:
+	*rxq = tmpl;
+	assert(err >= 0);
+	return err;
+}
+
+/**
  * Configure a RX queue.
  *
  * @param dev
diff --git a/drivers/net/mlx5/mlx5_rxtx.h b/drivers/net/mlx5/mlx5_rxtx.h
index 44170d7..0a2e650 100644
--- a/drivers/net/mlx5/mlx5_rxtx.h
+++ b/drivers/net/mlx5/mlx5_rxtx.h
@@ -147,6 +147,7 @@ struct txq {
 /* mlx5_rxq.c */
 
 void rxq_cleanup(struct rxq *);
+int rxq_rehash(struct rte_eth_dev *, struct rxq *);
 int rxq_setup(struct rte_eth_dev *, struct rxq *, uint16_t, unsigned int,
 	      const struct rte_eth_rxconf *, struct rte_mempool *);
 int mlx5_rx_queue_setup(struct rte_eth_dev *, uint16_t, uint16_t, unsigned int,
-- 
2.1.0

^ permalink raw reply	[flat|nested] 32+ messages in thread

* [dpdk-dev] [PATCH 07/13] mlx5: add software counters and related callbacks
  2015-10-05 17:52 [dpdk-dev] [PATCH 00/13] Mellanox ConnectX-4 PMD (mlx5) Adrien Mazarguil
                   ` (5 preceding siblings ...)
  2015-10-05 17:53 ` [dpdk-dev] [PATCH 06/13] mlx5: add MTU configuration support Adrien Mazarguil
@ 2015-10-05 17:53 ` Adrien Mazarguil
  2015-10-05 17:53 ` [dpdk-dev] [PATCH 08/13] mlx5: add promiscuous and allmulticast RX modes Adrien Mazarguil
                   ` (6 subsequent siblings)
  13 siblings, 0 replies; 32+ messages in thread
From: Adrien Mazarguil @ 2015-10-05 17:53 UTC (permalink / raw)
  To: dev

Hardware counters are not supported yet.

Signed-off-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>
Signed-off-by: Nelio Laranjeiro <nelio.laranjeiro@6wind.com>
---
 drivers/net/mlx5/Makefile     |   1 +
 drivers/net/mlx5/mlx5.c       |   2 +
 drivers/net/mlx5/mlx5.h       |   5 ++
 drivers/net/mlx5/mlx5_defs.h  |   8 +++
 drivers/net/mlx5/mlx5_rxq.c   |   1 +
 drivers/net/mlx5/mlx5_rxtx.c  |  43 +++++++++++++
 drivers/net/mlx5/mlx5_rxtx.h  |  21 ++++++
 drivers/net/mlx5/mlx5_stats.c | 144 ++++++++++++++++++++++++++++++++++++++++++
 drivers/net/mlx5/mlx5_txq.c   |   1 +
 9 files changed, 226 insertions(+)
 create mode 100644 drivers/net/mlx5/mlx5_stats.c

diff --git a/drivers/net/mlx5/Makefile b/drivers/net/mlx5/Makefile
index 028c22c..88b361c 100644
--- a/drivers/net/mlx5/Makefile
+++ b/drivers/net/mlx5/Makefile
@@ -48,6 +48,7 @@ SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5_rxtx.c
 SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5_trigger.c
 SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5_ethdev.c
 SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5_mac.c
+SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5_stats.c
 
 # Dependencies.
 DEPDIRS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += lib/librte_ether
diff --git a/drivers/net/mlx5/mlx5.c b/drivers/net/mlx5/mlx5.c
index 5bef742..262b458 100644
--- a/drivers/net/mlx5/mlx5.c
+++ b/drivers/net/mlx5/mlx5.c
@@ -133,6 +133,8 @@ static const struct eth_dev_ops mlx5_dev_ops = {
 	.dev_start = mlx5_dev_start,
 	.dev_stop = mlx5_dev_stop,
 	.dev_close = mlx5_dev_close,
+	.stats_get = mlx5_stats_get,
+	.stats_reset = mlx5_stats_reset,
 	.dev_infos_get = mlx5_dev_infos_get,
 	.rx_queue_setup = mlx5_rx_queue_setup,
 	.tx_queue_setup = mlx5_tx_queue_setup,
diff --git a/drivers/net/mlx5/mlx5.h b/drivers/net/mlx5/mlx5.h
index 14f55ba..261593e 100644
--- a/drivers/net/mlx5/mlx5.h
+++ b/drivers/net/mlx5/mlx5.h
@@ -184,6 +184,11 @@ int priv_mac_addr_add(struct priv *, unsigned int,
 void mlx5_mac_addr_add(struct rte_eth_dev *, struct ether_addr *, uint32_t,
 		       uint32_t);
 
+/* mlx5_stats.c */
+
+void mlx5_stats_get(struct rte_eth_dev *, struct rte_eth_stats *);
+void mlx5_stats_reset(struct rte_eth_dev *);
+
 /* mlx5_trigger.c */
 
 int mlx5_dev_start(struct rte_eth_dev *);
diff --git a/drivers/net/mlx5/mlx5_defs.h b/drivers/net/mlx5/mlx5_defs.h
index 4f13a4e..79de609 100644
--- a/drivers/net/mlx5/mlx5_defs.h
+++ b/drivers/net/mlx5/mlx5_defs.h
@@ -74,4 +74,12 @@
 #define MLX5_PMD_TX_MP_CACHE 8
 #endif
 
+/*
+ * If defined, only use software counters. The PMD will never ask the hardware
+ * for these, and many of them won't be available.
+ */
+#ifndef MLX5_PMD_SOFT_COUNTERS
+#define MLX5_PMD_SOFT_COUNTERS 1
+#endif
+
 #endif /* RTE_PMD_MLX5_DEFS_H_ */
diff --git a/drivers/net/mlx5/mlx5_rxq.c b/drivers/net/mlx5/mlx5_rxq.c
index 401b575..a7d5081 100644
--- a/drivers/net/mlx5/mlx5_rxq.c
+++ b/drivers/net/mlx5/mlx5_rxq.c
@@ -996,6 +996,7 @@ mlx5_rx_queue_setup(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc,
 	if (ret)
 		rte_free(rxq);
 	else {
+		rxq->stats.idx = idx;
 		DEBUG("%p: adding RX queue %p to list",
 		      (void *)dev, (void *)rxq);
 		(*priv->rxqs)[idx] = rxq;
diff --git a/drivers/net/mlx5/mlx5_rxtx.c b/drivers/net/mlx5/mlx5_rxtx.c
index 5042011..960a3e5 100644
--- a/drivers/net/mlx5/mlx5_rxtx.c
+++ b/drivers/net/mlx5/mlx5_rxtx.c
@@ -367,6 +367,9 @@ mlx5_tx_burst(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n)
 		struct txq_elt *elt_next = &(*txq->elts)[elts_head_next];
 		struct txq_elt *elt = &(*txq->elts)[elts_head];
 		unsigned int segs = NB_SEGS(buf);
+#ifdef MLX5_PMD_SOFT_COUNTERS
+		unsigned int sent_size = 0;
+#endif
 		uint32_t send_flags = 0;
 
 		/* Clean up old buffer. */
@@ -429,6 +432,9 @@ mlx5_tx_burst(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n)
 					 send_flags);
 			if (unlikely(err))
 				goto stop;
+#ifdef MLX5_PMD_SOFT_COUNTERS
+			sent_size += length;
+#endif
 		} else {
 #if MLX5_PMD_SGE_WR_N > 1
 			struct ibv_sge sges[MLX5_PMD_SGE_WR_N];
@@ -447,6 +453,9 @@ mlx5_tx_burst(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n)
 				 send_flags);
 			if (unlikely(err))
 				goto stop;
+#ifdef MLX5_PMD_SOFT_COUNTERS
+			sent_size += ret.length;
+#endif
 #else /* MLX5_PMD_SGE_WR_N > 1 */
 			DEBUG("%p: TX scattered buffers support not"
 			      " compiled in", (void *)txq);
@@ -454,11 +463,19 @@ mlx5_tx_burst(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n)
 #endif /* MLX5_PMD_SGE_WR_N > 1 */
 		}
 		elts_head = elts_head_next;
+#ifdef MLX5_PMD_SOFT_COUNTERS
+		/* Increment sent bytes counter. */
+		txq->stats.obytes += sent_size;
+#endif
 	}
 stop:
 	/* Take a shortcut if nothing must be sent. */
 	if (unlikely(i == 0))
 		return 0;
+#ifdef MLX5_PMD_SOFT_COUNTERS
+	/* Increment sent packets counter. */
+	txq->stats.opackets += i;
+#endif
 	/* Ring QP doorbell. */
 	err = txq->if_qp->send_flush(txq->qp);
 	if (unlikely(err)) {
@@ -549,6 +566,10 @@ mlx5_rx_burst_sp(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n)
 				      " completion status (%d): %s",
 				      (void *)rxq, wc.wr_id, wc.status,
 				      ibv_wc_status_str(wc.status));
+#ifdef MLX5_PMD_SOFT_COUNTERS
+				/* Increment dropped packets counter. */
+				++rxq->stats.idropped;
+#endif
 				/* Link completed WRs together for repost. */
 				*next = wr;
 				next = &wr->next;
@@ -592,6 +613,7 @@ mlx5_rx_burst_sp(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n)
 					rte_pktmbuf_free(pkt_buf);
 				}
 				/* Increment out of memory counters. */
+				++rxq->stats.rx_nombuf;
 				++rxq->priv->dev->data->rx_mbuf_alloc_failed;
 				goto repost;
 			}
@@ -651,6 +673,10 @@ mlx5_rx_burst_sp(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n)
 		/* Return packet. */
 		*(pkts++) = pkt_buf;
 		++pkts_ret;
+#ifdef MLX5_PMD_SOFT_COUNTERS
+		/* Increment bytes counter. */
+		rxq->stats.ibytes += pkt_buf_len;
+#endif
 repost:
 		if (++elts_head >= elts_n)
 			elts_head = 0;
@@ -673,6 +699,10 @@ repost:
 		abort();
 	}
 	rxq->elts_head = elts_head;
+#ifdef MLX5_PMD_SOFT_COUNTERS
+	/* Increment packets counter. */
+	rxq->stats.ipackets += pkts_ret;
+#endif
 	return pkts_ret;
 }
 
@@ -747,6 +777,10 @@ mlx5_rx_burst(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n)
 				      " completion status (%d): %s",
 				      (void *)rxq, wc.wr_id, wc.status,
 				      ibv_wc_status_str(wc.status));
+#ifdef MLX5_PMD_SOFT_COUNTERS
+				/* Increment dropped packets counter. */
+				++rxq->stats.idropped;
+#endif
 				/* Add SGE to array for repost. */
 				sges[i] = elt->sge;
 				goto repost;
@@ -771,6 +805,7 @@ mlx5_rx_burst(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n)
 			      " can't allocate a new mbuf",
 			      (void *)rxq, WR_ID(wr_id).id);
 			/* Increment out of memory counters. */
+			++rxq->stats.rx_nombuf;
 			++rxq->priv->dev->data->rx_mbuf_alloc_failed;
 			goto repost;
 		}
@@ -797,6 +832,10 @@ mlx5_rx_burst(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n)
 		/* Return packet. */
 		*(pkts++) = seg;
 		++pkts_ret;
+#ifdef MLX5_PMD_SOFT_COUNTERS
+		/* Increment bytes counter. */
+		rxq->stats.ibytes += len;
+#endif
 repost:
 		if (++elts_head >= elts_n)
 			elts_head = 0;
@@ -817,6 +856,10 @@ repost:
 		abort();
 	}
 	rxq->elts_head = elts_head;
+#ifdef MLX5_PMD_SOFT_COUNTERS
+	/* Increment packets counter. */
+	rxq->stats.ipackets += pkts_ret;
+#endif
 	return pkts_ret;
 }
 
diff --git a/drivers/net/mlx5/mlx5_rxtx.h b/drivers/net/mlx5/mlx5_rxtx.h
index 0a2e650..b37843b 100644
--- a/drivers/net/mlx5/mlx5_rxtx.h
+++ b/drivers/net/mlx5/mlx5_rxtx.h
@@ -60,6 +60,25 @@
 #include "mlx5.h"
 #include "mlx5_defs.h"
 
+struct mlx5_rxq_stats {
+	unsigned int idx; /**< Mapping index. */
+#ifdef MLX5_PMD_SOFT_COUNTERS
+	uint64_t ipackets; /**< Total of successfully received packets. */
+	uint64_t ibytes; /**< Total of successfully received bytes. */
+#endif
+	uint64_t idropped; /**< Total of packets dropped when RX ring full. */
+	uint64_t rx_nombuf; /**< Total of RX mbuf allocation failures. */
+};
+
+struct mlx5_txq_stats {
+	unsigned int idx; /**< Mapping index. */
+#ifdef MLX5_PMD_SOFT_COUNTERS
+	uint64_t opackets; /**< Total of successfully sent packets. */
+	uint64_t obytes; /**< Total of successfully sent bytes. */
+#endif
+	uint64_t odropped; /**< Total of packets not sent when TX ring full. */
+};
+
 /* RX element (scattered packets). */
 struct rxq_elt_sp {
 	struct ibv_recv_wr wr; /* Work Request. */
@@ -99,6 +118,7 @@ struct rxq {
 	} elts;
 	unsigned int sp:1; /* Use scattered RX elements. */
 	uint32_t mb_len; /* Length of a mp-issued mbuf. */
+	struct mlx5_rxq_stats stats; /* RX queue counters. */
 	unsigned int socket; /* CPU socket ID for allocations. */
 	struct ibv_exp_res_domain *rd; /* Resource Domain. */
 };
@@ -138,6 +158,7 @@ struct txq {
 	unsigned int elts_comp; /* Number of completion requests. */
 	unsigned int elts_comp_cd; /* Countdown for next completion request. */
 	unsigned int elts_comp_cd_init; /* Initial value for countdown. */
+	struct mlx5_txq_stats stats; /* TX queue counters. */
 	linear_t (*elts_linear)[]; /* Linearized buffers. */
 	struct ibv_mr *mr_linear; /* Memory Region for linearized buffers. */
 	unsigned int socket; /* CPU socket ID for allocations. */
diff --git a/drivers/net/mlx5/mlx5_stats.c b/drivers/net/mlx5/mlx5_stats.c
new file mode 100644
index 0000000..a51e945
--- /dev/null
+++ b/drivers/net/mlx5/mlx5_stats.c
@@ -0,0 +1,144 @@
+/*-
+ *   BSD LICENSE
+ *
+ *   Copyright 2015 6WIND S.A.
+ *   Copyright 2015 Mellanox.
+ *
+ *   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 6WIND S.A. 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.
+ */
+
+/* DPDK headers don't like -pedantic. */
+#ifdef PEDANTIC
+#pragma GCC diagnostic ignored "-pedantic"
+#endif
+#include <rte_ethdev.h>
+#ifdef PEDANTIC
+#pragma GCC diagnostic error "-pedantic"
+#endif
+
+#include "mlx5.h"
+#include "mlx5_rxtx.h"
+#include "mlx5_defs.h"
+
+/**
+ * DPDK callback to get device statistics.
+ *
+ * @param dev
+ *   Pointer to Ethernet device structure.
+ * @param[out] stats
+ *   Stats structure output buffer.
+ */
+void
+mlx5_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
+{
+	struct priv *priv = dev->data->dev_private;
+	struct rte_eth_stats tmp = {0};
+	unsigned int i;
+	unsigned int idx;
+
+	priv_lock(priv);
+	/* Add software counters. */
+	for (i = 0; (i != priv->rxqs_n); ++i) {
+		struct rxq *rxq = (*priv->rxqs)[i];
+
+		if (rxq == NULL)
+			continue;
+		idx = rxq->stats.idx;
+		if (idx < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
+#ifdef MLX5_PMD_SOFT_COUNTERS
+			tmp.q_ipackets[idx] += rxq->stats.ipackets;
+			tmp.q_ibytes[idx] += rxq->stats.ibytes;
+#endif
+			tmp.q_errors[idx] += (rxq->stats.idropped +
+					      rxq->stats.rx_nombuf);
+		}
+#ifdef MLX5_PMD_SOFT_COUNTERS
+		tmp.ipackets += rxq->stats.ipackets;
+		tmp.ibytes += rxq->stats.ibytes;
+#endif
+		tmp.ierrors += rxq->stats.idropped;
+		tmp.rx_nombuf += rxq->stats.rx_nombuf;
+	}
+	for (i = 0; (i != priv->txqs_n); ++i) {
+		struct txq *txq = (*priv->txqs)[i];
+
+		if (txq == NULL)
+			continue;
+		idx = txq->stats.idx;
+		if (idx < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
+#ifdef MLX5_PMD_SOFT_COUNTERS
+			tmp.q_opackets[idx] += txq->stats.opackets;
+			tmp.q_obytes[idx] += txq->stats.obytes;
+#endif
+			tmp.q_errors[idx] += txq->stats.odropped;
+		}
+#ifdef MLX5_PMD_SOFT_COUNTERS
+		tmp.opackets += txq->stats.opackets;
+		tmp.obytes += txq->stats.obytes;
+#endif
+		tmp.oerrors += txq->stats.odropped;
+	}
+#ifndef MLX5_PMD_SOFT_COUNTERS
+	/* FIXME: retrieve and add hardware counters. */
+#endif
+	*stats = tmp;
+	priv_unlock(priv);
+}
+
+/**
+ * DPDK callback to clear device statistics.
+ *
+ * @param dev
+ *   Pointer to Ethernet device structure.
+ */
+void
+mlx5_stats_reset(struct rte_eth_dev *dev)
+{
+	struct priv *priv = dev->data->dev_private;
+	unsigned int i;
+	unsigned int idx;
+
+	priv_lock(priv);
+	for (i = 0; (i != priv->rxqs_n); ++i) {
+		if ((*priv->rxqs)[i] == NULL)
+			continue;
+		idx = (*priv->rxqs)[i]->stats.idx;
+		(*priv->rxqs)[i]->stats =
+			(struct mlx5_rxq_stats){ .idx = idx };
+	}
+	for (i = 0; (i != priv->txqs_n); ++i) {
+		if ((*priv->txqs)[i] == NULL)
+			continue;
+		idx = (*priv->rxqs)[i]->stats.idx;
+		(*priv->txqs)[i]->stats =
+			(struct mlx5_txq_stats){ .idx = idx };
+	}
+#ifndef MLX5_PMD_SOFT_COUNTERS
+	/* FIXME: reset hardware counters. */
+#endif
+	priv_unlock(priv);
+}
diff --git a/drivers/net/mlx5/mlx5_txq.c b/drivers/net/mlx5/mlx5_txq.c
index 2bae61f..a53b128 100644
--- a/drivers/net/mlx5/mlx5_txq.c
+++ b/drivers/net/mlx5/mlx5_txq.c
@@ -472,6 +472,7 @@ mlx5_tx_queue_setup(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc,
 	if (ret)
 		rte_free(txq);
 	else {
+		txq->stats.idx = idx;
 		DEBUG("%p: adding TX queue %p to list",
 		      (void *)dev, (void *)txq);
 		(*priv->txqs)[idx] = txq;
-- 
2.1.0

^ permalink raw reply	[flat|nested] 32+ messages in thread

* [dpdk-dev] [PATCH 08/13] mlx5: add promiscuous and allmulticast RX modes
  2015-10-05 17:52 [dpdk-dev] [PATCH 00/13] Mellanox ConnectX-4 PMD (mlx5) Adrien Mazarguil
                   ` (6 preceding siblings ...)
  2015-10-05 17:53 ` [dpdk-dev] [PATCH 07/13] mlx5: add software counters and related callbacks Adrien Mazarguil
@ 2015-10-05 17:53 ` Adrien Mazarguil
  2015-10-05 17:53 ` [dpdk-dev] [PATCH 09/13] mlx5: add link update device operation Adrien Mazarguil
                   ` (5 subsequent siblings)
  13 siblings, 0 replies; 32+ messages in thread
From: Adrien Mazarguil @ 2015-10-05 17:53 UTC (permalink / raw)
  To: dev

These modes require special non-MAC flows.

Signed-off-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>
Signed-off-by: Nelio Laranjeiro <nelio.laranjeiro@6wind.com>
---
 drivers/net/mlx5/Makefile       |   1 +
 drivers/net/mlx5/mlx5.c         |   4 +
 drivers/net/mlx5/mlx5.h         |  11 ++
 drivers/net/mlx5/mlx5_ethdev.c  |   4 +
 drivers/net/mlx5/mlx5_rxmode.c  | 327 ++++++++++++++++++++++++++++++++++++++++
 drivers/net/mlx5/mlx5_rxq.c     |  12 ++
 drivers/net/mlx5/mlx5_rxtx.h    |   2 +
 drivers/net/mlx5/mlx5_trigger.c |   8 +
 8 files changed, 369 insertions(+)
 create mode 100644 drivers/net/mlx5/mlx5_rxmode.c

diff --git a/drivers/net/mlx5/Makefile b/drivers/net/mlx5/Makefile
index 88b361c..4d25d9c 100644
--- a/drivers/net/mlx5/Makefile
+++ b/drivers/net/mlx5/Makefile
@@ -48,6 +48,7 @@ SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5_rxtx.c
 SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5_trigger.c
 SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5_ethdev.c
 SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5_mac.c
+SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5_rxmode.c
 SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5_stats.c
 
 # Dependencies.
diff --git a/drivers/net/mlx5/mlx5.c b/drivers/net/mlx5/mlx5.c
index 262b458..7e60d6a 100644
--- a/drivers/net/mlx5/mlx5.c
+++ b/drivers/net/mlx5/mlx5.c
@@ -133,6 +133,10 @@ static const struct eth_dev_ops mlx5_dev_ops = {
 	.dev_start = mlx5_dev_start,
 	.dev_stop = mlx5_dev_stop,
 	.dev_close = mlx5_dev_close,
+	.promiscuous_enable = mlx5_promiscuous_enable,
+	.promiscuous_disable = mlx5_promiscuous_disable,
+	.allmulticast_enable = mlx5_allmulticast_enable,
+	.allmulticast_disable = mlx5_allmulticast_disable,
 	.stats_get = mlx5_stats_get,
 	.stats_reset = mlx5_stats_reset,
 	.dev_infos_get = mlx5_dev_infos_get,
diff --git a/drivers/net/mlx5/mlx5.h b/drivers/net/mlx5/mlx5.h
index 261593e..40dc1f9 100644
--- a/drivers/net/mlx5/mlx5.h
+++ b/drivers/net/mlx5/mlx5.h
@@ -184,6 +184,17 @@ int priv_mac_addr_add(struct priv *, unsigned int,
 void mlx5_mac_addr_add(struct rte_eth_dev *, struct ether_addr *, uint32_t,
 		       uint32_t);
 
+/* mlx5_rxmode.c */
+
+int rxq_promiscuous_enable(struct rxq *);
+void mlx5_promiscuous_enable(struct rte_eth_dev *);
+void rxq_promiscuous_disable(struct rxq *);
+void mlx5_promiscuous_disable(struct rte_eth_dev *);
+int rxq_allmulticast_enable(struct rxq *);
+void mlx5_allmulticast_enable(struct rte_eth_dev *);
+void rxq_allmulticast_disable(struct rxq *);
+void mlx5_allmulticast_disable(struct rte_eth_dev *);
+
 /* mlx5_stats.c */
 
 void mlx5_stats_get(struct rte_eth_dev *, struct rte_eth_stats *);
diff --git a/drivers/net/mlx5/mlx5_ethdev.c b/drivers/net/mlx5/mlx5_ethdev.c
index 4621282..0e95d26 100644
--- a/drivers/net/mlx5/mlx5_ethdev.c
+++ b/drivers/net/mlx5/mlx5_ethdev.c
@@ -604,6 +604,10 @@ mlx5_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
 		 * for errors at this stage. */
 		if (!priv->rss) {
 			rxq_mac_addrs_add(rxq);
+			if (priv->promisc)
+				rxq_promiscuous_enable(rxq);
+			if (priv->allmulti)
+				rxq_allmulticast_enable(rxq);
 		}
 		/* Scattered burst function takes priority. */
 		if (rxq->sp)
diff --git a/drivers/net/mlx5/mlx5_rxmode.c b/drivers/net/mlx5/mlx5_rxmode.c
new file mode 100644
index 0000000..b4e5493
--- /dev/null
+++ b/drivers/net/mlx5/mlx5_rxmode.c
@@ -0,0 +1,327 @@
+/*-
+ *   BSD LICENSE
+ *
+ *   Copyright 2015 6WIND S.A.
+ *   Copyright 2015 Mellanox.
+ *
+ *   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 6WIND S.A. 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 <stddef.h>
+#include <errno.h>
+#include <string.h>
+
+/* Verbs header. */
+/* ISO C doesn't support unnamed structs/unions, disabling -pedantic. */
+#ifdef PEDANTIC
+#pragma GCC diagnostic ignored "-pedantic"
+#endif
+#include <infiniband/verbs.h>
+#ifdef PEDANTIC
+#pragma GCC diagnostic error "-pedantic"
+#endif
+
+/* DPDK headers don't like -pedantic. */
+#ifdef PEDANTIC
+#pragma GCC diagnostic ignored "-pedantic"
+#endif
+#include <rte_ethdev.h>
+#ifdef PEDANTIC
+#pragma GCC diagnostic error "-pedantic"
+#endif
+
+#include "mlx5.h"
+#include "mlx5_rxtx.h"
+#include "mlx5_utils.h"
+
+/**
+ * Enable promiscuous mode in a RX queue.
+ *
+ * @param rxq
+ *   Pointer to RX queue structure.
+ *
+ * @return
+ *   0 on success, errno value on failure.
+ */
+int
+rxq_promiscuous_enable(struct rxq *rxq)
+{
+	struct ibv_flow *flow;
+	struct ibv_flow_attr attr = {
+		.type = IBV_FLOW_ATTR_ALL_DEFAULT,
+		.num_of_specs = 0,
+		.port = rxq->priv->port,
+		.flags = 0
+	};
+
+	if (rxq->priv->vf)
+		return 0;
+	DEBUG("%p: enabling promiscuous mode", (void *)rxq);
+	if (rxq->promisc_flow != NULL)
+		return EBUSY;
+	errno = 0;
+	flow = ibv_create_flow(rxq->qp, &attr);
+	if (flow == NULL) {
+		/* It's not clear whether errno is always set in this case. */
+		ERROR("%p: flow configuration failed, errno=%d: %s",
+		      (void *)rxq, errno,
+		      (errno ? strerror(errno) : "Unknown error"));
+		if (errno)
+			return errno;
+		return EINVAL;
+	}
+	rxq->promisc_flow = flow;
+	DEBUG("%p: promiscuous mode enabled", (void *)rxq);
+	return 0;
+}
+
+/**
+ * DPDK callback to enable promiscuous mode.
+ *
+ * @param dev
+ *   Pointer to Ethernet device structure.
+ */
+void
+mlx5_promiscuous_enable(struct rte_eth_dev *dev)
+{
+	struct priv *priv = dev->data->dev_private;
+	unsigned int i;
+	int ret;
+
+	priv_lock(priv);
+	if (priv->promisc) {
+		priv_unlock(priv);
+		return;
+	}
+	/* If device isn't started, this is all we need to do. */
+	if (!priv->started)
+		goto end;
+	if (priv->rss) {
+		ret = rxq_promiscuous_enable(&priv->rxq_parent);
+		if (ret) {
+			priv_unlock(priv);
+			return;
+		}
+		goto end;
+	}
+	for (i = 0; (i != priv->rxqs_n); ++i) {
+		if ((*priv->rxqs)[i] == NULL)
+			continue;
+		ret = rxq_promiscuous_enable((*priv->rxqs)[i]);
+		if (!ret)
+			continue;
+		/* Failure, rollback. */
+		while (i != 0)
+			if ((*priv->rxqs)[--i] != NULL)
+				rxq_promiscuous_disable((*priv->rxqs)[i]);
+		priv_unlock(priv);
+		return;
+	}
+end:
+	priv->promisc = 1;
+	priv_unlock(priv);
+}
+
+/**
+ * Disable promiscuous mode in a RX queue.
+ *
+ * @param rxq
+ *   Pointer to RX queue structure.
+ */
+void
+rxq_promiscuous_disable(struct rxq *rxq)
+{
+	if (rxq->priv->vf)
+		return;
+	DEBUG("%p: disabling promiscuous mode", (void *)rxq);
+	if (rxq->promisc_flow == NULL)
+		return;
+	claim_zero(ibv_destroy_flow(rxq->promisc_flow));
+	rxq->promisc_flow = NULL;
+	DEBUG("%p: promiscuous mode disabled", (void *)rxq);
+}
+
+/**
+ * DPDK callback to disable promiscuous mode.
+ *
+ * @param dev
+ *   Pointer to Ethernet device structure.
+ */
+void
+mlx5_promiscuous_disable(struct rte_eth_dev *dev)
+{
+	struct priv *priv = dev->data->dev_private;
+	unsigned int i;
+
+	priv_lock(priv);
+	if (!priv->promisc) {
+		priv_unlock(priv);
+		return;
+	}
+	if (priv->rss) {
+		rxq_promiscuous_disable(&priv->rxq_parent);
+		goto end;
+	}
+	for (i = 0; (i != priv->rxqs_n); ++i)
+		if ((*priv->rxqs)[i] != NULL)
+			rxq_promiscuous_disable((*priv->rxqs)[i]);
+end:
+	priv->promisc = 0;
+	priv_unlock(priv);
+}
+
+/**
+ * Enable allmulti mode in a RX queue.
+ *
+ * @param rxq
+ *   Pointer to RX queue structure.
+ *
+ * @return
+ *   0 on success, errno value on failure.
+ */
+int
+rxq_allmulticast_enable(struct rxq *rxq)
+{
+	struct ibv_flow *flow;
+	struct ibv_flow_attr attr = {
+		.type = IBV_FLOW_ATTR_MC_DEFAULT,
+		.num_of_specs = 0,
+		.port = rxq->priv->port,
+		.flags = 0
+	};
+
+	DEBUG("%p: enabling allmulticast mode", (void *)rxq);
+	if (rxq->allmulti_flow != NULL)
+		return EBUSY;
+	errno = 0;
+	flow = ibv_create_flow(rxq->qp, &attr);
+	if (flow == NULL) {
+		/* It's not clear whether errno is always set in this case. */
+		ERROR("%p: flow configuration failed, errno=%d: %s",
+		      (void *)rxq, errno,
+		      (errno ? strerror(errno) : "Unknown error"));
+		if (errno)
+			return errno;
+		return EINVAL;
+	}
+	rxq->allmulti_flow = flow;
+	DEBUG("%p: allmulticast mode enabled", (void *)rxq);
+	return 0;
+}
+
+/**
+ * DPDK callback to enable allmulti mode.
+ *
+ * @param dev
+ *   Pointer to Ethernet device structure.
+ */
+void
+mlx5_allmulticast_enable(struct rte_eth_dev *dev)
+{
+	struct priv *priv = dev->data->dev_private;
+	unsigned int i;
+	int ret;
+
+	priv_lock(priv);
+	if (priv->allmulti) {
+		priv_unlock(priv);
+		return;
+	}
+	/* If device isn't started, this is all we need to do. */
+	if (!priv->started)
+		goto end;
+	if (priv->rss) {
+		ret = rxq_allmulticast_enable(&priv->rxq_parent);
+		if (ret) {
+			priv_unlock(priv);
+			return;
+		}
+		goto end;
+	}
+	for (i = 0; (i != priv->rxqs_n); ++i) {
+		if ((*priv->rxqs)[i] == NULL)
+			continue;
+		ret = rxq_allmulticast_enable((*priv->rxqs)[i]);
+		if (!ret)
+			continue;
+		/* Failure, rollback. */
+		while (i != 0)
+			if ((*priv->rxqs)[--i] != NULL)
+				rxq_allmulticast_disable((*priv->rxqs)[i]);
+		priv_unlock(priv);
+		return;
+	}
+end:
+	priv->allmulti = 1;
+	priv_unlock(priv);
+}
+
+/**
+ * Disable allmulti mode in a RX queue.
+ *
+ * @param rxq
+ *   Pointer to RX queue structure.
+ */
+void
+rxq_allmulticast_disable(struct rxq *rxq)
+{
+	DEBUG("%p: disabling allmulticast mode", (void *)rxq);
+	if (rxq->allmulti_flow == NULL)
+		return;
+	claim_zero(ibv_destroy_flow(rxq->allmulti_flow));
+	rxq->allmulti_flow = NULL;
+	DEBUG("%p: allmulticast mode disabled", (void *)rxq);
+}
+
+/**
+ * DPDK callback to disable allmulti mode.
+ *
+ * @param dev
+ *   Pointer to Ethernet device structure.
+ */
+void
+mlx5_allmulticast_disable(struct rte_eth_dev *dev)
+{
+	struct priv *priv = dev->data->dev_private;
+	unsigned int i;
+
+	priv_lock(priv);
+	if (!priv->allmulti) {
+		priv_unlock(priv);
+		return;
+	}
+	if (priv->rss) {
+		rxq_allmulticast_disable(&priv->rxq_parent);
+		goto end;
+	}
+	for (i = 0; (i != priv->rxqs_n); ++i)
+		if ((*priv->rxqs)[i] != NULL)
+			rxq_allmulticast_disable((*priv->rxqs)[i]);
+end:
+	priv->allmulti = 0;
+	priv_unlock(priv);
+}
diff --git a/drivers/net/mlx5/mlx5_rxq.c b/drivers/net/mlx5/mlx5_rxq.c
index a7d5081..d44bb10 100644
--- a/drivers/net/mlx5/mlx5_rxq.c
+++ b/drivers/net/mlx5/mlx5_rxq.c
@@ -398,6 +398,8 @@ rxq_cleanup(struct rxq *rxq)
 						&params));
 	}
 	if (rxq->qp != NULL) {
+		rxq_promiscuous_disable(rxq);
+		rxq_allmulticast_disable(rxq);
 		rxq_mac_addrs_del(rxq);
 		claim_zero(ibv_destroy_qp(rxq->qp));
 	}
@@ -580,8 +582,12 @@ rxq_rehash(struct rte_eth_dev *dev, struct rxq *rxq)
 	}
 	/* Remove attached flows if RSS is disabled (no parent queue). */
 	if (!priv->rss) {
+		rxq_allmulticast_disable(&tmpl);
+		rxq_promiscuous_disable(&tmpl);
 		rxq_mac_addrs_del(&tmpl);
 		/* Update original queue in case of failure. */
+		rxq->allmulti_flow = tmpl.allmulti_flow;
+		rxq->promisc_flow = tmpl.promisc_flow;
 		memcpy(rxq->mac_configured, tmpl.mac_configured,
 		       sizeof(rxq->mac_configured));
 		memcpy(rxq->mac_flow, tmpl.mac_flow, sizeof(rxq->mac_flow));
@@ -622,7 +628,13 @@ rxq_rehash(struct rte_eth_dev *dev, struct rxq *rxq)
 	/* Reconfigure flows. Do not care for errors. */
 	if (!priv->rss) {
 		rxq_mac_addrs_add(&tmpl);
+		if (priv->promisc)
+			rxq_promiscuous_enable(&tmpl);
+		if (priv->allmulti)
+			rxq_allmulticast_enable(&tmpl);
 		/* Update original queue in case of failure. */
+		rxq->allmulti_flow = tmpl.allmulti_flow;
+		rxq->promisc_flow = tmpl.promisc_flow;
 		memcpy(rxq->mac_configured, tmpl.mac_configured,
 		       sizeof(rxq->mac_configured));
 		memcpy(rxq->mac_flow, tmpl.mac_flow, sizeof(rxq->mac_flow));
diff --git a/drivers/net/mlx5/mlx5_rxtx.h b/drivers/net/mlx5/mlx5_rxtx.h
index b37843b..228dff6 100644
--- a/drivers/net/mlx5/mlx5_rxtx.h
+++ b/drivers/net/mlx5/mlx5_rxtx.h
@@ -109,6 +109,8 @@ struct rxq {
 	 */
 	BITFIELD_DECLARE(mac_configured, uint32_t, MLX5_MAX_MAC_ADDRESSES);
 	struct ibv_flow *mac_flow[MLX5_MAX_MAC_ADDRESSES][MLX5_MAX_VLAN_IDS];
+	struct ibv_flow *promisc_flow; /* Promiscuous flow. */
+	struct ibv_flow *allmulti_flow; /* Multicast flow. */
 	unsigned int port_id; /* Port ID for incoming packets. */
 	unsigned int elts_n; /* (*elts)[] length. */
 	unsigned int elts_head; /* Current index in (*elts)[]. */
diff --git a/drivers/net/mlx5/mlx5_trigger.c b/drivers/net/mlx5/mlx5_trigger.c
index bcec957..fbc977c 100644
--- a/drivers/net/mlx5/mlx5_trigger.c
+++ b/drivers/net/mlx5/mlx5_trigger.c
@@ -86,6 +86,10 @@ mlx5_dev_start(struct rte_eth_dev *dev)
 		if (rxq == NULL)
 			continue;
 		ret = rxq_mac_addrs_add(rxq);
+		if (!ret && priv->promisc)
+			ret = rxq_promiscuous_enable(rxq);
+		if (!ret && priv->allmulti)
+			ret = rxq_allmulticast_enable(rxq);
 		if (!ret)
 			continue;
 		WARN("%p: QP flow attachment failed: %s",
@@ -94,6 +98,8 @@ mlx5_dev_start(struct rte_eth_dev *dev)
 		while (i != 0) {
 			rxq = (*priv->rxqs)[--i];
 			if (rxq != NULL) {
+				rxq_allmulticast_disable(rxq);
+				rxq_promiscuous_disable(rxq);
 				rxq_mac_addrs_del(rxq);
 			}
 		}
@@ -139,6 +145,8 @@ mlx5_dev_stop(struct rte_eth_dev *dev)
 		/* Ignore nonexistent RX queues. */
 		if (rxq == NULL)
 			continue;
+		rxq_allmulticast_disable(rxq);
+		rxq_promiscuous_disable(rxq);
 		rxq_mac_addrs_del(rxq);
 	} while ((--r) && ((rxq = (*priv->rxqs)[++i]), i));
 	priv_unlock(priv);
-- 
2.1.0

^ permalink raw reply	[flat|nested] 32+ messages in thread

* [dpdk-dev] [PATCH 09/13] mlx5: add link update device operation
  2015-10-05 17:52 [dpdk-dev] [PATCH 00/13] Mellanox ConnectX-4 PMD (mlx5) Adrien Mazarguil
                   ` (7 preceding siblings ...)
  2015-10-05 17:53 ` [dpdk-dev] [PATCH 08/13] mlx5: add promiscuous and allmulticast RX modes Adrien Mazarguil
@ 2015-10-05 17:53 ` Adrien Mazarguil
  2015-10-05 17:53 ` [dpdk-dev] [PATCH 10/13] mlx5: add flow control device operations Adrien Mazarguil
                   ` (4 subsequent siblings)
  13 siblings, 0 replies; 32+ messages in thread
From: Adrien Mazarguil @ 2015-10-05 17:53 UTC (permalink / raw)
  To: dev

Link information is retrieved using ethtool ioctls.

Signed-off-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>
Signed-off-by: Nelio Laranjeiro <nelio.laranjeiro@6wind.com>
---
 drivers/net/mlx5/mlx5.c        |  1 +
 drivers/net/mlx5/mlx5.h        |  1 +
 drivers/net/mlx5/mlx5_ethdev.c | 71 ++++++++++++++++++++++++++++++++++++++++++
 3 files changed, 73 insertions(+)

diff --git a/drivers/net/mlx5/mlx5.c b/drivers/net/mlx5/mlx5.c
index 7e60d6a..cb66186 100644
--- a/drivers/net/mlx5/mlx5.c
+++ b/drivers/net/mlx5/mlx5.c
@@ -137,6 +137,7 @@ static const struct eth_dev_ops mlx5_dev_ops = {
 	.promiscuous_disable = mlx5_promiscuous_disable,
 	.allmulticast_enable = mlx5_allmulticast_enable,
 	.allmulticast_disable = mlx5_allmulticast_disable,
+	.link_update = mlx5_link_update,
 	.stats_get = mlx5_stats_get,
 	.stats_reset = mlx5_stats_reset,
 	.dev_infos_get = mlx5_dev_infos_get,
diff --git a/drivers/net/mlx5/mlx5.h b/drivers/net/mlx5/mlx5.h
index 40dc1f9..3078f13 100644
--- a/drivers/net/mlx5/mlx5.h
+++ b/drivers/net/mlx5/mlx5.h
@@ -169,6 +169,7 @@ int priv_get_mtu(struct priv *, uint16_t *);
 int priv_set_flags(struct priv *, unsigned int, unsigned int);
 int mlx5_dev_configure(struct rte_eth_dev *);
 void mlx5_dev_infos_get(struct rte_eth_dev *, struct rte_eth_dev_info *);
+int mlx5_link_update(struct rte_eth_dev *, int);
 int mlx5_dev_set_mtu(struct rte_eth_dev *, uint16_t);
 int mlx5_ibv_device_to_pci_addr(const struct ibv_device *,
 				struct rte_pci_addr *);
diff --git a/drivers/net/mlx5/mlx5_ethdev.c b/drivers/net/mlx5/mlx5_ethdev.c
index 0e95d26..a665725 100644
--- a/drivers/net/mlx5/mlx5_ethdev.c
+++ b/drivers/net/mlx5/mlx5_ethdev.c
@@ -45,6 +45,8 @@
 #include <sys/socket.h>
 #include <netinet/in.h>
 #include <linux/if.h>
+#include <linux/ethtool.h>
+#include <linux/sockios.h>
 
 /* DPDK headers don't like -pedantic. */
 #ifdef PEDANTIC
@@ -535,6 +537,75 @@ mlx5_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *info)
 }
 
 /**
+ * DPDK callback to retrieve physical link information (unlocked version).
+ *
+ * @param dev
+ *   Pointer to Ethernet device structure.
+ * @param wait_to_complete
+ *   Wait for request completion (ignored).
+ */
+static int
+mlx5_link_update_unlocked(struct rte_eth_dev *dev, int wait_to_complete)
+{
+	struct priv *priv = dev->data->dev_private;
+	struct ethtool_cmd edata = {
+		.cmd = ETHTOOL_GSET
+	};
+	struct ifreq ifr;
+	struct rte_eth_link dev_link;
+	int link_speed = 0;
+
+	(void)wait_to_complete;
+	if (priv_ifreq(priv, SIOCGIFFLAGS, &ifr)) {
+		WARN("ioctl(SIOCGIFFLAGS) failed: %s", strerror(errno));
+		return -1;
+	}
+	memset(&dev_link, 0, sizeof(dev_link));
+	dev_link.link_status = ((ifr.ifr_flags & IFF_UP) &&
+				(ifr.ifr_flags & IFF_RUNNING));
+	ifr.ifr_data = &edata;
+	if (priv_ifreq(priv, SIOCETHTOOL, &ifr)) {
+		WARN("ioctl(SIOCETHTOOL, ETHTOOL_GSET) failed: %s",
+		     strerror(errno));
+		return -1;
+	}
+	link_speed = ethtool_cmd_speed(&edata);
+	if (link_speed == -1)
+		dev_link.link_speed = 0;
+	else
+		dev_link.link_speed = link_speed;
+	dev_link.link_duplex = ((edata.duplex == DUPLEX_HALF) ?
+				ETH_LINK_HALF_DUPLEX : ETH_LINK_FULL_DUPLEX);
+	if (memcmp(&dev_link, &dev->data->dev_link, sizeof(dev_link))) {
+		/* Link status changed. */
+		dev->data->dev_link = dev_link;
+		return 0;
+	}
+	/* Link status is still the same. */
+	return -1;
+}
+
+/**
+ * DPDK callback to retrieve physical link information.
+ *
+ * @param dev
+ *   Pointer to Ethernet device structure.
+ * @param wait_to_complete
+ *   Wait for request completion (ignored).
+ */
+int
+mlx5_link_update(struct rte_eth_dev *dev, int wait_to_complete)
+{
+	struct priv *priv = dev->data->dev_private;
+	int ret;
+
+	priv_lock(priv);
+	ret = mlx5_link_update_unlocked(dev, wait_to_complete);
+	priv_unlock(priv);
+	return ret;
+}
+
+/**
  * DPDK callback to change the MTU.
  *
  * Setting the MTU affects hardware MRU (packets larger than the MTU cannot be
-- 
2.1.0

^ permalink raw reply	[flat|nested] 32+ messages in thread

* [dpdk-dev] [PATCH 10/13] mlx5: add flow control device operations
  2015-10-05 17:52 [dpdk-dev] [PATCH 00/13] Mellanox ConnectX-4 PMD (mlx5) Adrien Mazarguil
                   ` (8 preceding siblings ...)
  2015-10-05 17:53 ` [dpdk-dev] [PATCH 09/13] mlx5: add link update device operation Adrien Mazarguil
@ 2015-10-05 17:53 ` Adrien Mazarguil
  2015-10-05 17:53 ` [dpdk-dev] [PATCH 11/13] mlx5: add VLAN filtering Adrien Mazarguil
                   ` (3 subsequent siblings)
  13 siblings, 0 replies; 32+ messages in thread
From: Adrien Mazarguil @ 2015-10-05 17:53 UTC (permalink / raw)
  To: dev

Like most other device control operations, those are handled by the related
kernel network device through syscalls.

Signed-off-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>
Signed-off-by: Nelio Laranjeiro <nelio.laranjeiro@6wind.com>
---
 drivers/net/mlx5/mlx5.c        |  2 +
 drivers/net/mlx5/mlx5.h        |  2 +
 drivers/net/mlx5/mlx5_ethdev.c | 99 ++++++++++++++++++++++++++++++++++++++++++
 3 files changed, 103 insertions(+)

diff --git a/drivers/net/mlx5/mlx5.c b/drivers/net/mlx5/mlx5.c
index cb66186..b182d0b 100644
--- a/drivers/net/mlx5/mlx5.c
+++ b/drivers/net/mlx5/mlx5.c
@@ -145,6 +145,8 @@ static const struct eth_dev_ops mlx5_dev_ops = {
 	.tx_queue_setup = mlx5_tx_queue_setup,
 	.rx_queue_release = mlx5_rx_queue_release,
 	.tx_queue_release = mlx5_tx_queue_release,
+	.flow_ctrl_get = mlx5_dev_get_flow_ctrl,
+	.flow_ctrl_set = mlx5_dev_set_flow_ctrl,
 	.mac_addr_remove = mlx5_mac_addr_remove,
 	.mac_addr_add = mlx5_mac_addr_add,
 	.mtu_set = mlx5_dev_set_mtu,
diff --git a/drivers/net/mlx5/mlx5.h b/drivers/net/mlx5/mlx5.h
index 3078f13..1d488fb 100644
--- a/drivers/net/mlx5/mlx5.h
+++ b/drivers/net/mlx5/mlx5.h
@@ -171,6 +171,8 @@ int mlx5_dev_configure(struct rte_eth_dev *);
 void mlx5_dev_infos_get(struct rte_eth_dev *, struct rte_eth_dev_info *);
 int mlx5_link_update(struct rte_eth_dev *, int);
 int mlx5_dev_set_mtu(struct rte_eth_dev *, uint16_t);
+int mlx5_dev_get_flow_ctrl(struct rte_eth_dev *, struct rte_eth_fc_conf *);
+int mlx5_dev_set_flow_ctrl(struct rte_eth_dev *, struct rte_eth_fc_conf *);
 int mlx5_ibv_device_to_pci_addr(const struct ibv_device *,
 				struct rte_pci_addr *);
 
diff --git a/drivers/net/mlx5/mlx5_ethdev.c b/drivers/net/mlx5/mlx5_ethdev.c
index a665725..181a877 100644
--- a/drivers/net/mlx5/mlx5_ethdev.c
+++ b/drivers/net/mlx5/mlx5_ethdev.c
@@ -694,6 +694,105 @@ out:
 }
 
 /**
+ * DPDK callback to get flow control status.
+ *
+ * @param dev
+ *   Pointer to Ethernet device structure.
+ * @param[out] fc_conf
+ *   Flow control output buffer.
+ *
+ * @return
+ *   0 on success, negative errno value on failure.
+ */
+int
+mlx5_dev_get_flow_ctrl(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
+{
+	struct priv *priv = dev->data->dev_private;
+	struct ifreq ifr;
+	struct ethtool_pauseparam ethpause = {
+		.cmd = ETHTOOL_GPAUSEPARAM
+	};
+	int ret;
+
+	ifr.ifr_data = &ethpause;
+	priv_lock(priv);
+	if (priv_ifreq(priv, SIOCETHTOOL, &ifr)) {
+		ret = errno;
+		WARN("ioctl(SIOCETHTOOL, ETHTOOL_GPAUSEPARAM)"
+		     " failed: %s",
+		     strerror(ret));
+		goto out;
+	}
+
+	fc_conf->autoneg = ethpause.autoneg;
+	if (ethpause.rx_pause && ethpause.tx_pause)
+		fc_conf->mode = RTE_FC_FULL;
+	else if (ethpause.rx_pause)
+		fc_conf->mode = RTE_FC_RX_PAUSE;
+	else if (ethpause.tx_pause)
+		fc_conf->mode = RTE_FC_TX_PAUSE;
+	else
+		fc_conf->mode = RTE_FC_NONE;
+	ret = 0;
+
+out:
+	priv_unlock(priv);
+	assert(ret >= 0);
+	return -ret;
+}
+
+/**
+ * DPDK callback to modify flow control parameters.
+ *
+ * @param dev
+ *   Pointer to Ethernet device structure.
+ * @param[in] fc_conf
+ *   Flow control parameters.
+ *
+ * @return
+ *   0 on success, negative errno value on failure.
+ */
+int
+mlx5_dev_set_flow_ctrl(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
+{
+	struct priv *priv = dev->data->dev_private;
+	struct ifreq ifr;
+	struct ethtool_pauseparam ethpause = {
+		.cmd = ETHTOOL_SPAUSEPARAM
+	};
+	int ret;
+
+	ifr.ifr_data = &ethpause;
+	ethpause.autoneg = fc_conf->autoneg;
+	if (((fc_conf->mode & RTE_FC_FULL) == RTE_FC_FULL) ||
+	    (fc_conf->mode & RTE_FC_RX_PAUSE))
+		ethpause.rx_pause = 1;
+	else
+		ethpause.rx_pause = 0;
+
+	if (((fc_conf->mode & RTE_FC_FULL) == RTE_FC_FULL) ||
+	    (fc_conf->mode & RTE_FC_TX_PAUSE))
+		ethpause.tx_pause = 1;
+	else
+		ethpause.tx_pause = 0;
+
+	priv_lock(priv);
+	if (priv_ifreq(priv, SIOCETHTOOL, &ifr)) {
+		ret = errno;
+		WARN("ioctl(SIOCETHTOOL, ETHTOOL_SPAUSEPARAM)"
+		     " failed: %s",
+		     strerror(ret));
+		goto out;
+	}
+	ret = 0;
+
+out:
+	priv_unlock(priv);
+	assert(ret >= 0);
+	return -ret;
+}
+
+/**
  * Get PCI information from struct ibv_device.
  *
  * @param device
-- 
2.1.0

^ permalink raw reply	[flat|nested] 32+ messages in thread

* [dpdk-dev] [PATCH 11/13] mlx5: add VLAN filtering
  2015-10-05 17:52 [dpdk-dev] [PATCH 00/13] Mellanox ConnectX-4 PMD (mlx5) Adrien Mazarguil
                   ` (9 preceding siblings ...)
  2015-10-05 17:53 ` [dpdk-dev] [PATCH 10/13] mlx5: add flow control device operations Adrien Mazarguil
@ 2015-10-05 17:53 ` Adrien Mazarguil
  2015-10-05 17:53 ` [dpdk-dev] [PATCH 12/13] mlx5: add checksum offloading support Adrien Mazarguil
                   ` (2 subsequent siblings)
  13 siblings, 0 replies; 32+ messages in thread
From: Adrien Mazarguil @ 2015-10-05 17:53 UTC (permalink / raw)
  To: dev

All MAC RX flows must be updated with VLAN information when configuring a
VLAN filter.

Signed-off-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>
Signed-off-by: Nelio Laranjeiro <nelio.laranjeiro@6wind.com>
---
 drivers/net/mlx5/Makefile    |   1 +
 drivers/net/mlx5/mlx5.c      |   1 +
 drivers/net/mlx5/mlx5.h      |   4 ++
 drivers/net/mlx5/mlx5_vlan.c | 166 +++++++++++++++++++++++++++++++++++++++++++
 4 files changed, 172 insertions(+)
 create mode 100644 drivers/net/mlx5/mlx5_vlan.c

diff --git a/drivers/net/mlx5/Makefile b/drivers/net/mlx5/Makefile
index 4d25d9c..8b1e32b 100644
--- a/drivers/net/mlx5/Makefile
+++ b/drivers/net/mlx5/Makefile
@@ -49,6 +49,7 @@ SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5_trigger.c
 SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5_ethdev.c
 SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5_mac.c
 SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5_rxmode.c
+SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5_vlan.c
 SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5_stats.c
 
 # Dependencies.
diff --git a/drivers/net/mlx5/mlx5.c b/drivers/net/mlx5/mlx5.c
index b182d0b..47070f8 100644
--- a/drivers/net/mlx5/mlx5.c
+++ b/drivers/net/mlx5/mlx5.c
@@ -141,6 +141,7 @@ static const struct eth_dev_ops mlx5_dev_ops = {
 	.stats_get = mlx5_stats_get,
 	.stats_reset = mlx5_stats_reset,
 	.dev_infos_get = mlx5_dev_infos_get,
+	.vlan_filter_set = mlx5_vlan_filter_set,
 	.rx_queue_setup = mlx5_rx_queue_setup,
 	.tx_queue_setup = mlx5_tx_queue_setup,
 	.rx_queue_release = mlx5_rx_queue_release,
diff --git a/drivers/net/mlx5/mlx5.h b/drivers/net/mlx5/mlx5.h
index 1d488fb..459dc3d 100644
--- a/drivers/net/mlx5/mlx5.h
+++ b/drivers/net/mlx5/mlx5.h
@@ -203,6 +203,10 @@ void mlx5_allmulticast_disable(struct rte_eth_dev *);
 void mlx5_stats_get(struct rte_eth_dev *, struct rte_eth_stats *);
 void mlx5_stats_reset(struct rte_eth_dev *);
 
+/* mlx5_vlan.c */
+
+int mlx5_vlan_filter_set(struct rte_eth_dev *, uint16_t, int);
+
 /* mlx5_trigger.c */
 
 int mlx5_dev_start(struct rte_eth_dev *);
diff --git a/drivers/net/mlx5/mlx5_vlan.c b/drivers/net/mlx5/mlx5_vlan.c
new file mode 100644
index 0000000..60fe06b
--- /dev/null
+++ b/drivers/net/mlx5/mlx5_vlan.c
@@ -0,0 +1,166 @@
+/*-
+ *   BSD LICENSE
+ *
+ *   Copyright 2015 6WIND S.A.
+ *   Copyright 2015 Mellanox.
+ *
+ *   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 6WIND S.A. 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 <stddef.h>
+#include <errno.h>
+#include <assert.h>
+#include <stdint.h>
+
+/* DPDK headers don't like -pedantic. */
+#ifdef PEDANTIC
+#pragma GCC diagnostic ignored "-pedantic"
+#endif
+#include <rte_ethdev.h>
+#include <rte_common.h>
+#ifdef PEDANTIC
+#pragma GCC diagnostic error "-pedantic"
+#endif
+
+#include "mlx5_utils.h"
+#include "mlx5.h"
+
+/**
+ * Configure a VLAN filter.
+ *
+ * @param dev
+ *   Pointer to Ethernet device structure.
+ * @param vlan_id
+ *   VLAN ID to filter.
+ * @param on
+ *   Toggle filter.
+ *
+ * @return
+ *   0 on success, errno value on failure.
+ */
+static int
+vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
+{
+	struct priv *priv = dev->data->dev_private;
+	unsigned int i;
+	unsigned int j = -1;
+
+	DEBUG("%p: %s VLAN filter ID %" PRIu16,
+	      (void *)dev, (on ? "enable" : "disable"), vlan_id);
+	for (i = 0; (i != RTE_DIM(priv->vlan_filter)); ++i) {
+		if (!priv->vlan_filter[i].enabled) {
+			/* Unused index, remember it. */
+			j = i;
+			continue;
+		}
+		if (priv->vlan_filter[i].id != vlan_id)
+			continue;
+		/* This VLAN ID is already known, use its index. */
+		j = i;
+		break;
+	}
+	/* Check if there's room for another VLAN filter. */
+	if (j == (unsigned int)-1)
+		return ENOMEM;
+	/*
+	 * VLAN filters apply to all configured MAC addresses, flow
+	 * specifications must be reconfigured accordingly.
+	 */
+	priv->vlan_filter[j].id = vlan_id;
+	if ((on) && (!priv->vlan_filter[j].enabled)) {
+		/*
+		 * Filter is disabled, enable it.
+		 * Rehashing flows in all RX queues is necessary.
+		 */
+		if (priv->rss)
+			rxq_mac_addrs_del(&priv->rxq_parent);
+		else
+			for (i = 0; (i != priv->rxqs_n); ++i)
+				if ((*priv->rxqs)[i] != NULL)
+					rxq_mac_addrs_del((*priv->rxqs)[i]);
+		priv->vlan_filter[j].enabled = 1;
+		if (priv->started) {
+			if (priv->rss)
+				rxq_mac_addrs_add(&priv->rxq_parent);
+			else
+				for (i = 0; (i != priv->rxqs_n); ++i) {
+					if ((*priv->rxqs)[i] == NULL)
+						continue;
+					rxq_mac_addrs_add((*priv->rxqs)[i]);
+				}
+		}
+	} else if ((!on) && (priv->vlan_filter[j].enabled)) {
+		/*
+		 * Filter is enabled, disable it.
+		 * Rehashing flows in all RX queues is necessary.
+		 */
+		if (priv->rss)
+			rxq_mac_addrs_del(&priv->rxq_parent);
+		else
+			for (i = 0; (i != priv->rxqs_n); ++i)
+				if ((*priv->rxqs)[i] != NULL)
+					rxq_mac_addrs_del((*priv->rxqs)[i]);
+		priv->vlan_filter[j].enabled = 0;
+		if (priv->started) {
+			if (priv->rss)
+				rxq_mac_addrs_add(&priv->rxq_parent);
+			else
+				for (i = 0; (i != priv->rxqs_n); ++i) {
+					if ((*priv->rxqs)[i] == NULL)
+						continue;
+					rxq_mac_addrs_add((*priv->rxqs)[i]);
+				}
+		}
+	}
+	return 0;
+}
+
+/**
+ * DPDK callback to configure a VLAN filter.
+ *
+ * @param dev
+ *   Pointer to Ethernet device structure.
+ * @param vlan_id
+ *   VLAN ID to filter.
+ * @param on
+ *   Toggle filter.
+ *
+ * @return
+ *   0 on success, negative errno value on failure.
+ */
+int
+mlx5_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
+{
+	struct priv *priv = dev->data->dev_private;
+	int ret;
+
+	priv_lock(priv);
+	ret = vlan_filter_set(dev, vlan_id, on);
+	priv_unlock(priv);
+	assert(ret >= 0);
+	return -ret;
+}
-- 
2.1.0

^ permalink raw reply	[flat|nested] 32+ messages in thread

* [dpdk-dev] [PATCH 12/13] mlx5: add checksum offloading support
  2015-10-05 17:52 [dpdk-dev] [PATCH 00/13] Mellanox ConnectX-4 PMD (mlx5) Adrien Mazarguil
                   ` (10 preceding siblings ...)
  2015-10-05 17:53 ` [dpdk-dev] [PATCH 11/13] mlx5: add VLAN filtering Adrien Mazarguil
@ 2015-10-05 17:53 ` Adrien Mazarguil
  2015-10-05 17:53 ` [dpdk-dev] [PATCH 13/13] doc: add mlx5 documentation and release notes for version 2.2 Adrien Mazarguil
  2015-10-30 18:52 ` [dpdk-dev] [PATCH v2 00/13] Mellanox ConnectX-4 PMD (mlx5) Adrien Mazarguil
  13 siblings, 0 replies; 32+ messages in thread
From: Adrien Mazarguil @ 2015-10-05 17:53 UTC (permalink / raw)
  To: dev

This is the same implementation as mlx4.

Signed-off-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>
Signed-off-by: Nelio Laranjeiro <nelio.laranjeiro@6wind.com>
---
 drivers/net/mlx5/mlx5_rxq.c   | 14 +++++++
 drivers/net/mlx5/mlx5_rxtx.c  | 94 +++++++++++++++++++++++++++++++++++++++++++
 drivers/net/mlx5/mlx5_rxtx.h  |  2 +
 drivers/net/mlx5/mlx5_utils.h |  6 +++
 4 files changed, 116 insertions(+)

diff --git a/drivers/net/mlx5/mlx5_rxq.c b/drivers/net/mlx5/mlx5_rxq.c
index d44bb10..8cfad17 100644
--- a/drivers/net/mlx5/mlx5_rxq.c
+++ b/drivers/net/mlx5/mlx5_rxq.c
@@ -565,6 +565,15 @@ rxq_rehash(struct rte_eth_dev *dev, struct rxq *rxq)
 	/* Number of descriptors and mbufs currently allocated. */
 	desc_n = (tmpl.elts_n * (tmpl.sp ? MLX5_PMD_SGE_WR_N : 1));
 	mbuf_n = desc_n;
+	/* Toggle RX checksum offload if hardware supports it. */
+	if (priv->hw_csum) {
+		tmpl.csum = !!dev->data->dev_conf.rxmode.hw_ip_checksum;
+		rxq->csum = tmpl.csum;
+	}
+	if (priv->hw_csum_l2tun) {
+		tmpl.csum_l2tun = !!dev->data->dev_conf.rxmode.hw_ip_checksum;
+		rxq->csum_l2tun = tmpl.csum_l2tun;
+	}
 	/* Enable scattered packets support for this queue if necessary. */
 	if ((dev->data->dev_conf.rxmode.jumbo_frame) &&
 	    (dev->data->dev_conf.rxmode.max_rx_pkt_len >
@@ -788,6 +797,11 @@ rxq_setup(struct rte_eth_dev *dev, struct rxq *rxq, uint16_t desc,
 		rte_pktmbuf_tailroom(buf)) == tmpl.mb_len);
 	assert(rte_pktmbuf_headroom(buf) == RTE_PKTMBUF_HEADROOM);
 	rte_pktmbuf_free(buf);
+	/* Toggle RX checksum offload if hardware supports it. */
+	if (priv->hw_csum)
+		tmpl.csum = !!dev->data->dev_conf.rxmode.hw_ip_checksum;
+	if (priv->hw_csum_l2tun)
+		tmpl.csum_l2tun = !!dev->data->dev_conf.rxmode.hw_ip_checksum;
 	/* Enable scattered packets support for this queue if necessary. */
 	if ((dev->data->dev_conf.rxmode.jumbo_frame) &&
 	    (dev->data->dev_conf.rxmode.max_rx_pkt_len >
diff --git a/drivers/net/mlx5/mlx5_rxtx.c b/drivers/net/mlx5/mlx5_rxtx.c
index 960a3e5..668aff0 100644
--- a/drivers/net/mlx5/mlx5_rxtx.c
+++ b/drivers/net/mlx5/mlx5_rxtx.c
@@ -390,6 +390,17 @@ mlx5_tx_burst(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n)
 			++elts_comp;
 			send_flags |= IBV_EXP_QP_BURST_SIGNALED;
 		}
+		/* Should we enable HW CKSUM offload */
+		if (buf->ol_flags &
+		    (PKT_TX_IP_CKSUM | PKT_TX_TCP_CKSUM | PKT_TX_UDP_CKSUM)) {
+			send_flags |= IBV_EXP_QP_BURST_IP_CSUM;
+			/* HW does not support checksum offloads at arbitrary
+			 * offsets but automatically recognizes the packet
+			 * type. For inner L3/L4 checksums, only VXLAN (UDP)
+			 * tunnels are currently supported. */
+			if (RTE_ETH_IS_TUNNEL_PKT(buf->packet_type))
+				send_flags |= IBV_EXP_QP_BURST_TUNNEL;
+		}
 		if (likely(segs == 1)) {
 			uintptr_t addr;
 			uint32_t length;
@@ -491,6 +502,85 @@ stop:
 }
 
 /**
+ * Translate RX completion flags to packet type.
+ *
+ * @param flags
+ *   RX completion flags returned by poll_length_flags().
+ *
+ * @return
+ *   Packet type for struct rte_mbuf.
+ */
+static inline uint32_t
+rxq_cq_to_pkt_type(uint32_t flags)
+{
+	uint32_t pkt_type;
+
+	if (flags & IBV_EXP_CQ_RX_TUNNEL_PACKET)
+		pkt_type =
+			TRANSPOSE(flags,
+				  IBV_EXP_CQ_RX_OUTER_IPV4_PACKET,
+				  RTE_PTYPE_L3_IPV4) |
+			TRANSPOSE(flags,
+				  IBV_EXP_CQ_RX_OUTER_IPV6_PACKET,
+				  RTE_PTYPE_L3_IPV6) |
+			TRANSPOSE(flags,
+				  IBV_EXP_CQ_RX_IPV4_PACKET,
+				  RTE_PTYPE_INNER_L3_IPV4) |
+			TRANSPOSE(flags,
+				  IBV_EXP_CQ_RX_IPV6_PACKET,
+				  RTE_PTYPE_INNER_L3_IPV6);
+	else
+		pkt_type =
+			TRANSPOSE(flags,
+				  IBV_EXP_CQ_RX_IPV4_PACKET,
+				  RTE_PTYPE_L3_IPV4) |
+			TRANSPOSE(flags,
+				  IBV_EXP_CQ_RX_IPV6_PACKET,
+				  RTE_PTYPE_L3_IPV6);
+	return pkt_type;
+}
+
+/**
+ * Translate RX completion flags to offload flags.
+ *
+ * @param[in] rxq
+ *   Pointer to RX queue structure.
+ * @param flags
+ *   RX completion flags returned by poll_length_flags().
+ *
+ * @return
+ *   Offload flags (ol_flags) for struct rte_mbuf.
+ */
+static inline uint32_t
+rxq_cq_to_ol_flags(const struct rxq *rxq, uint32_t flags)
+{
+	uint32_t ol_flags = 0;
+
+	if (rxq->csum)
+		ol_flags |=
+			TRANSPOSE(~flags,
+				  IBV_EXP_CQ_RX_IP_CSUM_OK,
+				  PKT_RX_IP_CKSUM_BAD) |
+			TRANSPOSE(~flags,
+				  IBV_EXP_CQ_RX_TCP_UDP_CSUM_OK,
+				  PKT_RX_L4_CKSUM_BAD);
+	/*
+	 * PKT_RX_IP_CKSUM_BAD and PKT_RX_L4_CKSUM_BAD are used in place
+	 * of PKT_RX_EIP_CKSUM_BAD because the latter is not functional
+	 * (its value is 0).
+	 */
+	if ((flags & IBV_EXP_CQ_RX_TUNNEL_PACKET) && (rxq->csum_l2tun))
+		ol_flags |=
+			TRANSPOSE(~flags,
+				  IBV_EXP_CQ_RX_OUTER_IP_CSUM_OK,
+				  PKT_RX_IP_CKSUM_BAD) |
+			TRANSPOSE(~flags,
+				  IBV_EXP_CQ_RX_OUTER_TCP_UDP_CSUM_OK,
+				  PKT_RX_L4_CKSUM_BAD);
+	return ol_flags;
+}
+
+/**
  * DPDK callback for RX with scattered packets support.
  *
  * @param dpdk_rxq
@@ -669,6 +759,8 @@ mlx5_rx_burst_sp(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n)
 		NB_SEGS(pkt_buf) = j;
 		PORT(pkt_buf) = rxq->port_id;
 		PKT_LEN(pkt_buf) = pkt_buf_len;
+		pkt_buf->packet_type = rxq_cq_to_pkt_type(flags);
+		pkt_buf->ol_flags = rxq_cq_to_ol_flags(rxq, flags);
 
 		/* Return packet. */
 		*(pkts++) = pkt_buf;
@@ -828,6 +920,8 @@ mlx5_rx_burst(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n)
 		NEXT(seg) = NULL;
 		PKT_LEN(seg) = len;
 		DATA_LEN(seg) = len;
+		seg->packet_type = rxq_cq_to_pkt_type(flags);
+		seg->ol_flags = rxq_cq_to_ol_flags(rxq, flags);
 
 		/* Return packet. */
 		*(pkts++) = seg;
diff --git a/drivers/net/mlx5/mlx5_rxtx.h b/drivers/net/mlx5/mlx5_rxtx.h
index 228dff6..0eb1e98 100644
--- a/drivers/net/mlx5/mlx5_rxtx.h
+++ b/drivers/net/mlx5/mlx5_rxtx.h
@@ -119,6 +119,8 @@ struct rxq {
 		struct rxq_elt (*no_sp)[]; /* RX elements. */
 	} elts;
 	unsigned int sp:1; /* Use scattered RX elements. */
+	unsigned int csum:1; /* Enable checksum offloading. */
+	unsigned int csum_l2tun:1; /* Same for L2 tunnels. */
 	uint32_t mb_len; /* Length of a mp-issued mbuf. */
 	struct mlx5_rxq_stats stats; /* RX queue counters. */
 	unsigned int socket; /* CPU socket ID for allocations. */
diff --git a/drivers/net/mlx5/mlx5_utils.h b/drivers/net/mlx5/mlx5_utils.h
index e48e6b6..8ff075b 100644
--- a/drivers/net/mlx5/mlx5_utils.h
+++ b/drivers/net/mlx5/mlx5_utils.h
@@ -149,6 +149,12 @@ pmd_drv_log_basename(const char *s)
 #define NB_SEGS(m) ((m)->nb_segs)
 #define PORT(m) ((m)->port)
 
+/* Transpose flags. Useful to convert IBV to DPDK flags. */
+#define TRANSPOSE(val, from, to) \
+	(((from) >= (to)) ? \
+	 (((val) & (from)) / ((from) / (to))) : \
+	 (((val) & (from)) * ((to) / (from))))
+
 /* Allocate a buffer on the stack and fill it with a printf format string. */
 #define MKSTR(name, ...) \
 	char name[snprintf(NULL, 0, __VA_ARGS__) + 1]; \
-- 
2.1.0

^ permalink raw reply	[flat|nested] 32+ messages in thread

* [dpdk-dev] [PATCH 13/13] doc: add mlx5 documentation and release notes for version 2.2
  2015-10-05 17:52 [dpdk-dev] [PATCH 00/13] Mellanox ConnectX-4 PMD (mlx5) Adrien Mazarguil
                   ` (11 preceding siblings ...)
  2015-10-05 17:53 ` [dpdk-dev] [PATCH 12/13] mlx5: add checksum offloading support Adrien Mazarguil
@ 2015-10-05 17:53 ` Adrien Mazarguil
  2015-10-30 18:52 ` [dpdk-dev] [PATCH v2 00/13] Mellanox ConnectX-4 PMD (mlx5) Adrien Mazarguil
  13 siblings, 0 replies; 32+ messages in thread
From: Adrien Mazarguil @ 2015-10-05 17:53 UTC (permalink / raw)
  To: dev

Signed-off-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>
---
 doc/guides/nics/mlx5.rst             | 308 +++++++++++++++++++++++++++++++++++
 doc/guides/rel_notes/release_2_2.rst |   8 +
 2 files changed, 316 insertions(+)
 create mode 100644 doc/guides/nics/mlx5.rst

diff --git a/doc/guides/nics/mlx5.rst b/doc/guides/nics/mlx5.rst
new file mode 100644
index 0000000..fdb621c
--- /dev/null
+++ b/doc/guides/nics/mlx5.rst
@@ -0,0 +1,308 @@
+..  BSD LICENSE
+    Copyright 2015 6WIND S.A.
+
+    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 6WIND S.A. 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.
+
+MLX5 poll mode driver
+=====================
+
+The MLX5 poll mode driver library (**librte_pmd_mlx5**) provides support for
+**Mellanox ConnectX-4 EN** and **Mellanox ConnectX-4 Lx EN** families of
+10/25/40/50/100 Gb/s adapters as well as their virtual functions (VF) in
+SR-IOV context.
+
+Information and documentation about these adapters can be found on the
+`Mellanox website <http://www.mellanox.com>`__. Help is also provided by the
+`Mellanox community <http://community.mellanox.com/welcome>`__.
+
+There is also a `section dedicated to this poll mode driver
+<http://www.mellanox.com/page/products_dyn?product_family=209&mtag=pmd_for_dpdk>`__.
+
+.. note::
+
+   Due to external dependencies, this driver is disabled by default. It must
+   be enabled manually by setting ``CONFIG_RTE_LIBRTE_MLX5_PMD=y`` and
+   recompiling DPDK.
+
+.. warning::
+
+   ``CONFIG_RTE_BUILD_COMBINE_LIBS`` with ``CONFIG_RTE_BUILD_SHARED_LIB``
+   is not supported and thus the compilation will fail with this configuration.
+
+Implementation details
+----------------------
+
+Besides its dependency on libibverbs (that implies libmlx5 and associated
+kernel support), librte_pmd_mlx5 relies heavily on system calls for control
+operations such as querying/updating the MTU and flow control parameters.
+
+For security reasons and robustness, this driver only deals with virtual
+memory addresses. The way resources allocations are handled by the kernel
+combined with hardware specifications that allow it to handle virtual memory
+addresses directly ensure that DPDK applications cannot access random
+physical memory (or memory that does not belong to the current process).
+
+This capability allows the PMD to coexist with kernel network interfaces
+which remain functional, although they stop receiving unicast packets as
+long as they share the same MAC address.
+
+Enabling librte_pmd_mlx5 causes DPDK applications to be linked against
+libibverbs.
+
+Configuration
+-------------
+
+Compilation options
+~~~~~~~~~~~~~~~~~~~
+
+These options can be modified in the ``.config`` file.
+
+- ``CONFIG_RTE_LIBRTE_MLX5_PMD`` (default **n**)
+
+  Toggle compilation of librte_pmd_mlx5 itself.
+
+- ``CONFIG_RTE_LIBRTE_MLX5_DEBUG`` (default **n**)
+
+  Toggle debugging code and stricter compilation flags. Enabling this option
+  adds additional run-time checks and debugging messages at the cost of
+  lower performance.
+
+- ``CONFIG_RTE_LIBRTE_MLX5_SGE_WR_N`` (default **4**)
+
+  Number of scatter/gather elements (SGEs) per work request (WR). Lowering
+  this number improves performance but also limits the ability to receive
+  scattered packets (packets that do not fit a single mbuf). The default
+  value is a safe tradeoff.
+
+- ``CONFIG_RTE_LIBRTE_MLX5_MAX_INLINE`` (default **0**)
+
+  Amount of data to be inlined during TX operations. Improves latency but
+  lowers throughput.
+
+- ``CONFIG_RTE_LIBRTE_MLX5_TX_MP_CACHE`` (default **8**)
+
+  Maximum number of cached memory pools (MPs) per TX queue. Each MP from
+  which buffers are to be transmitted must be associated to memory regions
+  (MRs). This is a slow operation that must be cached.
+
+  This value is always 1 for RX queues since they use a single MP.
+
+Run-time configuration
+~~~~~~~~~~~~~~~~~~~~~~
+
+- librte_pmd_mlx5 brings kernel network interfaces up during initialization
+  because it is affected by their state. Forcing them down prevents packets
+  reception.
+
+- **ethtool** operations on related kernel interfaces also affect the PMD.
+
+Prerequisites
+-------------
+
+This driver relies on external libraries and kernel drivers for resources
+allocations and initialization. The following dependencies are not part of
+DPDK and must be installed separately:
+
+- **libibverbs**
+
+  User space Verbs framework used by librte_pmd_mlx5. This library provides
+  a generic interface between the kernel and low-level user space drivers
+  such as libmlx5.
+
+  It allows slow and privileged operations (context initialization, hardware
+  resources allocations) to be managed by the kernel and fast operations to
+  never leave user space.
+
+- **libmlx5**
+
+  Low-level user space driver library for Mellanox ConnectX-4 devices,
+  it is automatically loaded by libibverbs.
+
+  This library basically implements send/receive calls to the hardware
+  queues.
+
+- **Kernel modules** (mlnx-ofed-kernel)
+
+  They provide the kernel-side Verbs API and low level device drivers that
+  manage actual hardware initialization and resources sharing with user
+  space processes.
+
+  Unlike most other PMDs, these modules must remain loaded and bound to
+  their devices:
+
+  - mlx5_core: hardware driver managing Mellanox ConnectX-4 devices and
+    related Ethernet kernel network devices.
+  - mlx5_ib: InifiniBand device driver.
+  - ib_uverbs: user space driver for Verbs (entry point for libibverbs).
+
+- **Firmware update**
+
+  Mellanox OFED releases include firmware updates for ConnectX-4 adapters.
+
+  Because each release provides new features, these updates must be applied to
+  match the kernel modules and libraries they come with.
+
+.. note::
+
+   Both libraries are BSD and GPL licensed. Linux kernel modules are GPL
+   licensed.
+
+Getting Mellanox OFED
+~~~~~~~~~~~~~~~~~~~~~
+
+While these libraries and kernel modules are available on OpenFabrics
+Alliance's `website <https://www.openfabrics.org/>`__ and provided by package
+managers on most distributions, this PMD requires Ethernet extensions that
+may not be supported at the moment (this is a work in progress).
+
+`Mellanox OFED
+<http://www.mellanox.com/page/products_dyn?product_family=26&mtag=linux>`__
+includes the necessary support and should be used in the meantime. For DPDK,
+only libibverbs, libmlx5, mlnx-ofed-kernel packages and firmware updates are
+required from that distribution.
+
+.. note::
+
+   Several versions of Mellanox OFED are available. Installing the version
+   this DPDK release was developed and tested against is strongly
+   recommended. Please check the `prerequisites`_.
+
+Usage example
+-------------
+
+This section demonstrates how to launch **testpmd** with Mellanox ConnectX-4
+devices managed by librte_pmd_mlx5.
+
+#. Load the kernel modules:
+
+   .. code-block:: console
+
+      modprobe -a ib_uverbs mlx5_core mlx5_ib
+
+   .. note::
+
+      User space I/O kernel modules (uio and igb_uio) are not used and do
+      not have to be loaded.
+
+#. Make sure Ethernet interfaces are in working order and linked to kernel
+   verbs. Related sysfs entries should be present:
+
+   .. code-block:: console
+
+      ls -d /sys/class/net/*/device/infiniband_verbs/uverbs* | cut -d / -f 5
+
+   Example output:
+
+   .. code-block:: console
+
+      eth30
+      eth31
+      eth32
+      eth33
+
+#. Optionally, retrieve their PCI bus addresses for whitelisting:
+
+   .. code-block:: console
+
+      {
+          for intf in eth2 eth3 eth4 eth5;
+          do
+              (cd "/sys/class/net/${intf}/device/" && pwd -P);
+          done;
+      } |
+      sed -n 's,.*/\(.*\),-w \1,p'
+
+   Example output:
+
+   .. code-block:: console
+
+      -w 0000:05:00.1
+      -w 0000:06:00.0
+      -w 0000:06:00.1
+      -w 0000:05:00.0
+
+#. Request huge pages:
+
+   .. code-block:: console
+
+      echo 1024 > /sys/kernel/mm/hugepages/hugepages-2048kB/nr_hugepages/nr_hugepages
+
+#. Start testpmd with basic parameters:
+
+   .. code-block:: console
+
+      testpmd -c 0xff00 -n 4 -w 05:00.0 -w 05:00.1 -w 06:00.0 -w 06:00.1 -- --rxq=2 --txq=2 -i
+
+   Example output:
+
+   .. code-block:: console
+
+      [...]
+      EAL: PCI device 0000:05:00.0 on NUMA socket 0
+      EAL:   probe driver: 15b3:1013 librte_pmd_mlx5
+      PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_0" (VF: false)
+      PMD: librte_pmd_mlx5: 1 port(s) detected
+      PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fe
+      EAL: PCI device 0000:05:00.1 on NUMA socket 0
+      EAL:   probe driver: 15b3:1013 librte_pmd_mlx5
+      PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_1" (VF: false)
+      PMD: librte_pmd_mlx5: 1 port(s) detected
+      PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:ff
+      EAL: PCI device 0000:06:00.0 on NUMA socket 0
+      EAL:   probe driver: 15b3:1013 librte_pmd_mlx5
+      PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_2" (VF: false)
+      PMD: librte_pmd_mlx5: 1 port(s) detected
+      PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fa
+      EAL: PCI device 0000:06:00.1 on NUMA socket 0
+      EAL:   probe driver: 15b3:1013 librte_pmd_mlx5
+      PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_3" (VF: false)
+      PMD: librte_pmd_mlx5: 1 port(s) detected
+      PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fb
+      Interactive-mode selected
+      Configuring Port 0 (socket 0)
+      PMD: librte_pmd_mlx5: 0x8cba80: TX queues number update: 0 -> 2
+      PMD: librte_pmd_mlx5: 0x8cba80: RX queues number update: 0 -> 2
+      Port 0: E4:1D:2D:E7:0C:FE
+      Configuring Port 1 (socket 0)
+      PMD: librte_pmd_mlx5: 0x8ccac8: TX queues number update: 0 -> 2
+      PMD: librte_pmd_mlx5: 0x8ccac8: RX queues number update: 0 -> 2
+      Port 1: E4:1D:2D:E7:0C:FF
+      Configuring Port 2 (socket 0)
+      PMD: librte_pmd_mlx5: 0x8cdb10: TX queues number update: 0 -> 2
+      PMD: librte_pmd_mlx5: 0x8cdb10: RX queues number update: 0 -> 2
+      Port 2: E4:1D:2D:E7:0C:FA
+      Configuring Port 3 (socket 0)
+      PMD: librte_pmd_mlx5: 0x8ceb58: TX queues number update: 0 -> 2
+      PMD: librte_pmd_mlx5: 0x8ceb58: RX queues number update: 0 -> 2
+      Port 3: E4:1D:2D:E7:0C:FB
+      Checking link statuses...
+      Port 0 Link Up - speed 40000 Mbps - full-duplex
+      Port 1 Link Up - speed 40000 Mbps - full-duplex
+      Port 2 Link Up - speed 10000 Mbps - full-duplex
+      Port 3 Link Up - speed 10000 Mbps - full-duplex
+      Done
+      testpmd>
diff --git a/doc/guides/rel_notes/release_2_2.rst b/doc/guides/rel_notes/release_2_2.rst
index 5687676..e562030 100644
--- a/doc/guides/rel_notes/release_2_2.rst
+++ b/doc/guides/rel_notes/release_2_2.rst
@@ -4,6 +4,14 @@ DPDK Release 2.2
 New Features
 ------------
 
+* **Added support for Mellanox ConnectX-4 adapters (mlx5).**
+
+  The mlx5 poll-mode driver implements support for Mellanox ConnectX-4 EN
+  and Mellanox ConnectX-4 Lx EN families of 10/25/40/50/100 Gb/s adapters.
+
+  Like mlx4, this PMD is only available for Linux and is disabled by default
+  due to external dependencies (libibverbs and libmlx5).
+
 
 Resolved Issues
 ---------------
-- 
2.1.0

^ permalink raw reply	[flat|nested] 32+ messages in thread

* [dpdk-dev] [PATCH v2 00/13] Mellanox ConnectX-4 PMD (mlx5)
  2015-10-05 17:52 [dpdk-dev] [PATCH 00/13] Mellanox ConnectX-4 PMD (mlx5) Adrien Mazarguil
                   ` (12 preceding siblings ...)
  2015-10-05 17:53 ` [dpdk-dev] [PATCH 13/13] doc: add mlx5 documentation and release notes for version 2.2 Adrien Mazarguil
@ 2015-10-30 18:52 ` Adrien Mazarguil
  2015-10-30 18:52   ` [dpdk-dev] [PATCH v2 01/13] mlx5: new poll-mode driver for Mellanox ConnectX-4 adapters Adrien Mazarguil
                     ` (13 more replies)
  13 siblings, 14 replies; 32+ messages in thread
From: Adrien Mazarguil @ 2015-10-30 18:52 UTC (permalink / raw)
  To: dev

This PMD adds basic support for Mellanox ConnectX-4 (mlx5) families of
10/25/40/50/100 Gb/s adapters through the Verbs framework.

Its design is very similar to that of mlx4 from which most of its code is
borrowed without the mistake of putting it all in a single huge file.

It is disabled by default due to its dependency on libibverbs.

Changes in v2:
- Removed useless port inactive warning.
- Simplified code by replacing configured MAC addresses RX queue bit-field
  with flow pointer checks.
- Replaced allmulti/promisc status bits with request bits to fix
  inconsistencies when restoring these modes.
- Updated comments about maximum number of MAC addresses and VLAN filters.
- Improved performance with better prefetching.
- Fixed deadlock in case of error during port start.
- Simplified VLAN filtering configuration storage using a basic list instead
  of a table (with holes).

Adrien Mazarguil (13):
  mlx5: new poll-mode driver for Mellanox ConnectX-4 adapters
  mlx5: add non-scattered TX and RX support
  mlx5: add MAC handling
  mlx5: add device configure/start/stop
  mlx5: add support for scattered RX and TX buffers
  mlx5: add MTU configuration support
  mlx5: add software counters and related callbacks
  mlx5: add promiscuous and allmulticast RX modes
  mlx5: add link update device operation
  mlx5: add flow control device operations
  mlx5: add VLAN filtering
  mlx5: add checksum offloading support
  doc: add mlx5 documentation and release notes for version 2.2

 MAINTAINERS                               |    4 +
 config/common_bsdapp                      |    9 +
 config/common_linuxapp                    |    9 +
 doc/guides/nics/mlx5.rst                  |  308 +++++++++
 doc/guides/rel_notes/release_2_2.rst      |    9 +
 drivers/net/Makefile                      |    1 +
 drivers/net/mlx5/Makefile                 |  128 ++++
 drivers/net/mlx5/mlx5.c                   |  553 +++++++++++++++
 drivers/net/mlx5/mlx5.h                   |  212 ++++++
 drivers/net/mlx5/mlx5_defs.h              |   78 +++
 drivers/net/mlx5/mlx5_ethdev.c            |  844 +++++++++++++++++++++++
 drivers/net/mlx5/mlx5_mac.c               |  464 +++++++++++++
 drivers/net/mlx5/mlx5_rxmode.c            |  311 +++++++++
 drivers/net/mlx5/mlx5_rxq.c               | 1064 +++++++++++++++++++++++++++++
 drivers/net/mlx5/mlx5_rxtx.c              | 1009 +++++++++++++++++++++++++++
 drivers/net/mlx5/mlx5_rxtx.h              |  194 ++++++
 drivers/net/mlx5/mlx5_stats.c             |  144 ++++
 drivers/net/mlx5/mlx5_trigger.c           |  154 +++++
 drivers/net/mlx5/mlx5_txq.c               |  513 ++++++++++++++
 drivers/net/mlx5/mlx5_utils.h             |  166 +++++
 drivers/net/mlx5/mlx5_vlan.c              |  156 +++++
 drivers/net/mlx5/rte_pmd_mlx5_version.map |    3 +
 mk/rte.app.mk                             |    5 +
 23 files changed, 6338 insertions(+)
 create mode 100644 doc/guides/nics/mlx5.rst
 create mode 100644 drivers/net/mlx5/Makefile
 create mode 100644 drivers/net/mlx5/mlx5.c
 create mode 100644 drivers/net/mlx5/mlx5.h
 create mode 100644 drivers/net/mlx5/mlx5_defs.h
 create mode 100644 drivers/net/mlx5/mlx5_ethdev.c
 create mode 100644 drivers/net/mlx5/mlx5_mac.c
 create mode 100644 drivers/net/mlx5/mlx5_rxmode.c
 create mode 100644 drivers/net/mlx5/mlx5_rxq.c
 create mode 100644 drivers/net/mlx5/mlx5_rxtx.c
 create mode 100644 drivers/net/mlx5/mlx5_rxtx.h
 create mode 100644 drivers/net/mlx5/mlx5_stats.c
 create mode 100644 drivers/net/mlx5/mlx5_trigger.c
 create mode 100644 drivers/net/mlx5/mlx5_txq.c
 create mode 100644 drivers/net/mlx5/mlx5_utils.h
 create mode 100644 drivers/net/mlx5/mlx5_vlan.c
 create mode 100644 drivers/net/mlx5/rte_pmd_mlx5_version.map

-- 
2.1.0

^ permalink raw reply	[flat|nested] 32+ messages in thread

* [dpdk-dev] [PATCH v2 01/13] mlx5: new poll-mode driver for Mellanox ConnectX-4 adapters
  2015-10-30 18:52 ` [dpdk-dev] [PATCH v2 00/13] Mellanox ConnectX-4 PMD (mlx5) Adrien Mazarguil
@ 2015-10-30 18:52   ` Adrien Mazarguil
  2015-10-30 18:52   ` [dpdk-dev] [PATCH v2 02/13] mlx5: add non-scattered TX and RX support Adrien Mazarguil
                     ` (12 subsequent siblings)
  13 siblings, 0 replies; 32+ messages in thread
From: Adrien Mazarguil @ 2015-10-30 18:52 UTC (permalink / raw)
  To: dev

In its current state, this driver implements the bare minimum to initialize
itself and Mellanox ConnectX-4 adapters without doing anything else
(no RX/TX for instance). It is disabled by default since it is based on the
mlx4 driver and also depends on libibverbs.

Signed-off-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>
Signed-off-by: Nelio Laranjeiro <nelio.laranjeiro@6wind.com>
Signed-off-by: Or Ami <ora@mellanox.com>
---
 MAINTAINERS                               |   4 +
 config/common_bsdapp                      |   6 +
 config/common_linuxapp                    |   6 +
 drivers/net/Makefile                      |   1 +
 drivers/net/mlx5/Makefile                 | 109 +++++++
 drivers/net/mlx5/mlx5.c                   | 496 ++++++++++++++++++++++++++++++
 drivers/net/mlx5/mlx5.h                   | 147 +++++++++
 drivers/net/mlx5/mlx5_defs.h              |  43 +++
 drivers/net/mlx5/mlx5_ethdev.c            | 420 +++++++++++++++++++++++++
 drivers/net/mlx5/mlx5_mac.c               | 150 +++++++++
 drivers/net/mlx5/mlx5_utils.h             | 149 +++++++++
 drivers/net/mlx5/rte_pmd_mlx5_version.map |   3 +
 mk/rte.app.mk                             |   5 +
 13 files changed, 1539 insertions(+)
 create mode 100644 drivers/net/mlx5/Makefile
 create mode 100644 drivers/net/mlx5/mlx5.c
 create mode 100644 drivers/net/mlx5/mlx5.h
 create mode 100644 drivers/net/mlx5/mlx5_defs.h
 create mode 100644 drivers/net/mlx5/mlx5_ethdev.c
 create mode 100644 drivers/net/mlx5/mlx5_mac.c
 create mode 100644 drivers/net/mlx5/mlx5_utils.h
 create mode 100644 drivers/net/mlx5/rte_pmd_mlx5_version.map

diff --git a/MAINTAINERS b/MAINTAINERS
index 080a8e8..9d11055 100644
--- a/MAINTAINERS
+++ b/MAINTAINERS
@@ -255,6 +255,10 @@ M: Adrien Mazarguil <adrien.mazarguil@6wind.com>
 F: drivers/net/mlx4/
 F: doc/guides/nics/mlx4.rst
 
+Mellanox mlx5
+M: Adrien Mazarguil <adrien.mazarguil@6wind.com>
+F: drivers/net/mlx5/
+
 RedHat virtio
 M: Huawei Xie <huawei.xie@intel.com>
 M: Changchun Ouyang <changchun.ouyang@intel.com>
diff --git a/config/common_bsdapp b/config/common_bsdapp
index b37dcf4..1e6885f 100644
--- a/config/common_bsdapp
+++ b/config/common_bsdapp
@@ -214,6 +214,12 @@ CONFIG_RTE_LIBRTE_MLX4_TX_MP_CACHE=8
 CONFIG_RTE_LIBRTE_MLX4_SOFT_COUNTERS=1
 
 #
+# Compile burst-oriented Mellanox ConnectX-4 (MLX5) PMD
+#
+CONFIG_RTE_LIBRTE_MLX5_PMD=n
+CONFIG_RTE_LIBRTE_MLX5_DEBUG=n
+
+#
 # Compile burst-oriented Broadcom PMD driver
 #
 CONFIG_RTE_LIBRTE_BNX2X_PMD=n
diff --git a/config/common_linuxapp b/config/common_linuxapp
index 0de43d5..7da7ba7 100644
--- a/config/common_linuxapp
+++ b/config/common_linuxapp
@@ -212,6 +212,12 @@ CONFIG_RTE_LIBRTE_MLX4_TX_MP_CACHE=8
 CONFIG_RTE_LIBRTE_MLX4_SOFT_COUNTERS=1
 
 #
+# Compile burst-oriented Mellanox ConnectX-4 (MLX5) PMD
+#
+CONFIG_RTE_LIBRTE_MLX5_PMD=n
+CONFIG_RTE_LIBRTE_MLX5_DEBUG=n
+
+#
 # Compile burst-oriented Broadcom PMD driver
 #
 CONFIG_RTE_LIBRTE_BNX2X_PMD=n
diff --git a/drivers/net/Makefile b/drivers/net/Makefile
index 5ebf963..6da1ce2 100644
--- a/drivers/net/Makefile
+++ b/drivers/net/Makefile
@@ -41,6 +41,7 @@ DIRS-$(CONFIG_RTE_LIBRTE_FM10K_PMD) += fm10k
 DIRS-$(CONFIG_RTE_LIBRTE_I40E_PMD) += i40e
 DIRS-$(CONFIG_RTE_LIBRTE_IXGBE_PMD) += ixgbe
 DIRS-$(CONFIG_RTE_LIBRTE_MLX4_PMD) += mlx4
+DIRS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5
 DIRS-$(CONFIG_RTE_LIBRTE_MPIPE_PMD) += mpipe
 DIRS-$(CONFIG_RTE_LIBRTE_PMD_NULL) += null
 DIRS-$(CONFIG_RTE_LIBRTE_PMD_PCAP) += pcap
diff --git a/drivers/net/mlx5/Makefile b/drivers/net/mlx5/Makefile
new file mode 100644
index 0000000..6e63073
--- /dev/null
+++ b/drivers/net/mlx5/Makefile
@@ -0,0 +1,109 @@
+#   BSD LICENSE
+#
+#   Copyright 2015 6WIND S.A.
+#   Copyright 2015 Mellanox.
+#
+#   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 6WIND S.A. 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 $(RTE_SDK)/mk/rte.vars.mk
+
+ifeq ($(CONFIG_RTE_BUILD_COMBINE_LIBS)$(CONFIG_RTE_BUILD_SHARED_LIB),yy)
+all:
+	@echo 'MLX5: Not supported in a combined shared library'
+	@false
+endif
+
+# Library name.
+LIB = librte_pmd_mlx5.a
+
+# Sources.
+SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5.c
+SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5_ethdev.c
+SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5_mac.c
+
+# Dependencies.
+DEPDIRS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += lib/librte_ether
+DEPDIRS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += lib/librte_mbuf
+DEPDIRS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += lib/librte_eal
+DEPDIRS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += lib/librte_mempool
+
+# Basic CFLAGS.
+CFLAGS += -O3
+CFLAGS += -std=gnu99 -Wall -Wextra
+CFLAGS += -g
+CFLAGS += -I.
+CFLAGS += -D_XOPEN_SOURCE=600
+CFLAGS += $(WERROR_FLAGS)
+LDLIBS += -libverbs
+
+# A few warnings cannot be avoided in external headers.
+CFLAGS += -Wno-error=cast-qual
+
+EXPORT_MAP := rte_pmd_mlx5_version.map
+LIBABIVER := 1
+
+# DEBUG which is usually provided on the command-line may enable
+# CONFIG_RTE_LIBRTE_MLX5_DEBUG.
+ifeq ($(DEBUG),1)
+CONFIG_RTE_LIBRTE_MLX5_DEBUG := y
+endif
+
+# User-defined CFLAGS.
+ifeq ($(CONFIG_RTE_LIBRTE_MLX5_DEBUG),y)
+CFLAGS += -pedantic -UNDEBUG -DPEDANTIC
+else
+CFLAGS += -DNDEBUG -UPEDANTIC
+endif
+
+include $(RTE_SDK)/mk/rte.lib.mk
+
+# Generate and clean-up mlx5_autoconf.h.
+
+export CC CFLAGS CPPFLAGS EXTRA_CFLAGS EXTRA_CPPFLAGS
+export AUTO_CONFIG_CFLAGS = -Wno-error
+
+ifndef V
+AUTOCONF_OUTPUT := >/dev/null
+endif
+
+mlx5_autoconf.h: $(RTE_SDK)/scripts/auto-config-h.sh
+	$Q $(RM) -f -- '$@'
+	$Q sh -- '$<' '$@' \
+		RSS_SUPPORT \
+		infiniband/verbs.h \
+		enum IBV_EXP_DEVICE_UD_RSS $(AUTOCONF_OUTPUT)
+	$Q sh -- '$<' '$@' \
+		HAVE_EXP_QUERY_DEVICE \
+		infiniband/verbs.h \
+		type 'struct ibv_exp_device_attr' $(AUTOCONF_OUTPUT)
+
+mlx5.o: mlx5_autoconf.h
+
+clean_mlx5: FORCE
+	$Q rm -f -- mlx5_autoconf.h
+
+clean: clean_mlx5
diff --git a/drivers/net/mlx5/mlx5.c b/drivers/net/mlx5/mlx5.c
new file mode 100644
index 0000000..6df486b
--- /dev/null
+++ b/drivers/net/mlx5/mlx5.c
@@ -0,0 +1,496 @@
+/*-
+ *   BSD LICENSE
+ *
+ *   Copyright 2015 6WIND S.A.
+ *   Copyright 2015 Mellanox.
+ *
+ *   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 6WIND S.A. 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 <stddef.h>
+#include <unistd.h>
+#include <string.h>
+#include <assert.h>
+#include <stdint.h>
+#include <stdlib.h>
+#include <net/if.h>
+
+/* Verbs header. */
+/* ISO C doesn't support unnamed structs/unions, disabling -pedantic. */
+#ifdef PEDANTIC
+#pragma GCC diagnostic ignored "-pedantic"
+#endif
+#include <infiniband/verbs.h>
+#ifdef PEDANTIC
+#pragma GCC diagnostic error "-pedantic"
+#endif
+
+/* DPDK headers don't like -pedantic. */
+#ifdef PEDANTIC
+#pragma GCC diagnostic ignored "-pedantic"
+#endif
+#include <rte_malloc.h>
+#include <rte_ethdev.h>
+#include <rte_pci.h>
+#include <rte_common.h>
+#ifdef PEDANTIC
+#pragma GCC diagnostic error "-pedantic"
+#endif
+
+#include "mlx5.h"
+#include "mlx5_utils.h"
+#include "mlx5_autoconf.h"
+
+/**
+ * DPDK callback to close the device.
+ *
+ * Destroy all queues and objects, free memory.
+ *
+ * @param dev
+ *   Pointer to Ethernet device structure.
+ */
+static void
+mlx5_dev_close(struct rte_eth_dev *dev)
+{
+	struct priv *priv = dev->data->dev_private;
+
+	priv_lock(priv);
+	DEBUG("%p: closing device \"%s\"",
+	      (void *)dev,
+	      ((priv->ctx != NULL) ? priv->ctx->device->name : ""));
+	if (priv->pd != NULL) {
+		assert(priv->ctx != NULL);
+		claim_zero(ibv_dealloc_pd(priv->pd));
+		claim_zero(ibv_close_device(priv->ctx));
+	} else
+		assert(priv->ctx == NULL);
+	priv_unlock(priv);
+	memset(priv, 0, sizeof(*priv));
+}
+
+static const struct eth_dev_ops mlx5_dev_ops = {
+	.dev_close = mlx5_dev_close,
+};
+
+static struct {
+	struct rte_pci_addr pci_addr; /* associated PCI address */
+	uint32_t ports; /* physical ports bitfield. */
+} mlx5_dev[32];
+
+/**
+ * Get device index in mlx5_dev[] from PCI bus address.
+ *
+ * @param[in] pci_addr
+ *   PCI bus address to look for.
+ *
+ * @return
+ *   mlx5_dev[] index on success, -1 on failure.
+ */
+static int
+mlx5_dev_idx(struct rte_pci_addr *pci_addr)
+{
+	unsigned int i;
+	int ret = -1;
+
+	assert(pci_addr != NULL);
+	for (i = 0; (i != RTE_DIM(mlx5_dev)); ++i) {
+		if ((mlx5_dev[i].pci_addr.domain == pci_addr->domain) &&
+		    (mlx5_dev[i].pci_addr.bus == pci_addr->bus) &&
+		    (mlx5_dev[i].pci_addr.devid == pci_addr->devid) &&
+		    (mlx5_dev[i].pci_addr.function == pci_addr->function))
+			return i;
+		if ((mlx5_dev[i].ports == 0) && (ret == -1))
+			ret = i;
+	}
+	return ret;
+}
+
+static struct eth_driver mlx5_driver;
+
+/**
+ * DPDK callback to register a PCI device.
+ *
+ * This function creates an Ethernet device for each port of a given
+ * PCI device.
+ *
+ * @param[in] pci_drv
+ *   PCI driver structure (mlx5_driver).
+ * @param[in] pci_dev
+ *   PCI device information.
+ *
+ * @return
+ *   0 on success, negative errno value on failure.
+ */
+static int
+mlx5_pci_devinit(struct rte_pci_driver *pci_drv, struct rte_pci_device *pci_dev)
+{
+	struct ibv_device **list;
+	struct ibv_device *ibv_dev;
+	int err = 0;
+	struct ibv_context *attr_ctx = NULL;
+	struct ibv_device_attr device_attr;
+	unsigned int vf;
+	int idx;
+	int i;
+
+	(void)pci_drv;
+	assert(pci_drv == &mlx5_driver.pci_drv);
+	/* Get mlx5_dev[] index. */
+	idx = mlx5_dev_idx(&pci_dev->addr);
+	if (idx == -1) {
+		ERROR("this driver cannot support any more adapters");
+		return -ENOMEM;
+	}
+	DEBUG("using driver device index %d", idx);
+
+	/* Save PCI address. */
+	mlx5_dev[idx].pci_addr = pci_dev->addr;
+	list = ibv_get_device_list(&i);
+	if (list == NULL) {
+		assert(errno);
+		if (errno == ENOSYS) {
+			WARN("cannot list devices, is ib_uverbs loaded?");
+			return 0;
+		}
+		return -errno;
+	}
+	assert(i >= 0);
+	/*
+	 * For each listed device, check related sysfs entry against
+	 * the provided PCI ID.
+	 */
+	while (i != 0) {
+		struct rte_pci_addr pci_addr;
+
+		--i;
+		DEBUG("checking device \"%s\"", list[i]->name);
+		if (mlx5_ibv_device_to_pci_addr(list[i], &pci_addr))
+			continue;
+		if ((pci_dev->addr.domain != pci_addr.domain) ||
+		    (pci_dev->addr.bus != pci_addr.bus) ||
+		    (pci_dev->addr.devid != pci_addr.devid) ||
+		    (pci_dev->addr.function != pci_addr.function))
+			continue;
+		vf = ((pci_dev->id.device_id ==
+		       PCI_DEVICE_ID_MELLANOX_CONNECTX4VF) ||
+		      (pci_dev->id.device_id ==
+		       PCI_DEVICE_ID_MELLANOX_CONNECTX4LXVF));
+		INFO("PCI information matches, using device \"%s\" (VF: %s)",
+		     list[i]->name, (vf ? "true" : "false"));
+		attr_ctx = ibv_open_device(list[i]);
+		err = errno;
+		break;
+	}
+	if (attr_ctx == NULL) {
+		ibv_free_device_list(list);
+		switch (err) {
+		case 0:
+			WARN("cannot access device, is mlx5_ib loaded?");
+			return 0;
+		case EINVAL:
+			WARN("cannot use device, are drivers up to date?");
+			return 0;
+		}
+		assert(err > 0);
+		return -err;
+	}
+	ibv_dev = list[i];
+
+	DEBUG("device opened");
+	if (ibv_query_device(attr_ctx, &device_attr))
+		goto error;
+	INFO("%u port(s) detected", device_attr.phys_port_cnt);
+
+	for (i = 0; i < device_attr.phys_port_cnt; i++) {
+		uint32_t port = i + 1; /* ports are indexed from one */
+		uint32_t test = (1 << i);
+		struct ibv_context *ctx = NULL;
+		struct ibv_port_attr port_attr;
+		struct ibv_pd *pd = NULL;
+		struct priv *priv = NULL;
+		struct rte_eth_dev *eth_dev;
+#ifdef HAVE_EXP_QUERY_DEVICE
+		struct ibv_exp_device_attr exp_device_attr;
+#endif /* HAVE_EXP_QUERY_DEVICE */
+		struct ether_addr mac;
+
+#ifdef HAVE_EXP_QUERY_DEVICE
+		exp_device_attr.comp_mask = IBV_EXP_DEVICE_ATTR_EXP_CAP_FLAGS;
+#ifdef RSS_SUPPORT
+		exp_device_attr.comp_mask |= IBV_EXP_DEVICE_ATTR_RSS_TBL_SZ;
+#endif /* RSS_SUPPORT */
+#endif /* HAVE_EXP_QUERY_DEVICE */
+
+		DEBUG("using port %u (%08" PRIx32 ")", port, test);
+
+		ctx = ibv_open_device(ibv_dev);
+		if (ctx == NULL)
+			goto port_error;
+
+		/* Check port status. */
+		err = ibv_query_port(ctx, port, &port_attr);
+		if (err) {
+			ERROR("port query failed: %s", strerror(err));
+			goto port_error;
+		}
+		if (port_attr.state != IBV_PORT_ACTIVE)
+			DEBUG("port %d is not active: \"%s\" (%d)",
+			      port, ibv_port_state_str(port_attr.state),
+			      port_attr.state);
+
+		/* Allocate protection domain. */
+		pd = ibv_alloc_pd(ctx);
+		if (pd == NULL) {
+			ERROR("PD allocation failure");
+			err = ENOMEM;
+			goto port_error;
+		}
+
+		mlx5_dev[idx].ports |= test;
+
+		/* from rte_ethdev.c */
+		priv = rte_zmalloc("ethdev private structure",
+				   sizeof(*priv),
+				   RTE_CACHE_LINE_SIZE);
+		if (priv == NULL) {
+			ERROR("priv allocation failure");
+			err = ENOMEM;
+			goto port_error;
+		}
+
+		priv->ctx = ctx;
+		priv->device_attr = device_attr;
+		priv->port = port;
+		priv->pd = pd;
+		priv->mtu = ETHER_MTU;
+#ifdef HAVE_EXP_QUERY_DEVICE
+		if (ibv_exp_query_device(ctx, &exp_device_attr)) {
+			ERROR("ibv_exp_query_device() failed");
+			goto port_error;
+		}
+#ifdef RSS_SUPPORT
+		if ((exp_device_attr.exp_device_cap_flags &
+		     IBV_EXP_DEVICE_QPG) &&
+		    (exp_device_attr.exp_device_cap_flags &
+		     IBV_EXP_DEVICE_UD_RSS) &&
+		    (exp_device_attr.comp_mask &
+		     IBV_EXP_DEVICE_ATTR_RSS_TBL_SZ) &&
+		    (exp_device_attr.max_rss_tbl_sz > 0)) {
+			priv->hw_qpg = 1;
+			priv->hw_rss = 1;
+			priv->max_rss_tbl_sz = exp_device_attr.max_rss_tbl_sz;
+		} else {
+			priv->hw_qpg = 0;
+			priv->hw_rss = 0;
+			priv->max_rss_tbl_sz = 0;
+		}
+		priv->hw_tss = !!(exp_device_attr.exp_device_cap_flags &
+				  IBV_EXP_DEVICE_UD_TSS);
+		DEBUG("device flags: %s%s%s",
+		      (priv->hw_qpg ? "IBV_DEVICE_QPG " : ""),
+		      (priv->hw_tss ? "IBV_DEVICE_TSS " : ""),
+		      (priv->hw_rss ? "IBV_DEVICE_RSS " : ""));
+		if (priv->hw_rss)
+			DEBUG("maximum RSS indirection table size: %u",
+			      exp_device_attr.max_rss_tbl_sz);
+#endif /* RSS_SUPPORT */
+
+		priv->hw_csum =
+			((exp_device_attr.exp_device_cap_flags &
+			  IBV_EXP_DEVICE_RX_CSUM_TCP_UDP_PKT) &&
+			 (exp_device_attr.exp_device_cap_flags &
+			  IBV_EXP_DEVICE_RX_CSUM_IP_PKT));
+		DEBUG("checksum offloading is %ssupported",
+		      (priv->hw_csum ? "" : "not "));
+
+		priv->hw_csum_l2tun = !!(exp_device_attr.exp_device_cap_flags &
+					 IBV_EXP_DEVICE_VXLAN_SUPPORT);
+		DEBUG("L2 tunnel checksum offloads are %ssupported",
+		      (priv->hw_csum_l2tun ? "" : "not "));
+
+#endif /* HAVE_EXP_QUERY_DEVICE */
+
+		priv->vf = vf;
+		/* Configure the first MAC address by default. */
+		if (priv_get_mac(priv, &mac.addr_bytes)) {
+			ERROR("cannot get MAC address, is mlx5_en loaded?"
+			      " (errno: %s)", strerror(errno));
+			goto port_error;
+		}
+		INFO("port %u MAC address is %02x:%02x:%02x:%02x:%02x:%02x",
+		     priv->port,
+		     mac.addr_bytes[0], mac.addr_bytes[1],
+		     mac.addr_bytes[2], mac.addr_bytes[3],
+		     mac.addr_bytes[4], mac.addr_bytes[5]);
+		/* Register MAC and broadcast addresses. */
+		claim_zero(priv_mac_addr_add(priv, 0,
+					     (const uint8_t (*)[ETHER_ADDR_LEN])
+					     mac.addr_bytes));
+		claim_zero(priv_mac_addr_add(priv, 1,
+					     &(const uint8_t [ETHER_ADDR_LEN])
+					     { "\xff\xff\xff\xff\xff\xff" }));
+#ifndef NDEBUG
+		{
+			char ifname[IF_NAMESIZE];
+
+			if (priv_get_ifname(priv, &ifname) == 0)
+				DEBUG("port %u ifname is \"%s\"",
+				      priv->port, ifname);
+			else
+				DEBUG("port %u ifname is unknown", priv->port);
+		}
+#endif
+		/* Get actual MTU if possible. */
+		priv_get_mtu(priv, &priv->mtu);
+		DEBUG("port %u MTU is %u", priv->port, priv->mtu);
+
+		/* from rte_ethdev.c */
+		{
+			char name[RTE_ETH_NAME_MAX_LEN];
+
+			snprintf(name, sizeof(name), "%s port %u",
+				 ibv_get_device_name(ibv_dev), port);
+			eth_dev = rte_eth_dev_allocate(name, RTE_ETH_DEV_PCI);
+		}
+		if (eth_dev == NULL) {
+			ERROR("can not allocate rte ethdev");
+			err = ENOMEM;
+			goto port_error;
+		}
+
+		eth_dev->data->dev_private = priv;
+		eth_dev->pci_dev = pci_dev;
+		eth_dev->driver = &mlx5_driver;
+		eth_dev->data->rx_mbuf_alloc_failed = 0;
+		eth_dev->data->mtu = ETHER_MTU;
+
+		priv->dev = eth_dev;
+		eth_dev->dev_ops = &mlx5_dev_ops;
+		eth_dev->data->mac_addrs = priv->mac;
+
+		/* Bring Ethernet device up. */
+		DEBUG("forcing Ethernet interface up");
+		priv_set_flags(priv, ~IFF_UP, IFF_UP);
+		continue;
+
+port_error:
+		rte_free(priv);
+		if (pd)
+			claim_zero(ibv_dealloc_pd(pd));
+		if (ctx)
+			claim_zero(ibv_close_device(ctx));
+		break;
+	}
+
+	/*
+	 * XXX if something went wrong in the loop above, there is a resource
+	 * leak (ctx, pd, priv, dpdk ethdev) but we can do nothing about it as
+	 * long as the dpdk does not provide a way to deallocate a ethdev and a
+	 * way to enumerate the registered ethdevs to free the previous ones.
+	 */
+
+	/* no port found, complain */
+	if (!mlx5_dev[idx].ports) {
+		err = ENODEV;
+		goto error;
+	}
+
+error:
+	if (attr_ctx)
+		claim_zero(ibv_close_device(attr_ctx));
+	if (list)
+		ibv_free_device_list(list);
+	assert(err >= 0);
+	return -err;
+}
+
+static const struct rte_pci_id mlx5_pci_id_map[] = {
+	{
+		.vendor_id = PCI_VENDOR_ID_MELLANOX,
+		.device_id = PCI_DEVICE_ID_MELLANOX_CONNECTX4,
+		.subsystem_vendor_id = PCI_ANY_ID,
+		.subsystem_device_id = PCI_ANY_ID
+	},
+	{
+		.vendor_id = PCI_VENDOR_ID_MELLANOX,
+		.device_id = PCI_DEVICE_ID_MELLANOX_CONNECTX4VF,
+		.subsystem_vendor_id = PCI_ANY_ID,
+		.subsystem_device_id = PCI_ANY_ID
+	},
+	{
+		.vendor_id = PCI_VENDOR_ID_MELLANOX,
+		.device_id = PCI_DEVICE_ID_MELLANOX_CONNECTX4LX,
+		.subsystem_vendor_id = PCI_ANY_ID,
+		.subsystem_device_id = PCI_ANY_ID
+	},
+	{
+		.vendor_id = PCI_VENDOR_ID_MELLANOX,
+		.device_id = PCI_DEVICE_ID_MELLANOX_CONNECTX4LXVF,
+		.subsystem_vendor_id = PCI_ANY_ID,
+		.subsystem_device_id = PCI_ANY_ID
+	},
+	{
+		.vendor_id = 0
+	}
+};
+
+static struct eth_driver mlx5_driver = {
+	.pci_drv = {
+		.name = MLX5_DRIVER_NAME,
+		.id_table = mlx5_pci_id_map,
+		.devinit = mlx5_pci_devinit,
+	},
+	.dev_private_size = sizeof(struct priv)
+};
+
+/**
+ * Driver initialization routine.
+ */
+static int
+rte_mlx5_pmd_init(const char *name, const char *args)
+{
+	(void)name;
+	(void)args;
+	/*
+	 * RDMAV_HUGEPAGES_SAFE tells ibv_fork_init() we intend to use
+	 * huge pages. Calling ibv_fork_init() during init allows
+	 * applications to use fork() safely for purposes other than
+	 * using this PMD, which is not supported in forked processes.
+	 */
+	setenv("RDMAV_HUGEPAGES_SAFE", "1", 1);
+	ibv_fork_init();
+	rte_eal_pci_register(&mlx5_driver.pci_drv);
+	return 0;
+}
+
+static struct rte_driver rte_mlx5_driver = {
+	.type = PMD_PDEV,
+	.name = MLX5_DRIVER_NAME,
+	.init = rte_mlx5_pmd_init,
+};
+
+PMD_REGISTER_DRIVER(rte_mlx5_driver)
diff --git a/drivers/net/mlx5/mlx5.h b/drivers/net/mlx5/mlx5.h
new file mode 100644
index 0000000..21db3cd
--- /dev/null
+++ b/drivers/net/mlx5/mlx5.h
@@ -0,0 +1,147 @@
+/*-
+ *   BSD LICENSE
+ *
+ *   Copyright 2015 6WIND S.A.
+ *   Copyright 2015 Mellanox.
+ *
+ *   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 6WIND S.A. 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 RTE_PMD_MLX5_H_
+#define RTE_PMD_MLX5_H_
+
+#include <stddef.h>
+#include <stdint.h>
+#include <limits.h>
+#include <net/if.h>
+#include <netinet/in.h>
+#include <linux/if.h>
+
+/* Verbs header. */
+/* ISO C doesn't support unnamed structs/unions, disabling -pedantic. */
+#ifdef PEDANTIC
+#pragma GCC diagnostic ignored "-pedantic"
+#endif
+#include <infiniband/verbs.h>
+#ifdef PEDANTIC
+#pragma GCC diagnostic error "-pedantic"
+#endif
+
+/* DPDK headers don't like -pedantic. */
+#ifdef PEDANTIC
+#pragma GCC diagnostic ignored "-pedantic"
+#endif
+#include <rte_ether.h>
+#include <rte_ethdev.h>
+#include <rte_spinlock.h>
+#ifdef PEDANTIC
+#pragma GCC diagnostic error "-pedantic"
+#endif
+
+#include "mlx5_utils.h"
+#include "mlx5_autoconf.h"
+#include "mlx5_defs.h"
+
+enum {
+	PCI_VENDOR_ID_MELLANOX = 0x15b3,
+};
+
+enum {
+	PCI_DEVICE_ID_MELLANOX_CONNECTX4 = 0x1013,
+	PCI_DEVICE_ID_MELLANOX_CONNECTX4VF = 0x1014,
+	PCI_DEVICE_ID_MELLANOX_CONNECTX4LX = 0x1015,
+	PCI_DEVICE_ID_MELLANOX_CONNECTX4LXVF = 0x1016,
+};
+
+struct priv {
+	struct rte_eth_dev *dev; /* Ethernet device. */
+	struct ibv_context *ctx; /* Verbs context. */
+	struct ibv_device_attr device_attr; /* Device properties. */
+	struct ibv_pd *pd; /* Protection Domain. */
+	/*
+	 * MAC addresses array and configuration bit-field.
+	 * An extra entry that cannot be modified by the DPDK is reserved
+	 * for broadcast frames (destination MAC address ff:ff:ff:ff:ff:ff).
+	 */
+	struct ether_addr mac[MLX5_MAX_MAC_ADDRESSES];
+	BITFIELD_DECLARE(mac_configured, uint32_t, MLX5_MAX_MAC_ADDRESSES);
+	/* Device properties. */
+	uint16_t mtu; /* Configured MTU. */
+	uint8_t port; /* Physical port number. */
+	unsigned int started:1; /* Device started, flows enabled. */
+	unsigned int hw_qpg:1; /* QP groups are supported. */
+	unsigned int hw_tss:1; /* TSS is supported. */
+	unsigned int hw_rss:1; /* RSS is supported. */
+	unsigned int hw_csum:1; /* Checksum offload is supported. */
+	unsigned int hw_csum_l2tun:1; /* Same for L2 tunnels. */
+	unsigned int rss:1; /* RSS is enabled. */
+	unsigned int vf:1; /* This is a VF device. */
+	unsigned int max_rss_tbl_sz; /* Maximum number of RSS queues. */
+	rte_spinlock_t lock; /* Lock for control functions. */
+};
+
+/**
+ * Lock private structure to protect it from concurrent access in the
+ * control path.
+ *
+ * @param priv
+ *   Pointer to private structure.
+ */
+static inline void
+priv_lock(struct priv *priv)
+{
+	rte_spinlock_lock(&priv->lock);
+}
+
+/**
+ * Unlock private structure.
+ *
+ * @param priv
+ *   Pointer to private structure.
+ */
+static inline void
+priv_unlock(struct priv *priv)
+{
+	rte_spinlock_unlock(&priv->lock);
+}
+
+/* mlx5_ethdev.c */
+
+int priv_get_ifname(const struct priv *, char (*)[IF_NAMESIZE]);
+int priv_ifreq(const struct priv *, int req, struct ifreq *);
+int priv_get_mtu(struct priv *, uint16_t *);
+int priv_set_flags(struct priv *, unsigned int, unsigned int);
+int mlx5_ibv_device_to_pci_addr(const struct ibv_device *,
+				struct rte_pci_addr *);
+
+/* mlx5_mac.c */
+
+int priv_get_mac(struct priv *, uint8_t (*)[ETHER_ADDR_LEN]);
+int priv_mac_addr_add(struct priv *, unsigned int,
+		      const uint8_t (*)[ETHER_ADDR_LEN]);
+
+#endif /* RTE_PMD_MLX5_H_ */
diff --git a/drivers/net/mlx5/mlx5_defs.h b/drivers/net/mlx5/mlx5_defs.h
new file mode 100644
index 0000000..c66a74f
--- /dev/null
+++ b/drivers/net/mlx5/mlx5_defs.h
@@ -0,0 +1,43 @@
+/*-
+ *   BSD LICENSE
+ *
+ *   Copyright 2015 6WIND S.A.
+ *   Copyright 2015 Mellanox.
+ *
+ *   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 6WIND S.A. 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 RTE_PMD_MLX5_DEFS_H_
+#define RTE_PMD_MLX5_DEFS_H_
+
+/* Reported driver name. */
+#define MLX5_DRIVER_NAME "librte_pmd_mlx5"
+
+/* Maximum number of simultaneous MAC addresses. */
+#define MLX5_MAX_MAC_ADDRESSES 128
+
+#endif /* RTE_PMD_MLX5_DEFS_H_ */
diff --git a/drivers/net/mlx5/mlx5_ethdev.c b/drivers/net/mlx5/mlx5_ethdev.c
new file mode 100644
index 0000000..b6c7d7a
--- /dev/null
+++ b/drivers/net/mlx5/mlx5_ethdev.c
@@ -0,0 +1,420 @@
+/*-
+ *   BSD LICENSE
+ *
+ *   Copyright 2015 6WIND S.A.
+ *   Copyright 2015 Mellanox.
+ *
+ *   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 6WIND S.A. 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 <stddef.h>
+#include <unistd.h>
+#include <stdint.h>
+#include <stdio.h>
+#include <string.h>
+#include <stdlib.h>
+#include <errno.h>
+#include <dirent.h>
+#include <net/if.h>
+#include <sys/ioctl.h>
+#include <sys/socket.h>
+#include <netinet/in.h>
+#include <linux/if.h>
+
+/* DPDK headers don't like -pedantic. */
+#ifdef PEDANTIC
+#pragma GCC diagnostic ignored "-pedantic"
+#endif
+#include <rte_atomic.h>
+#include <rte_ethdev.h>
+#include <rte_mbuf.h>
+#include <rte_common.h>
+#ifdef PEDANTIC
+#pragma GCC diagnostic error "-pedantic"
+#endif
+
+#include "mlx5.h"
+#include "mlx5_utils.h"
+
+/**
+ * Get interface name from private structure.
+ *
+ * @param[in] priv
+ *   Pointer to private structure.
+ * @param[out] ifname
+ *   Interface name output buffer.
+ *
+ * @return
+ *   0 on success, -1 on failure and errno is set.
+ */
+int
+priv_get_ifname(const struct priv *priv, char (*ifname)[IF_NAMESIZE])
+{
+	DIR *dir;
+	struct dirent *dent;
+	unsigned int dev_type = 0;
+	unsigned int dev_port_prev = ~0u;
+	char match[IF_NAMESIZE] = "";
+
+	{
+		MKSTR(path, "%s/device/net", priv->ctx->device->ibdev_path);
+
+		dir = opendir(path);
+		if (dir == NULL)
+			return -1;
+	}
+	while ((dent = readdir(dir)) != NULL) {
+		char *name = dent->d_name;
+		FILE *file;
+		unsigned int dev_port;
+		int r;
+
+		if ((name[0] == '.') &&
+		    ((name[1] == '\0') ||
+		     ((name[1] == '.') && (name[2] == '\0'))))
+			continue;
+
+		MKSTR(path, "%s/device/net/%s/%s",
+		      priv->ctx->device->ibdev_path, name,
+		      (dev_type ? "dev_id" : "dev_port"));
+
+		file = fopen(path, "rb");
+		if (file == NULL) {
+			if (errno != ENOENT)
+				continue;
+			/*
+			 * Switch to dev_id when dev_port does not exist as
+			 * is the case with Linux kernel versions < 3.15.
+			 */
+try_dev_id:
+			match[0] = '\0';
+			if (dev_type)
+				break;
+			dev_type = 1;
+			dev_port_prev = ~0u;
+			rewinddir(dir);
+			continue;
+		}
+		r = fscanf(file, (dev_type ? "%x" : "%u"), &dev_port);
+		fclose(file);
+		if (r != 1)
+			continue;
+		/*
+		 * Switch to dev_id when dev_port returns the same value for
+		 * all ports. May happen when using a MOFED release older than
+		 * 3.0 with a Linux kernel >= 3.15.
+		 */
+		if (dev_port == dev_port_prev)
+			goto try_dev_id;
+		dev_port_prev = dev_port;
+		if (dev_port == (priv->port - 1u))
+			snprintf(match, sizeof(match), "%s", name);
+	}
+	closedir(dir);
+	if (match[0] == '\0')
+		return -1;
+	strncpy(*ifname, match, sizeof(*ifname));
+	return 0;
+}
+
+/**
+ * Read from sysfs entry.
+ *
+ * @param[in] priv
+ *   Pointer to private structure.
+ * @param[in] entry
+ *   Entry name relative to sysfs path.
+ * @param[out] buf
+ *   Data output buffer.
+ * @param size
+ *   Buffer size.
+ *
+ * @return
+ *   0 on success, -1 on failure and errno is set.
+ */
+static int
+priv_sysfs_read(const struct priv *priv, const char *entry,
+		char *buf, size_t size)
+{
+	char ifname[IF_NAMESIZE];
+	FILE *file;
+	int ret;
+	int err;
+
+	if (priv_get_ifname(priv, &ifname))
+		return -1;
+
+	MKSTR(path, "%s/device/net/%s/%s", priv->ctx->device->ibdev_path,
+	      ifname, entry);
+
+	file = fopen(path, "rb");
+	if (file == NULL)
+		return -1;
+	ret = fread(buf, 1, size, file);
+	err = errno;
+	if (((size_t)ret < size) && (ferror(file)))
+		ret = -1;
+	else
+		ret = size;
+	fclose(file);
+	errno = err;
+	return ret;
+}
+
+/**
+ * Write to sysfs entry.
+ *
+ * @param[in] priv
+ *   Pointer to private structure.
+ * @param[in] entry
+ *   Entry name relative to sysfs path.
+ * @param[in] buf
+ *   Data buffer.
+ * @param size
+ *   Buffer size.
+ *
+ * @return
+ *   0 on success, -1 on failure and errno is set.
+ */
+static int
+priv_sysfs_write(const struct priv *priv, const char *entry,
+		 char *buf, size_t size)
+{
+	char ifname[IF_NAMESIZE];
+	FILE *file;
+	int ret;
+	int err;
+
+	if (priv_get_ifname(priv, &ifname))
+		return -1;
+
+	MKSTR(path, "%s/device/net/%s/%s", priv->ctx->device->ibdev_path,
+	      ifname, entry);
+
+	file = fopen(path, "wb");
+	if (file == NULL)
+		return -1;
+	ret = fwrite(buf, 1, size, file);
+	err = errno;
+	if (((size_t)ret < size) || (ferror(file)))
+		ret = -1;
+	else
+		ret = size;
+	fclose(file);
+	errno = err;
+	return ret;
+}
+
+/**
+ * Get unsigned long sysfs property.
+ *
+ * @param priv
+ *   Pointer to private structure.
+ * @param[in] name
+ *   Entry name relative to sysfs path.
+ * @param[out] value
+ *   Value output buffer.
+ *
+ * @return
+ *   0 on success, -1 on failure and errno is set.
+ */
+static int
+priv_get_sysfs_ulong(struct priv *priv, const char *name, unsigned long *value)
+{
+	int ret;
+	unsigned long value_ret;
+	char value_str[32];
+
+	ret = priv_sysfs_read(priv, name, value_str, (sizeof(value_str) - 1));
+	if (ret == -1) {
+		DEBUG("cannot read %s value from sysfs: %s",
+		      name, strerror(errno));
+		return -1;
+	}
+	value_str[ret] = '\0';
+	errno = 0;
+	value_ret = strtoul(value_str, NULL, 0);
+	if (errno) {
+		DEBUG("invalid %s value `%s': %s", name, value_str,
+		      strerror(errno));
+		return -1;
+	}
+	*value = value_ret;
+	return 0;
+}
+
+/**
+ * Set unsigned long sysfs property.
+ *
+ * @param priv
+ *   Pointer to private structure.
+ * @param[in] name
+ *   Entry name relative to sysfs path.
+ * @param value
+ *   Value to set.
+ *
+ * @return
+ *   0 on success, -1 on failure and errno is set.
+ */
+static int
+priv_set_sysfs_ulong(struct priv *priv, const char *name, unsigned long value)
+{
+	int ret;
+	MKSTR(value_str, "%lu", value);
+
+	ret = priv_sysfs_write(priv, name, value_str, (sizeof(value_str) - 1));
+	if (ret == -1) {
+		DEBUG("cannot write %s `%s' (%lu) to sysfs: %s",
+		      name, value_str, value, strerror(errno));
+		return -1;
+	}
+	return 0;
+}
+
+/**
+ * Perform ifreq ioctl() on associated Ethernet device.
+ *
+ * @param[in] priv
+ *   Pointer to private structure.
+ * @param req
+ *   Request number to pass to ioctl().
+ * @param[out] ifr
+ *   Interface request structure output buffer.
+ *
+ * @return
+ *   0 on success, -1 on failure and errno is set.
+ */
+int
+priv_ifreq(const struct priv *priv, int req, struct ifreq *ifr)
+{
+	int sock = socket(PF_INET, SOCK_DGRAM, IPPROTO_IP);
+	int ret = -1;
+
+	if (sock == -1)
+		return ret;
+	if (priv_get_ifname(priv, &ifr->ifr_name) == 0)
+		ret = ioctl(sock, req, ifr);
+	close(sock);
+	return ret;
+}
+
+/**
+ * Get device MTU.
+ *
+ * @param priv
+ *   Pointer to private structure.
+ * @param[out] mtu
+ *   MTU value output buffer.
+ *
+ * @return
+ *   0 on success, -1 on failure and errno is set.
+ */
+int
+priv_get_mtu(struct priv *priv, uint16_t *mtu)
+{
+	unsigned long ulong_mtu;
+
+	if (priv_get_sysfs_ulong(priv, "mtu", &ulong_mtu) == -1)
+		return -1;
+	*mtu = ulong_mtu;
+	return 0;
+}
+
+/**
+ * Set device flags.
+ *
+ * @param priv
+ *   Pointer to private structure.
+ * @param keep
+ *   Bitmask for flags that must remain untouched.
+ * @param flags
+ *   Bitmask for flags to modify.
+ *
+ * @return
+ *   0 on success, -1 on failure and errno is set.
+ */
+int
+priv_set_flags(struct priv *priv, unsigned int keep, unsigned int flags)
+{
+	unsigned long tmp;
+
+	if (priv_get_sysfs_ulong(priv, "flags", &tmp) == -1)
+		return -1;
+	tmp &= keep;
+	tmp |= flags;
+	return priv_set_sysfs_ulong(priv, "flags", tmp);
+}
+
+/**
+ * Get PCI information from struct ibv_device.
+ *
+ * @param device
+ *   Pointer to Ethernet device structure.
+ * @param[out] pci_addr
+ *   PCI bus address output buffer.
+ *
+ * @return
+ *   0 on success, -1 on failure and errno is set.
+ */
+int
+mlx5_ibv_device_to_pci_addr(const struct ibv_device *device,
+			    struct rte_pci_addr *pci_addr)
+{
+	FILE *file;
+	char line[32];
+	MKSTR(path, "%s/device/uevent", device->ibdev_path);
+
+	file = fopen(path, "rb");
+	if (file == NULL)
+		return -1;
+	while (fgets(line, sizeof(line), file) == line) {
+		size_t len = strlen(line);
+		int ret;
+
+		/* Truncate long lines. */
+		if (len == (sizeof(line) - 1))
+			while (line[(len - 1)] != '\n') {
+				ret = fgetc(file);
+				if (ret == EOF)
+					break;
+				line[(len - 1)] = ret;
+			}
+		/* Extract information. */
+		if (sscanf(line,
+			   "PCI_SLOT_NAME="
+			   "%" SCNx16 ":%" SCNx8 ":%" SCNx8 ".%" SCNx8 "\n",
+			   &pci_addr->domain,
+			   &pci_addr->bus,
+			   &pci_addr->devid,
+			   &pci_addr->function) == 4) {
+			ret = 0;
+			break;
+		}
+	}
+	fclose(file);
+	return 0;
+}
diff --git a/drivers/net/mlx5/mlx5_mac.c b/drivers/net/mlx5/mlx5_mac.c
new file mode 100644
index 0000000..f7e1cf6
--- /dev/null
+++ b/drivers/net/mlx5/mlx5_mac.c
@@ -0,0 +1,150 @@
+/*-
+ *   BSD LICENSE
+ *
+ *   Copyright 2015 6WIND S.A.
+ *   Copyright 2015 Mellanox.
+ *
+ *   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 6WIND S.A. 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 <stddef.h>
+#include <assert.h>
+#include <stdint.h>
+#include <string.h>
+#include <inttypes.h>
+#include <errno.h>
+#include <netinet/in.h>
+#include <linux/if.h>
+#include <sys/ioctl.h>
+#include <arpa/inet.h>
+
+/* Verbs header. */
+/* ISO C doesn't support unnamed structs/unions, disabling -pedantic. */
+#ifdef PEDANTIC
+#pragma GCC diagnostic ignored "-pedantic"
+#endif
+#include <infiniband/verbs.h>
+#ifdef PEDANTIC
+#pragma GCC diagnostic error "-pedantic"
+#endif
+
+/* DPDK headers don't like -pedantic. */
+#ifdef PEDANTIC
+#pragma GCC diagnostic ignored "-pedantic"
+#endif
+#include <rte_ether.h>
+#include <rte_ethdev.h>
+#include <rte_common.h>
+#ifdef PEDANTIC
+#pragma GCC diagnostic error "-pedantic"
+#endif
+
+#include "mlx5.h"
+#include "mlx5_utils.h"
+
+/**
+ * Get MAC address by querying netdevice.
+ *
+ * @param[in] priv
+ *   struct priv for the requested device.
+ * @param[out] mac
+ *   MAC address output buffer.
+ *
+ * @return
+ *   0 on success, -1 on failure and errno is set.
+ */
+int
+priv_get_mac(struct priv *priv, uint8_t (*mac)[ETHER_ADDR_LEN])
+{
+	struct ifreq request;
+
+	if (priv_ifreq(priv, SIOCGIFHWADDR, &request))
+		return -1;
+	memcpy(mac, request.ifr_hwaddr.sa_data, ETHER_ADDR_LEN);
+	return 0;
+}
+
+/**
+ * Unregister a MAC address.
+ *
+ * @param priv
+ *   Pointer to private structure.
+ * @param mac_index
+ *   MAC address index.
+ */
+static void
+priv_mac_addr_del(struct priv *priv, unsigned int mac_index)
+{
+	assert(mac_index < RTE_DIM(priv->mac));
+	if (!BITFIELD_ISSET(priv->mac_configured, mac_index))
+		return;
+	BITFIELD_RESET(priv->mac_configured, mac_index);
+}
+
+/**
+ * Register a MAC address.
+ *
+ * @param priv
+ *   Pointer to private structure.
+ * @param mac_index
+ *   MAC address index to use.
+ * @param mac
+ *   MAC address to register.
+ *
+ * @return
+ *   0 on success, errno value on failure.
+ */
+int
+priv_mac_addr_add(struct priv *priv, unsigned int mac_index,
+		  const uint8_t (*mac)[ETHER_ADDR_LEN])
+{
+	unsigned int i;
+
+	assert(mac_index < RTE_DIM(priv->mac));
+	/* First, make sure this address isn't already configured. */
+	for (i = 0; (i != RTE_DIM(priv->mac)); ++i) {
+		/* Skip this index, it's going to be reconfigured. */
+		if (i == mac_index)
+			continue;
+		if (!BITFIELD_ISSET(priv->mac_configured, i))
+			continue;
+		if (memcmp(priv->mac[i].addr_bytes, *mac, sizeof(*mac)))
+			continue;
+		/* Address already configured elsewhere, return with error. */
+		return EADDRINUSE;
+	}
+	if (BITFIELD_ISSET(priv->mac_configured, mac_index))
+		priv_mac_addr_del(priv, mac_index);
+	priv->mac[mac_index] = (struct ether_addr){
+		{
+			(*mac)[0], (*mac)[1], (*mac)[2],
+			(*mac)[3], (*mac)[4], (*mac)[5]
+		}
+	};
+	BITFIELD_SET(priv->mac_configured, mac_index);
+	return 0;
+}
diff --git a/drivers/net/mlx5/mlx5_utils.h b/drivers/net/mlx5/mlx5_utils.h
new file mode 100644
index 0000000..cc6aab6
--- /dev/null
+++ b/drivers/net/mlx5/mlx5_utils.h
@@ -0,0 +1,149 @@
+/*-
+ *   BSD LICENSE
+ *
+ *   Copyright 2015 6WIND S.A.
+ *   Copyright 2015 Mellanox.
+ *
+ *   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 6WIND S.A. 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 RTE_PMD_MLX5_UTILS_H_
+#define RTE_PMD_MLX5_UTILS_H_
+
+#include <stddef.h>
+#include <stdio.h>
+#include <limits.h>
+#include <assert.h>
+#include <errno.h>
+
+#include "mlx5_defs.h"
+
+/* Bit-field manipulation. */
+#define BITFIELD_DECLARE(bf, type, size) \
+	type bf[(((size_t)(size) / (sizeof(type) * CHAR_BIT)) + \
+		 !!((size_t)(size) % (sizeof(type) * CHAR_BIT)))]
+#define BITFIELD_DEFINE(bf, type, size) \
+	BITFIELD_DECLARE((bf), type, (size)) = { 0 }
+#define BITFIELD_SET(bf, b) \
+	(assert((size_t)(b) < (sizeof(bf) * CHAR_BIT)), \
+	 (void)((bf)[((b) / (sizeof((bf)[0]) * CHAR_BIT))] |= \
+		((size_t)1 << ((b) % (sizeof((bf)[0]) * CHAR_BIT)))))
+#define BITFIELD_RESET(bf, b) \
+	(assert((size_t)(b) < (sizeof(bf) * CHAR_BIT)), \
+	 (void)((bf)[((b) / (sizeof((bf)[0]) * CHAR_BIT))] &= \
+		~((size_t)1 << ((b) % (sizeof((bf)[0]) * CHAR_BIT)))))
+#define BITFIELD_ISSET(bf, b) \
+	(assert((size_t)(b) < (sizeof(bf) * CHAR_BIT)), \
+	 !!(((bf)[((b) / (sizeof((bf)[0]) * CHAR_BIT))] & \
+	     ((size_t)1 << ((b) % (sizeof((bf)[0]) * CHAR_BIT))))))
+
+/* Save and restore errno around argument evaluation. */
+#define ERRNO_SAFE(x) ((errno = (int []){ errno, ((x), 0) }[0]))
+
+/*
+ * Helper macros to work around __VA_ARGS__ limitations in a C99 compliant
+ * manner.
+ */
+#define PMD_DRV_LOG_STRIP(a, b) a
+#define PMD_DRV_LOG_OPAREN (
+#define PMD_DRV_LOG_CPAREN )
+#define PMD_DRV_LOG_COMMA ,
+
+/* Return the file name part of a path. */
+static inline const char *
+pmd_drv_log_basename(const char *s)
+{
+	const char *n = s;
+
+	while (*n)
+		if (*(n++) == '/')
+			s = n;
+	return s;
+}
+
+/*
+ * When debugging is enabled (NDEBUG not defined), file, line and function
+ * information replace the driver name (MLX5_DRIVER_NAME) in log messages.
+ */
+#ifndef NDEBUG
+
+#define PMD_DRV_LOG___(level, ...) \
+	ERRNO_SAFE(RTE_LOG(level, PMD, __VA_ARGS__))
+#define PMD_DRV_LOG__(level, ...) \
+	PMD_DRV_LOG___(level, "%s:%u: %s(): " __VA_ARGS__)
+#define PMD_DRV_LOG_(level, s, ...) \
+	PMD_DRV_LOG__(level, \
+		s "\n" PMD_DRV_LOG_COMMA \
+		pmd_drv_log_basename(__FILE__) PMD_DRV_LOG_COMMA \
+		__LINE__ PMD_DRV_LOG_COMMA \
+		__func__, \
+		__VA_ARGS__)
+
+#else /* NDEBUG */
+
+#define PMD_DRV_LOG___(level, ...) \
+	ERRNO_SAFE(RTE_LOG(level, PMD, MLX5_DRIVER_NAME ": " __VA_ARGS__))
+#define PMD_DRV_LOG__(level, ...) \
+	PMD_DRV_LOG___(level, __VA_ARGS__)
+#define PMD_DRV_LOG_(level, s, ...) \
+	PMD_DRV_LOG__(level, s "\n", __VA_ARGS__)
+
+#endif /* NDEBUG */
+
+/* Generic printf()-like logging macro with automatic line feed. */
+#define PMD_DRV_LOG(level, ...) \
+	PMD_DRV_LOG_(level, \
+		__VA_ARGS__ PMD_DRV_LOG_STRIP PMD_DRV_LOG_OPAREN, \
+		PMD_DRV_LOG_CPAREN)
+
+/*
+ * Like assert(), DEBUG() becomes a no-op and claim_zero() does not perform
+ * any check when debugging is disabled.
+ */
+#ifndef NDEBUG
+
+#define DEBUG(...) PMD_DRV_LOG(DEBUG, __VA_ARGS__)
+#define claim_zero(...) assert((__VA_ARGS__) == 0)
+
+#else /* NDEBUG */
+
+#define DEBUG(...) (void)0
+#define claim_zero(...) (__VA_ARGS__)
+
+#endif /* NDEBUG */
+
+#define INFO(...) PMD_DRV_LOG(INFO, __VA_ARGS__)
+#define WARN(...) PMD_DRV_LOG(WARNING, __VA_ARGS__)
+#define ERROR(...) PMD_DRV_LOG(ERR, __VA_ARGS__)
+
+/* Allocate a buffer on the stack and fill it with a printf format string. */
+#define MKSTR(name, ...) \
+	char name[snprintf(NULL, 0, __VA_ARGS__) + 1]; \
+	\
+	snprintf(name, sizeof(name), __VA_ARGS__)
+
+#endif /* RTE_PMD_MLX5_UTILS_H_ */
diff --git a/drivers/net/mlx5/rte_pmd_mlx5_version.map b/drivers/net/mlx5/rte_pmd_mlx5_version.map
new file mode 100644
index 0000000..ad607bb
--- /dev/null
+++ b/drivers/net/mlx5/rte_pmd_mlx5_version.map
@@ -0,0 +1,3 @@
+DPDK_2.2 {
+	local: *;
+};
diff --git a/mk/rte.app.mk b/mk/rte.app.mk
index 9e1909e..724efa7 100644
--- a/mk/rte.app.mk
+++ b/mk/rte.app.mk
@@ -104,6 +104,10 @@ ifeq ($(CONFIG_RTE_BUILD_SHARED_LIB),n)
 _LDLIBS-$(CONFIG_RTE_LIBRTE_MLX4_PMD)       += -libverbs
 endif # ! CONFIG_RTE_BUILD_SHARED_LIBS
 
+ifeq ($(CONFIG_RTE_BUILD_SHARED_LIB),n)
+_LDLIBS-$(CONFIG_RTE_LIBRTE_MLX5_PMD)       += -libverbs
+endif # ! CONFIG_RTE_BUILD_SHARED_LIBS
+
 _LDLIBS-$(CONFIG_RTE_LIBRTE_BNX2X_PMD)      += -lz
 
 _LDLIBS-y += --start-group
@@ -137,6 +141,7 @@ _LDLIBS-$(CONFIG_RTE_LIBRTE_FM10K_PMD)      += -lrte_pmd_fm10k
 _LDLIBS-$(CONFIG_RTE_LIBRTE_IXGBE_PMD)      += -lrte_pmd_ixgbe
 _LDLIBS-$(CONFIG_RTE_LIBRTE_E1000_PMD)      += -lrte_pmd_e1000
 _LDLIBS-$(CONFIG_RTE_LIBRTE_MLX4_PMD)       += -lrte_pmd_mlx4
+_LDLIBS-$(CONFIG_RTE_LIBRTE_MLX5_PMD)       += -lrte_pmd_mlx5
 _LDLIBS-$(CONFIG_RTE_LIBRTE_MPIPE_PMD)      += -lrte_pmd_mpipe -lgxio
 _LDLIBS-$(CONFIG_RTE_LIBRTE_PMD_RING)       += -lrte_pmd_ring
 _LDLIBS-$(CONFIG_RTE_LIBRTE_PMD_PCAP)       += -lrte_pmd_pcap
-- 
2.1.0

^ permalink raw reply	[flat|nested] 32+ messages in thread

* [dpdk-dev] [PATCH v2 02/13] mlx5: add non-scattered TX and RX support
  2015-10-30 18:52 ` [dpdk-dev] [PATCH v2 00/13] Mellanox ConnectX-4 PMD (mlx5) Adrien Mazarguil
  2015-10-30 18:52   ` [dpdk-dev] [PATCH v2 01/13] mlx5: new poll-mode driver for Mellanox ConnectX-4 adapters Adrien Mazarguil
@ 2015-10-30 18:52   ` Adrien Mazarguil
  2015-10-30 18:52   ` [dpdk-dev] [PATCH v2 03/13] mlx5: add MAC handling Adrien Mazarguil
                     ` (11 subsequent siblings)
  13 siblings, 0 replies; 32+ messages in thread
From: Adrien Mazarguil @ 2015-10-30 18:52 UTC (permalink / raw)
  To: dev

RSS implementation with parent/child QPs comes from mlx4 and is temporary.

Signed-off-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>
Signed-off-by: Nelio Laranjeiro <nelio.laranjeiro@6wind.com>
---
 config/common_bsdapp          |   3 +
 config/common_linuxapp        |   3 +
 drivers/net/mlx5/Makefile     |  15 +
 drivers/net/mlx5/mlx5.c       |  40 +++
 drivers/net/mlx5/mlx5.h       |  25 ++
 drivers/net/mlx5/mlx5_defs.h  |  24 ++
 drivers/net/mlx5/mlx5_rxq.c   | 682 ++++++++++++++++++++++++++++++++++++++++++
 drivers/net/mlx5/mlx5_rxtx.c  | 496 ++++++++++++++++++++++++++++++
 drivers/net/mlx5/mlx5_rxtx.h  | 156 ++++++++++
 drivers/net/mlx5/mlx5_txq.c   | 512 +++++++++++++++++++++++++++++++
 drivers/net/mlx5/mlx5_utils.h |  11 +
 11 files changed, 1967 insertions(+)
 create mode 100644 drivers/net/mlx5/mlx5_rxq.c
 create mode 100644 drivers/net/mlx5/mlx5_rxtx.c
 create mode 100644 drivers/net/mlx5/mlx5_rxtx.h
 create mode 100644 drivers/net/mlx5/mlx5_txq.c

diff --git a/config/common_bsdapp b/config/common_bsdapp
index 1e6885f..3b50ff9 100644
--- a/config/common_bsdapp
+++ b/config/common_bsdapp
@@ -218,6 +218,9 @@ CONFIG_RTE_LIBRTE_MLX4_SOFT_COUNTERS=1
 #
 CONFIG_RTE_LIBRTE_MLX5_PMD=n
 CONFIG_RTE_LIBRTE_MLX5_DEBUG=n
+CONFIG_RTE_LIBRTE_MLX5_SGE_WR_N=4
+CONFIG_RTE_LIBRTE_MLX5_MAX_INLINE=0
+CONFIG_RTE_LIBRTE_MLX5_TX_MP_CACHE=8
 
 #
 # Compile burst-oriented Broadcom PMD driver
diff --git a/config/common_linuxapp b/config/common_linuxapp
index 7da7ba7..eed8fc0 100644
--- a/config/common_linuxapp
+++ b/config/common_linuxapp
@@ -216,6 +216,9 @@ CONFIG_RTE_LIBRTE_MLX4_SOFT_COUNTERS=1
 #
 CONFIG_RTE_LIBRTE_MLX5_PMD=n
 CONFIG_RTE_LIBRTE_MLX5_DEBUG=n
+CONFIG_RTE_LIBRTE_MLX5_SGE_WR_N=4
+CONFIG_RTE_LIBRTE_MLX5_MAX_INLINE=0
+CONFIG_RTE_LIBRTE_MLX5_TX_MP_CACHE=8
 
 #
 # Compile burst-oriented Broadcom PMD driver
diff --git a/drivers/net/mlx5/Makefile b/drivers/net/mlx5/Makefile
index 6e63073..7b9c57b 100644
--- a/drivers/net/mlx5/Makefile
+++ b/drivers/net/mlx5/Makefile
@@ -42,6 +42,9 @@ LIB = librte_pmd_mlx5.a
 
 # Sources.
 SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5.c
+SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5_rxq.c
+SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5_txq.c
+SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5_rxtx.c
 SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5_ethdev.c
 SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5_mac.c
 
@@ -79,6 +82,18 @@ else
 CFLAGS += -DNDEBUG -UPEDANTIC
 endif
 
+ifdef CONFIG_RTE_LIBRTE_MLX5_SGE_WR_N
+CFLAGS += -DMLX5_PMD_SGE_WR_N=$(CONFIG_RTE_LIBRTE_MLX5_SGE_WR_N)
+endif
+
+ifdef CONFIG_RTE_LIBRTE_MLX5_MAX_INLINE
+CFLAGS += -DMLX5_PMD_MAX_INLINE=$(CONFIG_RTE_LIBRTE_MLX5_MAX_INLINE)
+endif
+
+ifdef CONFIG_RTE_LIBRTE_MLX5_TX_MP_CACHE
+CFLAGS += -DMLX5_PMD_TX_MP_CACHE=$(CONFIG_RTE_LIBRTE_MLX5_TX_MP_CACHE)
+endif
+
 include $(RTE_SDK)/mk/rte.lib.mk
 
 # Generate and clean-up mlx5_autoconf.h.
diff --git a/drivers/net/mlx5/mlx5.c b/drivers/net/mlx5/mlx5.c
index 6df486b..54bd6b9 100644
--- a/drivers/net/mlx5/mlx5.c
+++ b/drivers/net/mlx5/mlx5.c
@@ -63,6 +63,7 @@
 
 #include "mlx5.h"
 #include "mlx5_utils.h"
+#include "mlx5_rxtx.h"
 #include "mlx5_autoconf.h"
 
 /**
@@ -77,11 +78,46 @@ static void
 mlx5_dev_close(struct rte_eth_dev *dev)
 {
 	struct priv *priv = dev->data->dev_private;
+	void *tmp;
+	unsigned int i;
 
 	priv_lock(priv);
 	DEBUG("%p: closing device \"%s\"",
 	      (void *)dev,
 	      ((priv->ctx != NULL) ? priv->ctx->device->name : ""));
+	/* Prevent crashes when queues are still in use. */
+	dev->rx_pkt_burst = removed_rx_burst;
+	dev->tx_pkt_burst = removed_tx_burst;
+	if (priv->rxqs != NULL) {
+		/* XXX race condition if mlx5_rx_burst() is still running. */
+		usleep(1000);
+		for (i = 0; (i != priv->rxqs_n); ++i) {
+			tmp = (*priv->rxqs)[i];
+			if (tmp == NULL)
+				continue;
+			(*priv->rxqs)[i] = NULL;
+			rxq_cleanup(tmp);
+			rte_free(tmp);
+		}
+		priv->rxqs_n = 0;
+		priv->rxqs = NULL;
+	}
+	if (priv->txqs != NULL) {
+		/* XXX race condition if mlx5_tx_burst() is still running. */
+		usleep(1000);
+		for (i = 0; (i != priv->txqs_n); ++i) {
+			tmp = (*priv->txqs)[i];
+			if (tmp == NULL)
+				continue;
+			(*priv->txqs)[i] = NULL;
+			txq_cleanup(tmp);
+			rte_free(tmp);
+		}
+		priv->txqs_n = 0;
+		priv->txqs = NULL;
+	}
+	if (priv->rss)
+		rxq_cleanup(&priv->rxq_parent);
 	if (priv->pd != NULL) {
 		assert(priv->ctx != NULL);
 		claim_zero(ibv_dealloc_pd(priv->pd));
@@ -94,6 +130,10 @@ mlx5_dev_close(struct rte_eth_dev *dev)
 
 static const struct eth_dev_ops mlx5_dev_ops = {
 	.dev_close = mlx5_dev_close,
+	.rx_queue_setup = mlx5_rx_queue_setup,
+	.tx_queue_setup = mlx5_tx_queue_setup,
+	.rx_queue_release = mlx5_rx_queue_release,
+	.tx_queue_release = mlx5_tx_queue_release,
 };
 
 static struct {
diff --git a/drivers/net/mlx5/mlx5.h b/drivers/net/mlx5/mlx5.h
index 21db3cd..49978f5 100644
--- a/drivers/net/mlx5/mlx5.h
+++ b/drivers/net/mlx5/mlx5.h
@@ -63,6 +63,7 @@
 #endif
 
 #include "mlx5_utils.h"
+#include "mlx5_rxtx.h"
 #include "mlx5_autoconf.h"
 #include "mlx5_defs.h"
 
@@ -101,9 +102,33 @@ struct priv {
 	unsigned int rss:1; /* RSS is enabled. */
 	unsigned int vf:1; /* This is a VF device. */
 	unsigned int max_rss_tbl_sz; /* Maximum number of RSS queues. */
+	/* RX/TX queues. */
+	struct rxq rxq_parent; /* Parent queue when RSS is enabled. */
+	unsigned int rxqs_n; /* RX queues array size. */
+	unsigned int txqs_n; /* TX queues array size. */
+	struct rxq *(*rxqs)[]; /* RX queues. */
+	struct txq *(*txqs)[]; /* TX queues. */
 	rte_spinlock_t lock; /* Lock for control functions. */
 };
 
+/* Work Request ID data type (64 bit). */
+typedef union {
+	struct {
+		uint32_t id;
+		uint16_t offset;
+	} data;
+	uint64_t raw;
+} wr_id_t;
+
+/* Compile-time check. */
+static inline void wr_id_t_check(void)
+{
+	wr_id_t check[1 + (2 * -!(sizeof(wr_id_t) == sizeof(uint64_t)))];
+
+	(void)check;
+	(void)wr_id_t_check;
+}
+
 /**
  * Lock private structure to protect it from concurrent access in the
  * control path.
diff --git a/drivers/net/mlx5/mlx5_defs.h b/drivers/net/mlx5/mlx5_defs.h
index c66a74f..c85be9c 100644
--- a/drivers/net/mlx5/mlx5_defs.h
+++ b/drivers/net/mlx5/mlx5_defs.h
@@ -40,4 +40,28 @@
 /* Maximum number of simultaneous MAC addresses. */
 #define MLX5_MAX_MAC_ADDRESSES 128
 
+/* Request send completion once in every 64 sends, might be less. */
+#define MLX5_PMD_TX_PER_COMP_REQ 64
+
+/* Maximum number of Scatter/Gather Elements per Work Request. */
+#ifndef MLX5_PMD_SGE_WR_N
+#define MLX5_PMD_SGE_WR_N 4
+#endif
+
+/* Maximum size for inline data. */
+#ifndef MLX5_PMD_MAX_INLINE
+#define MLX5_PMD_MAX_INLINE 0
+#endif
+
+/*
+ * Maximum number of cached Memory Pools (MPs) per TX queue. Each RTE MP
+ * from which buffers are to be transmitted will have to be mapped by this
+ * driver to their own Memory Region (MR). This is a slow operation.
+ *
+ * This value is always 1 for RX queues.
+ */
+#ifndef MLX5_PMD_TX_MP_CACHE
+#define MLX5_PMD_TX_MP_CACHE 8
+#endif
+
 #endif /* RTE_PMD_MLX5_DEFS_H_ */
diff --git a/drivers/net/mlx5/mlx5_rxq.c b/drivers/net/mlx5/mlx5_rxq.c
new file mode 100644
index 0000000..01cc649
--- /dev/null
+++ b/drivers/net/mlx5/mlx5_rxq.c
@@ -0,0 +1,682 @@
+/*-
+ *   BSD LICENSE
+ *
+ *   Copyright 2015 6WIND S.A.
+ *   Copyright 2015 Mellanox.
+ *
+ *   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 6WIND S.A. 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 <stddef.h>
+#include <assert.h>
+#include <errno.h>
+#include <string.h>
+#include <stdint.h>
+
+/* Verbs header. */
+/* ISO C doesn't support unnamed structs/unions, disabling -pedantic. */
+#ifdef PEDANTIC
+#pragma GCC diagnostic ignored "-pedantic"
+#endif
+#include <infiniband/verbs.h>
+#ifdef PEDANTIC
+#pragma GCC diagnostic error "-pedantic"
+#endif
+
+/* DPDK headers don't like -pedantic. */
+#ifdef PEDANTIC
+#pragma GCC diagnostic ignored "-pedantic"
+#endif
+#include <rte_mbuf.h>
+#include <rte_malloc.h>
+#include <rte_ethdev.h>
+#include <rte_common.h>
+#ifdef PEDANTIC
+#pragma GCC diagnostic error "-pedantic"
+#endif
+
+#include "mlx5.h"
+#include "mlx5_rxtx.h"
+#include "mlx5_utils.h"
+#include "mlx5_defs.h"
+
+/**
+ * Allocate RX queue elements.
+ *
+ * @param rxq
+ *   Pointer to RX queue structure.
+ * @param elts_n
+ *   Number of elements to allocate.
+ * @param[in] pool
+ *   If not NULL, fetch buffers from this array instead of allocating them
+ *   with rte_pktmbuf_alloc().
+ *
+ * @return
+ *   0 on success, errno value on failure.
+ */
+static int
+rxq_alloc_elts(struct rxq *rxq, unsigned int elts_n, struct rte_mbuf **pool)
+{
+	unsigned int i;
+	struct rxq_elt (*elts)[elts_n] =
+		rte_calloc_socket("RXQ elements", 1, sizeof(*elts), 0,
+				  rxq->socket);
+	int ret = 0;
+
+	if (elts == NULL) {
+		ERROR("%p: can't allocate packets array", (void *)rxq);
+		ret = ENOMEM;
+		goto error;
+	}
+	/* For each WR (packet). */
+	for (i = 0; (i != elts_n); ++i) {
+		struct rxq_elt *elt = &(*elts)[i];
+		struct ibv_recv_wr *wr = &elt->wr;
+		struct ibv_sge *sge = &(*elts)[i].sge;
+		struct rte_mbuf *buf;
+
+		if (pool != NULL) {
+			buf = *(pool++);
+			assert(buf != NULL);
+			rte_pktmbuf_reset(buf);
+		} else
+			buf = rte_pktmbuf_alloc(rxq->mp);
+		if (buf == NULL) {
+			assert(pool == NULL);
+			ERROR("%p: empty mbuf pool", (void *)rxq);
+			ret = ENOMEM;
+			goto error;
+		}
+		/* Configure WR. Work request ID contains its own index in
+		 * the elts array and the offset between SGE buffer header and
+		 * its data. */
+		WR_ID(wr->wr_id).id = i;
+		WR_ID(wr->wr_id).offset =
+			(((uintptr_t)buf->buf_addr + RTE_PKTMBUF_HEADROOM) -
+			 (uintptr_t)buf);
+		wr->next = &(*elts)[(i + 1)].wr;
+		wr->sg_list = sge;
+		wr->num_sge = 1;
+		/* Headroom is reserved by rte_pktmbuf_alloc(). */
+		assert(DATA_OFF(buf) == RTE_PKTMBUF_HEADROOM);
+		/* Buffer is supposed to be empty. */
+		assert(rte_pktmbuf_data_len(buf) == 0);
+		assert(rte_pktmbuf_pkt_len(buf) == 0);
+		/* sge->addr must be able to store a pointer. */
+		assert(sizeof(sge->addr) >= sizeof(uintptr_t));
+		/* SGE keeps its headroom. */
+		sge->addr = (uintptr_t)
+			((uint8_t *)buf->buf_addr + RTE_PKTMBUF_HEADROOM);
+		sge->length = (buf->buf_len - RTE_PKTMBUF_HEADROOM);
+		sge->lkey = rxq->mr->lkey;
+		/* Redundant check for tailroom. */
+		assert(sge->length == rte_pktmbuf_tailroom(buf));
+		/* Make sure elts index and SGE mbuf pointer can be deduced
+		 * from WR ID. */
+		if ((WR_ID(wr->wr_id).id != i) ||
+		    ((void *)((uintptr_t)sge->addr -
+			WR_ID(wr->wr_id).offset) != buf)) {
+			ERROR("%p: cannot store index and offset in WR ID",
+			      (void *)rxq);
+			sge->addr = 0;
+			rte_pktmbuf_free(buf);
+			ret = EOVERFLOW;
+			goto error;
+		}
+	}
+	/* The last WR pointer must be NULL. */
+	(*elts)[(i - 1)].wr.next = NULL;
+	DEBUG("%p: allocated and configured %u single-segment WRs",
+	      (void *)rxq, elts_n);
+	rxq->elts_n = elts_n;
+	rxq->elts_head = 0;
+	rxq->elts.no_sp = elts;
+	assert(ret == 0);
+	return 0;
+error:
+	if (elts != NULL) {
+		assert(pool == NULL);
+		for (i = 0; (i != RTE_DIM(*elts)); ++i) {
+			struct rxq_elt *elt = &(*elts)[i];
+			struct rte_mbuf *buf;
+
+			if (elt->sge.addr == 0)
+				continue;
+			assert(WR_ID(elt->wr.wr_id).id == i);
+			buf = (void *)((uintptr_t)elt->sge.addr -
+				WR_ID(elt->wr.wr_id).offset);
+			rte_pktmbuf_free_seg(buf);
+		}
+		rte_free(elts);
+	}
+	DEBUG("%p: failed, freed everything", (void *)rxq);
+	assert(ret > 0);
+	return ret;
+}
+
+/**
+ * Free RX queue elements.
+ *
+ * @param rxq
+ *   Pointer to RX queue structure.
+ */
+static void
+rxq_free_elts(struct rxq *rxq)
+{
+	unsigned int i;
+	unsigned int elts_n = rxq->elts_n;
+	struct rxq_elt (*elts)[elts_n] = rxq->elts.no_sp;
+
+	DEBUG("%p: freeing WRs", (void *)rxq);
+	rxq->elts_n = 0;
+	rxq->elts.no_sp = NULL;
+	if (elts == NULL)
+		return;
+	for (i = 0; (i != RTE_DIM(*elts)); ++i) {
+		struct rxq_elt *elt = &(*elts)[i];
+		struct rte_mbuf *buf;
+
+		if (elt->sge.addr == 0)
+			continue;
+		assert(WR_ID(elt->wr.wr_id).id == i);
+		buf = (void *)((uintptr_t)elt->sge.addr -
+			WR_ID(elt->wr.wr_id).offset);
+		rte_pktmbuf_free_seg(buf);
+	}
+	rte_free(elts);
+}
+
+/**
+ * Clean up a RX queue.
+ *
+ * Destroy objects, free allocated memory and reset the structure for reuse.
+ *
+ * @param rxq
+ *   Pointer to RX queue structure.
+ */
+void
+rxq_cleanup(struct rxq *rxq)
+{
+	struct ibv_exp_release_intf_params params;
+
+	DEBUG("cleaning up %p", (void *)rxq);
+	rxq_free_elts(rxq);
+	if (rxq->if_qp != NULL) {
+		assert(rxq->priv != NULL);
+		assert(rxq->priv->ctx != NULL);
+		assert(rxq->qp != NULL);
+		params = (struct ibv_exp_release_intf_params){
+			.comp_mask = 0,
+		};
+		claim_zero(ibv_exp_release_intf(rxq->priv->ctx,
+						rxq->if_qp,
+						&params));
+	}
+	if (rxq->if_cq != NULL) {
+		assert(rxq->priv != NULL);
+		assert(rxq->priv->ctx != NULL);
+		assert(rxq->cq != NULL);
+		params = (struct ibv_exp_release_intf_params){
+			.comp_mask = 0,
+		};
+		claim_zero(ibv_exp_release_intf(rxq->priv->ctx,
+						rxq->if_cq,
+						&params));
+	}
+	if (rxq->qp != NULL) {
+		claim_zero(ibv_destroy_qp(rxq->qp));
+	}
+	if (rxq->cq != NULL)
+		claim_zero(ibv_destroy_cq(rxq->cq));
+	if (rxq->rd != NULL) {
+		struct ibv_exp_destroy_res_domain_attr attr = {
+			.comp_mask = 0,
+		};
+
+		assert(rxq->priv != NULL);
+		assert(rxq->priv->ctx != NULL);
+		claim_zero(ibv_exp_destroy_res_domain(rxq->priv->ctx,
+						      rxq->rd,
+						      &attr));
+	}
+	if (rxq->mr != NULL)
+		claim_zero(ibv_dereg_mr(rxq->mr));
+	memset(rxq, 0, sizeof(*rxq));
+}
+
+/**
+ * Allocate a Queue Pair.
+ * Optionally setup inline receive if supported.
+ *
+ * @param priv
+ *   Pointer to private structure.
+ * @param cq
+ *   Completion queue to associate with QP.
+ * @param desc
+ *   Number of descriptors in QP (hint only).
+ *
+ * @return
+ *   QP pointer or NULL in case of error.
+ */
+static struct ibv_qp *
+rxq_setup_qp(struct priv *priv, struct ibv_cq *cq, uint16_t desc,
+	     struct ibv_exp_res_domain *rd)
+{
+	struct ibv_exp_qp_init_attr attr = {
+		/* CQ to be associated with the send queue. */
+		.send_cq = cq,
+		/* CQ to be associated with the receive queue. */
+		.recv_cq = cq,
+		.cap = {
+			/* Max number of outstanding WRs. */
+			.max_recv_wr = ((priv->device_attr.max_qp_wr < desc) ?
+					priv->device_attr.max_qp_wr :
+					desc),
+			/* Max number of scatter/gather elements in a WR. */
+			.max_recv_sge = ((priv->device_attr.max_sge <
+					  MLX5_PMD_SGE_WR_N) ?
+					 priv->device_attr.max_sge :
+					 MLX5_PMD_SGE_WR_N),
+		},
+		.qp_type = IBV_QPT_RAW_PACKET,
+		.comp_mask = (IBV_EXP_QP_INIT_ATTR_PD |
+			      IBV_EXP_QP_INIT_ATTR_RES_DOMAIN),
+		.pd = priv->pd,
+		.res_domain = rd,
+	};
+
+	return ibv_exp_create_qp(priv->ctx, &attr);
+}
+
+#ifdef RSS_SUPPORT
+
+/**
+ * Allocate a RSS Queue Pair.
+ * Optionally setup inline receive if supported.
+ *
+ * @param priv
+ *   Pointer to private structure.
+ * @param cq
+ *   Completion queue to associate with QP.
+ * @param desc
+ *   Number of descriptors in QP (hint only).
+ * @param parent
+ *   If nonzero, create a parent QP, otherwise a child.
+ *
+ * @return
+ *   QP pointer or NULL in case of error.
+ */
+static struct ibv_qp *
+rxq_setup_qp_rss(struct priv *priv, struct ibv_cq *cq, uint16_t desc,
+		 int parent, struct ibv_exp_res_domain *rd)
+{
+	struct ibv_exp_qp_init_attr attr = {
+		/* CQ to be associated with the send queue. */
+		.send_cq = cq,
+		/* CQ to be associated with the receive queue. */
+		.recv_cq = cq,
+		.cap = {
+			/* Max number of outstanding WRs. */
+			.max_recv_wr = ((priv->device_attr.max_qp_wr < desc) ?
+					priv->device_attr.max_qp_wr :
+					desc),
+			/* Max number of scatter/gather elements in a WR. */
+			.max_recv_sge = ((priv->device_attr.max_sge <
+					  MLX5_PMD_SGE_WR_N) ?
+					 priv->device_attr.max_sge :
+					 MLX5_PMD_SGE_WR_N),
+		},
+		.qp_type = IBV_QPT_RAW_PACKET,
+		.comp_mask = (IBV_EXP_QP_INIT_ATTR_PD |
+			      IBV_EXP_QP_INIT_ATTR_RES_DOMAIN |
+			      IBV_EXP_QP_INIT_ATTR_QPG),
+		.pd = priv->pd,
+		.res_domain = rd,
+	};
+
+	if (parent) {
+		attr.qpg.qpg_type = IBV_EXP_QPG_PARENT;
+		/* TSS isn't necessary. */
+		attr.qpg.parent_attrib.tss_child_count = 0;
+		attr.qpg.parent_attrib.rss_child_count = priv->rxqs_n;
+		DEBUG("initializing parent RSS queue");
+	} else {
+		attr.qpg.qpg_type = IBV_EXP_QPG_CHILD_RX;
+		attr.qpg.qpg_parent = priv->rxq_parent.qp;
+		DEBUG("initializing child RSS queue");
+	}
+	return ibv_exp_create_qp(priv->ctx, &attr);
+}
+
+#endif /* RSS_SUPPORT */
+
+/**
+ * Configure a RX queue.
+ *
+ * @param dev
+ *   Pointer to Ethernet device structure.
+ * @param rxq
+ *   Pointer to RX queue structure.
+ * @param desc
+ *   Number of descriptors to configure in queue.
+ * @param socket
+ *   NUMA socket on which memory must be allocated.
+ * @param[in] conf
+ *   Thresholds parameters.
+ * @param mp
+ *   Memory pool for buffer allocations.
+ *
+ * @return
+ *   0 on success, errno value on failure.
+ */
+int
+rxq_setup(struct rte_eth_dev *dev, struct rxq *rxq, uint16_t desc,
+	  unsigned int socket, const struct rte_eth_rxconf *conf,
+	  struct rte_mempool *mp)
+{
+	struct priv *priv = dev->data->dev_private;
+	struct rxq tmpl = {
+		.priv = priv,
+		.mp = mp,
+		.socket = socket
+	};
+	struct ibv_exp_qp_attr mod;
+	union {
+		struct ibv_exp_query_intf_params params;
+		struct ibv_exp_cq_init_attr cq;
+		struct ibv_exp_res_domain_init_attr rd;
+	} attr;
+	enum ibv_exp_query_intf_status status;
+	struct ibv_recv_wr *bad_wr;
+	struct rte_mbuf *buf;
+	int ret = 0;
+	int parent = (rxq == &priv->rxq_parent);
+
+	(void)conf; /* Thresholds configuration (ignored). */
+	/*
+	 * If this is a parent queue, hardware must support RSS and
+	 * RSS must be enabled.
+	 */
+	assert((!parent) || ((priv->hw_rss) && (priv->rss)));
+	if (parent) {
+		/* Even if unused, ibv_create_cq() requires at least one
+		 * descriptor. */
+		desc = 1;
+		goto skip_mr;
+	}
+	if ((desc == 0) || (desc % MLX5_PMD_SGE_WR_N)) {
+		ERROR("%p: invalid number of RX descriptors (must be a"
+		      " multiple of %d)", (void *)dev, MLX5_PMD_SGE_WR_N);
+		return EINVAL;
+	}
+	/* Get mbuf length. */
+	buf = rte_pktmbuf_alloc(mp);
+	if (buf == NULL) {
+		ERROR("%p: unable to allocate mbuf", (void *)dev);
+		return ENOMEM;
+	}
+	tmpl.mb_len = buf->buf_len;
+	assert((rte_pktmbuf_headroom(buf) +
+		rte_pktmbuf_tailroom(buf)) == tmpl.mb_len);
+	assert(rte_pktmbuf_headroom(buf) == RTE_PKTMBUF_HEADROOM);
+	rte_pktmbuf_free(buf);
+	/* Use the entire RX mempool as the memory region. */
+	tmpl.mr = ibv_reg_mr(priv->pd,
+			     (void *)mp->elt_va_start,
+			     (mp->elt_va_end - mp->elt_va_start),
+			     (IBV_ACCESS_LOCAL_WRITE |
+			      IBV_ACCESS_REMOTE_WRITE));
+	if (tmpl.mr == NULL) {
+		ret = EINVAL;
+		ERROR("%p: MR creation failure: %s",
+		      (void *)dev, strerror(ret));
+		goto error;
+	}
+skip_mr:
+	attr.rd = (struct ibv_exp_res_domain_init_attr){
+		.comp_mask = (IBV_EXP_RES_DOMAIN_THREAD_MODEL |
+			      IBV_EXP_RES_DOMAIN_MSG_MODEL),
+		.thread_model = IBV_EXP_THREAD_SINGLE,
+		.msg_model = IBV_EXP_MSG_HIGH_BW,
+	};
+	tmpl.rd = ibv_exp_create_res_domain(priv->ctx, &attr.rd);
+	if (tmpl.rd == NULL) {
+		ret = ENOMEM;
+		ERROR("%p: RD creation failure: %s",
+		      (void *)dev, strerror(ret));
+		goto error;
+	}
+	attr.cq = (struct ibv_exp_cq_init_attr){
+		.comp_mask = IBV_EXP_CQ_INIT_ATTR_RES_DOMAIN,
+		.res_domain = tmpl.rd,
+	};
+	tmpl.cq = ibv_exp_create_cq(priv->ctx, desc, NULL, NULL, 0, &attr.cq);
+	if (tmpl.cq == NULL) {
+		ret = ENOMEM;
+		ERROR("%p: CQ creation failure: %s",
+		      (void *)dev, strerror(ret));
+		goto error;
+	}
+	DEBUG("priv->device_attr.max_qp_wr is %d",
+	      priv->device_attr.max_qp_wr);
+	DEBUG("priv->device_attr.max_sge is %d",
+	      priv->device_attr.max_sge);
+#ifdef RSS_SUPPORT
+	if (priv->rss)
+		tmpl.qp = rxq_setup_qp_rss(priv, tmpl.cq, desc, parent,
+					   tmpl.rd);
+	else
+#endif /* RSS_SUPPORT */
+		tmpl.qp = rxq_setup_qp(priv, tmpl.cq, desc, tmpl.rd);
+	if (tmpl.qp == NULL) {
+		ret = (errno ? errno : EINVAL);
+		ERROR("%p: QP creation failure: %s",
+		      (void *)dev, strerror(ret));
+		goto error;
+	}
+	mod = (struct ibv_exp_qp_attr){
+		/* Move the QP to this state. */
+		.qp_state = IBV_QPS_INIT,
+		/* Primary port number. */
+		.port_num = priv->port
+	};
+	ret = ibv_exp_modify_qp(tmpl.qp, &mod,
+				(IBV_EXP_QP_STATE |
+#ifdef RSS_SUPPORT
+				 (parent ? IBV_EXP_QP_GROUP_RSS : 0) |
+#endif /* RSS_SUPPORT */
+				 IBV_EXP_QP_PORT));
+	if (ret) {
+		ERROR("%p: QP state to IBV_QPS_INIT failed: %s",
+		      (void *)dev, strerror(ret));
+		goto error;
+	}
+	/* Allocate descriptors for RX queues, except for the RSS parent. */
+	if (parent)
+		goto skip_alloc;
+	ret = rxq_alloc_elts(&tmpl, desc, NULL);
+	if (ret) {
+		ERROR("%p: RXQ allocation failed: %s",
+		      (void *)dev, strerror(ret));
+		goto error;
+	}
+	ret = ibv_post_recv(tmpl.qp,
+			    &(*tmpl.elts.no_sp)[0].wr,
+			    &bad_wr);
+	if (ret) {
+		ERROR("%p: ibv_post_recv() failed for WR %p: %s",
+		      (void *)dev,
+		      (void *)bad_wr,
+		      strerror(ret));
+		goto error;
+	}
+skip_alloc:
+	mod = (struct ibv_exp_qp_attr){
+		.qp_state = IBV_QPS_RTR
+	};
+	ret = ibv_exp_modify_qp(tmpl.qp, &mod, IBV_EXP_QP_STATE);
+	if (ret) {
+		ERROR("%p: QP state to IBV_QPS_RTR failed: %s",
+		      (void *)dev, strerror(ret));
+		goto error;
+	}
+	/* Save port ID. */
+	tmpl.port_id = dev->data->port_id;
+	DEBUG("%p: RTE port ID: %u", (void *)rxq, tmpl.port_id);
+	attr.params = (struct ibv_exp_query_intf_params){
+		.intf_scope = IBV_EXP_INTF_GLOBAL,
+		.intf = IBV_EXP_INTF_CQ,
+		.obj = tmpl.cq,
+	};
+	tmpl.if_cq = ibv_exp_query_intf(priv->ctx, &attr.params, &status);
+	if (tmpl.if_cq == NULL) {
+		ERROR("%p: CQ interface family query failed with status %d",
+		      (void *)dev, status);
+		goto error;
+	}
+	attr.params = (struct ibv_exp_query_intf_params){
+		.intf_scope = IBV_EXP_INTF_GLOBAL,
+		.intf = IBV_EXP_INTF_QP_BURST,
+		.obj = tmpl.qp,
+	};
+	tmpl.if_qp = ibv_exp_query_intf(priv->ctx, &attr.params, &status);
+	if (tmpl.if_qp == NULL) {
+		ERROR("%p: QP interface family query failed with status %d",
+		      (void *)dev, status);
+		goto error;
+	}
+	/* Clean up rxq in case we're reinitializing it. */
+	DEBUG("%p: cleaning-up old rxq just in case", (void *)rxq);
+	rxq_cleanup(rxq);
+	*rxq = tmpl;
+	DEBUG("%p: rxq updated with %p", (void *)rxq, (void *)&tmpl);
+	assert(ret == 0);
+	return 0;
+error:
+	rxq_cleanup(&tmpl);
+	assert(ret > 0);
+	return ret;
+}
+
+/**
+ * DPDK callback to configure a RX queue.
+ *
+ * @param dev
+ *   Pointer to Ethernet device structure.
+ * @param idx
+ *   RX queue index.
+ * @param desc
+ *   Number of descriptors to configure in queue.
+ * @param socket
+ *   NUMA socket on which memory must be allocated.
+ * @param[in] conf
+ *   Thresholds parameters.
+ * @param mp
+ *   Memory pool for buffer allocations.
+ *
+ * @return
+ *   0 on success, negative errno value on failure.
+ */
+int
+mlx5_rx_queue_setup(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc,
+		    unsigned int socket, const struct rte_eth_rxconf *conf,
+		    struct rte_mempool *mp)
+{
+	struct priv *priv = dev->data->dev_private;
+	struct rxq *rxq = (*priv->rxqs)[idx];
+	int ret;
+
+	priv_lock(priv);
+	DEBUG("%p: configuring queue %u for %u descriptors",
+	      (void *)dev, idx, desc);
+	if (idx >= priv->rxqs_n) {
+		ERROR("%p: queue index out of range (%u >= %u)",
+		      (void *)dev, idx, priv->rxqs_n);
+		priv_unlock(priv);
+		return -EOVERFLOW;
+	}
+	if (rxq != NULL) {
+		DEBUG("%p: reusing already allocated queue index %u (%p)",
+		      (void *)dev, idx, (void *)rxq);
+		if (priv->started) {
+			priv_unlock(priv);
+			return -EEXIST;
+		}
+		(*priv->rxqs)[idx] = NULL;
+		rxq_cleanup(rxq);
+	} else {
+		rxq = rte_calloc_socket("RXQ", 1, sizeof(*rxq), 0, socket);
+		if (rxq == NULL) {
+			ERROR("%p: unable to allocate queue index %u",
+			      (void *)dev, idx);
+			priv_unlock(priv);
+			return -ENOMEM;
+		}
+	}
+	ret = rxq_setup(dev, rxq, desc, socket, conf, mp);
+	if (ret)
+		rte_free(rxq);
+	else {
+		DEBUG("%p: adding RX queue %p to list",
+		      (void *)dev, (void *)rxq);
+		(*priv->rxqs)[idx] = rxq;
+		/* Update receive callback. */
+		dev->rx_pkt_burst = mlx5_rx_burst;
+	}
+	priv_unlock(priv);
+	return -ret;
+}
+
+/**
+ * DPDK callback to release a RX queue.
+ *
+ * @param dpdk_rxq
+ *   Generic RX queue pointer.
+ */
+void
+mlx5_rx_queue_release(void *dpdk_rxq)
+{
+	struct rxq *rxq = (struct rxq *)dpdk_rxq;
+	struct priv *priv;
+	unsigned int i;
+
+	if (rxq == NULL)
+		return;
+	priv = rxq->priv;
+	priv_lock(priv);
+	assert(rxq != &priv->rxq_parent);
+	for (i = 0; (i != priv->rxqs_n); ++i)
+		if ((*priv->rxqs)[i] == rxq) {
+			DEBUG("%p: removing RX queue %p from list",
+			      (void *)priv->dev, (void *)rxq);
+			(*priv->rxqs)[i] = NULL;
+			break;
+		}
+	rxq_cleanup(rxq);
+	rte_free(rxq);
+	priv_unlock(priv);
+}
diff --git a/drivers/net/mlx5/mlx5_rxtx.c b/drivers/net/mlx5/mlx5_rxtx.c
new file mode 100644
index 0000000..0f1e541
--- /dev/null
+++ b/drivers/net/mlx5/mlx5_rxtx.c
@@ -0,0 +1,496 @@
+/*-
+ *   BSD LICENSE
+ *
+ *   Copyright 2015 6WIND S.A.
+ *   Copyright 2015 Mellanox.
+ *
+ *   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 6WIND S.A. 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 <assert.h>
+#include <stdint.h>
+#include <string.h>
+#include <stdlib.h>
+
+/* Verbs header. */
+/* ISO C doesn't support unnamed structs/unions, disabling -pedantic. */
+#ifdef PEDANTIC
+#pragma GCC diagnostic ignored "-pedantic"
+#endif
+#include <infiniband/verbs.h>
+#ifdef PEDANTIC
+#pragma GCC diagnostic error "-pedantic"
+#endif
+
+/* DPDK headers don't like -pedantic. */
+#ifdef PEDANTIC
+#pragma GCC diagnostic ignored "-pedantic"
+#endif
+#include <rte_mbuf.h>
+#include <rte_mempool.h>
+#include <rte_prefetch.h>
+#include <rte_common.h>
+#include <rte_branch_prediction.h>
+#ifdef PEDANTIC
+#pragma GCC diagnostic error "-pedantic"
+#endif
+
+#include "mlx5.h"
+#include "mlx5_utils.h"
+#include "mlx5_rxtx.h"
+#include "mlx5_defs.h"
+
+/**
+ * Manage TX completions.
+ *
+ * When sending a burst, mlx5_tx_burst() posts several WRs.
+ * To improve performance, a completion event is only required once every
+ * MLX5_PMD_TX_PER_COMP_REQ sends. Doing so discards completion information
+ * for other WRs, but this information would not be used anyway.
+ *
+ * @param txq
+ *   Pointer to TX queue structure.
+ *
+ * @return
+ *   0 on success, -1 on failure.
+ */
+static int
+txq_complete(struct txq *txq)
+{
+	unsigned int elts_comp = txq->elts_comp;
+	unsigned int elts_tail = txq->elts_tail;
+	const unsigned int elts_n = txq->elts_n;
+	int wcs_n;
+
+	if (unlikely(elts_comp == 0))
+		return 0;
+#ifdef DEBUG_SEND
+	DEBUG("%p: processing %u work requests completions",
+	      (void *)txq, elts_comp);
+#endif
+	wcs_n = txq->if_cq->poll_cnt(txq->cq, elts_comp);
+	if (unlikely(wcs_n == 0))
+		return 0;
+	if (unlikely(wcs_n < 0)) {
+		DEBUG("%p: ibv_poll_cq() failed (wcs_n=%d)",
+		      (void *)txq, wcs_n);
+		return -1;
+	}
+	elts_comp -= wcs_n;
+	assert(elts_comp <= txq->elts_comp);
+	/*
+	 * Assume WC status is successful as nothing can be done about it
+	 * anyway.
+	 */
+	elts_tail += wcs_n * txq->elts_comp_cd_init;
+	if (elts_tail >= elts_n)
+		elts_tail -= elts_n;
+	txq->elts_tail = elts_tail;
+	txq->elts_comp = elts_comp;
+	return 0;
+}
+
+/**
+ * Get Memory Region (MR) <-> Memory Pool (MP) association from txq->mp2mr[].
+ * Add MP to txq->mp2mr[] if it's not registered yet. If mp2mr[] is full,
+ * remove an entry first.
+ *
+ * @param txq
+ *   Pointer to TX queue structure.
+ * @param[in] mp
+ *   Memory Pool for which a Memory Region lkey must be returned.
+ *
+ * @return
+ *   mr->lkey on success, (uint32_t)-1 on failure.
+ */
+static uint32_t
+txq_mp2mr(struct txq *txq, struct rte_mempool *mp)
+{
+	unsigned int i;
+	struct ibv_mr *mr;
+
+	for (i = 0; (i != RTE_DIM(txq->mp2mr)); ++i) {
+		if (unlikely(txq->mp2mr[i].mp == NULL)) {
+			/* Unknown MP, add a new MR for it. */
+			break;
+		}
+		if (txq->mp2mr[i].mp == mp) {
+			assert(txq->mp2mr[i].lkey != (uint32_t)-1);
+			assert(txq->mp2mr[i].mr->lkey == txq->mp2mr[i].lkey);
+			return txq->mp2mr[i].lkey;
+		}
+	}
+	/* Add a new entry, register MR first. */
+	DEBUG("%p: discovered new memory pool %p", (void *)txq, (void *)mp);
+	mr = ibv_reg_mr(txq->priv->pd,
+			(void *)mp->elt_va_start,
+			(mp->elt_va_end - mp->elt_va_start),
+			(IBV_ACCESS_LOCAL_WRITE | IBV_ACCESS_REMOTE_WRITE));
+	if (unlikely(mr == NULL)) {
+		DEBUG("%p: unable to configure MR, ibv_reg_mr() failed.",
+		      (void *)txq);
+		return (uint32_t)-1;
+	}
+	if (unlikely(i == RTE_DIM(txq->mp2mr))) {
+		/* Table is full, remove oldest entry. */
+		DEBUG("%p: MR <-> MP table full, dropping oldest entry.",
+		      (void *)txq);
+		--i;
+		claim_zero(ibv_dereg_mr(txq->mp2mr[i].mr));
+		memmove(&txq->mp2mr[0], &txq->mp2mr[1],
+			(sizeof(txq->mp2mr) - sizeof(txq->mp2mr[0])));
+	}
+	/* Store the new entry. */
+	txq->mp2mr[i].mp = mp;
+	txq->mp2mr[i].mr = mr;
+	txq->mp2mr[i].lkey = mr->lkey;
+	DEBUG("%p: new MR lkey for MP %p: 0x%08" PRIu32,
+	      (void *)txq, (void *)mp, txq->mp2mr[i].lkey);
+	return txq->mp2mr[i].lkey;
+}
+
+/**
+ * DPDK callback for TX.
+ *
+ * @param dpdk_txq
+ *   Generic pointer to TX queue structure.
+ * @param[in] pkts
+ *   Packets to transmit.
+ * @param pkts_n
+ *   Number of packets in array.
+ *
+ * @return
+ *   Number of packets successfully transmitted (<= pkts_n).
+ */
+uint16_t
+mlx5_tx_burst(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n)
+{
+	struct txq *txq = (struct txq *)dpdk_txq;
+	unsigned int elts_head = txq->elts_head;
+	const unsigned int elts_tail = txq->elts_tail;
+	const unsigned int elts_n = txq->elts_n;
+	unsigned int elts_comp_cd = txq->elts_comp_cd;
+	unsigned int elts_comp = 0;
+	unsigned int i;
+	unsigned int max;
+	int err;
+
+	assert(elts_comp_cd != 0);
+	txq_complete(txq);
+	max = (elts_n - (elts_head - elts_tail));
+	if (max > elts_n)
+		max -= elts_n;
+	assert(max >= 1);
+	assert(max <= elts_n);
+	/* Always leave one free entry in the ring. */
+	--max;
+	if (max == 0)
+		return 0;
+	if (max > pkts_n)
+		max = pkts_n;
+	for (i = 0; (i != max); ++i) {
+		struct rte_mbuf *buf = pkts[i];
+		unsigned int elts_head_next =
+			(((elts_head + 1) == elts_n) ? 0 : elts_head + 1);
+		struct txq_elt *elt_next = &(*txq->elts)[elts_head_next];
+		struct txq_elt *elt = &(*txq->elts)[elts_head];
+		unsigned int segs = NB_SEGS(buf);
+		uint32_t send_flags = 0;
+
+		/* Clean up old buffer. */
+		if (likely(elt->buf != NULL)) {
+			struct rte_mbuf *tmp = elt->buf;
+
+			/* Faster than rte_pktmbuf_free(). */
+			do {
+				struct rte_mbuf *next = NEXT(tmp);
+
+				rte_pktmbuf_free_seg(tmp);
+				tmp = next;
+			} while (tmp != NULL);
+		}
+		/* Request TX completion. */
+		if (unlikely(--elts_comp_cd == 0)) {
+			elts_comp_cd = txq->elts_comp_cd_init;
+			++elts_comp;
+			send_flags |= IBV_EXP_QP_BURST_SIGNALED;
+		}
+		if (likely(segs == 1)) {
+			uintptr_t addr;
+			uint32_t length;
+			uint32_t lkey;
+
+			/* Retrieve buffer information. */
+			addr = rte_pktmbuf_mtod(buf, uintptr_t);
+			length = DATA_LEN(buf);
+			/* Retrieve Memory Region key for this memory pool. */
+			lkey = txq_mp2mr(txq, buf->pool);
+			if (unlikely(lkey == (uint32_t)-1)) {
+				/* MR does not exist. */
+				DEBUG("%p: unable to get MP <-> MR"
+				      " association", (void *)txq);
+				/* Clean up TX element. */
+				elt->buf = NULL;
+				goto stop;
+			}
+			/* Update element. */
+			elt->buf = buf;
+			if (txq->priv->vf)
+				rte_prefetch0((volatile void *)
+					      (uintptr_t)addr);
+			RTE_MBUF_PREFETCH_TO_FREE(elt_next->buf);
+			/* Put packet into send queue. */
+#if MLX5_PMD_MAX_INLINE > 0
+			if (length <= txq->max_inline)
+				err = txq->if_qp->send_pending_inline
+					(txq->qp,
+					 (void *)addr,
+					 length,
+					 send_flags);
+			else
+#endif
+				err = txq->if_qp->send_pending
+					(txq->qp,
+					 addr,
+					 length,
+					 lkey,
+					 send_flags);
+			if (unlikely(err))
+				goto stop;
+		} else {
+			DEBUG("%p: TX scattered buffers support not"
+			      " compiled in", (void *)txq);
+			goto stop;
+		}
+		elts_head = elts_head_next;
+	}
+stop:
+	/* Take a shortcut if nothing must be sent. */
+	if (unlikely(i == 0))
+		return 0;
+	/* Ring QP doorbell. */
+	err = txq->if_qp->send_flush(txq->qp);
+	if (unlikely(err)) {
+		/* A nonzero value is not supposed to be returned.
+		 * Nothing can be done about it. */
+		DEBUG("%p: send_flush() failed with error %d",
+		      (void *)txq, err);
+	}
+	txq->elts_head = elts_head;
+	txq->elts_comp += elts_comp;
+	txq->elts_comp_cd = elts_comp_cd;
+	return i;
+}
+
+/**
+ * DPDK callback for RX.
+ *
+ * @param dpdk_rxq
+ *   Generic pointer to RX queue structure.
+ * @param[out] pkts
+ *   Array to store received packets.
+ * @param pkts_n
+ *   Maximum number of packets in array.
+ *
+ * @return
+ *   Number of packets successfully received (<= pkts_n).
+ */
+uint16_t
+mlx5_rx_burst(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n)
+{
+	struct rxq *rxq = (struct rxq *)dpdk_rxq;
+	struct rxq_elt (*elts)[rxq->elts_n] = rxq->elts.no_sp;
+	const unsigned int elts_n = rxq->elts_n;
+	unsigned int elts_head = rxq->elts_head;
+	struct ibv_sge sges[pkts_n];
+	unsigned int i;
+	unsigned int pkts_ret = 0;
+	int ret;
+
+	for (i = 0; (i != pkts_n); ++i) {
+		struct rxq_elt *elt = &(*elts)[elts_head];
+		struct ibv_recv_wr *wr = &elt->wr;
+		uint64_t wr_id = wr->wr_id;
+		unsigned int len;
+		struct rte_mbuf *seg = (void *)((uintptr_t)elt->sge.addr -
+			WR_ID(wr_id).offset);
+		struct rte_mbuf *rep;
+		uint32_t flags;
+
+		/* Sanity checks. */
+		assert(WR_ID(wr_id).id < rxq->elts_n);
+		assert(wr->sg_list == &elt->sge);
+		assert(wr->num_sge == 1);
+		assert(elts_head < rxq->elts_n);
+		assert(rxq->elts_head < rxq->elts_n);
+		/*
+		 * Fetch initial bytes of packet descriptor into a
+		 * cacheline while allocating rep.
+		 */
+		rte_prefetch0(seg);
+		rte_prefetch0(&seg->cacheline1);
+		ret = rxq->if_cq->poll_length_flags(rxq->cq, NULL, NULL,
+						    &flags);
+		if (unlikely(ret < 0)) {
+			struct ibv_wc wc;
+			int wcs_n;
+
+			DEBUG("rxq=%p, poll_length() failed (ret=%d)",
+			      (void *)rxq, ret);
+			/* ibv_poll_cq() must be used in case of failure. */
+			wcs_n = ibv_poll_cq(rxq->cq, 1, &wc);
+			if (unlikely(wcs_n == 0))
+				break;
+			if (unlikely(wcs_n < 0)) {
+				DEBUG("rxq=%p, ibv_poll_cq() failed (wcs_n=%d)",
+				      (void *)rxq, wcs_n);
+				break;
+			}
+			assert(wcs_n == 1);
+			if (unlikely(wc.status != IBV_WC_SUCCESS)) {
+				/* Whatever, just repost the offending WR. */
+				DEBUG("rxq=%p, wr_id=%" PRIu64 ": bad work"
+				      " completion status (%d): %s",
+				      (void *)rxq, wc.wr_id, wc.status,
+				      ibv_wc_status_str(wc.status));
+				/* Add SGE to array for repost. */
+				sges[i] = elt->sge;
+				goto repost;
+			}
+			ret = wc.byte_len;
+		}
+		if (ret == 0)
+			break;
+		len = ret;
+		rep = __rte_mbuf_raw_alloc(rxq->mp);
+		if (unlikely(rep == NULL)) {
+			/*
+			 * Unable to allocate a replacement mbuf,
+			 * repost WR.
+			 */
+			DEBUG("rxq=%p, wr_id=%" PRIu32 ":"
+			      " can't allocate a new mbuf",
+			      (void *)rxq, WR_ID(wr_id).id);
+			/* Increment out of memory counters. */
+			++rxq->priv->dev->data->rx_mbuf_alloc_failed;
+			goto repost;
+		}
+
+		/* Reconfigure sge to use rep instead of seg. */
+		elt->sge.addr = (uintptr_t)rep->buf_addr + RTE_PKTMBUF_HEADROOM;
+		assert(elt->sge.lkey == rxq->mr->lkey);
+		WR_ID(wr->wr_id).offset =
+			(((uintptr_t)rep->buf_addr + RTE_PKTMBUF_HEADROOM) -
+			 (uintptr_t)rep);
+		assert(WR_ID(wr->wr_id).id == WR_ID(wr_id).id);
+
+		/* Add SGE to array for repost. */
+		sges[i] = elt->sge;
+
+		/* Update seg information. */
+		SET_DATA_OFF(seg, RTE_PKTMBUF_HEADROOM);
+		NB_SEGS(seg) = 1;
+		PORT(seg) = rxq->port_id;
+		NEXT(seg) = NULL;
+		PKT_LEN(seg) = len;
+		DATA_LEN(seg) = len;
+
+		/* Return packet. */
+		*(pkts++) = seg;
+		++pkts_ret;
+repost:
+		if (++elts_head >= elts_n)
+			elts_head = 0;
+		continue;
+	}
+	if (unlikely(i == 0))
+		return 0;
+	/* Repost WRs. */
+#ifdef DEBUG_RECV
+	DEBUG("%p: reposting %u WRs", (void *)rxq, i);
+#endif
+	ret = rxq->if_qp->recv_burst(rxq->qp, sges, i);
+	if (unlikely(ret)) {
+		/* Inability to repost WRs is fatal. */
+		DEBUG("%p: recv_burst(): failed (ret=%d)",
+		      (void *)rxq->priv,
+		      ret);
+		abort();
+	}
+	rxq->elts_head = elts_head;
+	return pkts_ret;
+}
+
+/**
+ * Dummy DPDK callback for TX.
+ *
+ * This function is used to temporarily replace the real callback during
+ * unsafe control operations on the queue, or in case of error.
+ *
+ * @param dpdk_txq
+ *   Generic pointer to TX queue structure.
+ * @param[in] pkts
+ *   Packets to transmit.
+ * @param pkts_n
+ *   Number of packets in array.
+ *
+ * @return
+ *   Number of packets successfully transmitted (<= pkts_n).
+ */
+uint16_t
+removed_tx_burst(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n)
+{
+	(void)dpdk_txq;
+	(void)pkts;
+	(void)pkts_n;
+	return 0;
+}
+
+/**
+ * Dummy DPDK callback for RX.
+ *
+ * This function is used to temporarily replace the real callback during
+ * unsafe control operations on the queue, or in case of error.
+ *
+ * @param dpdk_rxq
+ *   Generic pointer to RX queue structure.
+ * @param[out] pkts
+ *   Array to store received packets.
+ * @param pkts_n
+ *   Maximum number of packets in array.
+ *
+ * @return
+ *   Number of packets successfully received (<= pkts_n).
+ */
+uint16_t
+removed_rx_burst(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n)
+{
+	(void)dpdk_rxq;
+	(void)pkts;
+	(void)pkts_n;
+	return 0;
+}
diff --git a/drivers/net/mlx5/mlx5_rxtx.h b/drivers/net/mlx5/mlx5_rxtx.h
new file mode 100644
index 0000000..1459317
--- /dev/null
+++ b/drivers/net/mlx5/mlx5_rxtx.h
@@ -0,0 +1,156 @@
+/*-
+ *   BSD LICENSE
+ *
+ *   Copyright 2015 6WIND S.A.
+ *   Copyright 2015 Mellanox.
+ *
+ *   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 6WIND S.A. 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 RTE_PMD_MLX5_RXTX_H_
+#define RTE_PMD_MLX5_RXTX_H_
+
+#include <stdint.h>
+
+/* Verbs header. */
+/* ISO C doesn't support unnamed structs/unions, disabling -pedantic. */
+#ifdef PEDANTIC
+#pragma GCC diagnostic ignored "-pedantic"
+#endif
+#include <infiniband/verbs.h>
+#ifdef PEDANTIC
+#pragma GCC diagnostic error "-pedantic"
+#endif
+
+/* DPDK headers don't like -pedantic. */
+#ifdef PEDANTIC
+#pragma GCC diagnostic ignored "-pedantic"
+#endif
+#include <rte_mbuf.h>
+#include <rte_mempool.h>
+#ifdef PEDANTIC
+#pragma GCC diagnostic error "-pedantic"
+#endif
+
+#include "mlx5_utils.h"
+#include "mlx5.h"
+#include "mlx5_defs.h"
+
+/* RX element. */
+struct rxq_elt {
+	struct ibv_recv_wr wr; /* Work Request. */
+	struct ibv_sge sge; /* Scatter/Gather Element. */
+	/* mbuf pointer is derived from WR_ID(wr.wr_id).offset. */
+};
+
+struct priv;
+
+/* RX queue descriptor. */
+struct rxq {
+	struct priv *priv; /* Back pointer to private data. */
+	struct rte_mempool *mp; /* Memory Pool for allocations. */
+	struct ibv_mr *mr; /* Memory Region (for mp). */
+	struct ibv_cq *cq; /* Completion Queue. */
+	struct ibv_qp *qp; /* Queue Pair. */
+	struct ibv_exp_qp_burst_family *if_qp; /* QP burst interface. */
+	struct ibv_exp_cq_family *if_cq; /* CQ interface. */
+	unsigned int port_id; /* Port ID for incoming packets. */
+	unsigned int elts_n; /* (*elts)[] length. */
+	unsigned int elts_head; /* Current index in (*elts)[]. */
+	union {
+		struct rxq_elt (*no_sp)[]; /* RX elements. */
+	} elts;
+	uint32_t mb_len; /* Length of a mp-issued mbuf. */
+	unsigned int socket; /* CPU socket ID for allocations. */
+	struct ibv_exp_res_domain *rd; /* Resource Domain. */
+};
+
+/* TX element. */
+struct txq_elt {
+	struct rte_mbuf *buf;
+};
+
+/* Linear buffer type. It is used when transmitting buffers with too many
+ * segments that do not fit the hardware queue (see max_send_sge).
+ * Extra segments are copied (linearized) in such buffers, replacing the
+ * last SGE during TX.
+ * The size is arbitrary but large enough to hold a jumbo frame with
+ * 8 segments considering mbuf.buf_len is about 2048 bytes. */
+typedef uint8_t linear_t[16384];
+
+/* TX queue descriptor. */
+struct txq {
+	struct priv *priv; /* Back pointer to private data. */
+	struct {
+		struct rte_mempool *mp; /* Cached Memory Pool. */
+		struct ibv_mr *mr; /* Memory Region (for mp). */
+		uint32_t lkey; /* mr->lkey */
+	} mp2mr[MLX5_PMD_TX_MP_CACHE]; /* MP to MR translation table. */
+	struct ibv_cq *cq; /* Completion Queue. */
+	struct ibv_qp *qp; /* Queue Pair. */
+	struct ibv_exp_qp_burst_family *if_qp; /* QP burst interface. */
+	struct ibv_exp_cq_family *if_cq; /* CQ interface. */
+#if MLX5_PMD_MAX_INLINE > 0
+	uint32_t max_inline; /* Max inline send size <= MLX5_PMD_MAX_INLINE. */
+#endif
+	unsigned int elts_n; /* (*elts)[] length. */
+	struct txq_elt (*elts)[]; /* TX elements. */
+	unsigned int elts_head; /* Current index in (*elts)[]. */
+	unsigned int elts_tail; /* First element awaiting completion. */
+	unsigned int elts_comp; /* Number of completion requests. */
+	unsigned int elts_comp_cd; /* Countdown for next completion request. */
+	unsigned int elts_comp_cd_init; /* Initial value for countdown. */
+	linear_t (*elts_linear)[]; /* Linearized buffers. */
+	struct ibv_mr *mr_linear; /* Memory Region for linearized buffers. */
+	unsigned int socket; /* CPU socket ID for allocations. */
+	struct ibv_exp_res_domain *rd; /* Resource Domain. */
+};
+
+/* mlx5_rxq.c */
+
+void rxq_cleanup(struct rxq *);
+int rxq_setup(struct rte_eth_dev *, struct rxq *, uint16_t, unsigned int,
+	      const struct rte_eth_rxconf *, struct rte_mempool *);
+int mlx5_rx_queue_setup(struct rte_eth_dev *, uint16_t, uint16_t, unsigned int,
+			const struct rte_eth_rxconf *, struct rte_mempool *);
+void mlx5_rx_queue_release(void *);
+
+/* mlx5_txq.c */
+
+void txq_cleanup(struct txq *);
+int mlx5_tx_queue_setup(struct rte_eth_dev *, uint16_t, uint16_t, unsigned int,
+			const struct rte_eth_txconf *);
+void mlx5_tx_queue_release(void *);
+
+/* mlx5_rxtx.c */
+
+uint16_t mlx5_tx_burst(void *, struct rte_mbuf **, uint16_t);
+uint16_t mlx5_rx_burst(void *, struct rte_mbuf **, uint16_t);
+uint16_t removed_tx_burst(void *, struct rte_mbuf **, uint16_t);
+uint16_t removed_rx_burst(void *, struct rte_mbuf **, uint16_t);
+
+#endif /* RTE_PMD_MLX5_RXTX_H_ */
diff --git a/drivers/net/mlx5/mlx5_txq.c b/drivers/net/mlx5/mlx5_txq.c
new file mode 100644
index 0000000..2bae61f
--- /dev/null
+++ b/drivers/net/mlx5/mlx5_txq.c
@@ -0,0 +1,512 @@
+/*-
+ *   BSD LICENSE
+ *
+ *   Copyright 2015 6WIND S.A.
+ *   Copyright 2015 Mellanox.
+ *
+ *   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 6WIND S.A. 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 <stddef.h>
+#include <assert.h>
+#include <errno.h>
+#include <string.h>
+#include <stdint.h>
+
+/* Verbs header. */
+/* ISO C doesn't support unnamed structs/unions, disabling -pedantic. */
+#ifdef PEDANTIC
+#pragma GCC diagnostic ignored "-pedantic"
+#endif
+#include <infiniband/verbs.h>
+#ifdef PEDANTIC
+#pragma GCC diagnostic error "-pedantic"
+#endif
+
+/* DPDK headers don't like -pedantic. */
+#ifdef PEDANTIC
+#pragma GCC diagnostic ignored "-pedantic"
+#endif
+#include <rte_mbuf.h>
+#include <rte_malloc.h>
+#include <rte_ethdev.h>
+#include <rte_common.h>
+#ifdef PEDANTIC
+#pragma GCC diagnostic error "-pedantic"
+#endif
+
+#include "mlx5_utils.h"
+#include "mlx5.h"
+#include "mlx5_rxtx.h"
+#include "mlx5_autoconf.h"
+#include "mlx5_defs.h"
+
+/**
+ * Allocate TX queue elements.
+ *
+ * @param txq
+ *   Pointer to TX queue structure.
+ * @param elts_n
+ *   Number of elements to allocate.
+ *
+ * @return
+ *   0 on success, errno value on failure.
+ */
+static int
+txq_alloc_elts(struct txq *txq, unsigned int elts_n)
+{
+	unsigned int i;
+	struct txq_elt (*elts)[elts_n] =
+		rte_calloc_socket("TXQ", 1, sizeof(*elts), 0, txq->socket);
+	linear_t (*elts_linear)[elts_n] =
+		rte_calloc_socket("TXQ", 1, sizeof(*elts_linear), 0,
+				  txq->socket);
+	struct ibv_mr *mr_linear = NULL;
+	int ret = 0;
+
+	if ((elts == NULL) || (elts_linear == NULL)) {
+		ERROR("%p: can't allocate packets array", (void *)txq);
+		ret = ENOMEM;
+		goto error;
+	}
+	mr_linear =
+		ibv_reg_mr(txq->priv->pd, elts_linear, sizeof(*elts_linear),
+			   (IBV_ACCESS_LOCAL_WRITE | IBV_ACCESS_REMOTE_WRITE));
+	if (mr_linear == NULL) {
+		ERROR("%p: unable to configure MR, ibv_reg_mr() failed",
+		      (void *)txq);
+		ret = EINVAL;
+		goto error;
+	}
+	for (i = 0; (i != elts_n); ++i) {
+		struct txq_elt *elt = &(*elts)[i];
+
+		elt->buf = NULL;
+	}
+	DEBUG("%p: allocated and configured %u WRs", (void *)txq, elts_n);
+	txq->elts_n = elts_n;
+	txq->elts = elts;
+	txq->elts_head = 0;
+	txq->elts_tail = 0;
+	txq->elts_comp = 0;
+	/* Request send completion every MLX5_PMD_TX_PER_COMP_REQ packets or
+	 * at least 4 times per ring. */
+	txq->elts_comp_cd_init =
+		((MLX5_PMD_TX_PER_COMP_REQ < (elts_n / 4)) ?
+		 MLX5_PMD_TX_PER_COMP_REQ : (elts_n / 4));
+	txq->elts_comp_cd = txq->elts_comp_cd_init;
+	txq->elts_linear = elts_linear;
+	txq->mr_linear = mr_linear;
+	assert(ret == 0);
+	return 0;
+error:
+	if (mr_linear != NULL)
+		claim_zero(ibv_dereg_mr(mr_linear));
+
+	rte_free(elts_linear);
+	rte_free(elts);
+
+	DEBUG("%p: failed, freed everything", (void *)txq);
+	assert(ret > 0);
+	return ret;
+}
+
+/**
+ * Free TX queue elements.
+ *
+ * @param txq
+ *   Pointer to TX queue structure.
+ */
+static void
+txq_free_elts(struct txq *txq)
+{
+	unsigned int i;
+	unsigned int elts_n = txq->elts_n;
+	struct txq_elt (*elts)[elts_n] = txq->elts;
+	linear_t (*elts_linear)[elts_n] = txq->elts_linear;
+	struct ibv_mr *mr_linear = txq->mr_linear;
+
+	DEBUG("%p: freeing WRs", (void *)txq);
+	txq->elts_n = 0;
+	txq->elts = NULL;
+	txq->elts_linear = NULL;
+	txq->mr_linear = NULL;
+	if (mr_linear != NULL)
+		claim_zero(ibv_dereg_mr(mr_linear));
+
+	rte_free(elts_linear);
+	if (elts == NULL)
+		return;
+	for (i = 0; (i != RTE_DIM(*elts)); ++i) {
+		struct txq_elt *elt = &(*elts)[i];
+
+		if (elt->buf == NULL)
+			continue;
+		rte_pktmbuf_free(elt->buf);
+	}
+	rte_free(elts);
+}
+
+/**
+ * Clean up a TX queue.
+ *
+ * Destroy objects, free allocated memory and reset the structure for reuse.
+ *
+ * @param txq
+ *   Pointer to TX queue structure.
+ */
+void
+txq_cleanup(struct txq *txq)
+{
+	struct ibv_exp_release_intf_params params;
+	size_t i;
+
+	DEBUG("cleaning up %p", (void *)txq);
+	txq_free_elts(txq);
+	if (txq->if_qp != NULL) {
+		assert(txq->priv != NULL);
+		assert(txq->priv->ctx != NULL);
+		assert(txq->qp != NULL);
+		params = (struct ibv_exp_release_intf_params){
+			.comp_mask = 0,
+		};
+		claim_zero(ibv_exp_release_intf(txq->priv->ctx,
+						txq->if_qp,
+						&params));
+	}
+	if (txq->if_cq != NULL) {
+		assert(txq->priv != NULL);
+		assert(txq->priv->ctx != NULL);
+		assert(txq->cq != NULL);
+		params = (struct ibv_exp_release_intf_params){
+			.comp_mask = 0,
+		};
+		claim_zero(ibv_exp_release_intf(txq->priv->ctx,
+						txq->if_cq,
+						&params));
+	}
+	if (txq->qp != NULL)
+		claim_zero(ibv_destroy_qp(txq->qp));
+	if (txq->cq != NULL)
+		claim_zero(ibv_destroy_cq(txq->cq));
+	if (txq->rd != NULL) {
+		struct ibv_exp_destroy_res_domain_attr attr = {
+			.comp_mask = 0,
+		};
+
+		assert(txq->priv != NULL);
+		assert(txq->priv->ctx != NULL);
+		claim_zero(ibv_exp_destroy_res_domain(txq->priv->ctx,
+						      txq->rd,
+						      &attr));
+	}
+	for (i = 0; (i != RTE_DIM(txq->mp2mr)); ++i) {
+		if (txq->mp2mr[i].mp == NULL)
+			break;
+		assert(txq->mp2mr[i].mr != NULL);
+		claim_zero(ibv_dereg_mr(txq->mp2mr[i].mr));
+	}
+	memset(txq, 0, sizeof(*txq));
+}
+
+/**
+ * Configure a TX queue.
+ *
+ * @param dev
+ *   Pointer to Ethernet device structure.
+ * @param txq
+ *   Pointer to TX queue structure.
+ * @param desc
+ *   Number of descriptors to configure in queue.
+ * @param socket
+ *   NUMA socket on which memory must be allocated.
+ * @param[in] conf
+ *   Thresholds parameters.
+ *
+ * @return
+ *   0 on success, errno value on failure.
+ */
+static int
+txq_setup(struct rte_eth_dev *dev, struct txq *txq, uint16_t desc,
+	  unsigned int socket, const struct rte_eth_txconf *conf)
+{
+	struct priv *priv = dev->data->dev_private;
+	struct txq tmpl = {
+		.priv = priv,
+		.socket = socket
+	};
+	union {
+		struct ibv_exp_query_intf_params params;
+		struct ibv_exp_qp_init_attr init;
+		struct ibv_exp_res_domain_init_attr rd;
+		struct ibv_exp_cq_init_attr cq;
+		struct ibv_exp_qp_attr mod;
+	} attr;
+	enum ibv_exp_query_intf_status status;
+	int ret = 0;
+
+	(void)conf; /* Thresholds configuration (ignored). */
+	if ((desc == 0) || (desc % MLX5_PMD_SGE_WR_N)) {
+		ERROR("%p: invalid number of TX descriptors (must be a"
+		      " multiple of %d)", (void *)dev, MLX5_PMD_SGE_WR_N);
+		return EINVAL;
+	}
+	desc /= MLX5_PMD_SGE_WR_N;
+	/* MRs will be registered in mp2mr[] later. */
+	attr.rd = (struct ibv_exp_res_domain_init_attr){
+		.comp_mask = (IBV_EXP_RES_DOMAIN_THREAD_MODEL |
+			      IBV_EXP_RES_DOMAIN_MSG_MODEL),
+		.thread_model = IBV_EXP_THREAD_SINGLE,
+		.msg_model = IBV_EXP_MSG_HIGH_BW,
+	};
+	tmpl.rd = ibv_exp_create_res_domain(priv->ctx, &attr.rd);
+	if (tmpl.rd == NULL) {
+		ret = ENOMEM;
+		ERROR("%p: RD creation failure: %s",
+		      (void *)dev, strerror(ret));
+		goto error;
+	}
+	attr.cq = (struct ibv_exp_cq_init_attr){
+		.comp_mask = IBV_EXP_CQ_INIT_ATTR_RES_DOMAIN,
+		.res_domain = tmpl.rd,
+	};
+	tmpl.cq = ibv_exp_create_cq(priv->ctx, desc, NULL, NULL, 0, &attr.cq);
+	if (tmpl.cq == NULL) {
+		ret = ENOMEM;
+		ERROR("%p: CQ creation failure: %s",
+		      (void *)dev, strerror(ret));
+		goto error;
+	}
+	DEBUG("priv->device_attr.max_qp_wr is %d",
+	      priv->device_attr.max_qp_wr);
+	DEBUG("priv->device_attr.max_sge is %d",
+	      priv->device_attr.max_sge);
+	attr.init = (struct ibv_exp_qp_init_attr){
+		/* CQ to be associated with the send queue. */
+		.send_cq = tmpl.cq,
+		/* CQ to be associated with the receive queue. */
+		.recv_cq = tmpl.cq,
+		.cap = {
+			/* Max number of outstanding WRs. */
+			.max_send_wr = ((priv->device_attr.max_qp_wr < desc) ?
+					priv->device_attr.max_qp_wr :
+					desc),
+			/* Max number of scatter/gather elements in a WR. */
+			.max_send_sge = ((priv->device_attr.max_sge <
+					  MLX5_PMD_SGE_WR_N) ?
+					 priv->device_attr.max_sge :
+					 MLX5_PMD_SGE_WR_N),
+#if MLX5_PMD_MAX_INLINE > 0
+			.max_inline_data = MLX5_PMD_MAX_INLINE,
+#endif
+		},
+		.qp_type = IBV_QPT_RAW_PACKET,
+		/* Do *NOT* enable this, completions events are managed per
+		 * TX burst. */
+		.sq_sig_all = 0,
+		.pd = priv->pd,
+		.res_domain = tmpl.rd,
+		.comp_mask = (IBV_EXP_QP_INIT_ATTR_PD |
+			      IBV_EXP_QP_INIT_ATTR_RES_DOMAIN),
+	};
+	tmpl.qp = ibv_exp_create_qp(priv->ctx, &attr.init);
+	if (tmpl.qp == NULL) {
+		ret = (errno ? errno : EINVAL);
+		ERROR("%p: QP creation failure: %s",
+		      (void *)dev, strerror(ret));
+		goto error;
+	}
+#if MLX5_PMD_MAX_INLINE > 0
+	/* ibv_create_qp() updates this value. */
+	tmpl.max_inline = attr.init.cap.max_inline_data;
+#endif
+	attr.mod = (struct ibv_exp_qp_attr){
+		/* Move the QP to this state. */
+		.qp_state = IBV_QPS_INIT,
+		/* Primary port number. */
+		.port_num = priv->port
+	};
+	ret = ibv_exp_modify_qp(tmpl.qp, &attr.mod,
+				(IBV_EXP_QP_STATE | IBV_EXP_QP_PORT));
+	if (ret) {
+		ERROR("%p: QP state to IBV_QPS_INIT failed: %s",
+		      (void *)dev, strerror(ret));
+		goto error;
+	}
+	ret = txq_alloc_elts(&tmpl, desc);
+	if (ret) {
+		ERROR("%p: TXQ allocation failed: %s",
+		      (void *)dev, strerror(ret));
+		goto error;
+	}
+	attr.mod = (struct ibv_exp_qp_attr){
+		.qp_state = IBV_QPS_RTR
+	};
+	ret = ibv_exp_modify_qp(tmpl.qp, &attr.mod, IBV_EXP_QP_STATE);
+	if (ret) {
+		ERROR("%p: QP state to IBV_QPS_RTR failed: %s",
+		      (void *)dev, strerror(ret));
+		goto error;
+	}
+	attr.mod.qp_state = IBV_QPS_RTS;
+	ret = ibv_exp_modify_qp(tmpl.qp, &attr.mod, IBV_EXP_QP_STATE);
+	if (ret) {
+		ERROR("%p: QP state to IBV_QPS_RTS failed: %s",
+		      (void *)dev, strerror(ret));
+		goto error;
+	}
+	attr.params = (struct ibv_exp_query_intf_params){
+		.intf_scope = IBV_EXP_INTF_GLOBAL,
+		.intf = IBV_EXP_INTF_CQ,
+		.obj = tmpl.cq,
+	};
+	tmpl.if_cq = ibv_exp_query_intf(priv->ctx, &attr.params, &status);
+	if (tmpl.if_cq == NULL) {
+		ret = EINVAL;
+		ERROR("%p: CQ interface family query failed with status %d",
+		      (void *)dev, status);
+		goto error;
+	}
+	attr.params = (struct ibv_exp_query_intf_params){
+		.intf_scope = IBV_EXP_INTF_GLOBAL,
+		.intf = IBV_EXP_INTF_QP_BURST,
+		.obj = tmpl.qp,
+	};
+	tmpl.if_qp = ibv_exp_query_intf(priv->ctx, &attr.params, &status);
+	if (tmpl.if_qp == NULL) {
+		ret = EINVAL;
+		ERROR("%p: QP interface family query failed with status %d",
+		      (void *)dev, status);
+		goto error;
+	}
+	/* Clean up txq in case we're reinitializing it. */
+	DEBUG("%p: cleaning-up old txq just in case", (void *)txq);
+	txq_cleanup(txq);
+	*txq = tmpl;
+	DEBUG("%p: txq updated with %p", (void *)txq, (void *)&tmpl);
+	assert(ret == 0);
+	return 0;
+error:
+	txq_cleanup(&tmpl);
+	assert(ret > 0);
+	return ret;
+}
+
+/**
+ * DPDK callback to configure a TX queue.
+ *
+ * @param dev
+ *   Pointer to Ethernet device structure.
+ * @param idx
+ *   TX queue index.
+ * @param desc
+ *   Number of descriptors to configure in queue.
+ * @param socket
+ *   NUMA socket on which memory must be allocated.
+ * @param[in] conf
+ *   Thresholds parameters.
+ *
+ * @return
+ *   0 on success, negative errno value on failure.
+ */
+int
+mlx5_tx_queue_setup(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc,
+		    unsigned int socket, const struct rte_eth_txconf *conf)
+{
+	struct priv *priv = dev->data->dev_private;
+	struct txq *txq = (*priv->txqs)[idx];
+	int ret;
+
+	priv_lock(priv);
+	DEBUG("%p: configuring queue %u for %u descriptors",
+	      (void *)dev, idx, desc);
+	if (idx >= priv->txqs_n) {
+		ERROR("%p: queue index out of range (%u >= %u)",
+		      (void *)dev, idx, priv->txqs_n);
+		priv_unlock(priv);
+		return -EOVERFLOW;
+	}
+	if (txq != NULL) {
+		DEBUG("%p: reusing already allocated queue index %u (%p)",
+		      (void *)dev, idx, (void *)txq);
+		if (priv->started) {
+			priv_unlock(priv);
+			return -EEXIST;
+		}
+		(*priv->txqs)[idx] = NULL;
+		txq_cleanup(txq);
+	} else {
+		txq = rte_calloc_socket("TXQ", 1, sizeof(*txq), 0, socket);
+		if (txq == NULL) {
+			ERROR("%p: unable to allocate queue index %u",
+			      (void *)dev, idx);
+			priv_unlock(priv);
+			return -ENOMEM;
+		}
+	}
+	ret = txq_setup(dev, txq, desc, socket, conf);
+	if (ret)
+		rte_free(txq);
+	else {
+		DEBUG("%p: adding TX queue %p to list",
+		      (void *)dev, (void *)txq);
+		(*priv->txqs)[idx] = txq;
+		/* Update send callback. */
+		dev->tx_pkt_burst = mlx5_tx_burst;
+	}
+	priv_unlock(priv);
+	return -ret;
+}
+
+/**
+ * DPDK callback to release a TX queue.
+ *
+ * @param dpdk_txq
+ *   Generic TX queue pointer.
+ */
+void
+mlx5_tx_queue_release(void *dpdk_txq)
+{
+	struct txq *txq = (struct txq *)dpdk_txq;
+	struct priv *priv;
+	unsigned int i;
+
+	if (txq == NULL)
+		return;
+	priv = txq->priv;
+	priv_lock(priv);
+	for (i = 0; (i != priv->txqs_n); ++i)
+		if ((*priv->txqs)[i] == txq) {
+			DEBUG("%p: removing TX queue %p from list",
+			      (void *)priv->dev, (void *)txq);
+			(*priv->txqs)[i] = NULL;
+			break;
+		}
+	txq_cleanup(txq);
+	rte_free(txq);
+	priv_unlock(priv);
+}
diff --git a/drivers/net/mlx5/mlx5_utils.h b/drivers/net/mlx5/mlx5_utils.h
index cc6aab6..e48e6b6 100644
--- a/drivers/net/mlx5/mlx5_utils.h
+++ b/drivers/net/mlx5/mlx5_utils.h
@@ -140,10 +140,21 @@ pmd_drv_log_basename(const char *s)
 #define WARN(...) PMD_DRV_LOG(WARNING, __VA_ARGS__)
 #define ERROR(...) PMD_DRV_LOG(ERR, __VA_ARGS__)
 
+/* Convenience macros for accessing mbuf fields. */
+#define NEXT(m) ((m)->next)
+#define DATA_LEN(m) ((m)->data_len)
+#define PKT_LEN(m) ((m)->pkt_len)
+#define DATA_OFF(m) ((m)->data_off)
+#define SET_DATA_OFF(m, o) ((m)->data_off = (o))
+#define NB_SEGS(m) ((m)->nb_segs)
+#define PORT(m) ((m)->port)
+
 /* Allocate a buffer on the stack and fill it with a printf format string. */
 #define MKSTR(name, ...) \
 	char name[snprintf(NULL, 0, __VA_ARGS__) + 1]; \
 	\
 	snprintf(name, sizeof(name), __VA_ARGS__)
 
+#define WR_ID(o) (((wr_id_t *)&(o))->data)
+
 #endif /* RTE_PMD_MLX5_UTILS_H_ */
-- 
2.1.0

^ permalink raw reply	[flat|nested] 32+ messages in thread

* [dpdk-dev] [PATCH v2 03/13] mlx5: add MAC handling
  2015-10-30 18:52 ` [dpdk-dev] [PATCH v2 00/13] Mellanox ConnectX-4 PMD (mlx5) Adrien Mazarguil
  2015-10-30 18:52   ` [dpdk-dev] [PATCH v2 01/13] mlx5: new poll-mode driver for Mellanox ConnectX-4 adapters Adrien Mazarguil
  2015-10-30 18:52   ` [dpdk-dev] [PATCH v2 02/13] mlx5: add non-scattered TX and RX support Adrien Mazarguil
@ 2015-10-30 18:52   ` Adrien Mazarguil
  2015-10-30 18:52   ` [dpdk-dev] [PATCH v2 04/13] mlx5: add device configure/start/stop Adrien Mazarguil
                     ` (10 subsequent siblings)
  13 siblings, 0 replies; 32+ messages in thread
From: Adrien Mazarguil @ 2015-10-30 18:52 UTC (permalink / raw)
  To: dev

This commit adds support for MAC flow steering rules mandatory for the RX
path as well as the related callbacks to add/remove MAC addresses.

Signed-off-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>
Signed-off-by: Nelio Laranjeiro <nelio.laranjeiro@6wind.com>
Signed-off-by: Didier Pallard <didier.pallard@6wind.com>
---
 drivers/net/mlx5/mlx5.c      |   4 +-
 drivers/net/mlx5/mlx5.h      |   5 +
 drivers/net/mlx5/mlx5_mac.c  | 282 +++++++++++++++++++++++++++++++++++++++++++
 drivers/net/mlx5/mlx5_rxq.c  |  10 ++
 drivers/net/mlx5/mlx5_rxtx.h |   2 +
 5 files changed, 302 insertions(+), 1 deletion(-)

diff --git a/drivers/net/mlx5/mlx5.c b/drivers/net/mlx5/mlx5.c
index 54bd6b9..a241c28 100644
--- a/drivers/net/mlx5/mlx5.c
+++ b/drivers/net/mlx5/mlx5.c
@@ -134,6 +134,8 @@ static const struct eth_dev_ops mlx5_dev_ops = {
 	.tx_queue_setup = mlx5_tx_queue_setup,
 	.rx_queue_release = mlx5_rx_queue_release,
 	.tx_queue_release = mlx5_tx_queue_release,
+	.mac_addr_remove = mlx5_mac_addr_remove,
+	.mac_addr_add = mlx5_mac_addr_add,
 };
 
 static struct {
@@ -390,7 +392,7 @@ mlx5_pci_devinit(struct rte_pci_driver *pci_drv, struct rte_pci_device *pci_dev)
 		claim_zero(priv_mac_addr_add(priv, 0,
 					     (const uint8_t (*)[ETHER_ADDR_LEN])
 					     mac.addr_bytes));
-		claim_zero(priv_mac_addr_add(priv, 1,
+		claim_zero(priv_mac_addr_add(priv, (RTE_DIM(priv->mac) - 1),
 					     &(const uint8_t [ETHER_ADDR_LEN])
 					     { "\xff\xff\xff\xff\xff\xff" }));
 #ifndef NDEBUG
diff --git a/drivers/net/mlx5/mlx5.h b/drivers/net/mlx5/mlx5.h
index 49978f5..6ab31ad 100644
--- a/drivers/net/mlx5/mlx5.h
+++ b/drivers/net/mlx5/mlx5.h
@@ -166,7 +166,12 @@ int mlx5_ibv_device_to_pci_addr(const struct ibv_device *,
 /* mlx5_mac.c */
 
 int priv_get_mac(struct priv *, uint8_t (*)[ETHER_ADDR_LEN]);
+void rxq_mac_addrs_del(struct rxq *);
+void mlx5_mac_addr_remove(struct rte_eth_dev *, uint32_t);
+int rxq_mac_addrs_add(struct rxq *);
 int priv_mac_addr_add(struct priv *, unsigned int,
 		      const uint8_t (*)[ETHER_ADDR_LEN]);
+void mlx5_mac_addr_add(struct rte_eth_dev *, struct ether_addr *, uint32_t,
+		       uint32_t);
 
 #endif /* RTE_PMD_MLX5_H_ */
diff --git a/drivers/net/mlx5/mlx5_mac.c b/drivers/net/mlx5/mlx5_mac.c
index f7e1cf6..262d7c6 100644
--- a/drivers/net/mlx5/mlx5_mac.c
+++ b/drivers/net/mlx5/mlx5_mac.c
@@ -65,6 +65,8 @@
 
 #include "mlx5.h"
 #include "mlx5_utils.h"
+#include "mlx5_rxtx.h"
+#include "mlx5_defs.h"
 
 /**
  * Get MAC address by querying netdevice.
@@ -89,8 +91,69 @@ priv_get_mac(struct priv *priv, uint8_t (*mac)[ETHER_ADDR_LEN])
 }
 
 /**
+ * Delete MAC flow steering rule.
+ *
+ * @param rxq
+ *   Pointer to RX queue structure.
+ * @param mac_index
+ *   MAC address index.
+ */
+static void
+rxq_del_mac_flow(struct rxq *rxq, unsigned int mac_index)
+{
+#ifndef NDEBUG
+	const uint8_t (*mac)[ETHER_ADDR_LEN] =
+		(const uint8_t (*)[ETHER_ADDR_LEN])
+		rxq->priv->mac[mac_index].addr_bytes;
+#endif
+
+	assert(mac_index < RTE_DIM(rxq->mac_flow));
+	if (rxq->mac_flow[mac_index] == NULL)
+		return;
+	DEBUG("%p: removing MAC address %02x:%02x:%02x:%02x:%02x:%02x index %u",
+	      (void *)rxq,
+	      (*mac)[0], (*mac)[1], (*mac)[2], (*mac)[3], (*mac)[4], (*mac)[5],
+	      mac_index);
+	claim_zero(ibv_destroy_flow(rxq->mac_flow[mac_index]));
+	rxq->mac_flow[mac_index] = NULL;
+}
+
+/**
+ * Unregister a MAC address from a RX queue.
+ *
+ * @param rxq
+ *   Pointer to RX queue structure.
+ * @param mac_index
+ *   MAC address index.
+ */
+static void
+rxq_mac_addr_del(struct rxq *rxq, unsigned int mac_index)
+{
+	assert(mac_index < RTE_DIM(rxq->mac_flow));
+	rxq_del_mac_flow(rxq, mac_index);
+}
+
+/**
+ * Unregister all MAC addresses from a RX queue.
+ *
+ * @param rxq
+ *   Pointer to RX queue structure.
+ */
+void
+rxq_mac_addrs_del(struct rxq *rxq)
+{
+	unsigned int i;
+
+	for (i = 0; (i != RTE_DIM(rxq->mac_flow)); ++i)
+		rxq_mac_addr_del(rxq, i);
+}
+
+/**
  * Unregister a MAC address.
  *
+ * In RSS mode, the MAC address is unregistered from the parent queue,
+ * otherwise it is unregistered from each queue directly.
+ *
  * @param priv
  *   Pointer to private structure.
  * @param mac_index
@@ -99,15 +162,179 @@ priv_get_mac(struct priv *priv, uint8_t (*mac)[ETHER_ADDR_LEN])
 static void
 priv_mac_addr_del(struct priv *priv, unsigned int mac_index)
 {
+	unsigned int i;
+
 	assert(mac_index < RTE_DIM(priv->mac));
 	if (!BITFIELD_ISSET(priv->mac_configured, mac_index))
 		return;
+	if (priv->rss) {
+		rxq_mac_addr_del(&priv->rxq_parent, mac_index);
+		goto end;
+	}
+	for (i = 0; (i != priv->dev->data->nb_rx_queues); ++i)
+		rxq_mac_addr_del((*priv->rxqs)[i], mac_index);
+end:
 	BITFIELD_RESET(priv->mac_configured, mac_index);
 }
 
 /**
+ * DPDK callback to remove a MAC address.
+ *
+ * @param dev
+ *   Pointer to Ethernet device structure.
+ * @param index
+ *   MAC address index.
+ */
+void
+mlx5_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
+{
+	struct priv *priv = dev->data->dev_private;
+
+	priv_lock(priv);
+	DEBUG("%p: removing MAC address from index %" PRIu32,
+	      (void *)dev, index);
+	/* Last array entry is reserved for broadcast. */
+	if (index >= (RTE_DIM(priv->mac) - 1))
+		goto end;
+	priv_mac_addr_del(priv, index);
+end:
+	priv_unlock(priv);
+}
+
+/**
+ * Add MAC flow steering rule.
+ *
+ * @param rxq
+ *   Pointer to RX queue structure.
+ * @param mac_index
+ *   MAC address index to register.
+ *
+ * @return
+ *   0 on success, errno value on failure.
+ */
+static int
+rxq_add_mac_flow(struct rxq *rxq, unsigned int mac_index)
+{
+	struct ibv_flow *flow;
+	struct priv *priv = rxq->priv;
+	const uint8_t (*mac)[ETHER_ADDR_LEN] =
+			(const uint8_t (*)[ETHER_ADDR_LEN])
+			priv->mac[mac_index].addr_bytes;
+	struct __attribute__((packed)) {
+		struct ibv_flow_attr attr;
+		struct ibv_flow_spec_eth spec;
+	} data;
+	struct ibv_flow_attr *attr = &data.attr;
+	struct ibv_flow_spec_eth *spec = &data.spec;
+
+	assert(mac_index < RTE_DIM(rxq->mac_flow));
+	if (rxq->mac_flow[mac_index] != NULL)
+		return 0;
+	/*
+	 * No padding must be inserted by the compiler between attr and spec.
+	 * This layout is expected by libibverbs.
+	 */
+	assert(((uint8_t *)attr + sizeof(*attr)) == (uint8_t *)spec);
+	*attr = (struct ibv_flow_attr){
+		.type = IBV_FLOW_ATTR_NORMAL,
+		.num_of_specs = 1,
+		.port = priv->port,
+		.flags = 0
+	};
+	*spec = (struct ibv_flow_spec_eth){
+		.type = IBV_FLOW_SPEC_ETH,
+		.size = sizeof(*spec),
+		.val = {
+			.dst_mac = {
+				(*mac)[0], (*mac)[1], (*mac)[2],
+				(*mac)[3], (*mac)[4], (*mac)[5]
+			},
+		},
+		.mask = {
+			.dst_mac = "\xff\xff\xff\xff\xff\xff",
+		},
+	};
+	DEBUG("%p: adding MAC address %02x:%02x:%02x:%02x:%02x:%02x index %u",
+	      (void *)rxq,
+	      (*mac)[0], (*mac)[1], (*mac)[2], (*mac)[3], (*mac)[4], (*mac)[5],
+	      mac_index);
+	/* Create related flow. */
+	errno = 0;
+	flow = ibv_create_flow(rxq->qp, attr);
+	if (flow == NULL) {
+		/* It's not clear whether errno is always set in this case. */
+		ERROR("%p: flow configuration failed, errno=%d: %s",
+		      (void *)rxq, errno,
+		      (errno ? strerror(errno) : "Unknown error"));
+		if (errno)
+			return errno;
+		return EINVAL;
+	}
+	rxq->mac_flow[mac_index] = flow;
+	return 0;
+}
+
+/**
+ * Register a MAC address in a RX queue.
+ *
+ * @param rxq
+ *   Pointer to RX queue structure.
+ * @param mac_index
+ *   MAC address index to register.
+ *
+ * @return
+ *   0 on success, errno value on failure.
+ */
+static int
+rxq_mac_addr_add(struct rxq *rxq, unsigned int mac_index)
+{
+	int ret;
+
+	assert(mac_index < RTE_DIM(rxq->mac_flow));
+	ret = rxq_add_mac_flow(rxq, mac_index);
+	if (ret)
+		return ret;
+	return 0;
+}
+
+/**
+ * Register all MAC addresses in a RX queue.
+ *
+ * @param rxq
+ *   Pointer to RX queue structure.
+ *
+ * @return
+ *   0 on success, errno value on failure.
+ */
+int
+rxq_mac_addrs_add(struct rxq *rxq)
+{
+	struct priv *priv = rxq->priv;
+	unsigned int i;
+	int ret;
+
+	assert(RTE_DIM(priv->mac) == RTE_DIM(rxq->mac_flow));
+	for (i = 0; (i != RTE_DIM(priv->mac)); ++i) {
+		if (!BITFIELD_ISSET(priv->mac_configured, i))
+			continue;
+		ret = rxq_mac_addr_add(rxq, i);
+		if (!ret)
+			continue;
+		/* Failure, rollback. */
+		while (i != 0)
+			rxq_mac_addr_del(rxq, --i);
+		assert(ret > 0);
+		return ret;
+	}
+	return 0;
+}
+
+/**
  * Register a MAC address.
  *
+ * In RSS mode, the MAC address is registered in the parent queue,
+ * otherwise it is registered in each queue directly.
+ *
  * @param priv
  *   Pointer to private structure.
  * @param mac_index
@@ -123,6 +350,7 @@ priv_mac_addr_add(struct priv *priv, unsigned int mac_index,
 		  const uint8_t (*mac)[ETHER_ADDR_LEN])
 {
 	unsigned int i;
+	int ret;
 
 	assert(mac_index < RTE_DIM(priv->mac));
 	/* First, make sure this address isn't already configured. */
@@ -145,6 +373,60 @@ priv_mac_addr_add(struct priv *priv, unsigned int mac_index,
 			(*mac)[3], (*mac)[4], (*mac)[5]
 		}
 	};
+	/* If device isn't started, this is all we need to do. */
+	if (!priv->started)
+		goto end;
+	if (priv->rss) {
+		ret = rxq_mac_addr_add(&priv->rxq_parent, mac_index);
+		if (ret)
+			return ret;
+		goto end;
+	}
+	for (i = 0; (i != priv->rxqs_n); ++i) {
+		if ((*priv->rxqs)[i] == NULL)
+			continue;
+		ret = rxq_mac_addr_add((*priv->rxqs)[i], mac_index);
+		if (!ret)
+			continue;
+		/* Failure, rollback. */
+		while (i != 0)
+			if ((*priv->rxqs)[--i] != NULL)
+				rxq_mac_addr_del((*priv->rxqs)[i], mac_index);
+		return ret;
+	}
+end:
 	BITFIELD_SET(priv->mac_configured, mac_index);
 	return 0;
 }
+
+/**
+ * DPDK callback to add a MAC address.
+ *
+ * @param dev
+ *   Pointer to Ethernet device structure.
+ * @param mac_addr
+ *   MAC address to register.
+ * @param index
+ *   MAC address index.
+ * @param vmdq
+ *   VMDq pool index to associate address with (ignored).
+ */
+void
+mlx5_mac_addr_add(struct rte_eth_dev *dev, struct ether_addr *mac_addr,
+		  uint32_t index, uint32_t vmdq)
+{
+	struct priv *priv = dev->data->dev_private;
+
+	(void)vmdq;
+	priv_lock(priv);
+	DEBUG("%p: adding MAC address at index %" PRIu32,
+	      (void *)dev, index);
+	/* Last array entry is reserved for broadcast. */
+	if (index >= (RTE_DIM(priv->mac) - 1))
+		goto end;
+	priv_mac_addr_add(priv, index,
+			  (const uint8_t (*)[ETHER_ADDR_LEN])
+			  mac_addr->addr_bytes);
+end:
+	priv_unlock(priv);
+}
diff --git a/drivers/net/mlx5/mlx5_rxq.c b/drivers/net/mlx5/mlx5_rxq.c
index 01cc649..8450fe3 100644
--- a/drivers/net/mlx5/mlx5_rxq.c
+++ b/drivers/net/mlx5/mlx5_rxq.c
@@ -248,6 +248,7 @@ rxq_cleanup(struct rxq *rxq)
 						&params));
 	}
 	if (rxq->qp != NULL) {
+		rxq_mac_addrs_del(rxq);
 		claim_zero(ibv_destroy_qp(rxq->qp));
 	}
 	if (rxq->cq != NULL)
@@ -515,6 +516,15 @@ skip_mr:
 		      (void *)dev, strerror(ret));
 		goto error;
 	}
+	if ((parent) || (!priv->rss))  {
+		/* Configure MAC and broadcast addresses. */
+		ret = rxq_mac_addrs_add(&tmpl);
+		if (ret) {
+			ERROR("%p: QP flow attachment failed: %s",
+			      (void *)dev, strerror(ret));
+			goto error;
+		}
+	}
 	/* Allocate descriptors for RX queues, except for the RSS parent. */
 	if (parent)
 		goto skip_alloc;
diff --git a/drivers/net/mlx5/mlx5_rxtx.h b/drivers/net/mlx5/mlx5_rxtx.h
index 1459317..3733d3e 100644
--- a/drivers/net/mlx5/mlx5_rxtx.h
+++ b/drivers/net/mlx5/mlx5_rxtx.h
@@ -78,6 +78,8 @@ struct rxq {
 	struct ibv_qp *qp; /* Queue Pair. */
 	struct ibv_exp_qp_burst_family *if_qp; /* QP burst interface. */
 	struct ibv_exp_cq_family *if_cq; /* CQ interface. */
+	/* MAC flow steering rules. */
+	struct ibv_flow *mac_flow[MLX5_MAX_MAC_ADDRESSES];
 	unsigned int port_id; /* Port ID for incoming packets. */
 	unsigned int elts_n; /* (*elts)[] length. */
 	unsigned int elts_head; /* Current index in (*elts)[]. */
-- 
2.1.0

^ permalink raw reply	[flat|nested] 32+ messages in thread

* [dpdk-dev] [PATCH v2 04/13] mlx5: add device configure/start/stop
  2015-10-30 18:52 ` [dpdk-dev] [PATCH v2 00/13] Mellanox ConnectX-4 PMD (mlx5) Adrien Mazarguil
                     ` (2 preceding siblings ...)
  2015-10-30 18:52   ` [dpdk-dev] [PATCH v2 03/13] mlx5: add MAC handling Adrien Mazarguil
@ 2015-10-30 18:52   ` Adrien Mazarguil
  2015-10-30 18:52   ` [dpdk-dev] [PATCH v2 05/13] mlx5: add support for scattered RX and TX buffers Adrien Mazarguil
                     ` (9 subsequent siblings)
  13 siblings, 0 replies; 32+ messages in thread
From: Adrien Mazarguil @ 2015-10-30 18:52 UTC (permalink / raw)
  To: dev; +Cc: Francesco Santoro

This commit adds the remaining missing callbacks to make mlx5 usable.
Like mlx4, device start and stop are implemented on top of MAC RX flows.

Signed-off-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>
Signed-off-by: Nelio Laranjeiro <nelio.laranjeiro@6wind.com>
Signed-off-by: Francesco Santoro <francesco.santoro@6wind.com>
Signed-off-by: Didier Pallard <didier.pallard@6wind.com>
---
 drivers/net/mlx5/Makefile       |   1 +
 drivers/net/mlx5/mlx5.c         |   4 ++
 drivers/net/mlx5/mlx5.h         |   7 ++
 drivers/net/mlx5/mlx5_ethdev.c  | 148 ++++++++++++++++++++++++++++++++++++++++
 drivers/net/mlx5/mlx5_trigger.c | 146 +++++++++++++++++++++++++++++++++++++++
 5 files changed, 306 insertions(+)
 create mode 100644 drivers/net/mlx5/mlx5_trigger.c

diff --git a/drivers/net/mlx5/Makefile b/drivers/net/mlx5/Makefile
index 7b9c57b..028c22c 100644
--- a/drivers/net/mlx5/Makefile
+++ b/drivers/net/mlx5/Makefile
@@ -45,6 +45,7 @@ SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5.c
 SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5_rxq.c
 SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5_txq.c
 SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5_rxtx.c
+SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5_trigger.c
 SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5_ethdev.c
 SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5_mac.c
 
diff --git a/drivers/net/mlx5/mlx5.c b/drivers/net/mlx5/mlx5.c
index a241c28..aafa70b 100644
--- a/drivers/net/mlx5/mlx5.c
+++ b/drivers/net/mlx5/mlx5.c
@@ -129,7 +129,11 @@ mlx5_dev_close(struct rte_eth_dev *dev)
 }
 
 static const struct eth_dev_ops mlx5_dev_ops = {
+	.dev_configure = mlx5_dev_configure,
+	.dev_start = mlx5_dev_start,
+	.dev_stop = mlx5_dev_stop,
 	.dev_close = mlx5_dev_close,
+	.dev_infos_get = mlx5_dev_infos_get,
 	.rx_queue_setup = mlx5_rx_queue_setup,
 	.tx_queue_setup = mlx5_tx_queue_setup,
 	.rx_queue_release = mlx5_rx_queue_release,
diff --git a/drivers/net/mlx5/mlx5.h b/drivers/net/mlx5/mlx5.h
index 6ab31ad..3f47a15 100644
--- a/drivers/net/mlx5/mlx5.h
+++ b/drivers/net/mlx5/mlx5.h
@@ -160,6 +160,8 @@ int priv_get_ifname(const struct priv *, char (*)[IF_NAMESIZE]);
 int priv_ifreq(const struct priv *, int req, struct ifreq *);
 int priv_get_mtu(struct priv *, uint16_t *);
 int priv_set_flags(struct priv *, unsigned int, unsigned int);
+int mlx5_dev_configure(struct rte_eth_dev *);
+void mlx5_dev_infos_get(struct rte_eth_dev *, struct rte_eth_dev_info *);
 int mlx5_ibv_device_to_pci_addr(const struct ibv_device *,
 				struct rte_pci_addr *);
 
@@ -174,4 +176,9 @@ int priv_mac_addr_add(struct priv *, unsigned int,
 void mlx5_mac_addr_add(struct rte_eth_dev *, struct ether_addr *, uint32_t,
 		       uint32_t);
 
+/* mlx5_trigger.c */
+
+int mlx5_dev_start(struct rte_eth_dev *);
+void mlx5_dev_stop(struct rte_eth_dev *);
+
 #endif /* RTE_PMD_MLX5_H_ */
diff --git a/drivers/net/mlx5/mlx5_ethdev.c b/drivers/net/mlx5/mlx5_ethdev.c
index b6c7d7a..6b13cec 100644
--- a/drivers/net/mlx5/mlx5_ethdev.c
+++ b/drivers/net/mlx5/mlx5_ethdev.c
@@ -32,6 +32,7 @@
  */
 
 #include <stddef.h>
+#include <assert.h>
 #include <unistd.h>
 #include <stdint.h>
 #include <stdio.h>
@@ -58,6 +59,7 @@
 #endif
 
 #include "mlx5.h"
+#include "mlx5_rxtx.h"
 #include "mlx5_utils.h"
 
 /**
@@ -370,6 +372,152 @@ priv_set_flags(struct priv *priv, unsigned int keep, unsigned int flags)
 }
 
 /**
+ * Ethernet device configuration.
+ *
+ * Prepare the driver for a given number of TX and RX queues.
+ * Allocate parent RSS queue when several RX queues are requested.
+ *
+ * @param dev
+ *   Pointer to Ethernet device structure.
+ *
+ * @return
+ *   0 on success, errno value on failure.
+ */
+static int
+dev_configure(struct rte_eth_dev *dev)
+{
+	struct priv *priv = dev->data->dev_private;
+	unsigned int rxqs_n = dev->data->nb_rx_queues;
+	unsigned int txqs_n = dev->data->nb_tx_queues;
+	unsigned int tmp;
+	int ret;
+
+	priv->rxqs = (void *)dev->data->rx_queues;
+	priv->txqs = (void *)dev->data->tx_queues;
+	if (txqs_n != priv->txqs_n) {
+		INFO("%p: TX queues number update: %u -> %u",
+		     (void *)dev, priv->txqs_n, txqs_n);
+		priv->txqs_n = txqs_n;
+	}
+	if (rxqs_n == priv->rxqs_n)
+		return 0;
+	INFO("%p: RX queues number update: %u -> %u",
+	     (void *)dev, priv->rxqs_n, rxqs_n);
+	/* If RSS is enabled, disable it first. */
+	if (priv->rss) {
+		unsigned int i;
+
+		/* Only if there are no remaining child RX queues. */
+		for (i = 0; (i != priv->rxqs_n); ++i)
+			if ((*priv->rxqs)[i] != NULL)
+				return EINVAL;
+		rxq_cleanup(&priv->rxq_parent);
+		priv->rss = 0;
+		priv->rxqs_n = 0;
+	}
+	if (rxqs_n <= 1) {
+		/* Nothing else to do. */
+		priv->rxqs_n = rxqs_n;
+		return 0;
+	}
+	/* Allocate a new RSS parent queue if supported by hardware. */
+	if (!priv->hw_rss) {
+		ERROR("%p: only a single RX queue can be configured when"
+		      " hardware doesn't support RSS",
+		      (void *)dev);
+		return EINVAL;
+	}
+	/* Fail if hardware doesn't support that many RSS queues. */
+	if (rxqs_n >= priv->max_rss_tbl_sz) {
+		ERROR("%p: only %u RX queues can be configured for RSS",
+		      (void *)dev, priv->max_rss_tbl_sz);
+		return EINVAL;
+	}
+	priv->rss = 1;
+	tmp = priv->rxqs_n;
+	priv->rxqs_n = rxqs_n;
+	ret = rxq_setup(dev, &priv->rxq_parent, 0, 0, NULL, NULL);
+	if (!ret)
+		return 0;
+	/* Failure, rollback. */
+	priv->rss = 0;
+	priv->rxqs_n = tmp;
+	assert(ret > 0);
+	return ret;
+}
+
+/**
+ * DPDK callback for Ethernet device configuration.
+ *
+ * @param dev
+ *   Pointer to Ethernet device structure.
+ *
+ * @return
+ *   0 on success, negative errno value on failure.
+ */
+int
+mlx5_dev_configure(struct rte_eth_dev *dev)
+{
+	struct priv *priv = dev->data->dev_private;
+	int ret;
+
+	priv_lock(priv);
+	ret = dev_configure(dev);
+	assert(ret >= 0);
+	priv_unlock(priv);
+	return -ret;
+}
+
+/**
+ * DPDK callback to get information about the device.
+ *
+ * @param dev
+ *   Pointer to Ethernet device structure.
+ * @param[out] info
+ *   Info structure output buffer.
+ */
+void
+mlx5_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *info)
+{
+	struct priv *priv = dev->data->dev_private;
+	unsigned int max;
+	char ifname[IF_NAMESIZE];
+
+	priv_lock(priv);
+	/* FIXME: we should ask the device for these values. */
+	info->min_rx_bufsize = 32;
+	info->max_rx_pktlen = 65536;
+	/*
+	 * Since we need one CQ per QP, the limit is the minimum number
+	 * between the two values.
+	 */
+	max = ((priv->device_attr.max_cq > priv->device_attr.max_qp) ?
+	       priv->device_attr.max_qp : priv->device_attr.max_cq);
+	/* If max >= 65535 then max = 0, max_rx_queues is uint16_t. */
+	if (max >= 65535)
+		max = 65535;
+	info->max_rx_queues = max;
+	info->max_tx_queues = max;
+	/* Last array entry is reserved for broadcast. */
+	info->max_mac_addrs = (RTE_DIM(priv->mac) - 1);
+	info->rx_offload_capa =
+		(priv->hw_csum ?
+		 (DEV_RX_OFFLOAD_IPV4_CKSUM |
+		  DEV_RX_OFFLOAD_UDP_CKSUM |
+		  DEV_RX_OFFLOAD_TCP_CKSUM) :
+		 0);
+	info->tx_offload_capa =
+		(priv->hw_csum ?
+		 (DEV_TX_OFFLOAD_IPV4_CKSUM |
+		  DEV_TX_OFFLOAD_UDP_CKSUM |
+		  DEV_TX_OFFLOAD_TCP_CKSUM) :
+		 0);
+	if (priv_get_ifname(priv, &ifname) == 0)
+		info->if_index = if_nametoindex(ifname);
+	priv_unlock(priv);
+}
+
+/**
  * Get PCI information from struct ibv_device.
  *
  * @param device
diff --git a/drivers/net/mlx5/mlx5_trigger.c b/drivers/net/mlx5/mlx5_trigger.c
new file mode 100644
index 0000000..f5d965f
--- /dev/null
+++ b/drivers/net/mlx5/mlx5_trigger.c
@@ -0,0 +1,146 @@
+/*-
+ *   BSD LICENSE
+ *
+ *   Copyright 2015 6WIND S.A.
+ *   Copyright 2015 Mellanox.
+ *
+ *   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 6WIND S.A. 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.
+ */
+
+/* DPDK headers don't like -pedantic. */
+#ifdef PEDANTIC
+#pragma GCC diagnostic ignored "-pedantic"
+#endif
+#include <rte_ether.h>
+#include <rte_ethdev.h>
+#ifdef PEDANTIC
+#pragma GCC diagnostic error "-pedantic"
+#endif
+
+#include "mlx5.h"
+#include "mlx5_rxtx.h"
+#include "mlx5_utils.h"
+
+/**
+ * DPDK callback to start the device.
+ *
+ * Simulate device start by attaching all configured flows.
+ *
+ * @param dev
+ *   Pointer to Ethernet device structure.
+ *
+ * @return
+ *   0 on success, negative errno value on failure.
+ */
+int
+mlx5_dev_start(struct rte_eth_dev *dev)
+{
+	struct priv *priv = dev->data->dev_private;
+	unsigned int i = 0;
+	unsigned int r;
+	struct rxq *rxq;
+
+	priv_lock(priv);
+	if (priv->started) {
+		priv_unlock(priv);
+		return 0;
+	}
+	DEBUG("%p: attaching configured flows to all RX queues", (void *)dev);
+	priv->started = 1;
+	if (priv->rss) {
+		rxq = &priv->rxq_parent;
+		r = 1;
+	} else {
+		rxq = (*priv->rxqs)[0];
+		r = priv->rxqs_n;
+	}
+	/* Iterate only once when RSS is enabled. */
+	do {
+		int ret;
+
+		/* Ignore nonexistent RX queues. */
+		if (rxq == NULL)
+			continue;
+		ret = rxq_mac_addrs_add(rxq);
+		if (!ret)
+			continue;
+		WARN("%p: QP flow attachment failed: %s",
+		     (void *)dev, strerror(ret));
+		/* Rollback. */
+		while (i != 0) {
+			rxq = (*priv->rxqs)[--i];
+			if (rxq != NULL) {
+				rxq_mac_addrs_del(rxq);
+			}
+		}
+		priv->started = 0;
+		priv_unlock(priv);
+		return -ret;
+	} while ((--r) && ((rxq = (*priv->rxqs)[++i]), i));
+	priv_unlock(priv);
+	return 0;
+}
+
+/**
+ * DPDK callback to stop the device.
+ *
+ * Simulate device stop by detaching all configured flows.
+ *
+ * @param dev
+ *   Pointer to Ethernet device structure.
+ */
+void
+mlx5_dev_stop(struct rte_eth_dev *dev)
+{
+	struct priv *priv = dev->data->dev_private;
+	unsigned int i = 0;
+	unsigned int r;
+	struct rxq *rxq;
+
+	priv_lock(priv);
+	if (!priv->started) {
+		priv_unlock(priv);
+		return;
+	}
+	DEBUG("%p: detaching flows from all RX queues", (void *)dev);
+	priv->started = 0;
+	if (priv->rss) {
+		rxq = &priv->rxq_parent;
+		r = 1;
+	} else {
+		rxq = (*priv->rxqs)[0];
+		r = priv->rxqs_n;
+	}
+	/* Iterate only once when RSS is enabled. */
+	do {
+		/* Ignore nonexistent RX queues. */
+		if (rxq == NULL)
+			continue;
+		rxq_mac_addrs_del(rxq);
+	} while ((--r) && ((rxq = (*priv->rxqs)[++i]), i));
+	priv_unlock(priv);
+}
-- 
2.1.0

^ permalink raw reply	[flat|nested] 32+ messages in thread

* [dpdk-dev] [PATCH v2 05/13] mlx5: add support for scattered RX and TX buffers
  2015-10-30 18:52 ` [dpdk-dev] [PATCH v2 00/13] Mellanox ConnectX-4 PMD (mlx5) Adrien Mazarguil
                     ` (3 preceding siblings ...)
  2015-10-30 18:52   ` [dpdk-dev] [PATCH v2 04/13] mlx5: add device configure/start/stop Adrien Mazarguil
@ 2015-10-30 18:52   ` Adrien Mazarguil
  2015-10-30 18:52   ` [dpdk-dev] [PATCH v2 06/13] mlx5: add MTU configuration support Adrien Mazarguil
                     ` (8 subsequent siblings)
  13 siblings, 0 replies; 32+ messages in thread
From: Adrien Mazarguil @ 2015-10-30 18:52 UTC (permalink / raw)
  To: dev

A dedicated RX callback is added to handle scattered buffers. For better
performance, it is only used when jumbo frames are enabled and MTU is larger
than a single mbuf.

On the TX path, scattered buffers are also handled in a separate function.
When there are more than MLX5_PMD_SGE_WR_N segments in a given mbuf, the
remaining segments are linearized in the last SGE entry.

Signed-off-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>
Signed-off-by: Nelio Laranjeiro <nelio.laranjeiro@6wind.com>
---
 drivers/net/mlx5/mlx5_rxq.c  | 175 +++++++++++++++++++-
 drivers/net/mlx5/mlx5_rxtx.c | 376 +++++++++++++++++++++++++++++++++++++++++++
 drivers/net/mlx5/mlx5_rxtx.h |  10 ++
 3 files changed, 557 insertions(+), 4 deletions(-)

diff --git a/drivers/net/mlx5/mlx5_rxq.c b/drivers/net/mlx5/mlx5_rxq.c
index 8450fe3..1eddfc7 100644
--- a/drivers/net/mlx5/mlx5_rxq.c
+++ b/drivers/net/mlx5/mlx5_rxq.c
@@ -65,6 +65,153 @@
 #include "mlx5_defs.h"
 
 /**
+ * Allocate RX queue elements with scattered packets support.
+ *
+ * @param rxq
+ *   Pointer to RX queue structure.
+ * @param elts_n
+ *   Number of elements to allocate.
+ * @param[in] pool
+ *   If not NULL, fetch buffers from this array instead of allocating them
+ *   with rte_pktmbuf_alloc().
+ *
+ * @return
+ *   0 on success, errno value on failure.
+ */
+static int
+rxq_alloc_elts_sp(struct rxq *rxq, unsigned int elts_n,
+		  struct rte_mbuf **pool)
+{
+	unsigned int i;
+	struct rxq_elt_sp (*elts)[elts_n] =
+		rte_calloc_socket("RXQ elements", 1, sizeof(*elts), 0,
+				  rxq->socket);
+	int ret = 0;
+
+	if (elts == NULL) {
+		ERROR("%p: can't allocate packets array", (void *)rxq);
+		ret = ENOMEM;
+		goto error;
+	}
+	/* For each WR (packet). */
+	for (i = 0; (i != elts_n); ++i) {
+		unsigned int j;
+		struct rxq_elt_sp *elt = &(*elts)[i];
+		struct ibv_recv_wr *wr = &elt->wr;
+		struct ibv_sge (*sges)[RTE_DIM(elt->sges)] = &elt->sges;
+
+		/* These two arrays must have the same size. */
+		assert(RTE_DIM(elt->sges) == RTE_DIM(elt->bufs));
+		/* Configure WR. */
+		wr->wr_id = i;
+		wr->next = &(*elts)[(i + 1)].wr;
+		wr->sg_list = &(*sges)[0];
+		wr->num_sge = RTE_DIM(*sges);
+		/* For each SGE (segment). */
+		for (j = 0; (j != RTE_DIM(elt->bufs)); ++j) {
+			struct ibv_sge *sge = &(*sges)[j];
+			struct rte_mbuf *buf;
+
+			if (pool != NULL) {
+				buf = *(pool++);
+				assert(buf != NULL);
+				rte_pktmbuf_reset(buf);
+			} else
+				buf = rte_pktmbuf_alloc(rxq->mp);
+			if (buf == NULL) {
+				assert(pool == NULL);
+				ERROR("%p: empty mbuf pool", (void *)rxq);
+				ret = ENOMEM;
+				goto error;
+			}
+			elt->bufs[j] = buf;
+			/* Headroom is reserved by rte_pktmbuf_alloc(). */
+			assert(DATA_OFF(buf) == RTE_PKTMBUF_HEADROOM);
+			/* Buffer is supposed to be empty. */
+			assert(rte_pktmbuf_data_len(buf) == 0);
+			assert(rte_pktmbuf_pkt_len(buf) == 0);
+			/* sge->addr must be able to store a pointer. */
+			assert(sizeof(sge->addr) >= sizeof(uintptr_t));
+			if (j == 0) {
+				/* The first SGE keeps its headroom. */
+				sge->addr = rte_pktmbuf_mtod(buf, uintptr_t);
+				sge->length = (buf->buf_len -
+					       RTE_PKTMBUF_HEADROOM);
+			} else {
+				/* Subsequent SGEs lose theirs. */
+				assert(DATA_OFF(buf) == RTE_PKTMBUF_HEADROOM);
+				SET_DATA_OFF(buf, 0);
+				sge->addr = (uintptr_t)buf->buf_addr;
+				sge->length = buf->buf_len;
+			}
+			sge->lkey = rxq->mr->lkey;
+			/* Redundant check for tailroom. */
+			assert(sge->length == rte_pktmbuf_tailroom(buf));
+		}
+	}
+	/* The last WR pointer must be NULL. */
+	(*elts)[(i - 1)].wr.next = NULL;
+	DEBUG("%p: allocated and configured %u WRs (%zu segments)",
+	      (void *)rxq, elts_n, (elts_n * RTE_DIM((*elts)[0].sges)));
+	rxq->elts_n = elts_n;
+	rxq->elts_head = 0;
+	rxq->elts.sp = elts;
+	assert(ret == 0);
+	return 0;
+error:
+	if (elts != NULL) {
+		assert(pool == NULL);
+		for (i = 0; (i != RTE_DIM(*elts)); ++i) {
+			unsigned int j;
+			struct rxq_elt_sp *elt = &(*elts)[i];
+
+			for (j = 0; (j != RTE_DIM(elt->bufs)); ++j) {
+				struct rte_mbuf *buf = elt->bufs[j];
+
+				if (buf != NULL)
+					rte_pktmbuf_free_seg(buf);
+			}
+		}
+		rte_free(elts);
+	}
+	DEBUG("%p: failed, freed everything", (void *)rxq);
+	assert(ret > 0);
+	return ret;
+}
+
+/**
+ * Free RX queue elements with scattered packets support.
+ *
+ * @param rxq
+ *   Pointer to RX queue structure.
+ */
+static void
+rxq_free_elts_sp(struct rxq *rxq)
+{
+	unsigned int i;
+	unsigned int elts_n = rxq->elts_n;
+	struct rxq_elt_sp (*elts)[elts_n] = rxq->elts.sp;
+
+	DEBUG("%p: freeing WRs", (void *)rxq);
+	rxq->elts_n = 0;
+	rxq->elts.sp = NULL;
+	if (elts == NULL)
+		return;
+	for (i = 0; (i != RTE_DIM(*elts)); ++i) {
+		unsigned int j;
+		struct rxq_elt_sp *elt = &(*elts)[i];
+
+		for (j = 0; (j != RTE_DIM(elt->bufs)); ++j) {
+			struct rte_mbuf *buf = elt->bufs[j];
+
+			if (buf != NULL)
+				rte_pktmbuf_free_seg(buf);
+		}
+	}
+	rte_free(elts);
+}
+
+/**
  * Allocate RX queue elements.
  *
  * @param rxq
@@ -224,7 +371,10 @@ rxq_cleanup(struct rxq *rxq)
 	struct ibv_exp_release_intf_params params;
 
 	DEBUG("cleaning up %p", (void *)rxq);
-	rxq_free_elts(rxq);
+	if (rxq->sp)
+		rxq_free_elts_sp(rxq);
+	else
+		rxq_free_elts(rxq);
 	if (rxq->if_qp != NULL) {
 		assert(rxq->priv != NULL);
 		assert(rxq->priv->ctx != NULL);
@@ -445,6 +595,15 @@ rxq_setup(struct rte_eth_dev *dev, struct rxq *rxq, uint16_t desc,
 		rte_pktmbuf_tailroom(buf)) == tmpl.mb_len);
 	assert(rte_pktmbuf_headroom(buf) == RTE_PKTMBUF_HEADROOM);
 	rte_pktmbuf_free(buf);
+	/* Enable scattered packets support for this queue if necessary. */
+	if ((dev->data->dev_conf.rxmode.jumbo_frame) &&
+	    (dev->data->dev_conf.rxmode.max_rx_pkt_len >
+	     (tmpl.mb_len - RTE_PKTMBUF_HEADROOM))) {
+		tmpl.sp = 1;
+		desc /= MLX5_PMD_SGE_WR_N;
+	}
+	DEBUG("%p: %s scattered packets support (%u WRs)",
+	      (void *)dev, (tmpl.sp ? "enabling" : "disabling"), desc);
 	/* Use the entire RX mempool as the memory region. */
 	tmpl.mr = ibv_reg_mr(priv->pd,
 			     (void *)mp->elt_va_start,
@@ -528,14 +687,19 @@ skip_mr:
 	/* Allocate descriptors for RX queues, except for the RSS parent. */
 	if (parent)
 		goto skip_alloc;
-	ret = rxq_alloc_elts(&tmpl, desc, NULL);
+	if (tmpl.sp)
+		ret = rxq_alloc_elts_sp(&tmpl, desc, NULL);
+	else
+		ret = rxq_alloc_elts(&tmpl, desc, NULL);
 	if (ret) {
 		ERROR("%p: RXQ allocation failed: %s",
 		      (void *)dev, strerror(ret));
 		goto error;
 	}
 	ret = ibv_post_recv(tmpl.qp,
-			    &(*tmpl.elts.no_sp)[0].wr,
+			    (tmpl.sp ?
+			     &(*tmpl.elts.sp)[0].wr :
+			     &(*tmpl.elts.no_sp)[0].wr),
 			    &bad_wr);
 	if (ret) {
 		ERROR("%p: ibv_post_recv() failed for WR %p: %s",
@@ -655,7 +819,10 @@ mlx5_rx_queue_setup(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc,
 		      (void *)dev, (void *)rxq);
 		(*priv->rxqs)[idx] = rxq;
 		/* Update receive callback. */
-		dev->rx_pkt_burst = mlx5_rx_burst;
+		if (rxq->sp)
+			dev->rx_pkt_burst = mlx5_rx_burst_sp;
+		else
+			dev->rx_pkt_burst = mlx5_rx_burst;
 	}
 	priv_unlock(priv);
 	return -ret;
diff --git a/drivers/net/mlx5/mlx5_rxtx.c b/drivers/net/mlx5/mlx5_rxtx.c
index 0f1e541..ed6faa1 100644
--- a/drivers/net/mlx5/mlx5_rxtx.c
+++ b/drivers/net/mlx5/mlx5_rxtx.c
@@ -173,6 +173,154 @@ txq_mp2mr(struct txq *txq, struct rte_mempool *mp)
 	return txq->mp2mr[i].lkey;
 }
 
+#if MLX5_PMD_SGE_WR_N > 1
+
+/**
+ * Copy scattered mbuf contents to a single linear buffer.
+ *
+ * @param[out] linear
+ *   Linear output buffer.
+ * @param[in] buf
+ *   Scattered input buffer.
+ *
+ * @return
+ *   Number of bytes copied to the output buffer or 0 if not large enough.
+ */
+static unsigned int
+linearize_mbuf(linear_t *linear, struct rte_mbuf *buf)
+{
+	unsigned int size = 0;
+	unsigned int offset;
+
+	do {
+		unsigned int len = DATA_LEN(buf);
+
+		offset = size;
+		size += len;
+		if (unlikely(size > sizeof(*linear)))
+			return 0;
+		memcpy(&(*linear)[offset],
+		       rte_pktmbuf_mtod(buf, uint8_t *),
+		       len);
+		buf = NEXT(buf);
+	} while (buf != NULL);
+	return size;
+}
+
+/**
+ * Handle scattered buffers for mlx5_tx_burst().
+ *
+ * @param txq
+ *   TX queue structure.
+ * @param segs
+ *   Number of segments in buf.
+ * @param elt
+ *   TX queue element to fill.
+ * @param[in] buf
+ *   Buffer to process.
+ * @param elts_head
+ *   Index of the linear buffer to use if necessary (normally txq->elts_head).
+ * @param[out] sges
+ *   Array filled with SGEs on success.
+ *
+ * @return
+ *   A structure containing the processed packet size in bytes and the
+ *   number of SGEs. Both fields are set to (unsigned int)-1 in case of
+ *   failure.
+ */
+static struct tx_burst_sg_ret {
+	unsigned int length;
+	unsigned int num;
+}
+tx_burst_sg(struct txq *txq, unsigned int segs, struct txq_elt *elt,
+	    struct rte_mbuf *buf, unsigned int elts_head,
+	    struct ibv_sge (*sges)[MLX5_PMD_SGE_WR_N])
+{
+	unsigned int sent_size = 0;
+	unsigned int j;
+	int linearize = 0;
+
+	/* When there are too many segments, extra segments are
+	 * linearized in the last SGE. */
+	if (unlikely(segs > RTE_DIM(*sges))) {
+		segs = (RTE_DIM(*sges) - 1);
+		linearize = 1;
+	}
+	/* Update element. */
+	elt->buf = buf;
+	/* Register segments as SGEs. */
+	for (j = 0; (j != segs); ++j) {
+		struct ibv_sge *sge = &(*sges)[j];
+		uint32_t lkey;
+
+		/* Retrieve Memory Region key for this memory pool. */
+		lkey = txq_mp2mr(txq, buf->pool);
+		if (unlikely(lkey == (uint32_t)-1)) {
+			/* MR does not exist. */
+			DEBUG("%p: unable to get MP <-> MR association",
+			      (void *)txq);
+			/* Clean up TX element. */
+			elt->buf = NULL;
+			goto stop;
+		}
+		/* Update SGE. */
+		sge->addr = rte_pktmbuf_mtod(buf, uintptr_t);
+		if (txq->priv->vf)
+			rte_prefetch0((volatile void *)
+				      (uintptr_t)sge->addr);
+		sge->length = DATA_LEN(buf);
+		sge->lkey = lkey;
+		sent_size += sge->length;
+		buf = NEXT(buf);
+	}
+	/* If buf is not NULL here and is not going to be linearized,
+	 * nb_segs is not valid. */
+	assert(j == segs);
+	assert((buf == NULL) || (linearize));
+	/* Linearize extra segments. */
+	if (linearize) {
+		struct ibv_sge *sge = &(*sges)[segs];
+		linear_t *linear = &(*txq->elts_linear)[elts_head];
+		unsigned int size = linearize_mbuf(linear, buf);
+
+		assert(segs == (RTE_DIM(*sges) - 1));
+		if (size == 0) {
+			/* Invalid packet. */
+			DEBUG("%p: packet too large to be linearized.",
+			      (void *)txq);
+			/* Clean up TX element. */
+			elt->buf = NULL;
+			goto stop;
+		}
+		/* If MLX5_PMD_SGE_WR_N is 1, free mbuf immediately. */
+		if (RTE_DIM(*sges) == 1) {
+			do {
+				struct rte_mbuf *next = NEXT(buf);
+
+				rte_pktmbuf_free_seg(buf);
+				buf = next;
+			} while (buf != NULL);
+			elt->buf = NULL;
+		}
+		/* Update SGE. */
+		sge->addr = (uintptr_t)&(*linear)[0];
+		sge->length = size;
+		sge->lkey = txq->mr_linear->lkey;
+		sent_size += size;
+	}
+	return (struct tx_burst_sg_ret){
+		.length = sent_size,
+		.num = segs,
+	};
+stop:
+	return (struct tx_burst_sg_ret){
+		.length = -1,
+		.num = -1,
+	};
+}
+
+#endif /* MLX5_PMD_SGE_WR_N > 1 */
+
 /**
  * DPDK callback for TX.
  *
@@ -282,9 +430,28 @@ mlx5_tx_burst(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n)
 			if (unlikely(err))
 				goto stop;
 		} else {
+#if MLX5_PMD_SGE_WR_N > 1
+			struct ibv_sge sges[MLX5_PMD_SGE_WR_N];
+			struct tx_burst_sg_ret ret;
+
+			ret = tx_burst_sg(txq, segs, elt, buf, elts_head,
+					  &sges);
+			if (ret.length == (unsigned int)-1)
+				goto stop;
+			RTE_MBUF_PREFETCH_TO_FREE(elt_next->buf);
+			/* Put SG list into send queue. */
+			err = txq->if_qp->send_pending_sg_list
+				(txq->qp,
+				 sges,
+				 ret.num,
+				 send_flags);
+			if (unlikely(err))
+				goto stop;
+#else /* MLX5_PMD_SGE_WR_N > 1 */
 			DEBUG("%p: TX scattered buffers support not"
 			      " compiled in", (void *)txq);
 			goto stop;
+#endif /* MLX5_PMD_SGE_WR_N > 1 */
 		}
 		elts_head = elts_head_next;
 	}
@@ -307,8 +474,215 @@ stop:
 }
 
 /**
+ * DPDK callback for RX with scattered packets support.
+ *
+ * @param dpdk_rxq
+ *   Generic pointer to RX queue structure.
+ * @param[out] pkts
+ *   Array to store received packets.
+ * @param pkts_n
+ *   Maximum number of packets in array.
+ *
+ * @return
+ *   Number of packets successfully received (<= pkts_n).
+ */
+uint16_t
+mlx5_rx_burst_sp(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n)
+{
+	struct rxq *rxq = (struct rxq *)dpdk_rxq;
+	struct rxq_elt_sp (*elts)[rxq->elts_n] = rxq->elts.sp;
+	const unsigned int elts_n = rxq->elts_n;
+	unsigned int elts_head = rxq->elts_head;
+	struct ibv_recv_wr head;
+	struct ibv_recv_wr **next = &head.next;
+	struct ibv_recv_wr *bad_wr;
+	unsigned int i;
+	unsigned int pkts_ret = 0;
+	int ret;
+
+	if (unlikely(!rxq->sp))
+		return mlx5_rx_burst(dpdk_rxq, pkts, pkts_n);
+	if (unlikely(elts == NULL)) /* See RTE_DEV_CMD_SET_MTU. */
+		return 0;
+	for (i = 0; (i != pkts_n); ++i) {
+		struct rxq_elt_sp *elt = &(*elts)[elts_head];
+		struct ibv_recv_wr *wr = &elt->wr;
+		uint64_t wr_id = wr->wr_id;
+		unsigned int len;
+		unsigned int pkt_buf_len;
+		struct rte_mbuf *pkt_buf = NULL; /* Buffer returned in pkts. */
+		struct rte_mbuf **pkt_buf_next = &pkt_buf;
+		unsigned int seg_headroom = RTE_PKTMBUF_HEADROOM;
+		unsigned int j = 0;
+		uint32_t flags;
+
+		/* Sanity checks. */
+#ifdef NDEBUG
+		(void)wr_id;
+#endif
+		assert(wr_id < rxq->elts_n);
+		assert(wr->sg_list == elt->sges);
+		assert(wr->num_sge == RTE_DIM(elt->sges));
+		assert(elts_head < rxq->elts_n);
+		assert(rxq->elts_head < rxq->elts_n);
+		ret = rxq->if_cq->poll_length_flags(rxq->cq, NULL, NULL,
+						    &flags);
+		if (unlikely(ret < 0)) {
+			struct ibv_wc wc;
+			int wcs_n;
+
+			DEBUG("rxq=%p, poll_length() failed (ret=%d)",
+			      (void *)rxq, ret);
+			/* ibv_poll_cq() must be used in case of failure. */
+			wcs_n = ibv_poll_cq(rxq->cq, 1, &wc);
+			if (unlikely(wcs_n == 0))
+				break;
+			if (unlikely(wcs_n < 0)) {
+				DEBUG("rxq=%p, ibv_poll_cq() failed (wcs_n=%d)",
+				      (void *)rxq, wcs_n);
+				break;
+			}
+			assert(wcs_n == 1);
+			if (unlikely(wc.status != IBV_WC_SUCCESS)) {
+				/* Whatever, just repost the offending WR. */
+				DEBUG("rxq=%p, wr_id=%" PRIu64 ": bad work"
+				      " completion status (%d): %s",
+				      (void *)rxq, wc.wr_id, wc.status,
+				      ibv_wc_status_str(wc.status));
+				/* Link completed WRs together for repost. */
+				*next = wr;
+				next = &wr->next;
+				goto repost;
+			}
+			ret = wc.byte_len;
+		}
+		if (ret == 0)
+			break;
+		len = ret;
+		pkt_buf_len = len;
+		/* Link completed WRs together for repost. */
+		*next = wr;
+		next = &wr->next;
+		/*
+		 * Replace spent segments with new ones, concatenate and
+		 * return them as pkt_buf.
+		 */
+		while (1) {
+			struct ibv_sge *sge = &elt->sges[j];
+			struct rte_mbuf *seg = elt->bufs[j];
+			struct rte_mbuf *rep;
+			unsigned int seg_tailroom;
+
+			/*
+			 * Fetch initial bytes of packet descriptor into a
+			 * cacheline while allocating rep.
+			 */
+			rte_prefetch0(seg);
+			rep = __rte_mbuf_raw_alloc(rxq->mp);
+			if (unlikely(rep == NULL)) {
+				/*
+				 * Unable to allocate a replacement mbuf,
+				 * repost WR.
+				 */
+				DEBUG("rxq=%p, wr_id=%" PRIu64 ":"
+				      " can't allocate a new mbuf",
+				      (void *)rxq, wr_id);
+				if (pkt_buf != NULL) {
+					*pkt_buf_next = NULL;
+					rte_pktmbuf_free(pkt_buf);
+				}
+				/* Increment out of memory counters. */
+				++rxq->priv->dev->data->rx_mbuf_alloc_failed;
+				goto repost;
+			}
+#ifndef NDEBUG
+			/* Poison user-modifiable fields in rep. */
+			NEXT(rep) = (void *)((uintptr_t)-1);
+			SET_DATA_OFF(rep, 0xdead);
+			DATA_LEN(rep) = 0xd00d;
+			PKT_LEN(rep) = 0xdeadd00d;
+			NB_SEGS(rep) = 0x2a;
+			PORT(rep) = 0x2a;
+			rep->ol_flags = -1;
+#endif
+			assert(rep->buf_len == seg->buf_len);
+			assert(rep->buf_len == rxq->mb_len);
+			/* Reconfigure sge to use rep instead of seg. */
+			assert(sge->lkey == rxq->mr->lkey);
+			sge->addr = ((uintptr_t)rep->buf_addr + seg_headroom);
+			elt->bufs[j] = rep;
+			++j;
+			/* Update pkt_buf if it's the first segment, or link
+			 * seg to the previous one and update pkt_buf_next. */
+			*pkt_buf_next = seg;
+			pkt_buf_next = &NEXT(seg);
+			/* Update seg information. */
+			seg_tailroom = (seg->buf_len - seg_headroom);
+			assert(sge->length == seg_tailroom);
+			SET_DATA_OFF(seg, seg_headroom);
+			if (likely(len <= seg_tailroom)) {
+				/* Last segment. */
+				DATA_LEN(seg) = len;
+				PKT_LEN(seg) = len;
+				/* Sanity check. */
+				assert(rte_pktmbuf_headroom(seg) ==
+				       seg_headroom);
+				assert(rte_pktmbuf_tailroom(seg) ==
+				       (seg_tailroom - len));
+				break;
+			}
+			DATA_LEN(seg) = seg_tailroom;
+			PKT_LEN(seg) = seg_tailroom;
+			/* Sanity check. */
+			assert(rte_pktmbuf_headroom(seg) == seg_headroom);
+			assert(rte_pktmbuf_tailroom(seg) == 0);
+			/* Fix len and clear headroom for next segments. */
+			len -= seg_tailroom;
+			seg_headroom = 0;
+		}
+		/* Update head and tail segments. */
+		*pkt_buf_next = NULL;
+		assert(pkt_buf != NULL);
+		assert(j != 0);
+		NB_SEGS(pkt_buf) = j;
+		PORT(pkt_buf) = rxq->port_id;
+		PKT_LEN(pkt_buf) = pkt_buf_len;
+
+		/* Return packet. */
+		*(pkts++) = pkt_buf;
+		++pkts_ret;
+repost:
+		if (++elts_head >= elts_n)
+			elts_head = 0;
+		continue;
+	}
+	if (unlikely(i == 0))
+		return 0;
+	*next = NULL;
+	/* Repost WRs. */
+#ifdef DEBUG_RECV
+	DEBUG("%p: reposting %d WRs", (void *)rxq, i);
+#endif
+	ret = ibv_post_recv(rxq->qp, head.next, &bad_wr);
+	if (unlikely(ret)) {
+		/* Inability to repost WRs is fatal. */
+		DEBUG("%p: ibv_post_recv(): failed for WR %p: %s",
+		      (void *)rxq->priv,
+		      (void *)bad_wr,
+		      strerror(ret));
+		abort();
+	}
+	rxq->elts_head = elts_head;
+	return pkts_ret;
+}
+
+/**
  * DPDK callback for RX.
  *
+ * The following function is the same as mlx5_rx_burst_sp(), except it doesn't
+ * manage scattered packets. Improves performance when MRU is lower than the
+ * size of the first segment.
+ *
  * @param dpdk_rxq
  *   Generic pointer to RX queue structure.
  * @param[out] pkts
@@ -331,6 +705,8 @@ mlx5_rx_burst(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n)
 	unsigned int pkts_ret = 0;
 	int ret;
 
+	if (unlikely(rxq->sp))
+		return mlx5_rx_burst_sp(dpdk_rxq, pkts, pkts_n);
 	for (i = 0; (i != pkts_n); ++i) {
 		struct rxq_elt *elt = &(*elts)[elts_head];
 		struct ibv_recv_wr *wr = &elt->wr;
diff --git a/drivers/net/mlx5/mlx5_rxtx.h b/drivers/net/mlx5/mlx5_rxtx.h
index 3733d3e..c7f634e 100644
--- a/drivers/net/mlx5/mlx5_rxtx.h
+++ b/drivers/net/mlx5/mlx5_rxtx.h
@@ -60,6 +60,13 @@
 #include "mlx5.h"
 #include "mlx5_defs.h"
 
+/* RX element (scattered packets). */
+struct rxq_elt_sp {
+	struct ibv_recv_wr wr; /* Work Request. */
+	struct ibv_sge sges[MLX5_PMD_SGE_WR_N]; /* Scatter/Gather Elements. */
+	struct rte_mbuf *bufs[MLX5_PMD_SGE_WR_N]; /* SGEs buffers. */
+};
+
 /* RX element. */
 struct rxq_elt {
 	struct ibv_recv_wr wr; /* Work Request. */
@@ -84,8 +91,10 @@ struct rxq {
 	unsigned int elts_n; /* (*elts)[] length. */
 	unsigned int elts_head; /* Current index in (*elts)[]. */
 	union {
+		struct rxq_elt_sp (*sp)[]; /* Scattered RX elements. */
 		struct rxq_elt (*no_sp)[]; /* RX elements. */
 	} elts;
+	unsigned int sp:1; /* Use scattered RX elements. */
 	uint32_t mb_len; /* Length of a mp-issued mbuf. */
 	unsigned int socket; /* CPU socket ID for allocations. */
 	struct ibv_exp_res_domain *rd; /* Resource Domain. */
@@ -151,6 +160,7 @@ void mlx5_tx_queue_release(void *);
 /* mlx5_rxtx.c */
 
 uint16_t mlx5_tx_burst(void *, struct rte_mbuf **, uint16_t);
+uint16_t mlx5_rx_burst_sp(void *, struct rte_mbuf **, uint16_t);
 uint16_t mlx5_rx_burst(void *, struct rte_mbuf **, uint16_t);
 uint16_t removed_tx_burst(void *, struct rte_mbuf **, uint16_t);
 uint16_t removed_rx_burst(void *, struct rte_mbuf **, uint16_t);
-- 
2.1.0

^ permalink raw reply	[flat|nested] 32+ messages in thread

* [dpdk-dev] [PATCH v2 06/13] mlx5: add MTU configuration support
  2015-10-30 18:52 ` [dpdk-dev] [PATCH v2 00/13] Mellanox ConnectX-4 PMD (mlx5) Adrien Mazarguil
                     ` (4 preceding siblings ...)
  2015-10-30 18:52   ` [dpdk-dev] [PATCH v2 05/13] mlx5: add support for scattered RX and TX buffers Adrien Mazarguil
@ 2015-10-30 18:52   ` Adrien Mazarguil
  2015-10-30 18:52   ` [dpdk-dev] [PATCH v2 07/13] mlx5: add software counters and related callbacks Adrien Mazarguil
                     ` (7 subsequent siblings)
  13 siblings, 0 replies; 32+ messages in thread
From: Adrien Mazarguil @ 2015-10-30 18:52 UTC (permalink / raw)
  To: dev

Depending on the MTU and whether jumbo frames are enabled, RX queues may
switch between SG and non-SG modes for better performance.

Signed-off-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>
Signed-off-by: Nelio Laranjeiro <nelio.laranjeiro@6wind.com>
---
 drivers/net/mlx5/mlx5.c        |   1 +
 drivers/net/mlx5/mlx5.h        |   1 +
 drivers/net/mlx5/mlx5_ethdev.c | 102 +++++++++++++++++++++++
 drivers/net/mlx5/mlx5_rxq.c    | 178 +++++++++++++++++++++++++++++++++++++++++
 drivers/net/mlx5/mlx5_rxtx.h   |   1 +
 5 files changed, 283 insertions(+)

diff --git a/drivers/net/mlx5/mlx5.c b/drivers/net/mlx5/mlx5.c
index aafa70b..ddd74d0 100644
--- a/drivers/net/mlx5/mlx5.c
+++ b/drivers/net/mlx5/mlx5.c
@@ -140,6 +140,7 @@ static const struct eth_dev_ops mlx5_dev_ops = {
 	.tx_queue_release = mlx5_tx_queue_release,
 	.mac_addr_remove = mlx5_mac_addr_remove,
 	.mac_addr_add = mlx5_mac_addr_add,
+	.mtu_set = mlx5_dev_set_mtu,
 };
 
 static struct {
diff --git a/drivers/net/mlx5/mlx5.h b/drivers/net/mlx5/mlx5.h
index 3f47a15..0e2457a 100644
--- a/drivers/net/mlx5/mlx5.h
+++ b/drivers/net/mlx5/mlx5.h
@@ -162,6 +162,7 @@ int priv_get_mtu(struct priv *, uint16_t *);
 int priv_set_flags(struct priv *, unsigned int, unsigned int);
 int mlx5_dev_configure(struct rte_eth_dev *);
 void mlx5_dev_infos_get(struct rte_eth_dev *, struct rte_eth_dev_info *);
+int mlx5_dev_set_mtu(struct rte_eth_dev *, uint16_t);
 int mlx5_ibv_device_to_pci_addr(const struct ibv_device *,
 				struct rte_pci_addr *);
 
diff --git a/drivers/net/mlx5/mlx5_ethdev.c b/drivers/net/mlx5/mlx5_ethdev.c
index 6b13cec..0afc1bb 100644
--- a/drivers/net/mlx5/mlx5_ethdev.c
+++ b/drivers/net/mlx5/mlx5_ethdev.c
@@ -347,6 +347,23 @@ priv_get_mtu(struct priv *priv, uint16_t *mtu)
 }
 
 /**
+ * Set device MTU.
+ *
+ * @param priv
+ *   Pointer to private structure.
+ * @param mtu
+ *   MTU value to set.
+ *
+ * @return
+ *   0 on success, -1 on failure and errno is set.
+ */
+static int
+priv_set_mtu(struct priv *priv, uint16_t mtu)
+{
+	return priv_set_sysfs_ulong(priv, "mtu", mtu);
+}
+
+/**
  * Set device flags.
  *
  * @param priv
@@ -518,6 +535,91 @@ mlx5_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *info)
 }
 
 /**
+ * DPDK callback to change the MTU.
+ *
+ * Setting the MTU affects hardware MRU (packets larger than the MTU cannot be
+ * received). Use this as a hint to enable/disable scattered packets support
+ * and improve performance when not needed.
+ * Since failure is not an option, reconfiguring queues on the fly is not
+ * recommended.
+ *
+ * @param dev
+ *   Pointer to Ethernet device structure.
+ * @param in_mtu
+ *   New MTU.
+ *
+ * @return
+ *   0 on success, negative errno value on failure.
+ */
+int
+mlx5_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
+{
+	struct priv *priv = dev->data->dev_private;
+	int ret = 0;
+	unsigned int i;
+	uint16_t (*rx_func)(void *, struct rte_mbuf **, uint16_t) =
+		mlx5_rx_burst;
+
+	priv_lock(priv);
+	/* Set kernel interface MTU first. */
+	if (priv_set_mtu(priv, mtu)) {
+		ret = errno;
+		WARN("cannot set port %u MTU to %u: %s", priv->port, mtu,
+		     strerror(ret));
+		goto out;
+	} else
+		DEBUG("adapter port %u MTU set to %u", priv->port, mtu);
+	priv->mtu = mtu;
+	/* Temporarily replace RX handler with a fake one, assuming it has not
+	 * been copied elsewhere. */
+	dev->rx_pkt_burst = removed_rx_burst;
+	/* Make sure everyone has left mlx5_rx_burst() and uses
+	 * removed_rx_burst() instead. */
+	rte_wmb();
+	usleep(1000);
+	/* Reconfigure each RX queue. */
+	for (i = 0; (i != priv->rxqs_n); ++i) {
+		struct rxq *rxq = (*priv->rxqs)[i];
+		unsigned int max_frame_len;
+		int sp;
+
+		if (rxq == NULL)
+			continue;
+		/* Calculate new maximum frame length according to MTU and
+		 * toggle scattered support (sp) if necessary. */
+		max_frame_len = (priv->mtu + ETHER_HDR_LEN +
+				 (ETHER_MAX_VLAN_FRAME_LEN - ETHER_MAX_LEN));
+		sp = (max_frame_len > (rxq->mb_len - RTE_PKTMBUF_HEADROOM));
+		/* Provide new values to rxq_setup(). */
+		dev->data->dev_conf.rxmode.jumbo_frame = sp;
+		dev->data->dev_conf.rxmode.max_rx_pkt_len = max_frame_len;
+		ret = rxq_rehash(dev, rxq);
+		if (ret) {
+			/* Force SP RX if that queue requires it and abort. */
+			if (rxq->sp)
+				rx_func = mlx5_rx_burst_sp;
+			break;
+		}
+		/* Reenable non-RSS queue attributes. No need to check
+		 * for errors at this stage. */
+		if (!priv->rss) {
+			if (priv->started)
+				rxq_mac_addrs_add(rxq);
+		}
+		/* Scattered burst function takes priority. */
+		if (rxq->sp)
+			rx_func = mlx5_rx_burst_sp;
+	}
+	/* Burst functions can now be called again. */
+	rte_wmb();
+	dev->rx_pkt_burst = rx_func;
+out:
+	priv_unlock(priv);
+	assert(ret >= 0);
+	return -ret;
+}
+
+/**
  * Get PCI information from struct ibv_device.
  *
  * @param device
diff --git a/drivers/net/mlx5/mlx5_rxq.c b/drivers/net/mlx5/mlx5_rxq.c
index 1eddfc7..71d4470 100644
--- a/drivers/net/mlx5/mlx5_rxq.c
+++ b/drivers/net/mlx5/mlx5_rxq.c
@@ -526,6 +526,184 @@ rxq_setup_qp_rss(struct priv *priv, struct ibv_cq *cq, uint16_t desc,
 #endif /* RSS_SUPPORT */
 
 /**
+ * Reconfigure a RX queue with new parameters.
+ *
+ * rxq_rehash() does not allocate mbufs, which, if not done from the right
+ * thread (such as a control thread), may corrupt the pool.
+ * In case of failure, the queue is left untouched.
+ *
+ * @param dev
+ *   Pointer to Ethernet device structure.
+ * @param rxq
+ *   RX queue pointer.
+ *
+ * @return
+ *   0 on success, errno value on failure.
+ */
+int
+rxq_rehash(struct rte_eth_dev *dev, struct rxq *rxq)
+{
+	struct priv *priv = rxq->priv;
+	struct rxq tmpl = *rxq;
+	unsigned int mbuf_n;
+	unsigned int desc_n;
+	struct rte_mbuf **pool;
+	unsigned int i, k;
+	struct ibv_exp_qp_attr mod;
+	struct ibv_recv_wr *bad_wr;
+	int err;
+	int parent = (rxq == &priv->rxq_parent);
+
+	if (parent) {
+		ERROR("%p: cannot rehash parent queue %p",
+		      (void *)dev, (void *)rxq);
+		return EINVAL;
+	}
+	DEBUG("%p: rehashing queue %p", (void *)dev, (void *)rxq);
+	/* Number of descriptors and mbufs currently allocated. */
+	desc_n = (tmpl.elts_n * (tmpl.sp ? MLX5_PMD_SGE_WR_N : 1));
+	mbuf_n = desc_n;
+	/* Enable scattered packets support for this queue if necessary. */
+	if ((dev->data->dev_conf.rxmode.jumbo_frame) &&
+	    (dev->data->dev_conf.rxmode.max_rx_pkt_len >
+	     (tmpl.mb_len - RTE_PKTMBUF_HEADROOM))) {
+		tmpl.sp = 1;
+		desc_n /= MLX5_PMD_SGE_WR_N;
+	} else
+		tmpl.sp = 0;
+	DEBUG("%p: %s scattered packets support (%u WRs)",
+	      (void *)dev, (tmpl.sp ? "enabling" : "disabling"), desc_n);
+	/* If scatter mode is the same as before, nothing to do. */
+	if (tmpl.sp == rxq->sp) {
+		DEBUG("%p: nothing to do", (void *)dev);
+		return 0;
+	}
+	/* Remove attached flows if RSS is disabled (no parent queue). */
+	if (!priv->rss) {
+		rxq_mac_addrs_del(&tmpl);
+		/* Update original queue in case of failure. */
+		memcpy(rxq->mac_flow, tmpl.mac_flow, sizeof(rxq->mac_flow));
+	}
+	/* From now on, any failure will render the queue unusable.
+	 * Reinitialize QP. */
+	mod = (struct ibv_exp_qp_attr){ .qp_state = IBV_QPS_RESET };
+	err = ibv_exp_modify_qp(tmpl.qp, &mod, IBV_EXP_QP_STATE);
+	if (err) {
+		ERROR("%p: cannot reset QP: %s", (void *)dev, strerror(err));
+		assert(err > 0);
+		return err;
+	}
+	err = ibv_resize_cq(tmpl.cq, desc_n);
+	if (err) {
+		ERROR("%p: cannot resize CQ: %s", (void *)dev, strerror(err));
+		assert(err > 0);
+		return err;
+	}
+	mod = (struct ibv_exp_qp_attr){
+		/* Move the QP to this state. */
+		.qp_state = IBV_QPS_INIT,
+		/* Primary port number. */
+		.port_num = priv->port
+	};
+	err = ibv_exp_modify_qp(tmpl.qp, &mod,
+				(IBV_EXP_QP_STATE |
+#ifdef RSS_SUPPORT
+				 (parent ? IBV_EXP_QP_GROUP_RSS : 0) |
+#endif /* RSS_SUPPORT */
+				 IBV_EXP_QP_PORT));
+	if (err) {
+		ERROR("%p: QP state to IBV_QPS_INIT failed: %s",
+		      (void *)dev, strerror(err));
+		assert(err > 0);
+		return err;
+	};
+	/* Reconfigure flows. Do not care for errors. */
+	if (!priv->rss) {
+		if (priv->started)
+			rxq_mac_addrs_add(&tmpl);
+		/* Update original queue in case of failure. */
+		memcpy(rxq->mac_flow, tmpl.mac_flow, sizeof(rxq->mac_flow));
+	}
+	/* Allocate pool. */
+	pool = rte_malloc(__func__, (mbuf_n * sizeof(*pool)), 0);
+	if (pool == NULL) {
+		ERROR("%p: cannot allocate memory", (void *)dev);
+		return ENOBUFS;
+	}
+	/* Snatch mbufs from original queue. */
+	k = 0;
+	if (rxq->sp) {
+		struct rxq_elt_sp (*elts)[rxq->elts_n] = rxq->elts.sp;
+
+		for (i = 0; (i != RTE_DIM(*elts)); ++i) {
+			struct rxq_elt_sp *elt = &(*elts)[i];
+			unsigned int j;
+
+			for (j = 0; (j != RTE_DIM(elt->bufs)); ++j) {
+				assert(elt->bufs[j] != NULL);
+				pool[k++] = elt->bufs[j];
+			}
+		}
+	} else {
+		struct rxq_elt (*elts)[rxq->elts_n] = rxq->elts.no_sp;
+
+		for (i = 0; (i != RTE_DIM(*elts)); ++i) {
+			struct rxq_elt *elt = &(*elts)[i];
+			struct rte_mbuf *buf = (void *)
+				((uintptr_t)elt->sge.addr -
+				 WR_ID(elt->wr.wr_id).offset);
+
+			assert(WR_ID(elt->wr.wr_id).id == i);
+			pool[k++] = buf;
+		}
+	}
+	assert(k == mbuf_n);
+	tmpl.elts_n = 0;
+	tmpl.elts.sp = NULL;
+	assert((void *)&tmpl.elts.sp == (void *)&tmpl.elts.no_sp);
+	err = ((tmpl.sp) ?
+	       rxq_alloc_elts_sp(&tmpl, desc_n, pool) :
+	       rxq_alloc_elts(&tmpl, desc_n, pool));
+	if (err) {
+		ERROR("%p: cannot reallocate WRs, aborting", (void *)dev);
+		rte_free(pool);
+		assert(err > 0);
+		return err;
+	}
+	assert(tmpl.elts_n == desc_n);
+	assert(tmpl.elts.sp != NULL);
+	rte_free(pool);
+	/* Clean up original data. */
+	rxq->elts_n = 0;
+	rte_free(rxq->elts.sp);
+	rxq->elts.sp = NULL;
+	/* Post WRs. */
+	err = ibv_post_recv(tmpl.qp,
+			    (tmpl.sp ?
+			     &(*tmpl.elts.sp)[0].wr :
+			     &(*tmpl.elts.no_sp)[0].wr),
+			    &bad_wr);
+	if (err) {
+		ERROR("%p: ibv_post_recv() failed for WR %p: %s",
+		      (void *)dev,
+		      (void *)bad_wr,
+		      strerror(err));
+		goto skip_rtr;
+	}
+	mod = (struct ibv_exp_qp_attr){
+		.qp_state = IBV_QPS_RTR
+	};
+	err = ibv_exp_modify_qp(tmpl.qp, &mod, IBV_EXP_QP_STATE);
+	if (err)
+		ERROR("%p: QP state to IBV_QPS_RTR failed: %s",
+		      (void *)dev, strerror(err));
+skip_rtr:
+	*rxq = tmpl;
+	assert(err >= 0);
+	return err;
+}
+
+/**
  * Configure a RX queue.
  *
  * @param dev
diff --git a/drivers/net/mlx5/mlx5_rxtx.h b/drivers/net/mlx5/mlx5_rxtx.h
index c7f634e..b6f2128 100644
--- a/drivers/net/mlx5/mlx5_rxtx.h
+++ b/drivers/net/mlx5/mlx5_rxtx.h
@@ -144,6 +144,7 @@ struct txq {
 /* mlx5_rxq.c */
 
 void rxq_cleanup(struct rxq *);
+int rxq_rehash(struct rte_eth_dev *, struct rxq *);
 int rxq_setup(struct rte_eth_dev *, struct rxq *, uint16_t, unsigned int,
 	      const struct rte_eth_rxconf *, struct rte_mempool *);
 int mlx5_rx_queue_setup(struct rte_eth_dev *, uint16_t, uint16_t, unsigned int,
-- 
2.1.0

^ permalink raw reply	[flat|nested] 32+ messages in thread

* [dpdk-dev] [PATCH v2 07/13] mlx5: add software counters and related callbacks
  2015-10-30 18:52 ` [dpdk-dev] [PATCH v2 00/13] Mellanox ConnectX-4 PMD (mlx5) Adrien Mazarguil
                     ` (5 preceding siblings ...)
  2015-10-30 18:52   ` [dpdk-dev] [PATCH v2 06/13] mlx5: add MTU configuration support Adrien Mazarguil
@ 2015-10-30 18:52   ` Adrien Mazarguil
  2015-10-30 18:52   ` [dpdk-dev] [PATCH v2 08/13] mlx5: add promiscuous and allmulticast RX modes Adrien Mazarguil
                     ` (6 subsequent siblings)
  13 siblings, 0 replies; 32+ messages in thread
From: Adrien Mazarguil @ 2015-10-30 18:52 UTC (permalink / raw)
  To: dev

Hardware counters are not supported yet.

Signed-off-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>
Signed-off-by: Nelio Laranjeiro <nelio.laranjeiro@6wind.com>
---
 drivers/net/mlx5/Makefile     |   1 +
 drivers/net/mlx5/mlx5.c       |   2 +
 drivers/net/mlx5/mlx5.h       |   5 ++
 drivers/net/mlx5/mlx5_defs.h  |   8 +++
 drivers/net/mlx5/mlx5_rxq.c   |   1 +
 drivers/net/mlx5/mlx5_rxtx.c  |  43 +++++++++++++
 drivers/net/mlx5/mlx5_rxtx.h  |  21 ++++++
 drivers/net/mlx5/mlx5_stats.c | 144 ++++++++++++++++++++++++++++++++++++++++++
 drivers/net/mlx5/mlx5_txq.c   |   1 +
 9 files changed, 226 insertions(+)
 create mode 100644 drivers/net/mlx5/mlx5_stats.c

diff --git a/drivers/net/mlx5/Makefile b/drivers/net/mlx5/Makefile
index 028c22c..88b361c 100644
--- a/drivers/net/mlx5/Makefile
+++ b/drivers/net/mlx5/Makefile
@@ -48,6 +48,7 @@ SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5_rxtx.c
 SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5_trigger.c
 SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5_ethdev.c
 SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5_mac.c
+SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5_stats.c
 
 # Dependencies.
 DEPDIRS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += lib/librte_ether
diff --git a/drivers/net/mlx5/mlx5.c b/drivers/net/mlx5/mlx5.c
index ddd74d0..38f5199 100644
--- a/drivers/net/mlx5/mlx5.c
+++ b/drivers/net/mlx5/mlx5.c
@@ -133,6 +133,8 @@ static const struct eth_dev_ops mlx5_dev_ops = {
 	.dev_start = mlx5_dev_start,
 	.dev_stop = mlx5_dev_stop,
 	.dev_close = mlx5_dev_close,
+	.stats_get = mlx5_stats_get,
+	.stats_reset = mlx5_stats_reset,
 	.dev_infos_get = mlx5_dev_infos_get,
 	.rx_queue_setup = mlx5_rx_queue_setup,
 	.tx_queue_setup = mlx5_tx_queue_setup,
diff --git a/drivers/net/mlx5/mlx5.h b/drivers/net/mlx5/mlx5.h
index 0e2457a..79559bc 100644
--- a/drivers/net/mlx5/mlx5.h
+++ b/drivers/net/mlx5/mlx5.h
@@ -177,6 +177,11 @@ int priv_mac_addr_add(struct priv *, unsigned int,
 void mlx5_mac_addr_add(struct rte_eth_dev *, struct ether_addr *, uint32_t,
 		       uint32_t);
 
+/* mlx5_stats.c */
+
+void mlx5_stats_get(struct rte_eth_dev *, struct rte_eth_stats *);
+void mlx5_stats_reset(struct rte_eth_dev *);
+
 /* mlx5_trigger.c */
 
 int mlx5_dev_start(struct rte_eth_dev *);
diff --git a/drivers/net/mlx5/mlx5_defs.h b/drivers/net/mlx5/mlx5_defs.h
index c85be9c..d3a0d0e 100644
--- a/drivers/net/mlx5/mlx5_defs.h
+++ b/drivers/net/mlx5/mlx5_defs.h
@@ -64,4 +64,12 @@
 #define MLX5_PMD_TX_MP_CACHE 8
 #endif
 
+/*
+ * If defined, only use software counters. The PMD will never ask the hardware
+ * for these, and many of them won't be available.
+ */
+#ifndef MLX5_PMD_SOFT_COUNTERS
+#define MLX5_PMD_SOFT_COUNTERS 1
+#endif
+
 #endif /* RTE_PMD_MLX5_DEFS_H_ */
diff --git a/drivers/net/mlx5/mlx5_rxq.c b/drivers/net/mlx5/mlx5_rxq.c
index 71d4470..620ec70 100644
--- a/drivers/net/mlx5/mlx5_rxq.c
+++ b/drivers/net/mlx5/mlx5_rxq.c
@@ -993,6 +993,7 @@ mlx5_rx_queue_setup(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc,
 	if (ret)
 		rte_free(rxq);
 	else {
+		rxq->stats.idx = idx;
 		DEBUG("%p: adding RX queue %p to list",
 		      (void *)dev, (void *)rxq);
 		(*priv->rxqs)[idx] = rxq;
diff --git a/drivers/net/mlx5/mlx5_rxtx.c b/drivers/net/mlx5/mlx5_rxtx.c
index ed6faa1..eccbbb9 100644
--- a/drivers/net/mlx5/mlx5_rxtx.c
+++ b/drivers/net/mlx5/mlx5_rxtx.c
@@ -367,6 +367,9 @@ mlx5_tx_burst(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n)
 		struct txq_elt *elt_next = &(*txq->elts)[elts_head_next];
 		struct txq_elt *elt = &(*txq->elts)[elts_head];
 		unsigned int segs = NB_SEGS(buf);
+#ifdef MLX5_PMD_SOFT_COUNTERS
+		unsigned int sent_size = 0;
+#endif
 		uint32_t send_flags = 0;
 
 		/* Clean up old buffer. */
@@ -429,6 +432,9 @@ mlx5_tx_burst(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n)
 					 send_flags);
 			if (unlikely(err))
 				goto stop;
+#ifdef MLX5_PMD_SOFT_COUNTERS
+			sent_size += length;
+#endif
 		} else {
 #if MLX5_PMD_SGE_WR_N > 1
 			struct ibv_sge sges[MLX5_PMD_SGE_WR_N];
@@ -447,6 +453,9 @@ mlx5_tx_burst(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n)
 				 send_flags);
 			if (unlikely(err))
 				goto stop;
+#ifdef MLX5_PMD_SOFT_COUNTERS
+			sent_size += ret.length;
+#endif
 #else /* MLX5_PMD_SGE_WR_N > 1 */
 			DEBUG("%p: TX scattered buffers support not"
 			      " compiled in", (void *)txq);
@@ -454,11 +463,19 @@ mlx5_tx_burst(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n)
 #endif /* MLX5_PMD_SGE_WR_N > 1 */
 		}
 		elts_head = elts_head_next;
+#ifdef MLX5_PMD_SOFT_COUNTERS
+		/* Increment sent bytes counter. */
+		txq->stats.obytes += sent_size;
+#endif
 	}
 stop:
 	/* Take a shortcut if nothing must be sent. */
 	if (unlikely(i == 0))
 		return 0;
+#ifdef MLX5_PMD_SOFT_COUNTERS
+	/* Increment sent packets counter. */
+	txq->stats.opackets += i;
+#endif
 	/* Ring QP doorbell. */
 	err = txq->if_qp->send_flush(txq->qp);
 	if (unlikely(err)) {
@@ -549,6 +566,10 @@ mlx5_rx_burst_sp(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n)
 				      " completion status (%d): %s",
 				      (void *)rxq, wc.wr_id, wc.status,
 				      ibv_wc_status_str(wc.status));
+#ifdef MLX5_PMD_SOFT_COUNTERS
+				/* Increment dropped packets counter. */
+				++rxq->stats.idropped;
+#endif
 				/* Link completed WRs together for repost. */
 				*next = wr;
 				next = &wr->next;
@@ -592,6 +613,7 @@ mlx5_rx_burst_sp(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n)
 					rte_pktmbuf_free(pkt_buf);
 				}
 				/* Increment out of memory counters. */
+				++rxq->stats.rx_nombuf;
 				++rxq->priv->dev->data->rx_mbuf_alloc_failed;
 				goto repost;
 			}
@@ -651,6 +673,10 @@ mlx5_rx_burst_sp(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n)
 		/* Return packet. */
 		*(pkts++) = pkt_buf;
 		++pkts_ret;
+#ifdef MLX5_PMD_SOFT_COUNTERS
+		/* Increment bytes counter. */
+		rxq->stats.ibytes += pkt_buf_len;
+#endif
 repost:
 		if (++elts_head >= elts_n)
 			elts_head = 0;
@@ -673,6 +699,10 @@ repost:
 		abort();
 	}
 	rxq->elts_head = elts_head;
+#ifdef MLX5_PMD_SOFT_COUNTERS
+	/* Increment packets counter. */
+	rxq->stats.ipackets += pkts_ret;
+#endif
 	return pkts_ret;
 }
 
@@ -753,6 +783,10 @@ mlx5_rx_burst(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n)
 				      " completion status (%d): %s",
 				      (void *)rxq, wc.wr_id, wc.status,
 				      ibv_wc_status_str(wc.status));
+#ifdef MLX5_PMD_SOFT_COUNTERS
+				/* Increment dropped packets counter. */
+				++rxq->stats.idropped;
+#endif
 				/* Add SGE to array for repost. */
 				sges[i] = elt->sge;
 				goto repost;
@@ -772,6 +806,7 @@ mlx5_rx_burst(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n)
 			      " can't allocate a new mbuf",
 			      (void *)rxq, WR_ID(wr_id).id);
 			/* Increment out of memory counters. */
+			++rxq->stats.rx_nombuf;
 			++rxq->priv->dev->data->rx_mbuf_alloc_failed;
 			goto repost;
 		}
@@ -798,6 +833,10 @@ mlx5_rx_burst(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n)
 		/* Return packet. */
 		*(pkts++) = seg;
 		++pkts_ret;
+#ifdef MLX5_PMD_SOFT_COUNTERS
+		/* Increment bytes counter. */
+		rxq->stats.ibytes += len;
+#endif
 repost:
 		if (++elts_head >= elts_n)
 			elts_head = 0;
@@ -818,6 +857,10 @@ repost:
 		abort();
 	}
 	rxq->elts_head = elts_head;
+#ifdef MLX5_PMD_SOFT_COUNTERS
+	/* Increment packets counter. */
+	rxq->stats.ipackets += pkts_ret;
+#endif
 	return pkts_ret;
 }
 
diff --git a/drivers/net/mlx5/mlx5_rxtx.h b/drivers/net/mlx5/mlx5_rxtx.h
index b6f2128..4183820 100644
--- a/drivers/net/mlx5/mlx5_rxtx.h
+++ b/drivers/net/mlx5/mlx5_rxtx.h
@@ -60,6 +60,25 @@
 #include "mlx5.h"
 #include "mlx5_defs.h"
 
+struct mlx5_rxq_stats {
+	unsigned int idx; /**< Mapping index. */
+#ifdef MLX5_PMD_SOFT_COUNTERS
+	uint64_t ipackets; /**< Total of successfully received packets. */
+	uint64_t ibytes; /**< Total of successfully received bytes. */
+#endif
+	uint64_t idropped; /**< Total of packets dropped when RX ring full. */
+	uint64_t rx_nombuf; /**< Total of RX mbuf allocation failures. */
+};
+
+struct mlx5_txq_stats {
+	unsigned int idx; /**< Mapping index. */
+#ifdef MLX5_PMD_SOFT_COUNTERS
+	uint64_t opackets; /**< Total of successfully sent packets. */
+	uint64_t obytes; /**< Total of successfully sent bytes. */
+#endif
+	uint64_t odropped; /**< Total of packets not sent when TX ring full. */
+};
+
 /* RX element (scattered packets). */
 struct rxq_elt_sp {
 	struct ibv_recv_wr wr; /* Work Request. */
@@ -96,6 +115,7 @@ struct rxq {
 	} elts;
 	unsigned int sp:1; /* Use scattered RX elements. */
 	uint32_t mb_len; /* Length of a mp-issued mbuf. */
+	struct mlx5_rxq_stats stats; /* RX queue counters. */
 	unsigned int socket; /* CPU socket ID for allocations. */
 	struct ibv_exp_res_domain *rd; /* Resource Domain. */
 };
@@ -135,6 +155,7 @@ struct txq {
 	unsigned int elts_comp; /* Number of completion requests. */
 	unsigned int elts_comp_cd; /* Countdown for next completion request. */
 	unsigned int elts_comp_cd_init; /* Initial value for countdown. */
+	struct mlx5_txq_stats stats; /* TX queue counters. */
 	linear_t (*elts_linear)[]; /* Linearized buffers. */
 	struct ibv_mr *mr_linear; /* Memory Region for linearized buffers. */
 	unsigned int socket; /* CPU socket ID for allocations. */
diff --git a/drivers/net/mlx5/mlx5_stats.c b/drivers/net/mlx5/mlx5_stats.c
new file mode 100644
index 0000000..a51e945
--- /dev/null
+++ b/drivers/net/mlx5/mlx5_stats.c
@@ -0,0 +1,144 @@
+/*-
+ *   BSD LICENSE
+ *
+ *   Copyright 2015 6WIND S.A.
+ *   Copyright 2015 Mellanox.
+ *
+ *   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 6WIND S.A. 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.
+ */
+
+/* DPDK headers don't like -pedantic. */
+#ifdef PEDANTIC
+#pragma GCC diagnostic ignored "-pedantic"
+#endif
+#include <rte_ethdev.h>
+#ifdef PEDANTIC
+#pragma GCC diagnostic error "-pedantic"
+#endif
+
+#include "mlx5.h"
+#include "mlx5_rxtx.h"
+#include "mlx5_defs.h"
+
+/**
+ * DPDK callback to get device statistics.
+ *
+ * @param dev
+ *   Pointer to Ethernet device structure.
+ * @param[out] stats
+ *   Stats structure output buffer.
+ */
+void
+mlx5_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
+{
+	struct priv *priv = dev->data->dev_private;
+	struct rte_eth_stats tmp = {0};
+	unsigned int i;
+	unsigned int idx;
+
+	priv_lock(priv);
+	/* Add software counters. */
+	for (i = 0; (i != priv->rxqs_n); ++i) {
+		struct rxq *rxq = (*priv->rxqs)[i];
+
+		if (rxq == NULL)
+			continue;
+		idx = rxq->stats.idx;
+		if (idx < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
+#ifdef MLX5_PMD_SOFT_COUNTERS
+			tmp.q_ipackets[idx] += rxq->stats.ipackets;
+			tmp.q_ibytes[idx] += rxq->stats.ibytes;
+#endif
+			tmp.q_errors[idx] += (rxq->stats.idropped +
+					      rxq->stats.rx_nombuf);
+		}
+#ifdef MLX5_PMD_SOFT_COUNTERS
+		tmp.ipackets += rxq->stats.ipackets;
+		tmp.ibytes += rxq->stats.ibytes;
+#endif
+		tmp.ierrors += rxq->stats.idropped;
+		tmp.rx_nombuf += rxq->stats.rx_nombuf;
+	}
+	for (i = 0; (i != priv->txqs_n); ++i) {
+		struct txq *txq = (*priv->txqs)[i];
+
+		if (txq == NULL)
+			continue;
+		idx = txq->stats.idx;
+		if (idx < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
+#ifdef MLX5_PMD_SOFT_COUNTERS
+			tmp.q_opackets[idx] += txq->stats.opackets;
+			tmp.q_obytes[idx] += txq->stats.obytes;
+#endif
+			tmp.q_errors[idx] += txq->stats.odropped;
+		}
+#ifdef MLX5_PMD_SOFT_COUNTERS
+		tmp.opackets += txq->stats.opackets;
+		tmp.obytes += txq->stats.obytes;
+#endif
+		tmp.oerrors += txq->stats.odropped;
+	}
+#ifndef MLX5_PMD_SOFT_COUNTERS
+	/* FIXME: retrieve and add hardware counters. */
+#endif
+	*stats = tmp;
+	priv_unlock(priv);
+}
+
+/**
+ * DPDK callback to clear device statistics.
+ *
+ * @param dev
+ *   Pointer to Ethernet device structure.
+ */
+void
+mlx5_stats_reset(struct rte_eth_dev *dev)
+{
+	struct priv *priv = dev->data->dev_private;
+	unsigned int i;
+	unsigned int idx;
+
+	priv_lock(priv);
+	for (i = 0; (i != priv->rxqs_n); ++i) {
+		if ((*priv->rxqs)[i] == NULL)
+			continue;
+		idx = (*priv->rxqs)[i]->stats.idx;
+		(*priv->rxqs)[i]->stats =
+			(struct mlx5_rxq_stats){ .idx = idx };
+	}
+	for (i = 0; (i != priv->txqs_n); ++i) {
+		if ((*priv->txqs)[i] == NULL)
+			continue;
+		idx = (*priv->rxqs)[i]->stats.idx;
+		(*priv->txqs)[i]->stats =
+			(struct mlx5_txq_stats){ .idx = idx };
+	}
+#ifndef MLX5_PMD_SOFT_COUNTERS
+	/* FIXME: reset hardware counters. */
+#endif
+	priv_unlock(priv);
+}
diff --git a/drivers/net/mlx5/mlx5_txq.c b/drivers/net/mlx5/mlx5_txq.c
index 2bae61f..a53b128 100644
--- a/drivers/net/mlx5/mlx5_txq.c
+++ b/drivers/net/mlx5/mlx5_txq.c
@@ -472,6 +472,7 @@ mlx5_tx_queue_setup(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc,
 	if (ret)
 		rte_free(txq);
 	else {
+		txq->stats.idx = idx;
 		DEBUG("%p: adding TX queue %p to list",
 		      (void *)dev, (void *)txq);
 		(*priv->txqs)[idx] = txq;
-- 
2.1.0

^ permalink raw reply	[flat|nested] 32+ messages in thread

* [dpdk-dev] [PATCH v2 08/13] mlx5: add promiscuous and allmulticast RX modes
  2015-10-30 18:52 ` [dpdk-dev] [PATCH v2 00/13] Mellanox ConnectX-4 PMD (mlx5) Adrien Mazarguil
                     ` (6 preceding siblings ...)
  2015-10-30 18:52   ` [dpdk-dev] [PATCH v2 07/13] mlx5: add software counters and related callbacks Adrien Mazarguil
@ 2015-10-30 18:52   ` Adrien Mazarguil
  2015-10-30 18:52   ` [dpdk-dev] [PATCH v2 09/13] mlx5: add link update device operation Adrien Mazarguil
                     ` (5 subsequent siblings)
  13 siblings, 0 replies; 32+ messages in thread
From: Adrien Mazarguil @ 2015-10-30 18:52 UTC (permalink / raw)
  To: dev; +Cc: Yaacov Hazan

These modes require special non-MAC flows.

Signed-off-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>
Signed-off-by: Nelio Laranjeiro <nelio.laranjeiro@6wind.com>
Signed-off-by: Yaacov Hazan <yaacovh@mellanox.com>
---
 drivers/net/mlx5/Makefile       |   1 +
 drivers/net/mlx5/mlx5.c         |   4 +
 drivers/net/mlx5/mlx5.h         |  13 ++
 drivers/net/mlx5/mlx5_ethdev.c  |   4 +
 drivers/net/mlx5/mlx5_rxmode.c  | 311 ++++++++++++++++++++++++++++++++++++++++
 drivers/net/mlx5/mlx5_rxq.c     |  12 ++
 drivers/net/mlx5/mlx5_rxtx.h    |   2 +
 drivers/net/mlx5/mlx5_trigger.c |   8 ++
 8 files changed, 355 insertions(+)
 create mode 100644 drivers/net/mlx5/mlx5_rxmode.c

diff --git a/drivers/net/mlx5/Makefile b/drivers/net/mlx5/Makefile
index 88b361c..4d25d9c 100644
--- a/drivers/net/mlx5/Makefile
+++ b/drivers/net/mlx5/Makefile
@@ -48,6 +48,7 @@ SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5_rxtx.c
 SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5_trigger.c
 SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5_ethdev.c
 SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5_mac.c
+SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5_rxmode.c
 SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5_stats.c
 
 # Dependencies.
diff --git a/drivers/net/mlx5/mlx5.c b/drivers/net/mlx5/mlx5.c
index 38f5199..ee63bdf 100644
--- a/drivers/net/mlx5/mlx5.c
+++ b/drivers/net/mlx5/mlx5.c
@@ -133,6 +133,10 @@ static const struct eth_dev_ops mlx5_dev_ops = {
 	.dev_start = mlx5_dev_start,
 	.dev_stop = mlx5_dev_stop,
 	.dev_close = mlx5_dev_close,
+	.promiscuous_enable = mlx5_promiscuous_enable,
+	.promiscuous_disable = mlx5_promiscuous_disable,
+	.allmulticast_enable = mlx5_allmulticast_enable,
+	.allmulticast_disable = mlx5_allmulticast_disable,
 	.stats_get = mlx5_stats_get,
 	.stats_reset = mlx5_stats_reset,
 	.dev_infos_get = mlx5_dev_infos_get,
diff --git a/drivers/net/mlx5/mlx5.h b/drivers/net/mlx5/mlx5.h
index 79559bc..56da43c 100644
--- a/drivers/net/mlx5/mlx5.h
+++ b/drivers/net/mlx5/mlx5.h
@@ -94,6 +94,8 @@ struct priv {
 	uint16_t mtu; /* Configured MTU. */
 	uint8_t port; /* Physical port number. */
 	unsigned int started:1; /* Device started, flows enabled. */
+	unsigned int promisc_req:1; /* Promiscuous mode requested. */
+	unsigned int allmulti_req:1; /* All multicast mode requested. */
 	unsigned int hw_qpg:1; /* QP groups are supported. */
 	unsigned int hw_tss:1; /* TSS is supported. */
 	unsigned int hw_rss:1; /* RSS is supported. */
@@ -177,6 +179,17 @@ int priv_mac_addr_add(struct priv *, unsigned int,
 void mlx5_mac_addr_add(struct rte_eth_dev *, struct ether_addr *, uint32_t,
 		       uint32_t);
 
+/* mlx5_rxmode.c */
+
+int rxq_promiscuous_enable(struct rxq *);
+void mlx5_promiscuous_enable(struct rte_eth_dev *);
+void rxq_promiscuous_disable(struct rxq *);
+void mlx5_promiscuous_disable(struct rte_eth_dev *);
+int rxq_allmulticast_enable(struct rxq *);
+void mlx5_allmulticast_enable(struct rte_eth_dev *);
+void rxq_allmulticast_disable(struct rxq *);
+void mlx5_allmulticast_disable(struct rte_eth_dev *);
+
 /* mlx5_stats.c */
 
 void mlx5_stats_get(struct rte_eth_dev *, struct rte_eth_stats *);
diff --git a/drivers/net/mlx5/mlx5_ethdev.c b/drivers/net/mlx5/mlx5_ethdev.c
index 0afc1bb..26b6d73 100644
--- a/drivers/net/mlx5/mlx5_ethdev.c
+++ b/drivers/net/mlx5/mlx5_ethdev.c
@@ -605,6 +605,10 @@ mlx5_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
 		if (!priv->rss) {
 			if (priv->started)
 				rxq_mac_addrs_add(rxq);
+			if (priv->started && priv->promisc_req)
+				rxq_promiscuous_enable(rxq);
+			if (priv->started && priv->allmulti_req)
+				rxq_allmulticast_enable(rxq);
 		}
 		/* Scattered burst function takes priority. */
 		if (rxq->sp)
diff --git a/drivers/net/mlx5/mlx5_rxmode.c b/drivers/net/mlx5/mlx5_rxmode.c
new file mode 100644
index 0000000..7efa21b
--- /dev/null
+++ b/drivers/net/mlx5/mlx5_rxmode.c
@@ -0,0 +1,311 @@
+/*-
+ *   BSD LICENSE
+ *
+ *   Copyright 2015 6WIND S.A.
+ *   Copyright 2015 Mellanox.
+ *
+ *   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 6WIND S.A. 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 <stddef.h>
+#include <errno.h>
+#include <string.h>
+
+/* Verbs header. */
+/* ISO C doesn't support unnamed structs/unions, disabling -pedantic. */
+#ifdef PEDANTIC
+#pragma GCC diagnostic ignored "-pedantic"
+#endif
+#include <infiniband/verbs.h>
+#ifdef PEDANTIC
+#pragma GCC diagnostic error "-pedantic"
+#endif
+
+/* DPDK headers don't like -pedantic. */
+#ifdef PEDANTIC
+#pragma GCC diagnostic ignored "-pedantic"
+#endif
+#include <rte_ethdev.h>
+#ifdef PEDANTIC
+#pragma GCC diagnostic error "-pedantic"
+#endif
+
+#include "mlx5.h"
+#include "mlx5_rxtx.h"
+#include "mlx5_utils.h"
+
+/**
+ * Enable promiscuous mode in a RX queue.
+ *
+ * @param rxq
+ *   Pointer to RX queue structure.
+ *
+ * @return
+ *   0 on success, errno value on failure.
+ */
+int
+rxq_promiscuous_enable(struct rxq *rxq)
+{
+	struct ibv_flow *flow;
+	struct ibv_flow_attr attr = {
+		.type = IBV_FLOW_ATTR_ALL_DEFAULT,
+		.num_of_specs = 0,
+		.port = rxq->priv->port,
+		.flags = 0
+	};
+
+	if (rxq->priv->vf)
+		return 0;
+	if (rxq->promisc_flow != NULL)
+		return 0;
+	DEBUG("%p: enabling promiscuous mode", (void *)rxq);
+	errno = 0;
+	flow = ibv_create_flow(rxq->qp, &attr);
+	if (flow == NULL) {
+		/* It's not clear whether errno is always set in this case. */
+		ERROR("%p: flow configuration failed, errno=%d: %s",
+		      (void *)rxq, errno,
+		      (errno ? strerror(errno) : "Unknown error"));
+		if (errno)
+			return errno;
+		return EINVAL;
+	}
+	rxq->promisc_flow = flow;
+	DEBUG("%p: promiscuous mode enabled", (void *)rxq);
+	return 0;
+}
+
+/**
+ * DPDK callback to enable promiscuous mode.
+ *
+ * @param dev
+ *   Pointer to Ethernet device structure.
+ */
+void
+mlx5_promiscuous_enable(struct rte_eth_dev *dev)
+{
+	struct priv *priv = dev->data->dev_private;
+	unsigned int i;
+	int ret;
+
+	priv_lock(priv);
+	priv->promisc_req = 1;
+	/* If device isn't started, this is all we need to do. */
+	if (!priv->started)
+		goto end;
+	if (priv->rss) {
+		ret = rxq_promiscuous_enable(&priv->rxq_parent);
+		if (ret) {
+			priv_unlock(priv);
+			return;
+		}
+		goto end;
+	}
+	for (i = 0; (i != priv->rxqs_n); ++i) {
+		if ((*priv->rxqs)[i] == NULL)
+			continue;
+		ret = rxq_promiscuous_enable((*priv->rxqs)[i]);
+		if (!ret)
+			continue;
+		/* Failure, rollback. */
+		while (i != 0)
+			if ((*priv->rxqs)[--i] != NULL)
+				rxq_promiscuous_disable((*priv->rxqs)[i]);
+		priv_unlock(priv);
+		return;
+	}
+end:
+	priv_unlock(priv);
+}
+
+/**
+ * Disable promiscuous mode in a RX queue.
+ *
+ * @param rxq
+ *   Pointer to RX queue structure.
+ */
+void
+rxq_promiscuous_disable(struct rxq *rxq)
+{
+	if (rxq->priv->vf)
+		return;
+	if (rxq->promisc_flow == NULL)
+		return;
+	DEBUG("%p: disabling promiscuous mode", (void *)rxq);
+	claim_zero(ibv_destroy_flow(rxq->promisc_flow));
+	rxq->promisc_flow = NULL;
+	DEBUG("%p: promiscuous mode disabled", (void *)rxq);
+}
+
+/**
+ * DPDK callback to disable promiscuous mode.
+ *
+ * @param dev
+ *   Pointer to Ethernet device structure.
+ */
+void
+mlx5_promiscuous_disable(struct rte_eth_dev *dev)
+{
+	struct priv *priv = dev->data->dev_private;
+	unsigned int i;
+
+	priv_lock(priv);
+	priv->promisc_req = 0;
+	if (priv->rss) {
+		rxq_promiscuous_disable(&priv->rxq_parent);
+		goto end;
+	}
+	for (i = 0; (i != priv->rxqs_n); ++i)
+		if ((*priv->rxqs)[i] != NULL)
+			rxq_promiscuous_disable((*priv->rxqs)[i]);
+end:
+	priv_unlock(priv);
+}
+
+/**
+ * Enable allmulti mode in a RX queue.
+ *
+ * @param rxq
+ *   Pointer to RX queue structure.
+ *
+ * @return
+ *   0 on success, errno value on failure.
+ */
+int
+rxq_allmulticast_enable(struct rxq *rxq)
+{
+	struct ibv_flow *flow;
+	struct ibv_flow_attr attr = {
+		.type = IBV_FLOW_ATTR_MC_DEFAULT,
+		.num_of_specs = 0,
+		.port = rxq->priv->port,
+		.flags = 0
+	};
+
+	if (rxq->allmulti_flow != NULL)
+		return 0;
+	DEBUG("%p: enabling allmulticast mode", (void *)rxq);
+	errno = 0;
+	flow = ibv_create_flow(rxq->qp, &attr);
+	if (flow == NULL) {
+		/* It's not clear whether errno is always set in this case. */
+		ERROR("%p: flow configuration failed, errno=%d: %s",
+		      (void *)rxq, errno,
+		      (errno ? strerror(errno) : "Unknown error"));
+		if (errno)
+			return errno;
+		return EINVAL;
+	}
+	rxq->allmulti_flow = flow;
+	DEBUG("%p: allmulticast mode enabled", (void *)rxq);
+	return 0;
+}
+
+/**
+ * DPDK callback to enable allmulti mode.
+ *
+ * @param dev
+ *   Pointer to Ethernet device structure.
+ */
+void
+mlx5_allmulticast_enable(struct rte_eth_dev *dev)
+{
+	struct priv *priv = dev->data->dev_private;
+	unsigned int i;
+	int ret;
+
+	priv_lock(priv);
+	priv->allmulti_req = 1;
+	/* If device isn't started, this is all we need to do. */
+	if (!priv->started)
+		goto end;
+	if (priv->rss) {
+		ret = rxq_allmulticast_enable(&priv->rxq_parent);
+		if (ret) {
+			priv_unlock(priv);
+			return;
+		}
+		goto end;
+	}
+	for (i = 0; (i != priv->rxqs_n); ++i) {
+		if ((*priv->rxqs)[i] == NULL)
+			continue;
+		ret = rxq_allmulticast_enable((*priv->rxqs)[i]);
+		if (!ret)
+			continue;
+		/* Failure, rollback. */
+		while (i != 0)
+			if ((*priv->rxqs)[--i] != NULL)
+				rxq_allmulticast_disable((*priv->rxqs)[i]);
+		priv_unlock(priv);
+		return;
+	}
+end:
+	priv_unlock(priv);
+}
+
+/**
+ * Disable allmulti mode in a RX queue.
+ *
+ * @param rxq
+ *   Pointer to RX queue structure.
+ */
+void
+rxq_allmulticast_disable(struct rxq *rxq)
+{
+	if (rxq->allmulti_flow == NULL)
+		return;
+	DEBUG("%p: disabling allmulticast mode", (void *)rxq);
+	claim_zero(ibv_destroy_flow(rxq->allmulti_flow));
+	rxq->allmulti_flow = NULL;
+	DEBUG("%p: allmulticast mode disabled", (void *)rxq);
+}
+
+/**
+ * DPDK callback to disable allmulti mode.
+ *
+ * @param dev
+ *   Pointer to Ethernet device structure.
+ */
+void
+mlx5_allmulticast_disable(struct rte_eth_dev *dev)
+{
+	struct priv *priv = dev->data->dev_private;
+	unsigned int i;
+
+	priv_lock(priv);
+	priv->allmulti_req = 0;
+	if (priv->rss) {
+		rxq_allmulticast_disable(&priv->rxq_parent);
+		goto end;
+	}
+	for (i = 0; (i != priv->rxqs_n); ++i)
+		if ((*priv->rxqs)[i] != NULL)
+			rxq_allmulticast_disable((*priv->rxqs)[i]);
+end:
+	priv_unlock(priv);
+}
diff --git a/drivers/net/mlx5/mlx5_rxq.c b/drivers/net/mlx5/mlx5_rxq.c
index 620ec70..1cd28c2 100644
--- a/drivers/net/mlx5/mlx5_rxq.c
+++ b/drivers/net/mlx5/mlx5_rxq.c
@@ -398,6 +398,8 @@ rxq_cleanup(struct rxq *rxq)
 						&params));
 	}
 	if (rxq->qp != NULL) {
+		rxq_promiscuous_disable(rxq);
+		rxq_allmulticast_disable(rxq);
 		rxq_mac_addrs_del(rxq);
 		claim_zero(ibv_destroy_qp(rxq->qp));
 	}
@@ -580,8 +582,12 @@ rxq_rehash(struct rte_eth_dev *dev, struct rxq *rxq)
 	}
 	/* Remove attached flows if RSS is disabled (no parent queue). */
 	if (!priv->rss) {
+		rxq_allmulticast_disable(&tmpl);
+		rxq_promiscuous_disable(&tmpl);
 		rxq_mac_addrs_del(&tmpl);
 		/* Update original queue in case of failure. */
+		rxq->allmulti_flow = tmpl.allmulti_flow;
+		rxq->promisc_flow = tmpl.promisc_flow;
 		memcpy(rxq->mac_flow, tmpl.mac_flow, sizeof(rxq->mac_flow));
 	}
 	/* From now on, any failure will render the queue unusable.
@@ -621,7 +627,13 @@ rxq_rehash(struct rte_eth_dev *dev, struct rxq *rxq)
 	if (!priv->rss) {
 		if (priv->started)
 			rxq_mac_addrs_add(&tmpl);
+		if (priv->started && priv->promisc_req)
+			rxq_promiscuous_enable(&tmpl);
+		if (priv->started && priv->allmulti_req)
+			rxq_allmulticast_enable(&tmpl);
 		/* Update original queue in case of failure. */
+		rxq->allmulti_flow = tmpl.allmulti_flow;
+		rxq->promisc_flow = tmpl.promisc_flow;
 		memcpy(rxq->mac_flow, tmpl.mac_flow, sizeof(rxq->mac_flow));
 	}
 	/* Allocate pool. */
diff --git a/drivers/net/mlx5/mlx5_rxtx.h b/drivers/net/mlx5/mlx5_rxtx.h
index 4183820..020acf0 100644
--- a/drivers/net/mlx5/mlx5_rxtx.h
+++ b/drivers/net/mlx5/mlx5_rxtx.h
@@ -106,6 +106,8 @@ struct rxq {
 	struct ibv_exp_cq_family *if_cq; /* CQ interface. */
 	/* MAC flow steering rules. */
 	struct ibv_flow *mac_flow[MLX5_MAX_MAC_ADDRESSES];
+	struct ibv_flow *promisc_flow; /* Promiscuous flow. */
+	struct ibv_flow *allmulti_flow; /* Multicast flow. */
 	unsigned int port_id; /* Port ID for incoming packets. */
 	unsigned int elts_n; /* (*elts)[] length. */
 	unsigned int elts_head; /* Current index in (*elts)[]. */
diff --git a/drivers/net/mlx5/mlx5_trigger.c b/drivers/net/mlx5/mlx5_trigger.c
index f5d965f..dced025 100644
--- a/drivers/net/mlx5/mlx5_trigger.c
+++ b/drivers/net/mlx5/mlx5_trigger.c
@@ -86,6 +86,10 @@ mlx5_dev_start(struct rte_eth_dev *dev)
 		if (rxq == NULL)
 			continue;
 		ret = rxq_mac_addrs_add(rxq);
+		if (!ret && priv->promisc_req)
+			ret = rxq_promiscuous_enable(rxq);
+		if (!ret && priv->allmulti_req)
+			ret = rxq_allmulticast_enable(rxq);
 		if (!ret)
 			continue;
 		WARN("%p: QP flow attachment failed: %s",
@@ -94,6 +98,8 @@ mlx5_dev_start(struct rte_eth_dev *dev)
 		while (i != 0) {
 			rxq = (*priv->rxqs)[--i];
 			if (rxq != NULL) {
+				rxq_allmulticast_disable(rxq);
+				rxq_promiscuous_disable(rxq);
 				rxq_mac_addrs_del(rxq);
 			}
 		}
@@ -140,6 +146,8 @@ mlx5_dev_stop(struct rte_eth_dev *dev)
 		/* Ignore nonexistent RX queues. */
 		if (rxq == NULL)
 			continue;
+		rxq_allmulticast_disable(rxq);
+		rxq_promiscuous_disable(rxq);
 		rxq_mac_addrs_del(rxq);
 	} while ((--r) && ((rxq = (*priv->rxqs)[++i]), i));
 	priv_unlock(priv);
-- 
2.1.0

^ permalink raw reply	[flat|nested] 32+ messages in thread

* [dpdk-dev] [PATCH v2 09/13] mlx5: add link update device operation
  2015-10-30 18:52 ` [dpdk-dev] [PATCH v2 00/13] Mellanox ConnectX-4 PMD (mlx5) Adrien Mazarguil
                     ` (7 preceding siblings ...)
  2015-10-30 18:52   ` [dpdk-dev] [PATCH v2 08/13] mlx5: add promiscuous and allmulticast RX modes Adrien Mazarguil
@ 2015-10-30 18:52   ` Adrien Mazarguil
  2015-11-02 17:52     ` Stephen Hemminger
  2015-10-30 18:52   ` [dpdk-dev] [PATCH v2 10/13] mlx5: add flow control device operations Adrien Mazarguil
                     ` (4 subsequent siblings)
  13 siblings, 1 reply; 32+ messages in thread
From: Adrien Mazarguil @ 2015-10-30 18:52 UTC (permalink / raw)
  To: dev

Link information is retrieved using ethtool ioctls.

Signed-off-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>
Signed-off-by: Nelio Laranjeiro <nelio.laranjeiro@6wind.com>
---
 drivers/net/mlx5/mlx5.c        |  1 +
 drivers/net/mlx5/mlx5.h        |  1 +
 drivers/net/mlx5/mlx5_ethdev.c | 71 ++++++++++++++++++++++++++++++++++++++++++
 3 files changed, 73 insertions(+)

diff --git a/drivers/net/mlx5/mlx5.c b/drivers/net/mlx5/mlx5.c
index ee63bdf..5ed828d 100644
--- a/drivers/net/mlx5/mlx5.c
+++ b/drivers/net/mlx5/mlx5.c
@@ -137,6 +137,7 @@ static const struct eth_dev_ops mlx5_dev_ops = {
 	.promiscuous_disable = mlx5_promiscuous_disable,
 	.allmulticast_enable = mlx5_allmulticast_enable,
 	.allmulticast_disable = mlx5_allmulticast_disable,
+	.link_update = mlx5_link_update,
 	.stats_get = mlx5_stats_get,
 	.stats_reset = mlx5_stats_reset,
 	.dev_infos_get = mlx5_dev_infos_get,
diff --git a/drivers/net/mlx5/mlx5.h b/drivers/net/mlx5/mlx5.h
index 56da43c..1a18326 100644
--- a/drivers/net/mlx5/mlx5.h
+++ b/drivers/net/mlx5/mlx5.h
@@ -164,6 +164,7 @@ int priv_get_mtu(struct priv *, uint16_t *);
 int priv_set_flags(struct priv *, unsigned int, unsigned int);
 int mlx5_dev_configure(struct rte_eth_dev *);
 void mlx5_dev_infos_get(struct rte_eth_dev *, struct rte_eth_dev_info *);
+int mlx5_link_update(struct rte_eth_dev *, int);
 int mlx5_dev_set_mtu(struct rte_eth_dev *, uint16_t);
 int mlx5_ibv_device_to_pci_addr(const struct ibv_device *,
 				struct rte_pci_addr *);
diff --git a/drivers/net/mlx5/mlx5_ethdev.c b/drivers/net/mlx5/mlx5_ethdev.c
index 26b6d73..d01dee5 100644
--- a/drivers/net/mlx5/mlx5_ethdev.c
+++ b/drivers/net/mlx5/mlx5_ethdev.c
@@ -45,6 +45,8 @@
 #include <sys/socket.h>
 #include <netinet/in.h>
 #include <linux/if.h>
+#include <linux/ethtool.h>
+#include <linux/sockios.h>
 
 /* DPDK headers don't like -pedantic. */
 #ifdef PEDANTIC
@@ -535,6 +537,75 @@ mlx5_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *info)
 }
 
 /**
+ * DPDK callback to retrieve physical link information (unlocked version).
+ *
+ * @param dev
+ *   Pointer to Ethernet device structure.
+ * @param wait_to_complete
+ *   Wait for request completion (ignored).
+ */
+static int
+mlx5_link_update_unlocked(struct rte_eth_dev *dev, int wait_to_complete)
+{
+	struct priv *priv = dev->data->dev_private;
+	struct ethtool_cmd edata = {
+		.cmd = ETHTOOL_GSET
+	};
+	struct ifreq ifr;
+	struct rte_eth_link dev_link;
+	int link_speed = 0;
+
+	(void)wait_to_complete;
+	if (priv_ifreq(priv, SIOCGIFFLAGS, &ifr)) {
+		WARN("ioctl(SIOCGIFFLAGS) failed: %s", strerror(errno));
+		return -1;
+	}
+	memset(&dev_link, 0, sizeof(dev_link));
+	dev_link.link_status = ((ifr.ifr_flags & IFF_UP) &&
+				(ifr.ifr_flags & IFF_RUNNING));
+	ifr.ifr_data = &edata;
+	if (priv_ifreq(priv, SIOCETHTOOL, &ifr)) {
+		WARN("ioctl(SIOCETHTOOL, ETHTOOL_GSET) failed: %s",
+		     strerror(errno));
+		return -1;
+	}
+	link_speed = ethtool_cmd_speed(&edata);
+	if (link_speed == -1)
+		dev_link.link_speed = 0;
+	else
+		dev_link.link_speed = link_speed;
+	dev_link.link_duplex = ((edata.duplex == DUPLEX_HALF) ?
+				ETH_LINK_HALF_DUPLEX : ETH_LINK_FULL_DUPLEX);
+	if (memcmp(&dev_link, &dev->data->dev_link, sizeof(dev_link))) {
+		/* Link status changed. */
+		dev->data->dev_link = dev_link;
+		return 0;
+	}
+	/* Link status is still the same. */
+	return -1;
+}
+
+/**
+ * DPDK callback to retrieve physical link information.
+ *
+ * @param dev
+ *   Pointer to Ethernet device structure.
+ * @param wait_to_complete
+ *   Wait for request completion (ignored).
+ */
+int
+mlx5_link_update(struct rte_eth_dev *dev, int wait_to_complete)
+{
+	struct priv *priv = dev->data->dev_private;
+	int ret;
+
+	priv_lock(priv);
+	ret = mlx5_link_update_unlocked(dev, wait_to_complete);
+	priv_unlock(priv);
+	return ret;
+}
+
+/**
  * DPDK callback to change the MTU.
  *
  * Setting the MTU affects hardware MRU (packets larger than the MTU cannot be
-- 
2.1.0

^ permalink raw reply	[flat|nested] 32+ messages in thread

* [dpdk-dev] [PATCH v2 10/13] mlx5: add flow control device operations
  2015-10-30 18:52 ` [dpdk-dev] [PATCH v2 00/13] Mellanox ConnectX-4 PMD (mlx5) Adrien Mazarguil
                     ` (8 preceding siblings ...)
  2015-10-30 18:52   ` [dpdk-dev] [PATCH v2 09/13] mlx5: add link update device operation Adrien Mazarguil
@ 2015-10-30 18:52   ` Adrien Mazarguil
  2015-10-30 18:52   ` [dpdk-dev] [PATCH v2 11/13] mlx5: add VLAN filtering Adrien Mazarguil
                     ` (3 subsequent siblings)
  13 siblings, 0 replies; 32+ messages in thread
From: Adrien Mazarguil @ 2015-10-30 18:52 UTC (permalink / raw)
  To: dev

Like most other device control operations, those are handled by the related
kernel network device through syscalls.

Signed-off-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>
Signed-off-by: Nelio Laranjeiro <nelio.laranjeiro@6wind.com>
---
 drivers/net/mlx5/mlx5.c        |  2 +
 drivers/net/mlx5/mlx5.h        |  2 +
 drivers/net/mlx5/mlx5_ethdev.c | 99 ++++++++++++++++++++++++++++++++++++++++++
 3 files changed, 103 insertions(+)

diff --git a/drivers/net/mlx5/mlx5.c b/drivers/net/mlx5/mlx5.c
index 5ed828d..c454e93 100644
--- a/drivers/net/mlx5/mlx5.c
+++ b/drivers/net/mlx5/mlx5.c
@@ -145,6 +145,8 @@ static const struct eth_dev_ops mlx5_dev_ops = {
 	.tx_queue_setup = mlx5_tx_queue_setup,
 	.rx_queue_release = mlx5_rx_queue_release,
 	.tx_queue_release = mlx5_tx_queue_release,
+	.flow_ctrl_get = mlx5_dev_get_flow_ctrl,
+	.flow_ctrl_set = mlx5_dev_set_flow_ctrl,
 	.mac_addr_remove = mlx5_mac_addr_remove,
 	.mac_addr_add = mlx5_mac_addr_add,
 	.mtu_set = mlx5_dev_set_mtu,
diff --git a/drivers/net/mlx5/mlx5.h b/drivers/net/mlx5/mlx5.h
index 1a18326..c6c3d3f 100644
--- a/drivers/net/mlx5/mlx5.h
+++ b/drivers/net/mlx5/mlx5.h
@@ -166,6 +166,8 @@ int mlx5_dev_configure(struct rte_eth_dev *);
 void mlx5_dev_infos_get(struct rte_eth_dev *, struct rte_eth_dev_info *);
 int mlx5_link_update(struct rte_eth_dev *, int);
 int mlx5_dev_set_mtu(struct rte_eth_dev *, uint16_t);
+int mlx5_dev_get_flow_ctrl(struct rte_eth_dev *, struct rte_eth_fc_conf *);
+int mlx5_dev_set_flow_ctrl(struct rte_eth_dev *, struct rte_eth_fc_conf *);
 int mlx5_ibv_device_to_pci_addr(const struct ibv_device *,
 				struct rte_pci_addr *);
 
diff --git a/drivers/net/mlx5/mlx5_ethdev.c b/drivers/net/mlx5/mlx5_ethdev.c
index d01dee5..5df5fa1 100644
--- a/drivers/net/mlx5/mlx5_ethdev.c
+++ b/drivers/net/mlx5/mlx5_ethdev.c
@@ -695,6 +695,105 @@ out:
 }
 
 /**
+ * DPDK callback to get flow control status.
+ *
+ * @param dev
+ *   Pointer to Ethernet device structure.
+ * @param[out] fc_conf
+ *   Flow control output buffer.
+ *
+ * @return
+ *   0 on success, negative errno value on failure.
+ */
+int
+mlx5_dev_get_flow_ctrl(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
+{
+	struct priv *priv = dev->data->dev_private;
+	struct ifreq ifr;
+	struct ethtool_pauseparam ethpause = {
+		.cmd = ETHTOOL_GPAUSEPARAM
+	};
+	int ret;
+
+	ifr.ifr_data = &ethpause;
+	priv_lock(priv);
+	if (priv_ifreq(priv, SIOCETHTOOL, &ifr)) {
+		ret = errno;
+		WARN("ioctl(SIOCETHTOOL, ETHTOOL_GPAUSEPARAM)"
+		     " failed: %s",
+		     strerror(ret));
+		goto out;
+	}
+
+	fc_conf->autoneg = ethpause.autoneg;
+	if (ethpause.rx_pause && ethpause.tx_pause)
+		fc_conf->mode = RTE_FC_FULL;
+	else if (ethpause.rx_pause)
+		fc_conf->mode = RTE_FC_RX_PAUSE;
+	else if (ethpause.tx_pause)
+		fc_conf->mode = RTE_FC_TX_PAUSE;
+	else
+		fc_conf->mode = RTE_FC_NONE;
+	ret = 0;
+
+out:
+	priv_unlock(priv);
+	assert(ret >= 0);
+	return -ret;
+}
+
+/**
+ * DPDK callback to modify flow control parameters.
+ *
+ * @param dev
+ *   Pointer to Ethernet device structure.
+ * @param[in] fc_conf
+ *   Flow control parameters.
+ *
+ * @return
+ *   0 on success, negative errno value on failure.
+ */
+int
+mlx5_dev_set_flow_ctrl(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
+{
+	struct priv *priv = dev->data->dev_private;
+	struct ifreq ifr;
+	struct ethtool_pauseparam ethpause = {
+		.cmd = ETHTOOL_SPAUSEPARAM
+	};
+	int ret;
+
+	ifr.ifr_data = &ethpause;
+	ethpause.autoneg = fc_conf->autoneg;
+	if (((fc_conf->mode & RTE_FC_FULL) == RTE_FC_FULL) ||
+	    (fc_conf->mode & RTE_FC_RX_PAUSE))
+		ethpause.rx_pause = 1;
+	else
+		ethpause.rx_pause = 0;
+
+	if (((fc_conf->mode & RTE_FC_FULL) == RTE_FC_FULL) ||
+	    (fc_conf->mode & RTE_FC_TX_PAUSE))
+		ethpause.tx_pause = 1;
+	else
+		ethpause.tx_pause = 0;
+
+	priv_lock(priv);
+	if (priv_ifreq(priv, SIOCETHTOOL, &ifr)) {
+		ret = errno;
+		WARN("ioctl(SIOCETHTOOL, ETHTOOL_SPAUSEPARAM)"
+		     " failed: %s",
+		     strerror(ret));
+		goto out;
+	}
+	ret = 0;
+
+out:
+	priv_unlock(priv);
+	assert(ret >= 0);
+	return -ret;
+}
+
+/**
  * Get PCI information from struct ibv_device.
  *
  * @param device
-- 
2.1.0

^ permalink raw reply	[flat|nested] 32+ messages in thread

* [dpdk-dev] [PATCH v2 11/13] mlx5: add VLAN filtering
  2015-10-30 18:52 ` [dpdk-dev] [PATCH v2 00/13] Mellanox ConnectX-4 PMD (mlx5) Adrien Mazarguil
                     ` (9 preceding siblings ...)
  2015-10-30 18:52   ` [dpdk-dev] [PATCH v2 10/13] mlx5: add flow control device operations Adrien Mazarguil
@ 2015-10-30 18:52   ` Adrien Mazarguil
  2015-10-30 18:52   ` [dpdk-dev] [PATCH v2 12/13] mlx5: add checksum offloading support Adrien Mazarguil
                     ` (2 subsequent siblings)
  13 siblings, 0 replies; 32+ messages in thread
From: Adrien Mazarguil @ 2015-10-30 18:52 UTC (permalink / raw)
  To: dev

All MAC RX flows must be updated with VLAN information when configuring a
VLAN filter.

Signed-off-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>
Signed-off-by: Nelio Laranjeiro <nelio.laranjeiro@6wind.com>
---
 drivers/net/mlx5/Makefile    |   1 +
 drivers/net/mlx5/mlx5.c      |   1 +
 drivers/net/mlx5/mlx5.h      |   6 ++
 drivers/net/mlx5/mlx5_defs.h |   3 +
 drivers/net/mlx5/mlx5_mac.c  |  62 ++++++++++++-----
 drivers/net/mlx5/mlx5_rxtx.h |   4 +-
 drivers/net/mlx5/mlx5_vlan.c | 156 +++++++++++++++++++++++++++++++++++++++++++
 7 files changed, 216 insertions(+), 17 deletions(-)
 create mode 100644 drivers/net/mlx5/mlx5_vlan.c

diff --git a/drivers/net/mlx5/Makefile b/drivers/net/mlx5/Makefile
index 4d25d9c..8b1e32b 100644
--- a/drivers/net/mlx5/Makefile
+++ b/drivers/net/mlx5/Makefile
@@ -49,6 +49,7 @@ SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5_trigger.c
 SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5_ethdev.c
 SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5_mac.c
 SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5_rxmode.c
+SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5_vlan.c
 SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5_stats.c
 
 # Dependencies.
diff --git a/drivers/net/mlx5/mlx5.c b/drivers/net/mlx5/mlx5.c
index c454e93..8f75f76 100644
--- a/drivers/net/mlx5/mlx5.c
+++ b/drivers/net/mlx5/mlx5.c
@@ -141,6 +141,7 @@ static const struct eth_dev_ops mlx5_dev_ops = {
 	.stats_get = mlx5_stats_get,
 	.stats_reset = mlx5_stats_reset,
 	.dev_infos_get = mlx5_dev_infos_get,
+	.vlan_filter_set = mlx5_vlan_filter_set,
 	.rx_queue_setup = mlx5_rx_queue_setup,
 	.tx_queue_setup = mlx5_tx_queue_setup,
 	.rx_queue_release = mlx5_rx_queue_release,
diff --git a/drivers/net/mlx5/mlx5.h b/drivers/net/mlx5/mlx5.h
index c6c3d3f..3a1e7a6 100644
--- a/drivers/net/mlx5/mlx5.h
+++ b/drivers/net/mlx5/mlx5.h
@@ -90,6 +90,8 @@ struct priv {
 	 */
 	struct ether_addr mac[MLX5_MAX_MAC_ADDRESSES];
 	BITFIELD_DECLARE(mac_configured, uint32_t, MLX5_MAX_MAC_ADDRESSES);
+	uint16_t vlan_filter[MLX5_MAX_VLAN_IDS]; /* VLAN filters table. */
+	unsigned int vlan_filter_n; /* Number of configured VLAN filters. */
 	/* Device properties. */
 	uint16_t mtu; /* Configured MTU. */
 	uint8_t port; /* Physical port number. */
@@ -198,6 +200,10 @@ void mlx5_allmulticast_disable(struct rte_eth_dev *);
 void mlx5_stats_get(struct rte_eth_dev *, struct rte_eth_stats *);
 void mlx5_stats_reset(struct rte_eth_dev *);
 
+/* mlx5_vlan.c */
+
+int mlx5_vlan_filter_set(struct rte_eth_dev *, uint16_t, int);
+
 /* mlx5_trigger.c */
 
 int mlx5_dev_start(struct rte_eth_dev *);
diff --git a/drivers/net/mlx5/mlx5_defs.h b/drivers/net/mlx5/mlx5_defs.h
index d3a0d0e..369f8b6 100644
--- a/drivers/net/mlx5/mlx5_defs.h
+++ b/drivers/net/mlx5/mlx5_defs.h
@@ -40,6 +40,9 @@
 /* Maximum number of simultaneous MAC addresses. */
 #define MLX5_MAX_MAC_ADDRESSES 128
 
+/* Maximum number of simultaneous VLAN filters. */
+#define MLX5_MAX_VLAN_IDS 128
+
 /* Request send completion once in every 64 sends, might be less. */
 #define MLX5_PMD_TX_PER_COMP_REQ 64
 
diff --git a/drivers/net/mlx5/mlx5_mac.c b/drivers/net/mlx5/mlx5_mac.c
index 262d7c6..95afccf 100644
--- a/drivers/net/mlx5/mlx5_mac.c
+++ b/drivers/net/mlx5/mlx5_mac.c
@@ -97,9 +97,12 @@ priv_get_mac(struct priv *priv, uint8_t (*mac)[ETHER_ADDR_LEN])
  *   Pointer to RX queue structure.
  * @param mac_index
  *   MAC address index.
+ * @param vlan_index
+ *   VLAN index to use.
  */
 static void
-rxq_del_mac_flow(struct rxq *rxq, unsigned int mac_index)
+rxq_del_mac_flow(struct rxq *rxq, unsigned int mac_index,
+		 unsigned int vlan_index)
 {
 #ifndef NDEBUG
 	const uint8_t (*mac)[ETHER_ADDR_LEN] =
@@ -108,14 +111,17 @@ rxq_del_mac_flow(struct rxq *rxq, unsigned int mac_index)
 #endif
 
 	assert(mac_index < RTE_DIM(rxq->mac_flow));
-	if (rxq->mac_flow[mac_index] == NULL)
+	assert(vlan_index < RTE_DIM(rxq->mac_flow[mac_index]));
+	if (rxq->mac_flow[mac_index][vlan_index] == NULL)
 		return;
-	DEBUG("%p: removing MAC address %02x:%02x:%02x:%02x:%02x:%02x index %u",
+	DEBUG("%p: removing MAC address %02x:%02x:%02x:%02x:%02x:%02x index %u"
+	      " VLAN index %u",
 	      (void *)rxq,
 	      (*mac)[0], (*mac)[1], (*mac)[2], (*mac)[3], (*mac)[4], (*mac)[5],
-	      mac_index);
-	claim_zero(ibv_destroy_flow(rxq->mac_flow[mac_index]));
-	rxq->mac_flow[mac_index] = NULL;
+	      mac_index,
+	      vlan_index);
+	claim_zero(ibv_destroy_flow(rxq->mac_flow[mac_index][vlan_index]));
+	rxq->mac_flow[mac_index][vlan_index] = NULL;
 }
 
 /**
@@ -129,8 +135,11 @@ rxq_del_mac_flow(struct rxq *rxq, unsigned int mac_index)
 static void
 rxq_mac_addr_del(struct rxq *rxq, unsigned int mac_index)
 {
+	unsigned int i;
+
 	assert(mac_index < RTE_DIM(rxq->mac_flow));
-	rxq_del_mac_flow(rxq, mac_index);
+	for (i = 0; (i != RTE_DIM(rxq->mac_flow[mac_index])); ++i)
+		rxq_del_mac_flow(rxq, mac_index, i);
 }
 
 /**
@@ -208,12 +217,15 @@ end:
  *   Pointer to RX queue structure.
  * @param mac_index
  *   MAC address index to register.
+ * @param vlan_index
+ *   VLAN index to use.
  *
  * @return
  *   0 on success, errno value on failure.
  */
 static int
-rxq_add_mac_flow(struct rxq *rxq, unsigned int mac_index)
+rxq_add_mac_flow(struct rxq *rxq, unsigned int mac_index,
+		 unsigned int vlan_index)
 {
 	struct ibv_flow *flow;
 	struct priv *priv = rxq->priv;
@@ -226,9 +238,12 @@ rxq_add_mac_flow(struct rxq *rxq, unsigned int mac_index)
 	} data;
 	struct ibv_flow_attr *attr = &data.attr;
 	struct ibv_flow_spec_eth *spec = &data.spec;
+	unsigned int vlan_enabled = !!priv->vlan_filter_n;
+	unsigned int vlan_id = priv->vlan_filter[vlan_index];
 
 	assert(mac_index < RTE_DIM(rxq->mac_flow));
-	if (rxq->mac_flow[mac_index] != NULL)
+	assert(vlan_index < RTE_DIM(rxq->mac_flow[mac_index]));
+	if (rxq->mac_flow[mac_index][vlan_index] != NULL)
 		return 0;
 	/*
 	 * No padding must be inserted by the compiler between attr and spec.
@@ -249,15 +264,21 @@ rxq_add_mac_flow(struct rxq *rxq, unsigned int mac_index)
 				(*mac)[0], (*mac)[1], (*mac)[2],
 				(*mac)[3], (*mac)[4], (*mac)[5]
 			},
+			.vlan_tag = (vlan_enabled ? htons(vlan_id) : 0),
 		},
 		.mask = {
 			.dst_mac = "\xff\xff\xff\xff\xff\xff",
+			.vlan_tag = (vlan_enabled ? htons(0xfff) : 0),
 		},
 	};
-	DEBUG("%p: adding MAC address %02x:%02x:%02x:%02x:%02x:%02x index %u",
+	DEBUG("%p: adding MAC address %02x:%02x:%02x:%02x:%02x:%02x index %u"
+	      " VLAN index %u filtering %s, ID %u",
 	      (void *)rxq,
 	      (*mac)[0], (*mac)[1], (*mac)[2], (*mac)[3], (*mac)[4], (*mac)[5],
-	      mac_index);
+	      mac_index,
+	      vlan_index,
+	      (vlan_enabled ? "enabled" : "disabled"),
+	      vlan_id);
 	/* Create related flow. */
 	errno = 0;
 	flow = ibv_create_flow(rxq->qp, attr);
@@ -270,7 +291,7 @@ rxq_add_mac_flow(struct rxq *rxq, unsigned int mac_index)
 			return errno;
 		return EINVAL;
 	}
-	rxq->mac_flow[mac_index] = flow;
+	rxq->mac_flow[mac_index][vlan_index] = flow;
 	return 0;
 }
 
@@ -288,12 +309,23 @@ rxq_add_mac_flow(struct rxq *rxq, unsigned int mac_index)
 static int
 rxq_mac_addr_add(struct rxq *rxq, unsigned int mac_index)
 {
+	struct priv *priv = rxq->priv;
+	unsigned int i = 0;
 	int ret;
 
 	assert(mac_index < RTE_DIM(rxq->mac_flow));
-	ret = rxq_add_mac_flow(rxq, mac_index);
-	if (ret)
-		return ret;
+	assert(RTE_DIM(rxq->mac_flow[mac_index]) ==
+	       RTE_DIM(priv->vlan_filter));
+	/* Add a MAC address for each VLAN filter, or at least once. */
+	do {
+		ret = rxq_add_mac_flow(rxq, mac_index, i);
+		if (ret) {
+			/* Failure, rollback. */
+			while (i != 0)
+				rxq_del_mac_flow(rxq, mac_index, --i);
+			return ret;
+		}
+	} while (++i < priv->vlan_filter_n);
 	return 0;
 }
 
diff --git a/drivers/net/mlx5/mlx5_rxtx.h b/drivers/net/mlx5/mlx5_rxtx.h
index 020acf0..521aee0 100644
--- a/drivers/net/mlx5/mlx5_rxtx.h
+++ b/drivers/net/mlx5/mlx5_rxtx.h
@@ -104,8 +104,8 @@ struct rxq {
 	struct ibv_qp *qp; /* Queue Pair. */
 	struct ibv_exp_qp_burst_family *if_qp; /* QP burst interface. */
 	struct ibv_exp_cq_family *if_cq; /* CQ interface. */
-	/* MAC flow steering rules. */
-	struct ibv_flow *mac_flow[MLX5_MAX_MAC_ADDRESSES];
+	/* MAC flow steering rules, one per VLAN ID. */
+	struct ibv_flow *mac_flow[MLX5_MAX_MAC_ADDRESSES][MLX5_MAX_VLAN_IDS];
 	struct ibv_flow *promisc_flow; /* Promiscuous flow. */
 	struct ibv_flow *allmulti_flow; /* Multicast flow. */
 	unsigned int port_id; /* Port ID for incoming packets. */
diff --git a/drivers/net/mlx5/mlx5_vlan.c b/drivers/net/mlx5/mlx5_vlan.c
new file mode 100644
index 0000000..ca80571
--- /dev/null
+++ b/drivers/net/mlx5/mlx5_vlan.c
@@ -0,0 +1,156 @@
+/*-
+ *   BSD LICENSE
+ *
+ *   Copyright 2015 6WIND S.A.
+ *   Copyright 2015 Mellanox.
+ *
+ *   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 6WIND S.A. 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 <stddef.h>
+#include <errno.h>
+#include <assert.h>
+#include <stdint.h>
+
+/* DPDK headers don't like -pedantic. */
+#ifdef PEDANTIC
+#pragma GCC diagnostic ignored "-pedantic"
+#endif
+#include <rte_ethdev.h>
+#include <rte_common.h>
+#ifdef PEDANTIC
+#pragma GCC diagnostic error "-pedantic"
+#endif
+
+#include "mlx5_utils.h"
+#include "mlx5.h"
+
+/**
+ * Configure a VLAN filter.
+ *
+ * @param dev
+ *   Pointer to Ethernet device structure.
+ * @param vlan_id
+ *   VLAN ID to filter.
+ * @param on
+ *   Toggle filter.
+ *
+ * @return
+ *   0 on success, errno value on failure.
+ */
+static int
+vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
+{
+	struct priv *priv = dev->data->dev_private;
+	unsigned int i;
+	unsigned int r;
+	struct rxq *rxq;
+
+	DEBUG("%p: %s VLAN filter ID %" PRIu16,
+	      (void *)dev, (on ? "enable" : "disable"), vlan_id);
+	assert(priv->vlan_filter_n <= RTE_DIM(priv->vlan_filter));
+	for (i = 0; (i != priv->vlan_filter_n); ++i)
+		if (priv->vlan_filter[i] == vlan_id)
+			break;
+	/* Check if there's room for another VLAN filter. */
+	if (i == RTE_DIM(priv->vlan_filter))
+		return ENOMEM;
+	if (i < priv->vlan_filter_n) {
+		assert(priv->vlan_filter_n != 0);
+		/* Enabling an existing VLAN filter has no effect. */
+		if (on)
+			return 0;
+		/* Remove VLAN filter from list. */
+		--priv->vlan_filter_n;
+		memmove(&priv->vlan_filter[i],
+			&priv->vlan_filter[i + 1],
+			priv->vlan_filter_n - i);
+		priv->vlan_filter[priv->vlan_filter_n] = 0;
+	} else {
+		assert(i == priv->vlan_filter_n);
+		/* Disabling an unknown VLAN filter has no effect. */
+		if (!on)
+			return 0;
+		/* Add new VLAN filter. */
+		priv->vlan_filter[priv->vlan_filter_n] = vlan_id;
+		++priv->vlan_filter_n;
+	}
+	if (!priv->started)
+		return 0;
+	/* Rehash MAC flows in all RX queues. */
+	if (priv->rss) {
+		rxq = &priv->rxq_parent;
+		r = 1;
+	} else {
+		rxq = (*priv->rxqs)[0];
+		r = priv->rxqs_n;
+	}
+	for (i = 0; (i < r); rxq = (*priv->rxqs)[++i]) {
+		int ret;
+
+		if (rxq == NULL)
+			continue;
+		rxq_mac_addrs_del(rxq);
+		ret = rxq_mac_addrs_add(rxq);
+		if (!ret)
+			continue;
+		/* Rollback. */
+		while (i != 0) {
+			rxq = (*priv->rxqs)[--i];
+			if (rxq != NULL)
+				rxq_mac_addrs_del(rxq);
+		}
+		return ret;
+	}
+	return 0;
+}
+
+/**
+ * DPDK callback to configure a VLAN filter.
+ *
+ * @param dev
+ *   Pointer to Ethernet device structure.
+ * @param vlan_id
+ *   VLAN ID to filter.
+ * @param on
+ *   Toggle filter.
+ *
+ * @return
+ *   0 on success, negative errno value on failure.
+ */
+int
+mlx5_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
+{
+	struct priv *priv = dev->data->dev_private;
+	int ret;
+
+	priv_lock(priv);
+	ret = vlan_filter_set(dev, vlan_id, on);
+	priv_unlock(priv);
+	assert(ret >= 0);
+	return -ret;
+}
-- 
2.1.0

^ permalink raw reply	[flat|nested] 32+ messages in thread

* [dpdk-dev] [PATCH v2 12/13] mlx5: add checksum offloading support
  2015-10-30 18:52 ` [dpdk-dev] [PATCH v2 00/13] Mellanox ConnectX-4 PMD (mlx5) Adrien Mazarguil
                     ` (10 preceding siblings ...)
  2015-10-30 18:52   ` [dpdk-dev] [PATCH v2 11/13] mlx5: add VLAN filtering Adrien Mazarguil
@ 2015-10-30 18:52   ` Adrien Mazarguil
  2015-10-30 18:52   ` [dpdk-dev] [PATCH v2 13/13] doc: add mlx5 documentation and release notes for version 2.2 Adrien Mazarguil
  2015-10-30 23:18   ` [dpdk-dev] [PATCH v2 00/13] Mellanox ConnectX-4 PMD (mlx5) Thomas Monjalon
  13 siblings, 0 replies; 32+ messages in thread
From: Adrien Mazarguil @ 2015-10-30 18:52 UTC (permalink / raw)
  To: dev

This is the same implementation as mlx4.

Signed-off-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>
Signed-off-by: Nelio Laranjeiro <nelio.laranjeiro@6wind.com>
---
 drivers/net/mlx5/mlx5_rxq.c   | 14 +++++++
 drivers/net/mlx5/mlx5_rxtx.c  | 94 +++++++++++++++++++++++++++++++++++++++++++
 drivers/net/mlx5/mlx5_rxtx.h  |  2 +
 drivers/net/mlx5/mlx5_utils.h |  6 +++
 4 files changed, 116 insertions(+)

diff --git a/drivers/net/mlx5/mlx5_rxq.c b/drivers/net/mlx5/mlx5_rxq.c
index 1cd28c2..5a55886 100644
--- a/drivers/net/mlx5/mlx5_rxq.c
+++ b/drivers/net/mlx5/mlx5_rxq.c
@@ -565,6 +565,15 @@ rxq_rehash(struct rte_eth_dev *dev, struct rxq *rxq)
 	/* Number of descriptors and mbufs currently allocated. */
 	desc_n = (tmpl.elts_n * (tmpl.sp ? MLX5_PMD_SGE_WR_N : 1));
 	mbuf_n = desc_n;
+	/* Toggle RX checksum offload if hardware supports it. */
+	if (priv->hw_csum) {
+		tmpl.csum = !!dev->data->dev_conf.rxmode.hw_ip_checksum;
+		rxq->csum = tmpl.csum;
+	}
+	if (priv->hw_csum_l2tun) {
+		tmpl.csum_l2tun = !!dev->data->dev_conf.rxmode.hw_ip_checksum;
+		rxq->csum_l2tun = tmpl.csum_l2tun;
+	}
 	/* Enable scattered packets support for this queue if necessary. */
 	if ((dev->data->dev_conf.rxmode.jumbo_frame) &&
 	    (dev->data->dev_conf.rxmode.max_rx_pkt_len >
@@ -785,6 +794,11 @@ rxq_setup(struct rte_eth_dev *dev, struct rxq *rxq, uint16_t desc,
 		rte_pktmbuf_tailroom(buf)) == tmpl.mb_len);
 	assert(rte_pktmbuf_headroom(buf) == RTE_PKTMBUF_HEADROOM);
 	rte_pktmbuf_free(buf);
+	/* Toggle RX checksum offload if hardware supports it. */
+	if (priv->hw_csum)
+		tmpl.csum = !!dev->data->dev_conf.rxmode.hw_ip_checksum;
+	if (priv->hw_csum_l2tun)
+		tmpl.csum_l2tun = !!dev->data->dev_conf.rxmode.hw_ip_checksum;
 	/* Enable scattered packets support for this queue if necessary. */
 	if ((dev->data->dev_conf.rxmode.jumbo_frame) &&
 	    (dev->data->dev_conf.rxmode.max_rx_pkt_len >
diff --git a/drivers/net/mlx5/mlx5_rxtx.c b/drivers/net/mlx5/mlx5_rxtx.c
index eccbbb9..623219d 100644
--- a/drivers/net/mlx5/mlx5_rxtx.c
+++ b/drivers/net/mlx5/mlx5_rxtx.c
@@ -390,6 +390,17 @@ mlx5_tx_burst(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n)
 			++elts_comp;
 			send_flags |= IBV_EXP_QP_BURST_SIGNALED;
 		}
+		/* Should we enable HW CKSUM offload */
+		if (buf->ol_flags &
+		    (PKT_TX_IP_CKSUM | PKT_TX_TCP_CKSUM | PKT_TX_UDP_CKSUM)) {
+			send_flags |= IBV_EXP_QP_BURST_IP_CSUM;
+			/* HW does not support checksum offloads at arbitrary
+			 * offsets but automatically recognizes the packet
+			 * type. For inner L3/L4 checksums, only VXLAN (UDP)
+			 * tunnels are currently supported. */
+			if (RTE_ETH_IS_TUNNEL_PKT(buf->packet_type))
+				send_flags |= IBV_EXP_QP_BURST_TUNNEL;
+		}
 		if (likely(segs == 1)) {
 			uintptr_t addr;
 			uint32_t length;
@@ -491,6 +502,85 @@ stop:
 }
 
 /**
+ * Translate RX completion flags to packet type.
+ *
+ * @param flags
+ *   RX completion flags returned by poll_length_flags().
+ *
+ * @return
+ *   Packet type for struct rte_mbuf.
+ */
+static inline uint32_t
+rxq_cq_to_pkt_type(uint32_t flags)
+{
+	uint32_t pkt_type;
+
+	if (flags & IBV_EXP_CQ_RX_TUNNEL_PACKET)
+		pkt_type =
+			TRANSPOSE(flags,
+				  IBV_EXP_CQ_RX_OUTER_IPV4_PACKET,
+				  RTE_PTYPE_L3_IPV4) |
+			TRANSPOSE(flags,
+				  IBV_EXP_CQ_RX_OUTER_IPV6_PACKET,
+				  RTE_PTYPE_L3_IPV6) |
+			TRANSPOSE(flags,
+				  IBV_EXP_CQ_RX_IPV4_PACKET,
+				  RTE_PTYPE_INNER_L3_IPV4) |
+			TRANSPOSE(flags,
+				  IBV_EXP_CQ_RX_IPV6_PACKET,
+				  RTE_PTYPE_INNER_L3_IPV6);
+	else
+		pkt_type =
+			TRANSPOSE(flags,
+				  IBV_EXP_CQ_RX_IPV4_PACKET,
+				  RTE_PTYPE_L3_IPV4) |
+			TRANSPOSE(flags,
+				  IBV_EXP_CQ_RX_IPV6_PACKET,
+				  RTE_PTYPE_L3_IPV6);
+	return pkt_type;
+}
+
+/**
+ * Translate RX completion flags to offload flags.
+ *
+ * @param[in] rxq
+ *   Pointer to RX queue structure.
+ * @param flags
+ *   RX completion flags returned by poll_length_flags().
+ *
+ * @return
+ *   Offload flags (ol_flags) for struct rte_mbuf.
+ */
+static inline uint32_t
+rxq_cq_to_ol_flags(const struct rxq *rxq, uint32_t flags)
+{
+	uint32_t ol_flags = 0;
+
+	if (rxq->csum)
+		ol_flags |=
+			TRANSPOSE(~flags,
+				  IBV_EXP_CQ_RX_IP_CSUM_OK,
+				  PKT_RX_IP_CKSUM_BAD) |
+			TRANSPOSE(~flags,
+				  IBV_EXP_CQ_RX_TCP_UDP_CSUM_OK,
+				  PKT_RX_L4_CKSUM_BAD);
+	/*
+	 * PKT_RX_IP_CKSUM_BAD and PKT_RX_L4_CKSUM_BAD are used in place
+	 * of PKT_RX_EIP_CKSUM_BAD because the latter is not functional
+	 * (its value is 0).
+	 */
+	if ((flags & IBV_EXP_CQ_RX_TUNNEL_PACKET) && (rxq->csum_l2tun))
+		ol_flags |=
+			TRANSPOSE(~flags,
+				  IBV_EXP_CQ_RX_OUTER_IP_CSUM_OK,
+				  PKT_RX_IP_CKSUM_BAD) |
+			TRANSPOSE(~flags,
+				  IBV_EXP_CQ_RX_OUTER_TCP_UDP_CSUM_OK,
+				  PKT_RX_L4_CKSUM_BAD);
+	return ol_flags;
+}
+
+/**
  * DPDK callback for RX with scattered packets support.
  *
  * @param dpdk_rxq
@@ -669,6 +759,8 @@ mlx5_rx_burst_sp(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n)
 		NB_SEGS(pkt_buf) = j;
 		PORT(pkt_buf) = rxq->port_id;
 		PKT_LEN(pkt_buf) = pkt_buf_len;
+		pkt_buf->packet_type = rxq_cq_to_pkt_type(flags);
+		pkt_buf->ol_flags = rxq_cq_to_ol_flags(rxq, flags);
 
 		/* Return packet. */
 		*(pkts++) = pkt_buf;
@@ -829,6 +921,8 @@ mlx5_rx_burst(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n)
 		NEXT(seg) = NULL;
 		PKT_LEN(seg) = len;
 		DATA_LEN(seg) = len;
+		seg->packet_type = rxq_cq_to_pkt_type(flags);
+		seg->ol_flags = rxq_cq_to_ol_flags(rxq, flags);
 
 		/* Return packet. */
 		*(pkts++) = seg;
diff --git a/drivers/net/mlx5/mlx5_rxtx.h b/drivers/net/mlx5/mlx5_rxtx.h
index 521aee0..d86d623 100644
--- a/drivers/net/mlx5/mlx5_rxtx.h
+++ b/drivers/net/mlx5/mlx5_rxtx.h
@@ -116,6 +116,8 @@ struct rxq {
 		struct rxq_elt (*no_sp)[]; /* RX elements. */
 	} elts;
 	unsigned int sp:1; /* Use scattered RX elements. */
+	unsigned int csum:1; /* Enable checksum offloading. */
+	unsigned int csum_l2tun:1; /* Same for L2 tunnels. */
 	uint32_t mb_len; /* Length of a mp-issued mbuf. */
 	struct mlx5_rxq_stats stats; /* RX queue counters. */
 	unsigned int socket; /* CPU socket ID for allocations. */
diff --git a/drivers/net/mlx5/mlx5_utils.h b/drivers/net/mlx5/mlx5_utils.h
index e48e6b6..8ff075b 100644
--- a/drivers/net/mlx5/mlx5_utils.h
+++ b/drivers/net/mlx5/mlx5_utils.h
@@ -149,6 +149,12 @@ pmd_drv_log_basename(const char *s)
 #define NB_SEGS(m) ((m)->nb_segs)
 #define PORT(m) ((m)->port)
 
+/* Transpose flags. Useful to convert IBV to DPDK flags. */
+#define TRANSPOSE(val, from, to) \
+	(((from) >= (to)) ? \
+	 (((val) & (from)) / ((from) / (to))) : \
+	 (((val) & (from)) * ((to) / (from))))
+
 /* Allocate a buffer on the stack and fill it with a printf format string. */
 #define MKSTR(name, ...) \
 	char name[snprintf(NULL, 0, __VA_ARGS__) + 1]; \
-- 
2.1.0

^ permalink raw reply	[flat|nested] 32+ messages in thread

* [dpdk-dev] [PATCH v2 13/13] doc: add mlx5 documentation and release notes for version 2.2
  2015-10-30 18:52 ` [dpdk-dev] [PATCH v2 00/13] Mellanox ConnectX-4 PMD (mlx5) Adrien Mazarguil
                     ` (11 preceding siblings ...)
  2015-10-30 18:52   ` [dpdk-dev] [PATCH v2 12/13] mlx5: add checksum offloading support Adrien Mazarguil
@ 2015-10-30 18:52   ` Adrien Mazarguil
  2015-10-30 23:18   ` [dpdk-dev] [PATCH v2 00/13] Mellanox ConnectX-4 PMD (mlx5) Thomas Monjalon
  13 siblings, 0 replies; 32+ messages in thread
From: Adrien Mazarguil @ 2015-10-30 18:52 UTC (permalink / raw)
  To: dev

Signed-off-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>
---
 doc/guides/nics/mlx5.rst             | 308 +++++++++++++++++++++++++++++++++++
 doc/guides/rel_notes/release_2_2.rst |   9 +
 2 files changed, 317 insertions(+)
 create mode 100644 doc/guides/nics/mlx5.rst

diff --git a/doc/guides/nics/mlx5.rst b/doc/guides/nics/mlx5.rst
new file mode 100644
index 0000000..fdb621c
--- /dev/null
+++ b/doc/guides/nics/mlx5.rst
@@ -0,0 +1,308 @@
+..  BSD LICENSE
+    Copyright 2015 6WIND S.A.
+
+    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 6WIND S.A. 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.
+
+MLX5 poll mode driver
+=====================
+
+The MLX5 poll mode driver library (**librte_pmd_mlx5**) provides support for
+**Mellanox ConnectX-4 EN** and **Mellanox ConnectX-4 Lx EN** families of
+10/25/40/50/100 Gb/s adapters as well as their virtual functions (VF) in
+SR-IOV context.
+
+Information and documentation about these adapters can be found on the
+`Mellanox website <http://www.mellanox.com>`__. Help is also provided by the
+`Mellanox community <http://community.mellanox.com/welcome>`__.
+
+There is also a `section dedicated to this poll mode driver
+<http://www.mellanox.com/page/products_dyn?product_family=209&mtag=pmd_for_dpdk>`__.
+
+.. note::
+
+   Due to external dependencies, this driver is disabled by default. It must
+   be enabled manually by setting ``CONFIG_RTE_LIBRTE_MLX5_PMD=y`` and
+   recompiling DPDK.
+
+.. warning::
+
+   ``CONFIG_RTE_BUILD_COMBINE_LIBS`` with ``CONFIG_RTE_BUILD_SHARED_LIB``
+   is not supported and thus the compilation will fail with this configuration.
+
+Implementation details
+----------------------
+
+Besides its dependency on libibverbs (that implies libmlx5 and associated
+kernel support), librte_pmd_mlx5 relies heavily on system calls for control
+operations such as querying/updating the MTU and flow control parameters.
+
+For security reasons and robustness, this driver only deals with virtual
+memory addresses. The way resources allocations are handled by the kernel
+combined with hardware specifications that allow it to handle virtual memory
+addresses directly ensure that DPDK applications cannot access random
+physical memory (or memory that does not belong to the current process).
+
+This capability allows the PMD to coexist with kernel network interfaces
+which remain functional, although they stop receiving unicast packets as
+long as they share the same MAC address.
+
+Enabling librte_pmd_mlx5 causes DPDK applications to be linked against
+libibverbs.
+
+Configuration
+-------------
+
+Compilation options
+~~~~~~~~~~~~~~~~~~~
+
+These options can be modified in the ``.config`` file.
+
+- ``CONFIG_RTE_LIBRTE_MLX5_PMD`` (default **n**)
+
+  Toggle compilation of librte_pmd_mlx5 itself.
+
+- ``CONFIG_RTE_LIBRTE_MLX5_DEBUG`` (default **n**)
+
+  Toggle debugging code and stricter compilation flags. Enabling this option
+  adds additional run-time checks and debugging messages at the cost of
+  lower performance.
+
+- ``CONFIG_RTE_LIBRTE_MLX5_SGE_WR_N`` (default **4**)
+
+  Number of scatter/gather elements (SGEs) per work request (WR). Lowering
+  this number improves performance but also limits the ability to receive
+  scattered packets (packets that do not fit a single mbuf). The default
+  value is a safe tradeoff.
+
+- ``CONFIG_RTE_LIBRTE_MLX5_MAX_INLINE`` (default **0**)
+
+  Amount of data to be inlined during TX operations. Improves latency but
+  lowers throughput.
+
+- ``CONFIG_RTE_LIBRTE_MLX5_TX_MP_CACHE`` (default **8**)
+
+  Maximum number of cached memory pools (MPs) per TX queue. Each MP from
+  which buffers are to be transmitted must be associated to memory regions
+  (MRs). This is a slow operation that must be cached.
+
+  This value is always 1 for RX queues since they use a single MP.
+
+Run-time configuration
+~~~~~~~~~~~~~~~~~~~~~~
+
+- librte_pmd_mlx5 brings kernel network interfaces up during initialization
+  because it is affected by their state. Forcing them down prevents packets
+  reception.
+
+- **ethtool** operations on related kernel interfaces also affect the PMD.
+
+Prerequisites
+-------------
+
+This driver relies on external libraries and kernel drivers for resources
+allocations and initialization. The following dependencies are not part of
+DPDK and must be installed separately:
+
+- **libibverbs**
+
+  User space Verbs framework used by librte_pmd_mlx5. This library provides
+  a generic interface between the kernel and low-level user space drivers
+  such as libmlx5.
+
+  It allows slow and privileged operations (context initialization, hardware
+  resources allocations) to be managed by the kernel and fast operations to
+  never leave user space.
+
+- **libmlx5**
+
+  Low-level user space driver library for Mellanox ConnectX-4 devices,
+  it is automatically loaded by libibverbs.
+
+  This library basically implements send/receive calls to the hardware
+  queues.
+
+- **Kernel modules** (mlnx-ofed-kernel)
+
+  They provide the kernel-side Verbs API and low level device drivers that
+  manage actual hardware initialization and resources sharing with user
+  space processes.
+
+  Unlike most other PMDs, these modules must remain loaded and bound to
+  their devices:
+
+  - mlx5_core: hardware driver managing Mellanox ConnectX-4 devices and
+    related Ethernet kernel network devices.
+  - mlx5_ib: InifiniBand device driver.
+  - ib_uverbs: user space driver for Verbs (entry point for libibverbs).
+
+- **Firmware update**
+
+  Mellanox OFED releases include firmware updates for ConnectX-4 adapters.
+
+  Because each release provides new features, these updates must be applied to
+  match the kernel modules and libraries they come with.
+
+.. note::
+
+   Both libraries are BSD and GPL licensed. Linux kernel modules are GPL
+   licensed.
+
+Getting Mellanox OFED
+~~~~~~~~~~~~~~~~~~~~~
+
+While these libraries and kernel modules are available on OpenFabrics
+Alliance's `website <https://www.openfabrics.org/>`__ and provided by package
+managers on most distributions, this PMD requires Ethernet extensions that
+may not be supported at the moment (this is a work in progress).
+
+`Mellanox OFED
+<http://www.mellanox.com/page/products_dyn?product_family=26&mtag=linux>`__
+includes the necessary support and should be used in the meantime. For DPDK,
+only libibverbs, libmlx5, mlnx-ofed-kernel packages and firmware updates are
+required from that distribution.
+
+.. note::
+
+   Several versions of Mellanox OFED are available. Installing the version
+   this DPDK release was developed and tested against is strongly
+   recommended. Please check the `prerequisites`_.
+
+Usage example
+-------------
+
+This section demonstrates how to launch **testpmd** with Mellanox ConnectX-4
+devices managed by librte_pmd_mlx5.
+
+#. Load the kernel modules:
+
+   .. code-block:: console
+
+      modprobe -a ib_uverbs mlx5_core mlx5_ib
+
+   .. note::
+
+      User space I/O kernel modules (uio and igb_uio) are not used and do
+      not have to be loaded.
+
+#. Make sure Ethernet interfaces are in working order and linked to kernel
+   verbs. Related sysfs entries should be present:
+
+   .. code-block:: console
+
+      ls -d /sys/class/net/*/device/infiniband_verbs/uverbs* | cut -d / -f 5
+
+   Example output:
+
+   .. code-block:: console
+
+      eth30
+      eth31
+      eth32
+      eth33
+
+#. Optionally, retrieve their PCI bus addresses for whitelisting:
+
+   .. code-block:: console
+
+      {
+          for intf in eth2 eth3 eth4 eth5;
+          do
+              (cd "/sys/class/net/${intf}/device/" && pwd -P);
+          done;
+      } |
+      sed -n 's,.*/\(.*\),-w \1,p'
+
+   Example output:
+
+   .. code-block:: console
+
+      -w 0000:05:00.1
+      -w 0000:06:00.0
+      -w 0000:06:00.1
+      -w 0000:05:00.0
+
+#. Request huge pages:
+
+   .. code-block:: console
+
+      echo 1024 > /sys/kernel/mm/hugepages/hugepages-2048kB/nr_hugepages/nr_hugepages
+
+#. Start testpmd with basic parameters:
+
+   .. code-block:: console
+
+      testpmd -c 0xff00 -n 4 -w 05:00.0 -w 05:00.1 -w 06:00.0 -w 06:00.1 -- --rxq=2 --txq=2 -i
+
+   Example output:
+
+   .. code-block:: console
+
+      [...]
+      EAL: PCI device 0000:05:00.0 on NUMA socket 0
+      EAL:   probe driver: 15b3:1013 librte_pmd_mlx5
+      PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_0" (VF: false)
+      PMD: librte_pmd_mlx5: 1 port(s) detected
+      PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fe
+      EAL: PCI device 0000:05:00.1 on NUMA socket 0
+      EAL:   probe driver: 15b3:1013 librte_pmd_mlx5
+      PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_1" (VF: false)
+      PMD: librte_pmd_mlx5: 1 port(s) detected
+      PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:ff
+      EAL: PCI device 0000:06:00.0 on NUMA socket 0
+      EAL:   probe driver: 15b3:1013 librte_pmd_mlx5
+      PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_2" (VF: false)
+      PMD: librte_pmd_mlx5: 1 port(s) detected
+      PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fa
+      EAL: PCI device 0000:06:00.1 on NUMA socket 0
+      EAL:   probe driver: 15b3:1013 librte_pmd_mlx5
+      PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_3" (VF: false)
+      PMD: librte_pmd_mlx5: 1 port(s) detected
+      PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fb
+      Interactive-mode selected
+      Configuring Port 0 (socket 0)
+      PMD: librte_pmd_mlx5: 0x8cba80: TX queues number update: 0 -> 2
+      PMD: librte_pmd_mlx5: 0x8cba80: RX queues number update: 0 -> 2
+      Port 0: E4:1D:2D:E7:0C:FE
+      Configuring Port 1 (socket 0)
+      PMD: librte_pmd_mlx5: 0x8ccac8: TX queues number update: 0 -> 2
+      PMD: librte_pmd_mlx5: 0x8ccac8: RX queues number update: 0 -> 2
+      Port 1: E4:1D:2D:E7:0C:FF
+      Configuring Port 2 (socket 0)
+      PMD: librte_pmd_mlx5: 0x8cdb10: TX queues number update: 0 -> 2
+      PMD: librte_pmd_mlx5: 0x8cdb10: RX queues number update: 0 -> 2
+      Port 2: E4:1D:2D:E7:0C:FA
+      Configuring Port 3 (socket 0)
+      PMD: librte_pmd_mlx5: 0x8ceb58: TX queues number update: 0 -> 2
+      PMD: librte_pmd_mlx5: 0x8ceb58: RX queues number update: 0 -> 2
+      Port 3: E4:1D:2D:E7:0C:FB
+      Checking link statuses...
+      Port 0 Link Up - speed 40000 Mbps - full-duplex
+      Port 1 Link Up - speed 40000 Mbps - full-duplex
+      Port 2 Link Up - speed 10000 Mbps - full-duplex
+      Port 3 Link Up - speed 10000 Mbps - full-duplex
+      Done
+      testpmd>
diff --git a/doc/guides/rel_notes/release_2_2.rst b/doc/guides/rel_notes/release_2_2.rst
index 89e4d58..e0f8681 100644
--- a/doc/guides/rel_notes/release_2_2.rst
+++ b/doc/guides/rel_notes/release_2_2.rst
@@ -31,6 +31,15 @@ New Features
 
 * **Added vhost-user multiple queue support.**
 
+* **Added support for Mellanox ConnectX-4 adapters (mlx5).**
+
+  The mlx5 poll-mode driver implements support for Mellanox ConnectX-4 EN
+  and Mellanox ConnectX-4 Lx EN families of 10/25/40/50/100 Gb/s adapters.
+
+  Like mlx4, this PMD is only available for Linux and is disabled by default
+  due to external dependencies (libibverbs and libmlx5).
+
+
 Resolved Issues
 ---------------
 
-- 
2.1.0

^ permalink raw reply	[flat|nested] 32+ messages in thread

* Re: [dpdk-dev] [PATCH v2 00/13] Mellanox ConnectX-4 PMD (mlx5)
  2015-10-30 18:52 ` [dpdk-dev] [PATCH v2 00/13] Mellanox ConnectX-4 PMD (mlx5) Adrien Mazarguil
                     ` (12 preceding siblings ...)
  2015-10-30 18:52   ` [dpdk-dev] [PATCH v2 13/13] doc: add mlx5 documentation and release notes for version 2.2 Adrien Mazarguil
@ 2015-10-30 23:18   ` Thomas Monjalon
  13 siblings, 0 replies; 32+ messages in thread
From: Thomas Monjalon @ 2015-10-30 23:18 UTC (permalink / raw)
  To: Adrien Mazarguil; +Cc: dev

2015-10-30 19:52, Adrien Mazarguil:
> This PMD adds basic support for Mellanox ConnectX-4 (mlx5) families of
> 10/25/40/50/100 Gb/s adapters through the Verbs framework.
> 
> Its design is very similar to that of mlx4 from which most of its code is
> borrowed without the mistake of putting it all in a single huge file.
> 
> It is disabled by default due to its dependency on libibverbs.
> 
> Changes in v2:
> - Removed useless port inactive warning.
> - Simplified code by replacing configured MAC addresses RX queue bit-field
>   with flow pointer checks.
> - Replaced allmulti/promisc status bits with request bits to fix
>   inconsistencies when restoring these modes.
> - Updated comments about maximum number of MAC addresses and VLAN filters.
> - Improved performance with better prefetching.
> - Fixed deadlock in case of error during port start.
> - Simplified VLAN filtering configuration storage using a basic list instead
>   of a table (with holes).

Applied, thanks

^ permalink raw reply	[flat|nested] 32+ messages in thread

* Re: [dpdk-dev] [PATCH v2 09/13] mlx5: add link update device operation
  2015-10-30 18:52   ` [dpdk-dev] [PATCH v2 09/13] mlx5: add link update device operation Adrien Mazarguil
@ 2015-11-02 17:52     ` Stephen Hemminger
  2015-11-02 18:27       ` Adrien Mazarguil
  0 siblings, 1 reply; 32+ messages in thread
From: Stephen Hemminger @ 2015-11-02 17:52 UTC (permalink / raw)
  To: Adrien Mazarguil; +Cc: dev

On Fri, 30 Oct 2015 19:52:38 +0100
Adrien Mazarguil <adrien.mazarguil@6wind.com> wrote:

> +static int
> +mlx5_link_update_unlocked(struct rte_eth_dev *dev, int wait_to_complete)
> +{
> +	struct priv *priv = dev->data->dev_private;
> +	struct ethtool_cmd edata = {
> +		.cmd = ETHTOOL_GSET
> +	};
> +	struct ifreq ifr;
> +	struct rte_eth_link dev_link;
> +	int link_speed = 0;
> +
> +	(void)wait_to_complete;

DPDK style is to use the __rte_unused attribute rather than dummy statements
to avoid unused warnings.

^ permalink raw reply	[flat|nested] 32+ messages in thread

* Re: [dpdk-dev] [PATCH v2 09/13] mlx5: add link update device operation
  2015-11-02 17:52     ` Stephen Hemminger
@ 2015-11-02 18:27       ` Adrien Mazarguil
  2015-11-02 18:43         ` Stephen Hemminger
  0 siblings, 1 reply; 32+ messages in thread
From: Adrien Mazarguil @ 2015-11-02 18:27 UTC (permalink / raw)
  To: Stephen Hemminger; +Cc: dev

On Mon, Nov 02, 2015 at 09:52:17AM -0800, Stephen Hemminger wrote:
> On Fri, 30 Oct 2015 19:52:38 +0100
> Adrien Mazarguil <adrien.mazarguil@6wind.com> wrote:
> 
> > +static int
> > +mlx5_link_update_unlocked(struct rte_eth_dev *dev, int wait_to_complete)
> > +{
> > +	struct priv *priv = dev->data->dev_private;
> > +	struct ethtool_cmd edata = {
> > +		.cmd = ETHTOOL_GSET
> > +	};
> > +	struct ifreq ifr;
> > +	struct rte_eth_link dev_link;
> > +	int link_speed = 0;
> > +
> > +	(void)wait_to_complete;
> 
> DPDK style is to use the __rte_unused attribute rather than dummy statements
> to avoid unused warnings.

Thanks for pointing this out, I'm used to avoiding C extensions whenever
possible but will stick to DPDK style next time.

Still, it would be nice if we could steer DPDK away from such extensions as
much as possible. As a library, we should allow user applications to compile
with flags we can't control (such as -pedantic -std=c99, and various
-Wsomething).

-- 
Adrien Mazarguil
6WIND

^ permalink raw reply	[flat|nested] 32+ messages in thread

* Re: [dpdk-dev] [PATCH v2 09/13] mlx5: add link update device operation
  2015-11-02 18:27       ` Adrien Mazarguil
@ 2015-11-02 18:43         ` Stephen Hemminger
  0 siblings, 0 replies; 32+ messages in thread
From: Stephen Hemminger @ 2015-11-02 18:43 UTC (permalink / raw)
  To: Adrien Mazarguil; +Cc: dev

On Mon, 2 Nov 2015 19:27:40 +0100
Adrien Mazarguil <adrien.mazarguil@6wind.com> wrote:

> Thanks for pointing this out, I'm used to avoiding C extensions whenever
> possible but will stick to DPDK style next time.
> 
> Still, it would be nice if we could steer DPDK away from such extensions as
> much as possible. As a library, we should allow user applications to compile
> with flags we can't control (such as -pedantic -std=c99, and various
> -Wsomething)

No. The extensions are very useful, catch errors, and generate more readable code.
For example the extension to check printf formats.

^ permalink raw reply	[flat|nested] 32+ messages in thread

end of thread, other threads:[~2015-11-02 18:42 UTC | newest]

Thread overview: 32+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2015-10-05 17:52 [dpdk-dev] [PATCH 00/13] Mellanox ConnectX-4 PMD (mlx5) Adrien Mazarguil
2015-10-05 17:52 ` [dpdk-dev] [PATCH 01/13] mlx5: new poll-mode driver for Mellanox ConnectX-4 adapters Adrien Mazarguil
2015-10-05 17:52 ` [dpdk-dev] [PATCH 02/13] mlx5: add non-scattered TX and RX support Adrien Mazarguil
2015-10-05 17:52 ` [dpdk-dev] [PATCH 03/13] mlx5: add MAC handling Adrien Mazarguil
2015-10-05 17:53 ` [dpdk-dev] [PATCH 04/13] mlx5: add device configure/start/stop Adrien Mazarguil
2015-10-05 17:53 ` [dpdk-dev] [PATCH 05/13] mlx5: add support for scattered RX and TX buffers Adrien Mazarguil
2015-10-05 17:53 ` [dpdk-dev] [PATCH 06/13] mlx5: add MTU configuration support Adrien Mazarguil
2015-10-05 17:53 ` [dpdk-dev] [PATCH 07/13] mlx5: add software counters and related callbacks Adrien Mazarguil
2015-10-05 17:53 ` [dpdk-dev] [PATCH 08/13] mlx5: add promiscuous and allmulticast RX modes Adrien Mazarguil
2015-10-05 17:53 ` [dpdk-dev] [PATCH 09/13] mlx5: add link update device operation Adrien Mazarguil
2015-10-05 17:53 ` [dpdk-dev] [PATCH 10/13] mlx5: add flow control device operations Adrien Mazarguil
2015-10-05 17:53 ` [dpdk-dev] [PATCH 11/13] mlx5: add VLAN filtering Adrien Mazarguil
2015-10-05 17:53 ` [dpdk-dev] [PATCH 12/13] mlx5: add checksum offloading support Adrien Mazarguil
2015-10-05 17:53 ` [dpdk-dev] [PATCH 13/13] doc: add mlx5 documentation and release notes for version 2.2 Adrien Mazarguil
2015-10-30 18:52 ` [dpdk-dev] [PATCH v2 00/13] Mellanox ConnectX-4 PMD (mlx5) Adrien Mazarguil
2015-10-30 18:52   ` [dpdk-dev] [PATCH v2 01/13] mlx5: new poll-mode driver for Mellanox ConnectX-4 adapters Adrien Mazarguil
2015-10-30 18:52   ` [dpdk-dev] [PATCH v2 02/13] mlx5: add non-scattered TX and RX support Adrien Mazarguil
2015-10-30 18:52   ` [dpdk-dev] [PATCH v2 03/13] mlx5: add MAC handling Adrien Mazarguil
2015-10-30 18:52   ` [dpdk-dev] [PATCH v2 04/13] mlx5: add device configure/start/stop Adrien Mazarguil
2015-10-30 18:52   ` [dpdk-dev] [PATCH v2 05/13] mlx5: add support for scattered RX and TX buffers Adrien Mazarguil
2015-10-30 18:52   ` [dpdk-dev] [PATCH v2 06/13] mlx5: add MTU configuration support Adrien Mazarguil
2015-10-30 18:52   ` [dpdk-dev] [PATCH v2 07/13] mlx5: add software counters and related callbacks Adrien Mazarguil
2015-10-30 18:52   ` [dpdk-dev] [PATCH v2 08/13] mlx5: add promiscuous and allmulticast RX modes Adrien Mazarguil
2015-10-30 18:52   ` [dpdk-dev] [PATCH v2 09/13] mlx5: add link update device operation Adrien Mazarguil
2015-11-02 17:52     ` Stephen Hemminger
2015-11-02 18:27       ` Adrien Mazarguil
2015-11-02 18:43         ` Stephen Hemminger
2015-10-30 18:52   ` [dpdk-dev] [PATCH v2 10/13] mlx5: add flow control device operations Adrien Mazarguil
2015-10-30 18:52   ` [dpdk-dev] [PATCH v2 11/13] mlx5: add VLAN filtering Adrien Mazarguil
2015-10-30 18:52   ` [dpdk-dev] [PATCH v2 12/13] mlx5: add checksum offloading support Adrien Mazarguil
2015-10-30 18:52   ` [dpdk-dev] [PATCH v2 13/13] doc: add mlx5 documentation and release notes for version 2.2 Adrien Mazarguil
2015-10-30 23:18   ` [dpdk-dev] [PATCH v2 00/13] Mellanox ConnectX-4 PMD (mlx5) Thomas Monjalon

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).