mlx4_en: Add driver for Mellanox ConnectX 10GbE NIC
authorYevgeny Petrilin <yevgenyp@mellanox.co.il>
Wed, 22 Oct 2008 22:47:49 +0000 (15:47 -0700)
committerRoland Dreier <rolandd@cisco.com>
Wed, 22 Oct 2008 22:47:49 +0000 (15:47 -0700)
The Mellanox ConnectX can operate as an InfiniBand adapter, as an
Ethernet NIC, or as a Fibre Channel (FC) HBA.  The kernel has a
low-level driver, mlx4_core, which handles multiplexing access to the
device, and there is also already an InfiniBad driver, mlx4_ib.

This patch adds a new driver, mlx4_en, which implements a standard
Ethernet NIC driver.

Signed-off-by: Liran Liss <liranl@mellanox.co.il>
Signed-off-by: Yevgeny Petrilin <yevgenyp@mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
12 files changed:
drivers/net/Kconfig
drivers/net/mlx4/Makefile
drivers/net/mlx4/en_cq.c [new file with mode: 0644]
drivers/net/mlx4/en_main.c [new file with mode: 0644]
drivers/net/mlx4/en_netdev.c [new file with mode: 0644]
drivers/net/mlx4/en_params.c [new file with mode: 0644]
drivers/net/mlx4/en_port.c [new file with mode: 0644]
drivers/net/mlx4/en_port.h [new file with mode: 0644]
drivers/net/mlx4/en_resources.c [new file with mode: 0644]
drivers/net/mlx4/en_rx.c [new file with mode: 0644]
drivers/net/mlx4/en_tx.c [new file with mode: 0644]
drivers/net/mlx4/mlx4_en.h [new file with mode: 0644]

index 4a11296..69e2c10 100644 (file)
@@ -2465,6 +2465,15 @@ config PASEMI_MAC
          This driver supports the on-chip 1/10Gbit Ethernet controller on
          PA Semi's PWRficient line of chips.
 
+config MLX4_EN
+       tristate "Mellanox Technologies 10Gbit Ethernet support"
+       depends on PCI && INET
+       select MLX4_CORE
+       select INET_LRO
+       help
+         This driver supports Mellanox Technologies ConnectX Ethernet
+         devices.
+
 config MLX4_CORE
        tristate
        depends on PCI
index 9f49366..a7a97bf 100644 (file)
@@ -2,3 +2,8 @@ obj-$(CONFIG_MLX4_CORE)         += mlx4_core.o
 
 mlx4_core-y := alloc.o catas.o cmd.o cq.o eq.o fw.o icm.o intf.o main.o mcg.o \
                mr.o pd.o port.o profile.o qp.o reset.o srq.o
+
+obj-$(CONFIG_MLX4_EN)               += mlx4_en.o
+
+mlx4_en-y :=   en_main.o en_tx.o en_rx.o en_params.o en_port.o en_cq.o \
+               en_resources.o en_netdev.o
diff --git a/drivers/net/mlx4/en_cq.c b/drivers/net/mlx4/en_cq.c
new file mode 100644 (file)
index 0000000..1368a80
--- /dev/null
@@ -0,0 +1,146 @@
+/*
+ * Copyright (c) 2007 Mellanox Technologies. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ */
+
+#include <linux/mlx4/cq.h>
+#include <linux/mlx4/qp.h>
+#include <linux/mlx4/cmd.h>
+
+#include "mlx4_en.h"
+
+static void mlx4_en_cq_event(struct mlx4_cq *cq, enum mlx4_event event)
+{
+       return;
+}
+
+
+int mlx4_en_create_cq(struct mlx4_en_priv *priv,
+                     struct mlx4_en_cq *cq,
+                     int entries, int ring, enum cq_type mode)
+{
+       struct mlx4_en_dev *mdev = priv->mdev;
+       int err;
+
+       cq->size = entries;
+       if (mode == RX)
+               cq->buf_size = cq->size * sizeof(struct mlx4_cqe);
+       else
+               cq->buf_size = sizeof(struct mlx4_cqe);
+
+       cq->ring = ring;
+       cq->is_tx = mode;
+       spin_lock_init(&cq->lock);
+
+       err = mlx4_alloc_hwq_res(mdev->dev, &cq->wqres,
+                               cq->buf_size, 2 * PAGE_SIZE);
+       if (err)
+               return err;
+
+       err = mlx4_en_map_buffer(&cq->wqres.buf);
+       if (err)
+               mlx4_free_hwq_res(mdev->dev, &cq->wqres, cq->buf_size);
+
+       return err;
+}
+
+int mlx4_en_activate_cq(struct mlx4_en_priv *priv, struct mlx4_en_cq *cq)
+{
+       struct mlx4_en_dev *mdev = priv->mdev;
+       int err;
+
+       cq->dev = mdev->pndev[priv->port];
+       cq->mcq.set_ci_db  = cq->wqres.db.db;
+       cq->mcq.arm_db     = cq->wqres.db.db + 1;
+       *cq->mcq.set_ci_db = 0;
+       *cq->mcq.arm_db    = 0;
+       cq->buf = (struct mlx4_cqe *) cq->wqres.buf.direct.buf;
+       memset(cq->buf, 0, cq->buf_size);
+
+       err = mlx4_cq_alloc(mdev->dev, cq->size, &cq->wqres.mtt, &mdev->priv_uar,
+                           cq->wqres.db.dma, &cq->mcq, cq->is_tx);
+       if (err)
+               return err;
+
+       cq->mcq.comp  = cq->is_tx ? mlx4_en_tx_irq : mlx4_en_rx_irq;
+       cq->mcq.event = mlx4_en_cq_event;
+
+       if (cq->is_tx) {
+               init_timer(&cq->timer);
+               cq->timer.function = mlx4_en_poll_tx_cq;
+               cq->timer.data = (unsigned long) cq;
+       } else {
+               netif_napi_add(cq->dev, &cq->napi, mlx4_en_poll_rx_cq, 64);
+               napi_enable(&cq->napi);
+       }
+
+       return 0;
+}
+
+void mlx4_en_destroy_cq(struct mlx4_en_priv *priv, struct mlx4_en_cq *cq)
+{
+       struct mlx4_en_dev *mdev = priv->mdev;
+
+       mlx4_en_unmap_buffer(&cq->wqres.buf);
+       mlx4_free_hwq_res(mdev->dev, &cq->wqres, cq->buf_size);
+       cq->buf_size = 0;
+       cq->buf = NULL;
+}
+
+void mlx4_en_deactivate_cq(struct mlx4_en_priv *priv, struct mlx4_en_cq *cq)
+{
+       struct mlx4_en_dev *mdev = priv->mdev;
+
+       if (cq->is_tx)
+               del_timer(&cq->timer);
+       else
+               napi_disable(&cq->napi);
+
+       mlx4_cq_free(mdev->dev, &cq->mcq);
+}
+
+/* Set rx cq moderation parameters */
+int mlx4_en_set_cq_moder(struct mlx4_en_priv *priv, struct mlx4_en_cq *cq)
+{
+       return mlx4_cq_modify(priv->mdev->dev, &cq->mcq,
+                             cq->moder_cnt, cq->moder_time);
+}
+
+int mlx4_en_arm_cq(struct mlx4_en_priv *priv, struct mlx4_en_cq *cq)
+{
+       cq->armed = 1;
+       mlx4_cq_arm(&cq->mcq, MLX4_CQ_DB_REQ_NOT, priv->mdev->uar_map,
+                   &priv->mdev->uar_lock);
+
+       return 0;
+}
+
+
diff --git a/drivers/net/mlx4/en_main.c b/drivers/net/mlx4/en_main.c
new file mode 100644 (file)
index 0000000..1b0eebf
--- /dev/null
@@ -0,0 +1,254 @@
+/*
+ * Copyright (c) 2007 Mellanox Technologies. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ */
+
+#include <linux/cpumask.h>
+#include <linux/module.h>
+#include <linux/delay.h>
+#include <linux/netdevice.h>
+#include <linux/cpumask.h>
+
+#include <linux/mlx4/driver.h>
+#include <linux/mlx4/device.h>
+#include <linux/mlx4/cmd.h>
+
+#include "mlx4_en.h"
+
+MODULE_AUTHOR("Liran Liss, Yevgeny Petrilin");
+MODULE_DESCRIPTION("Mellanox ConnectX HCA Ethernet driver");
+MODULE_LICENSE("Dual BSD/GPL");
+MODULE_VERSION(DRV_VERSION " ("DRV_RELDATE")");
+
+static const char mlx4_en_version[] =
+       DRV_NAME ": Mellanox ConnectX HCA Ethernet driver v"
+       DRV_VERSION " (" DRV_RELDATE ")\n";
+
+static void mlx4_en_event(struct mlx4_dev *dev, void *endev_ptr,
+                         enum mlx4_dev_event event, int port)
+{
+       struct mlx4_en_dev *mdev = (struct mlx4_en_dev *) endev_ptr;
+       struct mlx4_en_priv *priv;
+
+       if (!mdev->pndev[port])
+               return;
+
+       priv = netdev_priv(mdev->pndev[port]);
+       switch (event) {
+       case MLX4_DEV_EVENT_PORT_UP:
+       case MLX4_DEV_EVENT_PORT_DOWN:
+               /* To prevent races, we poll the link state in a separate
+                 task rather than changing it here */
+               priv->link_state = event;
+               queue_work(mdev->workqueue, &priv->linkstate_task);
+               break;
+
+       case MLX4_DEV_EVENT_CATASTROPHIC_ERROR:
+               mlx4_err(mdev, "Internal error detected, restarting device\n");
+               break;
+
+       default:
+               mlx4_warn(mdev, "Unhandled event: %d\n", event);
+       }
+}
+
+static void mlx4_en_remove(struct mlx4_dev *dev, void *endev_ptr)
+{
+       struct mlx4_en_dev *mdev = endev_ptr;
+       int i;
+
+       mutex_lock(&mdev->state_lock);
+       mdev->device_up = false;
+       mutex_unlock(&mdev->state_lock);
+
+       mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_ETH)
+               if (mdev->pndev[i])
+                       mlx4_en_destroy_netdev(mdev->pndev[i]);
+
+       flush_workqueue(mdev->workqueue);
+       destroy_workqueue(mdev->workqueue);
+       mlx4_mr_free(dev, &mdev->mr);
+       mlx4_uar_free(dev, &mdev->priv_uar);
+       mlx4_pd_free(dev, mdev->priv_pdn);
+       kfree(mdev);
+}
+
+static void *mlx4_en_add(struct mlx4_dev *dev)
+{
+       static int mlx4_en_version_printed;
+       struct mlx4_en_dev *mdev;
+       int i;
+       int err;
+
+       if (!mlx4_en_version_printed) {
+               printk(KERN_INFO "%s", mlx4_en_version);
+               mlx4_en_version_printed++;
+       }
+
+       mdev = kzalloc(sizeof *mdev, GFP_KERNEL);
+       if (!mdev) {
+               dev_err(&dev->pdev->dev, "Device struct alloc failed, "
+                       "aborting.\n");
+               err = -ENOMEM;
+               goto err_free_res;
+       }
+
+       if (mlx4_pd_alloc(dev, &mdev->priv_pdn))
+               goto err_free_dev;
+
+       if (mlx4_uar_alloc(dev, &mdev->priv_uar))
+               goto err_pd;
+
+       mdev->uar_map = ioremap(mdev->priv_uar.pfn << PAGE_SHIFT, PAGE_SIZE);
+       if (!mdev->uar_map)
+               goto err_uar;
+       spin_lock_init(&mdev->uar_lock);
+
+       mdev->dev = dev;
+       mdev->dma_device = &(dev->pdev->dev);
+       mdev->pdev = dev->pdev;
+       mdev->device_up = false;
+
+       mdev->LSO_support = !!(dev->caps.flags & (1 << 15));
+       if (!mdev->LSO_support)
+               mlx4_warn(mdev, "LSO not supported, please upgrade to later "
+                               "FW version to enable LSO\n");
+
+       if (mlx4_mr_alloc(mdev->dev, mdev->priv_pdn, 0, ~0ull,
+                        MLX4_PERM_LOCAL_WRITE |  MLX4_PERM_LOCAL_READ,
+                        0, 0, &mdev->mr)) {
+               mlx4_err(mdev, "Failed allocating memory region\n");
+               goto err_uar;
+       }
+       if (mlx4_mr_enable(mdev->dev, &mdev->mr)) {
+               mlx4_err(mdev, "Failed enabling memory region\n");
+               goto err_mr;
+       }
+
+       /* Build device profile according to supplied module parameters */
+       err = mlx4_en_get_profile(mdev);
+       if (err) {
+               mlx4_err(mdev, "Bad module parameters, aborting.\n");
+               goto err_mr;
+       }
+
+       /* Configure wich ports to start according to module parameters */
+       mdev->port_cnt = 0;
+       mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_ETH)
+               mdev->port_cnt++;
+
+       /* If we did not receive an explicit number of Rx rings, default to
+        * the number of completion vectors populated by the mlx4_core */
+       mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_ETH) {
+               mlx4_info(mdev, "Using %d tx rings for port:%d\n",
+                         mdev->profile.prof[i].tx_ring_num, i);
+               if (!mdev->profile.prof[i].rx_ring_num) {
+                       mdev->profile.prof[i].rx_ring_num = 1;
+                       mlx4_info(mdev, "Defaulting to %d rx rings for port:%d\n",
+                                 1, i);
+               } else
+                       mlx4_info(mdev, "Using %d rx rings for port:%d\n",
+                                 mdev->profile.prof[i].rx_ring_num, i);
+       }
+
+       /* Create our own workqueue for reset/multicast tasks
+        * Note: we cannot use the shared workqueue because of deadlocks caused
+        *       by the rtnl lock */
+       mdev->workqueue = create_singlethread_workqueue("mlx4_en");
+       if (!mdev->workqueue) {
+               err = -ENOMEM;
+               goto err_close_nic;
+       }
+
+       /* At this stage all non-port specific tasks are complete:
+        * mark the card state as up */
+       mutex_init(&mdev->state_lock);
+       mdev->device_up = true;
+
+       /* Setup ports */
+
+       /* Create a netdev for each port */
+       mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_ETH) {
+               mlx4_info(mdev, "Activating port:%d\n", i);
+               if (mlx4_en_init_netdev(mdev, i, &mdev->profile.prof[i])) {
+                       mdev->pndev[i] = NULL;
+                       goto err_free_netdev;
+               }
+       }
+       return mdev;
+
+
+err_free_netdev:
+       mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_ETH) {
+               if (mdev->pndev[i])
+                       mlx4_en_destroy_netdev(mdev->pndev[i]);
+       }
+
+       mutex_lock(&mdev->state_lock);
+       mdev->device_up = false;
+       mutex_unlock(&mdev->state_lock);
+       flush_workqueue(mdev->workqueue);
+
+       /* Stop event queue before we drop down to release shared SW state */
+
+err_close_nic:
+       destroy_workqueue(mdev->workqueue);
+err_mr:
+       mlx4_mr_free(dev, &mdev->mr);
+err_uar:
+       mlx4_uar_free(dev, &mdev->priv_uar);
+err_pd:
+       mlx4_pd_free(dev, mdev->priv_pdn);
+err_free_dev:
+       kfree(mdev);
+err_free_res:
+       return NULL;
+}
+
+static struct mlx4_interface mlx4_en_interface = {
+       .add    = mlx4_en_add,
+       .remove = mlx4_en_remove,
+       .event  = mlx4_en_event,
+};
+
+static int __init mlx4_en_init(void)
+{
+       return mlx4_register_interface(&mlx4_en_interface);
+}
+
+static void __exit mlx4_en_cleanup(void)
+{
+       mlx4_unregister_interface(&mlx4_en_interface);
+}
+
+module_init(mlx4_en_init);
+module_exit(mlx4_en_cleanup);
+
diff --git a/drivers/net/mlx4/en_netdev.c b/drivers/net/mlx4/en_netdev.c
new file mode 100644 (file)
index 0000000..a339afb
--- /dev/null
@@ -0,0 +1,1088 @@
+/*
+ * Copyright (c) 2007 Mellanox Technologies. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ */
+
+#include <linux/etherdevice.h>
+#include <linux/tcp.h>
+#include <linux/if_vlan.h>
+#include <linux/delay.h>
+
+#include <linux/mlx4/driver.h>
+#include <linux/mlx4/device.h>
+#include <linux/mlx4/cmd.h>
+#include <linux/mlx4/cq.h>
+
+#include "mlx4_en.h"
+#include "en_port.h"
+
+
+static void mlx4_en_vlan_rx_register(struct net_device *dev, struct vlan_group *grp)
+{
+       struct mlx4_en_priv *priv = netdev_priv(dev);
+       struct mlx4_en_dev *mdev = priv->mdev;
+       int err;
+
+       mlx4_dbg(HW, priv, "Registering VLAN group:%p\n", grp);
+       priv->vlgrp = grp;
+
+       mutex_lock(&mdev->state_lock);
+       if (mdev->device_up && priv->port_up) {
+               err = mlx4_SET_VLAN_FLTR(mdev->dev, priv->port, grp);
+               if (err)
+                       mlx4_err(mdev, "Failed configuring VLAN filter\n");
+       }
+       mutex_unlock(&mdev->state_lock);
+}
+
+static void mlx4_en_vlan_rx_add_vid(struct net_device *dev, unsigned short vid)
+{
+       struct mlx4_en_priv *priv = netdev_priv(dev);
+       struct mlx4_en_dev *mdev = priv->mdev;
+       int err;
+
+       if (!priv->vlgrp)
+               return;
+
+       mlx4_dbg(HW, priv, "adding VLAN:%d (vlgrp entry:%p)\n",
+                vid, vlan_group_get_device(priv->vlgrp, vid));
+
+       /* Add VID to port VLAN filter */
+       mutex_lock(&mdev->state_lock);
+       if (mdev->device_up && priv->port_up) {
+               err = mlx4_SET_VLAN_FLTR(mdev->dev, priv->port, priv->vlgrp);
+               if (err)
+                       mlx4_err(mdev, "Failed configuring VLAN filter\n");
+       }
+       mutex_unlock(&mdev->state_lock);
+}
+
+static void mlx4_en_vlan_rx_kill_vid(struct net_device *dev, unsigned short vid)
+{
+       struct mlx4_en_priv *priv = netdev_priv(dev);
+       struct mlx4_en_dev *mdev = priv->mdev;
+       int err;
+
+       if (!priv->vlgrp)
+               return;
+
+       mlx4_dbg(HW, priv, "Killing VID:%d (vlgrp:%p vlgrp "
+                "entry:%p)\n", vid, priv->vlgrp,
+                vlan_group_get_device(priv->vlgrp, vid));
+       vlan_group_set_device(priv->vlgrp, vid, NULL);
+
+       /* Remove VID from port VLAN filter */
+       mutex_lock(&mdev->state_lock);
+       if (mdev->device_up && priv->port_up) {
+               err = mlx4_SET_VLAN_FLTR(mdev->dev, priv->port, priv->vlgrp);
+               if (err)
+                       mlx4_err(mdev, "Failed configuring VLAN filter\n");
+       }
+       mutex_unlock(&mdev->state_lock);
+}
+
+static u64 mlx4_en_mac_to_u64(u8 *addr)
+{
+       u64 mac = 0;
+       int i;
+
+       for (i = 0; i < ETH_ALEN; i++) {
+               mac <<= 8;
+               mac |= addr[i];
+       }
+       return mac;
+}
+
+static int mlx4_en_set_mac(struct net_device *dev, void *addr)
+{
+       struct mlx4_en_priv *priv = netdev_priv(dev);
+       struct mlx4_en_dev *mdev = priv->mdev;
+       struct sockaddr *saddr = addr;
+
+       if (!is_valid_ether_addr(saddr->sa_data))
+               return -EADDRNOTAVAIL;
+
+       memcpy(dev->dev_addr, saddr->sa_data, ETH_ALEN);
+       priv->mac = mlx4_en_mac_to_u64(dev->dev_addr);
+       queue_work(mdev->workqueue, &priv->mac_task);
+       return 0;
+}
+
+static void mlx4_en_do_set_mac(struct work_struct *work)
+{
+       struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
+                                                mac_task);
+       struct mlx4_en_dev *mdev = priv->mdev;
+       int err = 0;
+
+       mutex_lock(&mdev->state_lock);
+       if (priv->port_up) {
+               /* Remove old MAC and insert the new one */
+               mlx4_unregister_mac(mdev->dev, priv->port, priv->mac_index);
+               err = mlx4_register_mac(mdev->dev, priv->port,
+                                       priv->mac, &priv->mac_index);
+               if (err)
+                       mlx4_err(mdev, "Failed changing HW MAC address\n");
+       } else
+               mlx4_dbg(HW, priv, "Port is down, exiting...\n");
+
+       mutex_unlock(&mdev->state_lock);
+}
+
+static void mlx4_en_clear_list(struct net_device *dev)
+{
+       struct mlx4_en_priv *priv = netdev_priv(dev);
+       struct dev_mc_list *plist = priv->mc_list;
+       struct dev_mc_list *next;
+
+       while (plist) {
+               next = plist->next;
+               kfree(plist);
+               plist = next;
+       }
+       priv->mc_list = NULL;
+}
+
+static void mlx4_en_cache_mclist(struct net_device *dev)
+{
+       struct mlx4_en_priv *priv = netdev_priv(dev);
+       struct mlx4_en_dev *mdev = priv->mdev;
+       struct dev_mc_list *mclist;
+       struct dev_mc_list *tmp;
+       struct dev_mc_list *plist = NULL;
+
+       for (mclist = dev->mc_list; mclist; mclist = mclist->next) {
+               tmp = kmalloc(sizeof(struct dev_mc_list), GFP_ATOMIC);
+               if (!tmp) {
+                       mlx4_err(mdev, "failed to allocate multicast list\n");
+                       mlx4_en_clear_list(dev);
+                       return;
+               }
+               memcpy(tmp, mclist, sizeof(struct dev_mc_list));
+               tmp->next = NULL;
+               if (plist)
+                       plist->next = tmp;
+               else
+                       priv->mc_list = tmp;
+               plist = tmp;
+       }
+}
+
+
+static void mlx4_en_set_multicast(struct net_device *dev)
+{
+       struct mlx4_en_priv *priv = netdev_priv(dev);
+
+       if (!priv->port_up)
+               return;
+
+       queue_work(priv->mdev->workqueue, &priv->mcast_task);
+}
+
+static void mlx4_en_do_set_multicast(struct work_struct *work)
+{
+       struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
+                                                mcast_task);
+       struct mlx4_en_dev *mdev = priv->mdev;
+       struct net_device *dev = priv->dev;
+       struct dev_mc_list *mclist;
+       u64 mcast_addr = 0;
+       int err;
+
+       mutex_lock(&mdev->state_lock);
+       if (!mdev->device_up) {
+               mlx4_dbg(HW, priv, "Card is not up, ignoring "
+                                  "multicast change.\n");
+               goto out;
+       }
+       if (!priv->port_up) {
+               mlx4_dbg(HW, priv, "Port is down, ignoring "
+                                  "multicast change.\n");
+               goto out;
+       }
+
+       /*
+        * Promsicuous mode: disable all filters
+        */
+
+       if (dev->flags & IFF_PROMISC) {
+               if (!(priv->flags & MLX4_EN_FLAG_PROMISC)) {
+                       if (netif_msg_rx_status(priv))
+                               mlx4_warn(mdev, "Port:%d entering promiscuous mode\n",
+                                         priv->port);
+                       priv->flags |= MLX4_EN_FLAG_PROMISC;
+
+                       /* Enable promiscouos mode */
+                       err = mlx4_SET_PORT_qpn_calc(mdev->dev, priv->port,
+                                                    priv->base_qpn, 1);
+                       if (err)
+                               mlx4_err(mdev, "Failed enabling "
+                                        "promiscous mode\n");
+
+                       /* Disable port multicast filter (unconditionally) */
+                       err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
+                                                 0, MLX4_MCAST_DISABLE);
+                       if (err)
+                               mlx4_err(mdev, "Failed disabling "
+                                        "multicast filter\n");
+
+                       /* Disable port VLAN filter */
+                       err = mlx4_SET_VLAN_FLTR(mdev->dev, priv->port, NULL);
+                       if (err)
+                               mlx4_err(mdev, "Failed disabling "
+                                        "VLAN filter\n");
+               }
+               goto out;
+       }
+
+       /*
+        * Not in promiscous mode
+        */
+
+       if (priv->flags & MLX4_EN_FLAG_PROMISC) {
+               if (netif_msg_rx_status(priv))
+                       mlx4_warn(mdev, "Port:%d leaving promiscuous mode\n",
+                                 priv->port);
+               priv->flags &= ~MLX4_EN_FLAG_PROMISC;
+
+               /* Disable promiscouos mode */
+               err = mlx4_SET_PORT_qpn_calc(mdev->dev, priv->port,
+                                            priv->base_qpn, 0);
+               if (err)
+                       mlx4_err(mdev, "Failed disabling promiscous mode\n");
+
+               /* Enable port VLAN filter */
+               err = mlx4_SET_VLAN_FLTR(mdev->dev, priv->port, priv->vlgrp);
+               if (err)
+                       mlx4_err(mdev, "Failed enabling VLAN filter\n");
+       }
+
+       /* Enable/disable the multicast filter according to IFF_ALLMULTI */
+       if (dev->flags & IFF_ALLMULTI) {
+               err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
+                                         0, MLX4_MCAST_DISABLE);
+               if (err)
+                       mlx4_err(mdev, "Failed disabling multicast filter\n");
+       } else {
+               err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
+                                         0, MLX4_MCAST_DISABLE);
+               if (err)
+                       mlx4_err(mdev, "Failed disabling multicast filter\n");
+
+               /* Flush mcast filter and init it with broadcast address */
+               mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, ETH_BCAST,
+                                   1, MLX4_MCAST_CONFIG);
+
+               /* Update multicast list - we cache all addresses so they won't
+                * change while HW is updated holding the command semaphor */
+               netif_tx_lock_bh(dev);
+               mlx4_en_cache_mclist(dev);
+               netif_tx_unlock_bh(dev);
+               for (mclist = priv->mc_list; mclist; mclist = mclist->next) {
+                       mcast_addr = mlx4_en_mac_to_u64(mclist->dmi_addr);
+                       mlx4_SET_MCAST_FLTR(mdev->dev, priv->port,
+                                           mcast_addr, 0, MLX4_MCAST_CONFIG);
+               }
+               err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
+                                         0, MLX4_MCAST_ENABLE);
+               if (err)
+                       mlx4_err(mdev, "Failed enabling multicast filter\n");
+
+               mlx4_en_clear_list(dev);
+       }
+out:
+       mutex_unlock(&mdev->state_lock);
+}
+
+#ifdef CONFIG_NET_POLL_CONTROLLER
+static void mlx4_en_netpoll(struct net_device *dev)
+{
+       struct mlx4_en_priv *priv = netdev_priv(dev);
+       struct mlx4_en_cq *cq;
+       unsigned long flags;
+       int i;
+
+       for (i = 0; i < priv->rx_ring_num; i++) {
+               cq = &priv->rx_cq[i];
+               spin_lock_irqsave(&cq->lock, flags);
+               napi_synchronize(&cq->napi);
+               mlx4_en_process_rx_cq(dev, cq, 0);
+               spin_unlock_irqrestore(&cq->lock, flags);
+       }
+}
+#endif
+
+static void mlx4_en_tx_timeout(struct net_device *dev)
+{
+       struct mlx4_en_priv *priv = netdev_priv(dev);
+       struct mlx4_en_dev *mdev = priv->mdev;
+
+       if (netif_msg_timer(priv))
+               mlx4_warn(mdev, "Tx timeout called on port:%d\n", priv->port);
+
+       if (netif_carrier_ok(dev)) {
+               priv->port_stats.tx_timeout++;
+               mlx4_dbg(DRV, priv, "Scheduling watchdog\n");
+               queue_work(mdev->workqueue, &priv->watchdog_task);
+       }
+}
+
+
+static struct net_device_stats *mlx4_en_get_stats(struct net_device *dev)
+{
+       struct mlx4_en_priv *priv = netdev_priv(dev);
+
+       spin_lock_bh(&priv->stats_lock);
+       memcpy(&priv->ret_stats, &priv->stats, sizeof(priv->stats));
+       spin_unlock_bh(&priv->stats_lock);
+
+       return &priv->ret_stats;
+}
+
+static void mlx4_en_set_default_moderation(struct mlx4_en_priv *priv)
+{
+       struct mlx4_en_dev *mdev = priv->mdev;
+       struct mlx4_en_cq *cq;
+       int i;
+
+       /* If we haven't received a specific coalescing setting
+        * (module param), we set the moderation paramters as follows:
+        * - moder_cnt is set to the number of mtu sized packets to
+        *   satisfy our coelsing target.
+        * - moder_time is set to a fixed value.
+        */
+       priv->rx_frames = (mdev->profile.rx_moder_cnt ==
+                          MLX4_EN_AUTO_CONF) ?
+                               MLX4_EN_RX_COAL_TARGET /
+                               priv->dev->mtu + 1 :
+                               mdev->profile.rx_moder_cnt;
+       priv->rx_usecs = (mdev->profile.rx_moder_time ==
+                         MLX4_EN_AUTO_CONF) ?
+                               MLX4_EN_RX_COAL_TIME :
+                               mdev->profile.rx_moder_time;
+       mlx4_dbg(INTR, priv, "Default coalesing params for mtu:%d - "
+                            "rx_frames:%d rx_usecs:%d\n",
+                priv->dev->mtu, priv->rx_frames, priv->rx_usecs);
+
+       /* Setup cq moderation params */
+       for (i = 0; i < priv->rx_ring_num; i++) {
+               cq = &priv->rx_cq[i];
+               cq->moder_cnt = priv->rx_frames;
+               cq->moder_time = priv->rx_usecs;
+       }
+
+       for (i = 0; i < priv->tx_ring_num; i++) {
+               cq = &priv->tx_cq[i];
+               cq->moder_cnt = MLX4_EN_TX_COAL_PKTS;
+               cq->moder_time = MLX4_EN_TX_COAL_TIME;
+       }
+
+       /* Reset auto-moderation params */
+       priv->pkt_rate_low = MLX4_EN_RX_RATE_LOW;
+       priv->rx_usecs_low = MLX4_EN_RX_COAL_TIME_LOW;
+       priv->pkt_rate_high = MLX4_EN_RX_RATE_HIGH;
+       priv->rx_usecs_high = MLX4_EN_RX_COAL_TIME_HIGH;
+       priv->sample_interval = MLX4_EN_SAMPLE_INTERVAL;
+       priv->adaptive_rx_coal = mdev->profile.auto_moder;
+       priv->last_moder_time = MLX4_EN_AUTO_CONF;
+       priv->last_moder_jiffies = 0;
+       priv->last_moder_packets = 0;
+       priv->last_moder_tx_packets = 0;
+       priv->last_moder_bytes = 0;
+}
+
+static void mlx4_en_auto_moderation(struct mlx4_en_priv *priv)
+{
+       unsigned long period = (unsigned long) (jiffies - priv->last_moder_jiffies);
+       struct mlx4_en_dev *mdev = priv->mdev;
+       struct mlx4_en_cq *cq;
+       unsigned long packets;
+       unsigned long rate;
+       unsigned long avg_pkt_size;
+       unsigned long rx_packets;
+       unsigned long rx_bytes;
+       unsigned long tx_packets;
+       unsigned long tx_pkt_diff;
+       unsigned long rx_pkt_diff;
+       int moder_time;
+       int i, err;
+
+       if (!priv->adaptive_rx_coal || period < priv->sample_interval * HZ)
+               return;
+
+       spin_lock_bh(&priv->stats_lock);
+       rx_packets = priv->stats.rx_packets;
+       rx_bytes = priv->stats.rx_bytes;
+       tx_packets = priv->stats.tx_packets;
+       spin_unlock_bh(&priv->stats_lock);
+
+       if (!priv->last_moder_jiffies || !period)
+               goto out;
+
+       tx_pkt_diff = ((unsigned long) (tx_packets -
+                                       priv->last_moder_tx_packets));
+       rx_pkt_diff = ((unsigned long) (rx_packets -
+                                       priv->last_moder_packets));
+       packets = max(tx_pkt_diff, rx_pkt_diff);
+       rate = packets * HZ / period;
+       avg_pkt_size = packets ? ((unsigned long) (rx_bytes -
+                                priv->last_moder_bytes)) / packets : 0;
+
+       /* Apply auto-moderation only when packet rate exceeds a rate that
+        * it matters */
+       if (rate > MLX4_EN_RX_RATE_THRESH) {
+               /* If tx and rx packet rates are not balanced, assume that
+                * traffic is mainly BW bound and apply maximum moderation.
+                * Otherwise, moderate according to packet rate */
+               if (2 * tx_pkt_diff > 3 * rx_pkt_diff ||
+                   2 * rx_pkt_diff > 3 * tx_pkt_diff) {
+                       moder_time = priv->rx_usecs_high;
+               } else {
+                       if (rate < priv->pkt_rate_low)
+                               moder_time = priv->rx_usecs_low;
+                       else if (rate > priv->pkt_rate_high)
+                               moder_time = priv->rx_usecs_high;
+                       else
+                               moder_time = (rate - priv->pkt_rate_low) *
+                                       (priv->rx_usecs_high - priv->rx_usecs_low) /
+                                       (priv->pkt_rate_high - priv->pkt_rate_low) +
+                                       priv->rx_usecs_low;
+               }
+       } else {
+               /* When packet rate is low, use default moderation rather than
+                * 0 to prevent interrupt storms if traffic suddenly increases */
+               moder_time = priv->rx_usecs;
+       }
+
+       mlx4_dbg(INTR, priv, "tx rate:%lu rx_rate:%lu\n",
+                tx_pkt_diff * HZ / period, rx_pkt_diff * HZ / period);
+
+       mlx4_dbg(INTR, priv, "Rx moder_time changed from:%d to %d period:%lu "
+                "[jiff] packets:%lu avg_pkt_size:%lu rate:%lu [p/s])\n",
+                priv->last_moder_time, moder_time, period, packets,
+                avg_pkt_size, rate);
+
+       if (moder_time != priv->last_moder_time) {
+               priv->last_moder_time = moder_time;
+               for (i = 0; i < priv->rx_ring_num; i++) {
+                       cq = &priv->rx_cq[i];
+                       cq->moder_time = moder_time;
+                       err = mlx4_en_set_cq_moder(priv, cq);
+                       if (err) {
+                               mlx4_err(mdev, "Failed modifying moderation for cq:%d "
+                                        "on port:%d\n", i, priv->port);
+                               break;
+                       }
+               }
+       }
+
+out:
+       priv->last_moder_packets = rx_packets;
+       priv->last_moder_tx_packets = tx_packets;
+       priv->last_moder_bytes = rx_bytes;
+       priv->last_moder_jiffies = jiffies;
+}
+
+static void mlx4_en_do_get_stats(struct work_struct *work)
+{
+       struct delayed_work *delay = container_of(work, struct delayed_work, work);
+       struct mlx4_en_priv *priv = container_of(delay, struct mlx4_en_priv,
+                                                stats_task);
+       struct mlx4_en_dev *mdev = priv->mdev;
+       int err;
+
+       err = mlx4_en_DUMP_ETH_STATS(mdev, priv->port, 0);
+       if (err)
+               mlx4_dbg(HW, priv, "Could not update stats for "
+                                  "port:%d\n", priv->port);
+
+       mutex_lock(&mdev->state_lock);
+       if (mdev->device_up) {
+               if (priv->port_up)
+                       mlx4_en_auto_moderation(priv);
+
+               queue_delayed_work(mdev->workqueue, &priv->stats_task, STATS_DELAY);
+       }
+       mutex_unlock(&mdev->state_lock);
+}
+
+static void mlx4_en_linkstate(struct work_struct *work)
+{
+       struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
+                                                linkstate_task);
+       struct mlx4_en_dev *mdev = priv->mdev;
+       int linkstate = priv->link_state;
+
+       mutex_lock(&mdev->state_lock);
+       /* If observable port state changed set carrier state and
+        * report to system log */
+       if (priv->last_link_state != linkstate) {
+               if (linkstate == MLX4_DEV_EVENT_PORT_DOWN) {
+                       if (netif_msg_link(priv))
+                               mlx4_info(mdev, "Port %d - link down\n", priv->port);
+                       netif_carrier_off(priv->dev);
+               } else {
+                       if (netif_msg_link(priv))
+                               mlx4_info(mdev, "Port %d - link up\n", priv->port);
+                       netif_carrier_on(priv->dev);
+               }
+       }
+       priv->last_link_state = linkstate;
+       mutex_unlock(&mdev->state_lock);
+}
+
+
+static int mlx4_en_start_port(struct net_device *dev)
+{
+       struct mlx4_en_priv *priv = netdev_priv(dev);
+       struct mlx4_en_dev *mdev = priv->mdev;
+       struct mlx4_en_cq *cq;
+       struct mlx4_en_tx_ring *tx_ring;
+       struct mlx4_en_rx_ring *rx_ring;
+       int rx_index = 0;
+       int tx_index = 0;
+       u16 stride;
+       int err = 0;
+       int i;
+       int j;
+
+       if (priv->port_up) {
+               mlx4_dbg(DRV, priv, "start port called while port already up\n");
+               return 0;
+       }
+
+       /* Calculate Rx buf size */
+       dev->mtu = min(dev->mtu, priv->max_mtu);
+       mlx4_en_calc_rx_buf(dev);
+       mlx4_dbg(DRV, priv, "Rx buf size:%d\n", priv->rx_skb_size);
+       stride = roundup_pow_of_two(sizeof(struct mlx4_en_rx_desc) +
+                                   DS_SIZE * priv->num_frags);
+       /* Configure rx cq's and rings */
+       for (i = 0; i < priv->rx_ring_num; i++) {
+               cq = &priv->rx_cq[i];
+               rx_ring = &priv->rx_ring[i];
+
+               err = mlx4_en_activate_cq(priv, cq);
+               if (err) {
+                       mlx4_err(mdev, "Failed activating Rx CQ\n");
+                       goto rx_err;
+               }
+               for (j = 0; j < cq->size; j++)
+                       cq->buf[j].owner_sr_opcode = MLX4_CQE_OWNER_MASK;
+               err = mlx4_en_set_cq_moder(priv, cq);
+               if (err) {
+                       mlx4_err(mdev, "Failed setting cq moderation parameters");
+                       mlx4_en_deactivate_cq(priv, cq);
+                       goto cq_err;
+               }
+               mlx4_en_arm_cq(priv, cq);
+
+               ++rx_index;
+       }
+
+       err = mlx4_en_activate_rx_rings(priv);
+       if (err) {
+               mlx4_err(mdev, "Failed to activate RX rings\n");
+               goto cq_err;
+       }
+
+       err = mlx4_en_config_rss_steer(priv);
+       if (err) {
+               mlx4_err(mdev, "Failed configuring rss steering\n");
+               goto rx_err;
+       }
+
+       /* Configure tx cq's and rings */
+       for (i = 0; i < priv->tx_ring_num; i++) {
+               /* Configure cq */
+               cq = &priv->tx_cq[i];
+               err = mlx4_en_activate_cq(priv, cq);
+               if (err) {
+                       mlx4_err(mdev, "Failed allocating Tx CQ\n");
+                       goto tx_err;
+               }
+               err = mlx4_en_set_cq_moder(priv, cq);
+               if (err) {
+                       mlx4_err(mdev, "Failed setting cq moderation parameters");
+                       mlx4_en_deactivate_cq(priv, cq);
+                       goto tx_err;
+               }
+               mlx4_dbg(DRV, priv, "Resetting index of collapsed CQ:%d to -1\n", i);
+               cq->buf->wqe_index = cpu_to_be16(0xffff);
+
+               /* Configure ring */
+               tx_ring = &priv->tx_ring[i];
+               err = mlx4_en_activate_tx_ring(priv, tx_ring, cq->mcq.cqn,
+                                              priv->rx_ring[0].srq.srqn);
+               if (err) {
+                       mlx4_err(mdev, "Failed allocating Tx ring\n");
+                       mlx4_en_deactivate_cq(priv, cq);
+                       goto tx_err;
+               }
+               /* Set initial ownership of all Tx TXBBs to SW (1) */
+               for (j = 0; j < tx_ring->buf_size; j += STAMP_STRIDE)
+                       *((u32 *) (tx_ring->buf + j)) = 0xffffffff;
+               ++tx_index;
+       }
+
+       /* Configure port */
+       err = mlx4_SET_PORT_general(mdev->dev, priv->port,
+                                   priv->rx_skb_size + ETH_FCS_LEN,
+                                   mdev->profile.tx_pause,
+                                   mdev->profile.tx_ppp,
+                                   mdev->profile.rx_pause,
+                                   mdev->profile.rx_ppp);
+       if (err) {
+               mlx4_err(mdev, "Failed setting port general configurations"
+                              " for port %d, with error %d\n", priv->port, err);
+               goto tx_err;
+       }
+       /* Set default qp number */
+       err = mlx4_SET_PORT_qpn_calc(mdev->dev, priv->port, priv->base_qpn, 0);
+       if (err) {
+               mlx4_err(mdev, "Failed setting default qp numbers\n");
+               goto tx_err;
+       }
+       /* Set port mac number */
+       mlx4_dbg(DRV, priv, "Setting mac for port %d\n", priv->port);
+       err = mlx4_register_mac(mdev->dev, priv->port,
+                               priv->mac, &priv->mac_index);
+       if (err) {
+               mlx4_err(mdev, "Failed setting port mac\n");
+               goto tx_err;
+       }
+
+       /* Init port */
+       mlx4_dbg(HW, priv, "Initializing port\n");
+       err = mlx4_INIT_PORT(mdev->dev, priv->port);
+       if (err) {
+               mlx4_err(mdev, "Failed Initializing port\n");
+               goto mac_err;
+       }
+
+       /* Schedule multicast task to populate multicast list */
+       queue_work(mdev->workqueue, &priv->mcast_task);
+
+       priv->port_up = true;
+       netif_start_queue(dev);
+       return 0;
+
+mac_err:
+       mlx4_unregister_mac(mdev->dev, priv->port, priv->mac_index);
+tx_err:
+       while (tx_index--) {
+               mlx4_en_deactivate_tx_ring(priv, &priv->tx_ring[tx_index]);
+               mlx4_en_deactivate_cq(priv, &priv->tx_cq[tx_index]);
+       }
+
+       mlx4_en_release_rss_steer(priv);
+rx_err:
+       for (i = 0; i < priv->rx_ring_num; i++)
+               mlx4_en_deactivate_rx_ring(priv, &priv->rx_ring[rx_index]);
+cq_err:
+       while (rx_index--)
+               mlx4_en_deactivate_cq(priv, &priv->rx_cq[rx_index]);
+
+       return err; /* need to close devices */
+}
+
+
+static void mlx4_en_stop_port(struct net_device *dev)
+{
+       struct mlx4_en_priv *priv = netdev_priv(dev);
+       struct mlx4_en_dev *mdev = priv->mdev;
+       int i;
+
+       if (!priv->port_up) {
+               mlx4_dbg(DRV, priv, "stop port (%d) called while port already down\n",
+                        priv->port);
+               return;
+       }
+       netif_stop_queue(dev);
+
+       /* Synchronize with tx routine */
+       netif_tx_lock_bh(dev);
+       priv->port_up = false;
+       netif_tx_unlock_bh(dev);
+
+       /* close port*/
+       mlx4_CLOSE_PORT(mdev->dev, priv->port);
+
+       /* Unregister Mac address for the port */
+       mlx4_unregister_mac(mdev->dev, priv->port, priv->mac_index);
+
+       /* Free TX Rings */
+       for (i = 0; i < priv->tx_ring_num; i++) {
+               mlx4_en_deactivate_tx_ring(priv, &priv->tx_ring[i]);
+               mlx4_en_deactivate_cq(priv, &priv->tx_cq[i]);
+       }
+       msleep(10);
+
+       for (i = 0; i < priv->tx_ring_num; i++)
+               mlx4_en_free_tx_buf(dev, &priv->tx_ring[i]);
+
+       /* Free RSS qps */
+       mlx4_en_release_rss_steer(priv);
+
+       /* Free RX Rings */
+       for (i = 0; i < priv->rx_ring_num; i++) {
+               mlx4_en_deactivate_rx_ring(priv, &priv->rx_ring[i]);
+               while (test_bit(NAPI_STATE_SCHED, &priv->rx_cq[i].napi.state))
+                       msleep(1);
+               mlx4_en_deactivate_cq(priv, &priv->rx_cq[i]);
+       }
+}
+
+static void mlx4_en_restart(struct work_struct *work)
+{
+       struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
+                                                watchdog_task);
+       struct mlx4_en_dev *mdev = priv->mdev;
+       struct net_device *dev = priv->dev;
+
+       mlx4_dbg(DRV, priv, "Watchdog task called for port %d\n", priv->port);
+       mlx4_en_stop_port(dev);
+       if (mlx4_en_start_port(dev))
+           mlx4_err(mdev, "Failed restarting port %d\n", priv->port);
+}
+
+
+static int mlx4_en_open(struct net_device *dev)
+{
+       struct mlx4_en_priv *priv = netdev_priv(dev);
+       struct mlx4_en_dev *mdev = priv->mdev;
+       int i;
+       int err = 0;
+
+       mutex_lock(&mdev->state_lock);
+
+       if (!mdev->device_up) {
+               mlx4_err(mdev, "Cannot open - device down/disabled\n");
+               err = -EBUSY;
+               goto out;
+       }
+
+       /* Reset HW statistics and performance counters */
+       if (mlx4_en_DUMP_ETH_STATS(mdev, priv->port, 1))
+               mlx4_dbg(HW, priv, "Failed dumping statistics\n");
+
+       memset(&priv->stats, 0, sizeof(priv->stats));
+       memset(&priv->pstats, 0, sizeof(priv->pstats));
+
+       for (i = 0; i < priv->tx_ring_num; i++) {
+               priv->tx_ring[i].bytes = 0;
+               priv->tx_ring[i].packets = 0;
+       }
+       for (i = 0; i < priv->rx_ring_num; i++) {
+               priv->rx_ring[i].bytes = 0;
+               priv->rx_ring[i].packets = 0;
+       }
+
+       mlx4_en_set_default_moderation(priv);
+       err = mlx4_en_start_port(dev);
+       if (err)
+               mlx4_err(mdev, "Failed starting port:%d\n", priv->port);
+
+out:
+       mutex_unlock(&mdev->state_lock);
+       return err;
+}
+
+
+static int mlx4_en_close(struct net_device *dev)
+{
+       struct mlx4_en_priv *priv = netdev_priv(dev);
+       struct mlx4_en_dev *mdev = priv->mdev;
+
+       if (netif_msg_ifdown(priv))
+               mlx4_info(mdev, "Close called for port:%d\n", priv->port);
+
+       mutex_lock(&mdev->state_lock);
+
+       mlx4_en_stop_port(dev);
+       netif_carrier_off(dev);
+
+       mutex_unlock(&mdev->state_lock);
+       return 0;
+}
+
+static void mlx4_en_free_resources(struct mlx4_en_priv *priv)
+{
+       int i;
+
+       for (i = 0; i < priv->tx_ring_num; i++) {
+               if (priv->tx_ring[i].tx_info)
+                       mlx4_en_destroy_tx_ring(priv, &priv->tx_ring[i]);
+               if (priv->tx_cq[i].buf)
+                       mlx4_en_destroy_cq(priv, &priv->tx_cq[i]);
+       }
+
+       for (i = 0; i < priv->rx_ring_num; i++) {
+               if (priv->rx_ring[i].rx_info)
+                       mlx4_en_destroy_rx_ring(priv, &priv->rx_ring[i]);
+               if (priv->rx_cq[i].buf)
+                       mlx4_en_destroy_cq(priv, &priv->rx_cq[i]);
+       }
+}
+
+static int mlx4_en_alloc_resources(struct mlx4_en_priv *priv)
+{
+       struct mlx4_en_dev *mdev = priv->mdev;
+       struct mlx4_en_port_profile *prof = priv->prof;
+       int i;
+
+       /* Create tx Rings */
+       for (i = 0; i < priv->tx_ring_num; i++) {
+               if (mlx4_en_create_cq(priv, &priv->tx_cq[i],
+                                     prof->tx_ring_size, i, TX))
+                       goto err;
+
+               if (mlx4_en_create_tx_ring(priv, &priv->tx_ring[i],
+                                          prof->tx_ring_size, TXBB_SIZE))
+                       goto err;
+       }
+
+       /* Create rx Rings */
+       for (i = 0; i < priv->rx_ring_num; i++) {
+               if (mlx4_en_create_cq(priv, &priv->rx_cq[i],
+                                     prof->rx_ring_size, i, RX))
+                       goto err;
+
+               if (mlx4_en_create_rx_ring(priv, &priv->rx_ring[i],
+                                          prof->rx_ring_size, priv->stride))
+                       goto err;
+       }
+
+       return 0;
+
+err:
+       mlx4_err(mdev, "Failed to allocate NIC resources\n");
+       return -ENOMEM;
+}
+
+
+void mlx4_en_destroy_netdev(struct net_device *dev)
+{
+       struct mlx4_en_priv *priv = netdev_priv(dev);
+       struct mlx4_en_dev *mdev = priv->mdev;
+
+       mlx4_dbg(DRV, priv, "Destroying netdev on port:%d\n", priv->port);
+
+       /* Unregister device - this will close the port if it was up */
+       if (priv->registered)
+               unregister_netdev(dev);
+
+       if (priv->allocated)
+               mlx4_free_hwq_res(mdev->dev, &priv->res, MLX4_EN_PAGE_SIZE);
+
+       cancel_delayed_work(&priv->stats_task);
+       cancel_delayed_work(&priv->refill_task);
+       /* flush any pending task for this netdev */
+       flush_workqueue(mdev->workqueue);
+
+       /* Detach the netdev so tasks would not attempt to access it */
+       mutex_lock(&mdev->state_lock);
+       mdev->pndev[priv->port] = NULL;
+       mutex_unlock(&mdev->state_lock);
+
+       mlx4_en_free_resources(priv);
+       free_netdev(dev);
+}
+
+static int mlx4_en_change_mtu(struct net_device *dev, int new_mtu)
+{
+       struct mlx4_en_priv *priv = netdev_priv(dev);
+       struct mlx4_en_dev *mdev = priv->mdev;
+       int err = 0;
+
+       mlx4_dbg(DRV, priv, "Change MTU called - current:%d new:%d\n",
+                dev->mtu, new_mtu);
+
+       if ((new_mtu < MLX4_EN_MIN_MTU) || (new_mtu > priv->max_mtu)) {
+               mlx4_err(mdev, "Bad MTU size:%d.\n", new_mtu);
+               return -EPERM;
+       }
+       dev->mtu = new_mtu;
+
+       if (netif_running(dev)) {
+               mutex_lock(&mdev->state_lock);
+               if (!mdev->device_up) {
+                       /* NIC is probably restarting - let watchdog task reset
+                        * the port */
+                       mlx4_dbg(DRV, priv, "Change MTU called with card down!?\n");
+               } else {
+                       mlx4_en_stop_port(dev);
+                       mlx4_en_set_default_moderation(priv);
+                       err = mlx4_en_start_port(dev);
+                       if (err) {
+                               mlx4_err(mdev, "Failed restarting port:%d\n",
+                                        priv->port);
+                               queue_work(mdev->workqueue, &priv->watchdog_task);
+                       }
+               }
+               mutex_unlock(&mdev->state_lock);
+       }
+       return 0;
+}
+
+int mlx4_en_init_netdev(struct mlx4_en_dev *mdev, int port,
+                       struct mlx4_en_port_profile *prof)
+{
+       struct net_device *dev;
+       struct mlx4_en_priv *priv;
+       int i;
+       int err;
+
+       dev = alloc_etherdev(sizeof(struct mlx4_en_priv));
+       if (dev == NULL) {
+               mlx4_err(mdev, "Net device allocation failed\n");
+               return -ENOMEM;
+       }
+
+       SET_NETDEV_DEV(dev, &mdev->dev->pdev->dev);
+
+       /*
+        * Initialize driver private data
+        */
+
+       priv = netdev_priv(dev);
+       memset(priv, 0, sizeof(struct mlx4_en_priv));
+       priv->dev = dev;
+       priv->mdev = mdev;
+       priv->prof = prof;
+       priv->port = port;
+       priv->port_up = false;
+       priv->rx_csum = 1;
+       priv->flags = prof->flags;
+       priv->tx_ring_num = prof->tx_ring_num;
+       priv->rx_ring_num = prof->rx_ring_num;
+       priv->mc_list = NULL;
+       priv->mac_index = -1;
+       priv->msg_enable = MLX4_EN_MSG_LEVEL;
+       spin_lock_init(&priv->stats_lock);
+       INIT_WORK(&priv->mcast_task, mlx4_en_do_set_multicast);
+       INIT_WORK(&priv->mac_task, mlx4_en_do_set_mac);
+       INIT_DELAYED_WORK(&priv->refill_task, mlx4_en_rx_refill);
+       INIT_WORK(&priv->watchdog_task, mlx4_en_restart);
+       INIT_WORK(&priv->linkstate_task, mlx4_en_linkstate);
+       INIT_DELAYED_WORK(&priv->stats_task, mlx4_en_do_get_stats);
+
+       /* Query for default mac and max mtu */
+       priv->max_mtu = mdev->dev->caps.eth_mtu_cap[priv->port];
+       priv->mac = mdev->dev->caps.def_mac[priv->port];
+       if (ILLEGAL_MAC(priv->mac)) {
+               mlx4_err(mdev, "Port: %d, invalid mac burned: 0x%llx, quiting\n",
+                        priv->port, priv->mac);
+               err = -EINVAL;
+               goto out;
+       }
+
+       priv->stride = roundup_pow_of_two(sizeof(struct mlx4_en_rx_desc) +
+                                         DS_SIZE * MLX4_EN_MAX_RX_FRAGS);
+       err = mlx4_en_alloc_resources(priv);
+       if (err)
+               goto out;
+
+       /* Populate Rx default RSS mappings */
+       mlx4_en_set_default_rss_map(priv, &priv->rss_map, priv->rx_ring_num *
+                                               RSS_FACTOR, priv->rx_ring_num);
+       /* Allocate page for receive rings */
+       err = mlx4_alloc_hwq_res(mdev->dev, &priv->res,
+                               MLX4_EN_PAGE_SIZE, MLX4_EN_PAGE_SIZE);
+       if (err) {
+               mlx4_err(mdev, "Failed to allocate page for rx qps\n");
+               goto out;
+       }
+       priv->allocated = 1;
+
+       /* Populate Tx priority mappings */
+       mlx4_en_set_prio_map(priv, priv->tx_prio_map, prof->tx_ring_num);
+
+       /*
+        * Initialize netdev entry points
+        */
+
+       dev->open = &mlx4_en_open;
+       dev->stop = &mlx4_en_close;
+       dev->hard_start_xmit = &mlx4_en_xmit;
+       dev->get_stats = &mlx4_en_get_stats;
+       dev->set_multicast_list = &mlx4_en_set_multicast;
+       dev->set_mac_address = &mlx4_en_set_mac;
+       dev->change_mtu = &mlx4_en_change_mtu;
+       dev->tx_timeout = &mlx4_en_tx_timeout;
+       dev->watchdog_timeo = MLX4_EN_WATCHDOG_TIMEOUT;
+       dev->vlan_rx_register = mlx4_en_vlan_rx_register;
+       dev->vlan_rx_add_vid = mlx4_en_vlan_rx_add_vid;
+       dev->vlan_rx_kill_vid = mlx4_en_vlan_rx_kill_vid;
+#ifdef CONFIG_NET_POLL_CONTROLLER
+       dev->poll_controller = mlx4_en_netpoll;
+#endif
+       SET_ETHTOOL_OPS(dev, &mlx4_en_ethtool_ops);
+
+       /* Set defualt MAC */
+       dev->addr_len = ETH_ALEN;
+       for (i = 0; i < ETH_ALEN; i++)
+               dev->dev_addr[ETH_ALEN - 1 - i] =
+               (u8) (priv->mac >> (8 * i));
+
+       /*
+        * Set driver features
+        */
+       dev->features |= NETIF_F_SG;
+       dev->features |= NETIF_F_HW_CSUM;
+       dev->features |= NETIF_F_HIGHDMA;
+       dev->features |= NETIF_F_HW_VLAN_TX |
+                        NETIF_F_HW_VLAN_RX |
+                        NETIF_F_HW_VLAN_FILTER;
+       if (mdev->profile.num_lro)
+               dev->features |= NETIF_F_LRO;
+       if (mdev->LSO_support) {
+               dev->features |= NETIF_F_TSO;
+               dev->features |= NETIF_F_TSO6;
+       }
+
+       mdev->pndev[port] = dev;
+
+       netif_carrier_off(dev);
+       err = register_netdev(dev);
+       if (err) {
+               mlx4_err(mdev, "Netdev registration failed\n");
+               goto out;
+       }
+       priv->registered = 1;
+       queue_delayed_work(mdev->workqueue, &priv->stats_task, STATS_DELAY);
+       return 0;
+
+out:
+       mlx4_en_destroy_netdev(dev);
+       return err;
+}
+
Simple merge
Simple merge
Simple merge
Simple merge
Simple merge
Simple merge
Simple merge