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 4a11296a951438df0730456206473bc40b363059..69e2c10cb9f44d1acfd7b349c6e139db02c31a2b 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 9f493666e27b3451503e09029716ef86dc966ee2..a7a97bf998f866a8bdf4b471ed7ed140968a145d 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;
+}
+
diff --git a/drivers/net/mlx4/en_params.c b/drivers/net/mlx4/en_params.c
new file mode 100644 (file)
index 0000000..c2e69b1
--- /dev/null
@@ -0,0 +1,480 @@
+/*
+ * 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/kernel.h>
+#include <linux/ethtool.h>
+#include <linux/netdevice.h>
+
+#include "mlx4_en.h"
+#include "en_port.h"
+
+#define MLX4_EN_PARM_INT(X, def_val, desc) \
+       static unsigned int X = def_val;\
+       module_param(X , uint, 0444); \
+       MODULE_PARM_DESC(X, desc);
+
+
+/*
+ * Device scope module parameters
+ */
+
+
+/* Use a XOR rathern than Toeplitz hash function for RSS */
+MLX4_EN_PARM_INT(rss_xor, 0, "Use XOR hash function for RSS");
+
+/* RSS hash type mask - default to <saddr, daddr, sport, dport> */
+MLX4_EN_PARM_INT(rss_mask, 0xf, "RSS hash type bitmask");
+
+/* Number of LRO sessions per Rx ring (rounded up to a power of two) */
+MLX4_EN_PARM_INT(num_lro, MLX4_EN_MAX_LRO_DESCRIPTORS,
+                "Number of LRO sessions per ring or disabled (0)");
+
+/* Priority pausing */
+MLX4_EN_PARM_INT(pptx, MLX4_EN_DEF_TX_PAUSE,
+                "Pause policy on TX: 0 never generate pause frames "
+                "1 generate pause frames according to RX buffer threshold");
+MLX4_EN_PARM_INT(pprx, MLX4_EN_DEF_RX_PAUSE,
+                "Pause policy on RX: 0 ignore received pause frames "
+                "1 respect received pause frames");
+MLX4_EN_PARM_INT(pfctx, 0, "Priority based Flow Control policy on TX[7:0]."
+                          " Per priority bit mask");
+MLX4_EN_PARM_INT(pfcrx, 0, "Priority based Flow Control policy on RX[7:0]."
+                          " Per priority bit mask");
+
+/* Interrupt moderation tunning */
+MLX4_EN_PARM_INT(rx_moder_cnt, MLX4_EN_AUTO_CONF,
+              "Max coalesced descriptors for Rx interrupt moderation");
+MLX4_EN_PARM_INT(rx_moder_time, MLX4_EN_AUTO_CONF,
+              "Timeout following last packet for Rx interrupt moderation");
+MLX4_EN_PARM_INT(auto_moder, 1, "Enable dynamic interrupt moderation");
+
+MLX4_EN_PARM_INT(rx_ring_num1, 0, "Number or Rx rings for port 1 (0 = #cores)");
+MLX4_EN_PARM_INT(rx_ring_num2, 0, "Number or Rx rings for port 2 (0 = #cores)");
+
+MLX4_EN_PARM_INT(tx_ring_size1, MLX4_EN_AUTO_CONF, "Tx ring size for port 1");
+MLX4_EN_PARM_INT(tx_ring_size2, MLX4_EN_AUTO_CONF, "Tx ring size for port 2");
+MLX4_EN_PARM_INT(rx_ring_size1, MLX4_EN_AUTO_CONF, "Rx ring size for port 1");
+MLX4_EN_PARM_INT(rx_ring_size2, MLX4_EN_AUTO_CONF, "Rx ring size for port 2");
+
+
+int mlx4_en_get_profile(struct mlx4_en_dev *mdev)
+{
+       struct mlx4_en_profile *params = &mdev->profile;
+
+       params->rx_moder_cnt = min_t(int, rx_moder_cnt, MLX4_EN_AUTO_CONF);
+       params->rx_moder_time = min_t(int, rx_moder_time, MLX4_EN_AUTO_CONF);
+       params->auto_moder = auto_moder;
+       params->rss_xor = (rss_xor != 0);
+       params->rss_mask = rss_mask & 0x1f;
+       params->num_lro = min_t(int, num_lro , MLX4_EN_MAX_LRO_DESCRIPTORS);
+       params->rx_pause = pprx;
+       params->rx_ppp = pfcrx;
+       params->tx_pause = pptx;
+       params->tx_ppp = pfctx;
+       if (params->rx_ppp || params->tx_ppp) {
+               params->prof[1].tx_ring_num = MLX4_EN_TX_RING_NUM;
+               params->prof[2].tx_ring_num = MLX4_EN_TX_RING_NUM;
+       } else {
+               params->prof[1].tx_ring_num = 1;
+               params->prof[2].tx_ring_num = 1;
+       }
+       params->prof[1].rx_ring_num = min_t(int, rx_ring_num1, MAX_RX_RINGS);
+       params->prof[2].rx_ring_num = min_t(int, rx_ring_num2, MAX_RX_RINGS);
+
+       if (tx_ring_size1 == MLX4_EN_AUTO_CONF)
+               tx_ring_size1 = MLX4_EN_DEF_TX_RING_SIZE;
+       params->prof[1].tx_ring_size =
+               (tx_ring_size1 < MLX4_EN_MIN_TX_SIZE) ?
+                MLX4_EN_MIN_TX_SIZE : roundup_pow_of_two(tx_ring_size1);
+
+       if (tx_ring_size2 == MLX4_EN_AUTO_CONF)
+               tx_ring_size2 = MLX4_EN_DEF_TX_RING_SIZE;
+       params->prof[2].tx_ring_size =
+               (tx_ring_size2 < MLX4_EN_MIN_TX_SIZE) ?
+                MLX4_EN_MIN_TX_SIZE : roundup_pow_of_two(tx_ring_size2);
+
+       if (rx_ring_size1 == MLX4_EN_AUTO_CONF)
+               rx_ring_size1 = MLX4_EN_DEF_RX_RING_SIZE;
+       params->prof[1].rx_ring_size =
+               (rx_ring_size1 < MLX4_EN_MIN_RX_SIZE) ?
+                MLX4_EN_MIN_RX_SIZE : roundup_pow_of_two(rx_ring_size1);
+
+       if (rx_ring_size2 == MLX4_EN_AUTO_CONF)
+               rx_ring_size2 = MLX4_EN_DEF_RX_RING_SIZE;
+       params->prof[2].rx_ring_size =
+               (rx_ring_size2 < MLX4_EN_MIN_RX_SIZE) ?
+                MLX4_EN_MIN_RX_SIZE : roundup_pow_of_two(rx_ring_size2);
+       return 0;
+}
+
+
+/*
+ * Ethtool support
+ */
+
+static void mlx4_en_update_lro_stats(struct mlx4_en_priv *priv)
+{
+       int i;
+
+       priv->port_stats.lro_aggregated = 0;
+       priv->port_stats.lro_flushed = 0;
+       priv->port_stats.lro_no_desc = 0;
+
+       for (i = 0; i < priv->rx_ring_num; i++) {
+               priv->port_stats.lro_aggregated += priv->rx_ring[i].lro.stats.aggregated;
+               priv->port_stats.lro_flushed += priv->rx_ring[i].lro.stats.flushed;
+               priv->port_stats.lro_no_desc += priv->rx_ring[i].lro.stats.no_desc;
+       }
+}
+
+static void
+mlx4_en_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *drvinfo)
+{
+       struct mlx4_en_priv *priv = netdev_priv(dev);
+       struct mlx4_en_dev *mdev = priv->mdev;
+
+       sprintf(drvinfo->driver, DRV_NAME " (%s)", mdev->dev->board_id);
+       strncpy(drvinfo->version, DRV_VERSION " (" DRV_RELDATE ")", 32);
+       sprintf(drvinfo->fw_version, "%d.%d.%d",
+               (u16) (mdev->dev->caps.fw_ver >> 32),
+               (u16) ((mdev->dev->caps.fw_ver >> 16) & 0xffff),
+               (u16) (mdev->dev->caps.fw_ver & 0xffff));
+       strncpy(drvinfo->bus_info, pci_name(mdev->dev->pdev), 32);
+       drvinfo->n_stats = 0;
+       drvinfo->regdump_len = 0;
+       drvinfo->eedump_len = 0;
+}
+
+static u32 mlx4_en_get_tso(struct net_device *dev)
+{
+       return (dev->features & NETIF_F_TSO) != 0;
+}
+
+static int mlx4_en_set_tso(struct net_device *dev, u32 data)
+{
+       struct mlx4_en_priv *priv = netdev_priv(dev);
+
+       if (data) {
+               if (!priv->mdev->LSO_support)
+                       return -EPERM;
+               dev->features |= (NETIF_F_TSO | NETIF_F_TSO6);
+       } else
+               dev->features &= ~(NETIF_F_TSO | NETIF_F_TSO6);
+       return 0;
+}
+
+static u32 mlx4_en_get_rx_csum(struct net_device *dev)
+{
+       struct mlx4_en_priv *priv = netdev_priv(dev);
+       return priv->rx_csum;
+}
+
+static int mlx4_en_set_rx_csum(struct net_device *dev, u32 data)
+{
+       struct mlx4_en_priv *priv = netdev_priv(dev);
+       priv->rx_csum = (data != 0);
+       return 0;
+}
+
+static const char main_strings[][ETH_GSTRING_LEN] = {
+       "rx_packets", "tx_packets", "rx_bytes", "tx_bytes", "rx_errors",
+       "tx_errors", "rx_dropped", "tx_dropped", "multicast", "collisions",
+       "rx_length_errors", "rx_over_errors", "rx_crc_errors",
+       "rx_frame_errors", "rx_fifo_errors", "rx_missed_errors",
+       "tx_aborted_errors", "tx_carrier_errors", "tx_fifo_errors",
+       "tx_heartbeat_errors", "tx_window_errors",
+
+       /* port statistics */
+       "lro_aggregated", "lro_flushed", "lro_no_desc", "tso_packets",
+       "queue_stopped", "wake_queue", "tx_timeout", "rx_alloc_failed",
+       "rx_csum_good", "rx_csum_none", "tx_chksum_offload",
+
+       /* packet statistics */
+       "broadcast", "rx_prio_0", "rx_prio_1", "rx_prio_2", "rx_prio_3",
+       "rx_prio_4", "rx_prio_5", "rx_prio_6", "rx_prio_7", "tx_prio_0",
+       "tx_prio_1", "tx_prio_2", "tx_prio_3", "tx_prio_4", "tx_prio_5",
+       "tx_prio_6", "tx_prio_7",
+};
+#define NUM_MAIN_STATS 21
+#define NUM_ALL_STATS  (NUM_MAIN_STATS + NUM_PORT_STATS + NUM_PKT_STATS + NUM_PERF_STATS)
+
+static u32 mlx4_en_get_msglevel(struct net_device *dev)
+{
+       return ((struct mlx4_en_priv *) netdev_priv(dev))->msg_enable;
+}
+
+static void mlx4_en_set_msglevel(struct net_device *dev, u32 val)
+{
+       ((struct mlx4_en_priv *) netdev_priv(dev))->msg_enable = val;
+}
+
+static void mlx4_en_get_wol(struct net_device *netdev,
+                           struct ethtool_wolinfo *wol)
+{
+       wol->supported = 0;
+       wol->wolopts = 0;
+
+       return;
+}
+
+static int mlx4_en_get_sset_count(struct net_device *dev, int sset)
+{
+       struct mlx4_en_priv *priv = netdev_priv(dev);
+
+       if (sset != ETH_SS_STATS)
+               return -EOPNOTSUPP;
+
+       return NUM_ALL_STATS + (priv->tx_ring_num + priv->rx_ring_num) * 2;
+}
+
+static void mlx4_en_get_ethtool_stats(struct net_device *dev,
+               struct ethtool_stats *stats, uint64_t *data)
+{
+       struct mlx4_en_priv *priv = netdev_priv(dev);
+       int index = 0;
+       int i;
+
+       spin_lock_bh(&priv->stats_lock);
+
+       mlx4_en_update_lro_stats(priv);
+
+       for (i = 0; i < NUM_MAIN_STATS; i++)
+               data[index++] = ((unsigned long *) &priv->stats)[i];
+       for (i = 0; i < NUM_PORT_STATS; i++)
+               data[index++] = ((unsigned long *) &priv->port_stats)[i];
+       for (i = 0; i < priv->tx_ring_num; i++) {
+               data[index++] = priv->tx_ring[i].packets;
+               data[index++] = priv->tx_ring[i].bytes;
+       }
+       for (i = 0; i < priv->rx_ring_num; i++) {
+               data[index++] = priv->rx_ring[i].packets;
+               data[index++] = priv->rx_ring[i].bytes;
+       }
+       for (i = 0; i < NUM_PKT_STATS; i++)
+               data[index++] = ((unsigned long *) &priv->pkstats)[i];
+       spin_unlock_bh(&priv->stats_lock);
+
+}
+
+static void mlx4_en_get_strings(struct net_device *dev,
+                               uint32_t stringset, uint8_t *data)
+{
+       struct mlx4_en_priv *priv = netdev_priv(dev);
+       int index = 0;
+       int i;
+
+       if (stringset != ETH_SS_STATS)
+               return;
+
+       /* Add main counters */
+       for (i = 0; i < NUM_MAIN_STATS; i++)
+               strcpy(data + (index++) * ETH_GSTRING_LEN, main_strings[i]);
+       for (i = 0; i < NUM_PORT_STATS; i++)
+               strcpy(data + (index++) * ETH_GSTRING_LEN,
+                       main_strings[i + NUM_MAIN_STATS]);
+       for (i = 0; i < priv->tx_ring_num; i++) {
+               sprintf(data + (index++) * ETH_GSTRING_LEN,
+                       "tx%d_packets", i);
+               sprintf(data + (index++) * ETH_GSTRING_LEN,
+                       "tx%d_bytes", i);
+       }
+       for (i = 0; i < priv->rx_ring_num; i++) {
+               sprintf(data + (index++) * ETH_GSTRING_LEN,
+                       "rx%d_packets", i);
+               sprintf(data + (index++) * ETH_GSTRING_LEN,
+                       "rx%d_bytes", i);
+       }
+       for (i = 0; i < NUM_PKT_STATS; i++)
+               strcpy(data + (index++) * ETH_GSTRING_LEN,
+                       main_strings[i + NUM_MAIN_STATS + NUM_PORT_STATS]);
+}
+
+static int mlx4_en_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
+{
+       cmd->autoneg = AUTONEG_DISABLE;
+       cmd->supported = SUPPORTED_10000baseT_Full;
+       cmd->advertising = SUPPORTED_10000baseT_Full;
+       if (netif_carrier_ok(dev)) {
+               cmd->speed = SPEED_10000;
+               cmd->duplex = DUPLEX_FULL;
+       } else {
+               cmd->speed = -1;
+               cmd->duplex = -1;
+       }
+       return 0;
+}
+
+static int mlx4_en_set_settings(struct net_device *dev, struct ethtool_cmd *cmd)
+{
+       if ((cmd->autoneg == AUTONEG_ENABLE) ||
+           (cmd->speed != SPEED_10000) || (cmd->duplex != DUPLEX_FULL))
+               return -EINVAL;
+
+       /* Nothing to change */
+       return 0;
+}
+
+static int mlx4_en_get_coalesce(struct net_device *dev,
+                             struct ethtool_coalesce *coal)
+{
+       struct mlx4_en_priv *priv = netdev_priv(dev);
+
+       coal->tx_coalesce_usecs = 0;
+       coal->tx_max_coalesced_frames = 0;
+       coal->rx_coalesce_usecs = priv->rx_usecs;
+       coal->rx_max_coalesced_frames = priv->rx_frames;
+
+       coal->pkt_rate_low = priv->pkt_rate_low;
+       coal->rx_coalesce_usecs_low = priv->rx_usecs_low;
+       coal->pkt_rate_high = priv->pkt_rate_high;
+       coal->rx_coalesce_usecs_high = priv->rx_usecs_high;
+       coal->rate_sample_interval = priv->sample_interval;
+       coal->use_adaptive_rx_coalesce = priv->adaptive_rx_coal;
+       return 0;
+}
+
+static int mlx4_en_set_coalesce(struct net_device *dev,
+                             struct ethtool_coalesce *coal)
+{
+       struct mlx4_en_priv *priv = netdev_priv(dev);
+       int err, i;
+
+       priv->rx_frames = (coal->rx_max_coalesced_frames ==
+                          MLX4_EN_AUTO_CONF) ?
+                               MLX4_EN_RX_COAL_TARGET /
+                               priv->dev->mtu + 1 :
+                               coal->rx_max_coalesced_frames;
+       priv->rx_usecs = (coal->rx_coalesce_usecs ==
+                         MLX4_EN_AUTO_CONF) ?
+                               MLX4_EN_RX_COAL_TIME :
+                               coal->rx_coalesce_usecs;
+
+       /* Set adaptive coalescing params */
+       priv->pkt_rate_low = coal->pkt_rate_low;
+       priv->rx_usecs_low = coal->rx_coalesce_usecs_low;
+       priv->pkt_rate_high = coal->pkt_rate_high;
+       priv->rx_usecs_high = coal->rx_coalesce_usecs_high;
+       priv->sample_interval = coal->rate_sample_interval;
+       priv->adaptive_rx_coal = coal->use_adaptive_rx_coalesce;
+       priv->last_moder_time = MLX4_EN_AUTO_CONF;
+       if (priv->adaptive_rx_coal)
+               return 0;
+
+       for (i = 0; i < priv->rx_ring_num; i++) {
+               priv->rx_cq[i].moder_cnt = priv->rx_frames;
+               priv->rx_cq[i].moder_time = priv->rx_usecs;
+               err = mlx4_en_set_cq_moder(priv, &priv->rx_cq[i]);
+               if (err)
+                       return err;
+       }
+       return 0;
+}
+
+static int mlx4_en_set_pauseparam(struct net_device *dev,
+                               struct ethtool_pauseparam *pause)
+{
+       struct mlx4_en_priv *priv = netdev_priv(dev);
+       struct mlx4_en_dev *mdev = priv->mdev;
+       int err;
+
+       mdev->profile.tx_pause = pause->tx_pause != 0;
+       mdev->profile.rx_pause = pause->rx_pause != 0;
+       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 pause params to\n");
+
+       return err;
+}
+
+static void mlx4_en_get_pauseparam(struct net_device *dev,
+                                struct ethtool_pauseparam *pause)
+{
+       struct mlx4_en_priv *priv = netdev_priv(dev);
+       struct mlx4_en_dev *mdev = priv->mdev;
+
+       pause->tx_pause = mdev->profile.tx_pause;
+       pause->rx_pause = mdev->profile.rx_pause;
+}
+
+static void mlx4_en_get_ringparam(struct net_device *dev,
+                                 struct ethtool_ringparam *param)
+{
+       struct mlx4_en_priv *priv = netdev_priv(dev);
+       struct mlx4_en_dev *mdev = priv->mdev;
+
+       memset(param, 0, sizeof(*param));
+       param->rx_max_pending = mdev->dev->caps.max_rq_sg;
+       param->tx_max_pending = mdev->dev->caps.max_sq_sg;
+       param->rx_pending = mdev->profile.prof[priv->port].rx_ring_size;
+       param->tx_pending = mdev->profile.prof[priv->port].tx_ring_size;
+}
+
+const struct ethtool_ops mlx4_en_ethtool_ops = {
+       .get_drvinfo = mlx4_en_get_drvinfo,
+       .get_settings = mlx4_en_get_settings,
+       .set_settings = mlx4_en_set_settings,
+#ifdef NETIF_F_TSO
+       .get_tso = mlx4_en_get_tso,
+       .set_tso = mlx4_en_set_tso,
+#endif
+       .get_sg = ethtool_op_get_sg,
+       .set_sg = ethtool_op_set_sg,
+       .get_link = ethtool_op_get_link,
+       .get_rx_csum = mlx4_en_get_rx_csum,
+       .set_rx_csum = mlx4_en_set_rx_csum,
+       .get_tx_csum = ethtool_op_get_tx_csum,
+       .set_tx_csum = ethtool_op_set_tx_ipv6_csum,
+       .get_strings = mlx4_en_get_strings,
+       .get_sset_count = mlx4_en_get_sset_count,
+       .get_ethtool_stats = mlx4_en_get_ethtool_stats,
+       .get_wol = mlx4_en_get_wol,
+       .get_msglevel = mlx4_en_get_msglevel,
+       .set_msglevel = mlx4_en_set_msglevel,
+       .get_coalesce = mlx4_en_get_coalesce,
+       .set_coalesce = mlx4_en_set_coalesce,
+       .get_pauseparam = mlx4_en_get_pauseparam,
+       .set_pauseparam = mlx4_en_set_pauseparam,
+       .get_ringparam = mlx4_en_get_ringparam,
+       .get_flags = ethtool_op_get_flags,
+       .set_flags = ethtool_op_set_flags,
+};
+
+
+
+
+
diff --git a/drivers/net/mlx4/en_port.c b/drivers/net/mlx4/en_port.c
new file mode 100644 (file)
index 0000000..c5a4c03
--- /dev/null
@@ -0,0 +1,261 @@
+/*
+ * 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/if_vlan.h>
+
+#include <linux/mlx4/device.h>
+#include <linux/mlx4/cmd.h>
+
+#include "en_port.h"
+#include "mlx4_en.h"
+
+
+int mlx4_SET_MCAST_FLTR(struct mlx4_dev *dev, u8 port,
+                       u64 mac, u64 clear, u8 mode)
+{
+       return mlx4_cmd(dev, (mac | (clear << 63)), port, mode,
+                       MLX4_CMD_SET_MCAST_FLTR, MLX4_CMD_TIME_CLASS_B);
+}
+
+int mlx4_SET_VLAN_FLTR(struct mlx4_dev *dev, u8 port, struct vlan_group *grp)
+{
+       struct mlx4_cmd_mailbox *mailbox;
+       struct mlx4_set_vlan_fltr_mbox *filter;
+       int i;
+       int j;
+       int index = 0;
+       u32 entry;
+       int err = 0;
+
+       mailbox = mlx4_alloc_cmd_mailbox(dev);
+       if (IS_ERR(mailbox))
+               return PTR_ERR(mailbox);
+
+       filter = mailbox->buf;
+       if (grp) {
+               memset(filter, 0, sizeof *filter);
+               for (i = VLAN_FLTR_SIZE - 1; i >= 0; i--) {
+                       entry = 0;
+                       for (j = 0; j < 32; j++)
+                               if (vlan_group_get_device(grp, index++))
+                                       entry |= 1 << j;
+                       filter->entry[i] = cpu_to_be32(entry);
+               }
+       } else {
+               /* When no vlans are configured we block all vlans */
+               memset(filter, 0, sizeof(*filter));
+       }
+       err = mlx4_cmd(dev, mailbox->dma, port, 0, MLX4_CMD_SET_VLAN_FLTR,
+                      MLX4_CMD_TIME_CLASS_B);
+       mlx4_free_cmd_mailbox(dev, mailbox);
+       return err;
+}
+
+
+int mlx4_SET_PORT_general(struct mlx4_dev *dev, u8 port, int mtu,
+                         u8 pptx, u8 pfctx, u8 pprx, u8 pfcrx)
+{
+       struct mlx4_cmd_mailbox *mailbox;
+       struct mlx4_set_port_general_context *context;
+       int err;
+       u32 in_mod;
+
+       mailbox = mlx4_alloc_cmd_mailbox(dev);
+       if (IS_ERR(mailbox))
+               return PTR_ERR(mailbox);
+       context = mailbox->buf;
+       memset(context, 0, sizeof *context);
+
+       context->flags = SET_PORT_GEN_ALL_VALID;
+       context->mtu = cpu_to_be16(mtu);
+       context->pptx = (pptx * (!pfctx)) << 7;
+       context->pfctx = pfctx;
+       context->pprx = (pprx * (!pfcrx)) << 7;
+       context->pfcrx = pfcrx;
+
+       in_mod = MLX4_SET_PORT_GENERAL << 8 | port;
+       err = mlx4_cmd(dev, mailbox->dma, in_mod, 1, MLX4_CMD_SET_PORT,
+                      MLX4_CMD_TIME_CLASS_B);
+
+       mlx4_free_cmd_mailbox(dev, mailbox);
+       return err;
+}
+
+int mlx4_SET_PORT_qpn_calc(struct mlx4_dev *dev, u8 port, u32 base_qpn,
+                          u8 promisc)
+{
+       struct mlx4_cmd_mailbox *mailbox;
+       struct mlx4_set_port_rqp_calc_context *context;
+       int err;
+       u32 in_mod;
+
+       mailbox = mlx4_alloc_cmd_mailbox(dev);
+       if (IS_ERR(mailbox))
+               return PTR_ERR(mailbox);
+       context = mailbox->buf;
+       memset(context, 0, sizeof *context);
+
+       context->base_qpn = cpu_to_be32(base_qpn);
+       context->promisc = cpu_to_be32(promisc << SET_PORT_PROMISC_SHIFT | base_qpn);
+       context->mcast = cpu_to_be32(1 << SET_PORT_PROMISC_SHIFT | base_qpn);
+       context->intra_no_vlan = 0;
+       context->no_vlan = MLX4_NO_VLAN_IDX;
+       context->intra_vlan_miss = 0;
+       context->vlan_miss = MLX4_VLAN_MISS_IDX;
+
+       in_mod = MLX4_SET_PORT_RQP_CALC << 8 | port;
+       err = mlx4_cmd(dev, mailbox->dma, in_mod, 1, MLX4_CMD_SET_PORT,
+                      MLX4_CMD_TIME_CLASS_B);
+
+       mlx4_free_cmd_mailbox(dev, mailbox);
+       return err;
+}
+
+
+int mlx4_en_DUMP_ETH_STATS(struct mlx4_en_dev *mdev, u8 port, u8 reset)
+{
+       struct mlx4_en_stat_out_mbox *mlx4_en_stats;
+       struct mlx4_en_priv *priv = netdev_priv(mdev->pndev[port]);
+       struct net_device_stats *stats = &priv->stats;
+       struct mlx4_cmd_mailbox *mailbox;
+       u64 in_mod = reset << 8 | port;
+       int err;
+
+       mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
+       if (IS_ERR(mailbox))
+               return PTR_ERR(mailbox);
+       memset(mailbox->buf, 0, sizeof(*mlx4_en_stats));
+       err = mlx4_cmd_box(mdev->dev, 0, mailbox->dma, in_mod, 0,
+                          MLX4_CMD_DUMP_ETH_STATS, MLX4_CMD_TIME_CLASS_B);
+       if (err)
+               goto out;
+
+       mlx4_en_stats = mailbox->buf;
+
+       spin_lock_bh(&priv->stats_lock);
+
+       stats->rx_packets = be32_to_cpu(mlx4_en_stats->RTOTFRMS) -
+                           be32_to_cpu(mlx4_en_stats->RDROP);
+       stats->tx_packets = be64_to_cpu(mlx4_en_stats->TTOT_prio_0) +
+                           be64_to_cpu(mlx4_en_stats->TTOT_prio_1) +
+                           be64_to_cpu(mlx4_en_stats->TTOT_prio_2) +
+                           be64_to_cpu(mlx4_en_stats->TTOT_prio_3) +
+                           be64_to_cpu(mlx4_en_stats->TTOT_prio_4) +
+                           be64_to_cpu(mlx4_en_stats->TTOT_prio_5) +
+                           be64_to_cpu(mlx4_en_stats->TTOT_prio_6) +
+                           be64_to_cpu(mlx4_en_stats->TTOT_prio_7) +
+                           be64_to_cpu(mlx4_en_stats->TTOT_novlan) +
+                           be64_to_cpu(mlx4_en_stats->TTOT_loopbk);
+       stats->rx_bytes = be64_to_cpu(mlx4_en_stats->ROCT_prio_0) +
+                         be64_to_cpu(mlx4_en_stats->ROCT_prio_1) +
+                         be64_to_cpu(mlx4_en_stats->ROCT_prio_2) +
+                         be64_to_cpu(mlx4_en_stats->ROCT_prio_3) +
+                         be64_to_cpu(mlx4_en_stats->ROCT_prio_4) +
+                         be64_to_cpu(mlx4_en_stats->ROCT_prio_5) +
+                         be64_to_cpu(mlx4_en_stats->ROCT_prio_6) +
+                         be64_to_cpu(mlx4_en_stats->ROCT_prio_7) +
+                         be64_to_cpu(mlx4_en_stats->ROCT_novlan);
+
+       stats->tx_bytes = be64_to_cpu(mlx4_en_stats->TTTLOCT_prio_0) +
+                         be64_to_cpu(mlx4_en_stats->TTTLOCT_prio_1) +
+                         be64_to_cpu(mlx4_en_stats->TTTLOCT_prio_2) +
+                         be64_to_cpu(mlx4_en_stats->TTTLOCT_prio_3) +
+                         be64_to_cpu(mlx4_en_stats->TTTLOCT_prio_4) +
+                         be64_to_cpu(mlx4_en_stats->TTTLOCT_prio_5) +
+                         be64_to_cpu(mlx4_en_stats->TTTLOCT_prio_6) +
+                         be64_to_cpu(mlx4_en_stats->TTTLOCT_prio_7) +
+                         be64_to_cpu(mlx4_en_stats->TTTLOCT_novlan) +
+                         be64_to_cpu(mlx4_en_stats->TTTLOCT_loopbk);
+
+       stats->rx_errors = be64_to_cpu(mlx4_en_stats->PCS) +
+                          be32_to_cpu(mlx4_en_stats->RdropLength) +
+                          be32_to_cpu(mlx4_en_stats->RJBBR) +
+                          be32_to_cpu(mlx4_en_stats->RCRC) +
+                          be32_to_cpu(mlx4_en_stats->RRUNT);
+       stats->tx_errors = be32_to_cpu(mlx4_en_stats->TDROP);
+       stats->multicast = be64_to_cpu(mlx4_en_stats->MCAST_prio_0) +
+                          be64_to_cpu(mlx4_en_stats->MCAST_prio_1) +
+                          be64_to_cpu(mlx4_en_stats->MCAST_prio_2) +
+                          be64_to_cpu(mlx4_en_stats->MCAST_prio_3) +
+                          be64_to_cpu(mlx4_en_stats->MCAST_prio_4) +
+                          be64_to_cpu(mlx4_en_stats->MCAST_prio_5) +
+                          be64_to_cpu(mlx4_en_stats->MCAST_prio_6) +
+                          be64_to_cpu(mlx4_en_stats->MCAST_prio_7) +
+                          be64_to_cpu(mlx4_en_stats->MCAST_novlan);
+       stats->collisions = 0;
+       stats->rx_length_errors = be32_to_cpu(mlx4_en_stats->RdropLength);
+       stats->rx_over_errors = be32_to_cpu(mlx4_en_stats->RdropOvflw);
+       stats->rx_crc_errors = be32_to_cpu(mlx4_en_stats->RCRC);
+       stats->rx_frame_errors = 0;
+       stats->rx_fifo_errors = be32_to_cpu(mlx4_en_stats->RdropOvflw);
+       stats->rx_missed_errors = be32_to_cpu(mlx4_en_stats->RdropOvflw);
+       stats->tx_aborted_errors = 0;
+       stats->tx_carrier_errors = 0;
+       stats->tx_fifo_errors = 0;
+       stats->tx_heartbeat_errors = 0;
+       stats->tx_window_errors = 0;
+
+       priv->pkstats.broadcast =
+                               be64_to_cpu(mlx4_en_stats->RBCAST_prio_0) +
+                               be64_to_cpu(mlx4_en_stats->RBCAST_prio_1) +
+                               be64_to_cpu(mlx4_en_stats->RBCAST_prio_2) +
+                               be64_to_cpu(mlx4_en_stats->RBCAST_prio_3) +
+                               be64_to_cpu(mlx4_en_stats->RBCAST_prio_4) +
+                               be64_to_cpu(mlx4_en_stats->RBCAST_prio_5) +
+                               be64_to_cpu(mlx4_en_stats->RBCAST_prio_6) +
+                               be64_to_cpu(mlx4_en_stats->RBCAST_prio_7) +
+                               be64_to_cpu(mlx4_en_stats->RBCAST_novlan);
+       priv->pkstats.rx_prio[0] = be64_to_cpu(mlx4_en_stats->RTOT_prio_0);
+       priv->pkstats.rx_prio[1] = be64_to_cpu(mlx4_en_stats->RTOT_prio_1);
+       priv->pkstats.rx_prio[2] = be64_to_cpu(mlx4_en_stats->RTOT_prio_2);
+       priv->pkstats.rx_prio[3] = be64_to_cpu(mlx4_en_stats->RTOT_prio_3);
+       priv->pkstats.rx_prio[4] = be64_to_cpu(mlx4_en_stats->RTOT_prio_4);
+       priv->pkstats.rx_prio[5] = be64_to_cpu(mlx4_en_stats->RTOT_prio_5);
+       priv->pkstats.rx_prio[6] = be64_to_cpu(mlx4_en_stats->RTOT_prio_6);
+       priv->pkstats.rx_prio[7] = be64_to_cpu(mlx4_en_stats->RTOT_prio_7);
+       priv->pkstats.tx_prio[0] = be64_to_cpu(mlx4_en_stats->TTOT_prio_0);
+       priv->pkstats.tx_prio[1] = be64_to_cpu(mlx4_en_stats->TTOT_prio_1);
+       priv->pkstats.tx_prio[2] = be64_to_cpu(mlx4_en_stats->TTOT_prio_2);
+       priv->pkstats.tx_prio[3] = be64_to_cpu(mlx4_en_stats->TTOT_prio_3);
+       priv->pkstats.tx_prio[4] = be64_to_cpu(mlx4_en_stats->TTOT_prio_4);
+       priv->pkstats.tx_prio[5] = be64_to_cpu(mlx4_en_stats->TTOT_prio_5);
+       priv->pkstats.tx_prio[6] = be64_to_cpu(mlx4_en_stats->TTOT_prio_6);
+       priv->pkstats.tx_prio[7] = be64_to_cpu(mlx4_en_stats->TTOT_prio_7);
+       spin_unlock_bh(&priv->stats_lock);
+
+out:
+       mlx4_free_cmd_mailbox(mdev->dev, mailbox);
+       return err;
+}
+
diff --git a/drivers/net/mlx4/en_port.h b/drivers/net/mlx4/en_port.h
new file mode 100644 (file)
index 0000000..e6477f1
--- /dev/null
@@ -0,0 +1,570 @@
+/*
+ * 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.
+ *
+ */
+
+#ifndef _MLX4_EN_PORT_H_
+#define _MLX4_EN_PORT_H_
+
+
+#define SET_PORT_GEN_ALL_VALID 0x7
+#define SET_PORT_PROMISC_SHIFT 31
+
+enum {
+       MLX4_CMD_SET_VLAN_FLTR  = 0x47,
+       MLX4_CMD_SET_MCAST_FLTR = 0x48,
+       MLX4_CMD_DUMP_ETH_STATS = 0x49,
+};
+
+struct mlx4_set_port_general_context {
+       u8 reserved[3];
+       u8 flags;
+       u16 reserved2;
+       __be16 mtu;
+       u8 pptx;
+       u8 pfctx;
+       u16 reserved3;
+       u8 pprx;
+       u8 pfcrx;
+       u16 reserved4;
+};
+
+struct mlx4_set_port_rqp_calc_context {
+       __be32 base_qpn;
+       __be32 flags;
+       u8 reserved[3];
+       u8 mac_miss;
+       u8 intra_no_vlan;
+       u8 no_vlan;
+       u8 intra_vlan_miss;
+       u8 vlan_miss;
+       u8 reserved2[3];
+       u8 no_vlan_prio;
+       __be32 promisc;
+       __be32 mcast;
+};
+
+#define VLAN_FLTR_SIZE 128
+struct mlx4_set_vlan_fltr_mbox {
+       __be32 entry[VLAN_FLTR_SIZE];
+};
+
+
+enum {
+       MLX4_MCAST_CONFIG       = 0,
+       MLX4_MCAST_DISABLE      = 1,
+       MLX4_MCAST_ENABLE       = 2,
+};
+
+
+struct mlx4_en_stat_out_mbox {
+       /* Received frames with a length of 64 octets */
+       __be64 R64_prio_0;
+       __be64 R64_prio_1;
+       __be64 R64_prio_2;
+       __be64 R64_prio_3;
+       __be64 R64_prio_4;
+       __be64 R64_prio_5;
+       __be64 R64_prio_6;
+       __be64 R64_prio_7;
+       __be64 R64_novlan;
+       /* Received frames with a length of 127 octets */
+       __be64 R127_prio_0;
+       __be64 R127_prio_1;
+       __be64 R127_prio_2;
+       __be64 R127_prio_3;
+       __be64 R127_prio_4;
+       __be64 R127_prio_5;
+       __be64 R127_prio_6;
+       __be64 R127_prio_7;
+       __be64 R127_novlan;
+       /* Received frames with a length of 255 octets */
+       __be64 R255_prio_0;
+       __be64 R255_prio_1;
+       __be64 R255_prio_2;
+       __be64 R255_prio_3;
+       __be64 R255_prio_4;
+       __be64 R255_prio_5;
+       __be64 R255_prio_6;
+       __be64 R255_prio_7;
+       __be64 R255_novlan;
+       /* Received frames with a length of 511 octets */
+       __be64 R511_prio_0;
+       __be64 R511_prio_1;
+       __be64 R511_prio_2;
+       __be64 R511_prio_3;
+       __be64 R511_prio_4;
+       __be64 R511_prio_5;
+       __be64 R511_prio_6;
+       __be64 R511_prio_7;
+       __be64 R511_novlan;
+       /* Received frames with a length of 1023 octets */
+       __be64 R1023_prio_0;
+       __be64 R1023_prio_1;
+       __be64 R1023_prio_2;
+       __be64 R1023_prio_3;
+       __be64 R1023_prio_4;
+       __be64 R1023_prio_5;
+       __be64 R1023_prio_6;
+       __be64 R1023_prio_7;
+       __be64 R1023_novlan;
+       /* Received frames with a length of 1518 octets */
+       __be64 R1518_prio_0;
+       __be64 R1518_prio_1;
+       __be64 R1518_prio_2;
+       __be64 R1518_prio_3;
+       __be64 R1518_prio_4;
+       __be64 R1518_prio_5;
+       __be64 R1518_prio_6;
+       __be64 R1518_prio_7;
+       __be64 R1518_novlan;
+       /* Received frames with a length of 1522 octets */
+       __be64 R1522_prio_0;
+       __be64 R1522_prio_1;
+       __be64 R1522_prio_2;
+       __be64 R1522_prio_3;
+       __be64 R1522_prio_4;
+       __be64 R1522_prio_5;
+       __be64 R1522_prio_6;
+       __be64 R1522_prio_7;
+       __be64 R1522_novlan;
+       /* Received frames with a length of 1548 octets */
+       __be64 R1548_prio_0;
+       __be64 R1548_prio_1;
+       __be64 R1548_prio_2;
+       __be64 R1548_prio_3;
+       __be64 R1548_prio_4;
+       __be64 R1548_prio_5;
+       __be64 R1548_prio_6;
+       __be64 R1548_prio_7;
+       __be64 R1548_novlan;
+       /* Received frames with a length of 1548 < octets < MTU */
+       __be64 R2MTU_prio_0;
+       __be64 R2MTU_prio_1;
+       __be64 R2MTU_prio_2;
+       __be64 R2MTU_prio_3;
+       __be64 R2MTU_prio_4;
+       __be64 R2MTU_prio_5;
+       __be64 R2MTU_prio_6;
+       __be64 R2MTU_prio_7;
+       __be64 R2MTU_novlan;
+       /* Received frames with a length of MTU< octets and good CRC */
+       __be64 RGIANT_prio_0;
+       __be64 RGIANT_prio_1;
+       __be64 RGIANT_prio_2;
+       __be64 RGIANT_prio_3;
+       __be64 RGIANT_prio_4;
+       __be64 RGIANT_prio_5;
+       __be64 RGIANT_prio_6;
+       __be64 RGIANT_prio_7;
+       __be64 RGIANT_novlan;
+       /* Received broadcast frames with good CRC */
+       __be64 RBCAST_prio_0;
+       __be64 RBCAST_prio_1;
+       __be64 RBCAST_prio_2;
+       __be64 RBCAST_prio_3;
+       __be64 RBCAST_prio_4;
+       __be64 RBCAST_prio_5;
+       __be64 RBCAST_prio_6;
+       __be64 RBCAST_prio_7;
+       __be64 RBCAST_novlan;
+       /* Received multicast frames with good CRC */
+       __be64 MCAST_prio_0;
+       __be64 MCAST_prio_1;
+       __be64 MCAST_prio_2;
+       __be64 MCAST_prio_3;
+       __be64 MCAST_prio_4;
+       __be64 MCAST_prio_5;
+       __be64 MCAST_prio_6;
+       __be64 MCAST_prio_7;
+       __be64 MCAST_novlan;
+       /* Received unicast not short or GIANT frames with good CRC */
+       __be64 RTOTG_prio_0;
+       __be64 RTOTG_prio_1;
+       __be64 RTOTG_prio_2;
+       __be64 RTOTG_prio_3;
+       __be64 RTOTG_prio_4;
+       __be64 RTOTG_prio_5;
+       __be64 RTOTG_prio_6;
+       __be64 RTOTG_prio_7;
+       __be64 RTOTG_novlan;
+
+       /* Count of total octets of received frames, includes framing characters */
+       __be64 RTTLOCT_prio_0;
+       /* Count of total octets of received frames, not including framing
+          characters */
+       __be64 RTTLOCT_NOFRM_prio_0;
+       /* Count of Total number of octets received
+          (only for frames without errors) */
+       __be64 ROCT_prio_0;
+
+       __be64 RTTLOCT_prio_1;
+       __be64 RTTLOCT_NOFRM_prio_1;
+       __be64 ROCT_prio_1;
+
+       __be64 RTTLOCT_prio_2;
+       __be64 RTTLOCT_NOFRM_prio_2;
+       __be64 ROCT_prio_2;
+
+       __be64 RTTLOCT_prio_3;
+       __be64 RTTLOCT_NOFRM_prio_3;
+       __be64 ROCT_prio_3;
+
+       __be64 RTTLOCT_prio_4;
+       __be64 RTTLOCT_NOFRM_prio_4;
+       __be64 ROCT_prio_4;
+
+       __be64 RTTLOCT_prio_5;
+       __be64 RTTLOCT_NOFRM_prio_5;
+       __be64 ROCT_prio_5;
+
+       __be64 RTTLOCT_prio_6;
+       __be64 RTTLOCT_NOFRM_prio_6;
+       __be64 ROCT_prio_6;
+
+       __be64 RTTLOCT_prio_7;
+       __be64 RTTLOCT_NOFRM_prio_7;
+       __be64 ROCT_prio_7;
+
+       __be64 RTTLOCT_novlan;
+       __be64 RTTLOCT_NOFRM_novlan;
+       __be64 ROCT_novlan;
+
+       /* Count of Total received frames including bad frames */
+       __be64 RTOT_prio_0;
+       /* Count of  Total number of received frames with 802.1Q encapsulation */
+       __be64 R1Q_prio_0;
+       __be64 reserved1;
+
+       __be64 RTOT_prio_1;
+       __be64 R1Q_prio_1;
+       __be64 reserved2;
+
+       __be64 RTOT_prio_2;
+       __be64 R1Q_prio_2;
+       __be64 reserved3;
+
+       __be64 RTOT_prio_3;
+       __be64 R1Q_prio_3;
+       __be64 reserved4;
+
+       __be64 RTOT_prio_4;
+       __be64 R1Q_prio_4;
+       __be64 reserved5;
+
+       __be64 RTOT_prio_5;
+       __be64 R1Q_prio_5;
+       __be64 reserved6;
+
+       __be64 RTOT_prio_6;
+       __be64 R1Q_prio_6;
+       __be64 reserved7;
+
+       __be64 RTOT_prio_7;
+       __be64 R1Q_prio_7;
+       __be64 reserved8;
+
+       __be64 RTOT_novlan;
+       __be64 R1Q_novlan;
+       __be64 reserved9;
+
+       /* Total number of Successfully Received Control Frames */
+       __be64 RCNTL;
+       __be64 reserved10;
+       __be64 reserved11;
+       __be64 reserved12;
+       /* Count of received frames with a length/type field  value between 46
+          (42 for VLANtagged frames) and 1500 (also 1500 for VLAN-tagged frames),
+          inclusive */
+       __be64 RInRangeLengthErr;
+       /* Count of received frames with length/type field between 1501 and 1535
+          decimal, inclusive */
+       __be64 ROutRangeLengthErr;
+       /* Count of received frames that are longer than max allowed size for
+          802.3 frames (1518/1522) */
+       __be64 RFrmTooLong;
+       /* Count frames received with PCS error */
+       __be64 PCS;
+
+       /* Transmit frames with a length of 64 octets */
+       __be64 T64_prio_0;
+       __be64 T64_prio_1;
+       __be64 T64_prio_2;
+       __be64 T64_prio_3;
+       __be64 T64_prio_4;
+       __be64 T64_prio_5;
+       __be64 T64_prio_6;
+       __be64 T64_prio_7;
+       __be64 T64_novlan;
+       __be64 T64_loopbk;
+       /* Transmit frames with a length of 65 to 127 octets. */
+       __be64 T127_prio_0;
+       __be64 T127_prio_1;
+       __be64 T127_prio_2;
+       __be64 T127_prio_3;
+       __be64 T127_prio_4;
+       __be64 T127_prio_5;
+       __be64 T127_prio_6;
+       __be64 T127_prio_7;
+       __be64 T127_novlan;
+       __be64 T127_loopbk;
+       /* Transmit frames with a length of 128 to 255 octets */
+       __be64 T255_prio_0;
+       __be64 T255_prio_1;
+       __be64 T255_prio_2;
+       __be64 T255_prio_3;
+       __be64 T255_prio_4;
+       __be64 T255_prio_5;
+       __be64 T255_prio_6;
+       __be64 T255_prio_7;
+       __be64 T255_novlan;
+       __be64 T255_loopbk;
+       /* Transmit frames with a length of 256 to 511 octets */
+       __be64 T511_prio_0;
+       __be64 T511_prio_1;
+       __be64 T511_prio_2;
+       __be64 T511_prio_3;
+       __be64 T511_prio_4;
+       __be64 T511_prio_5;
+       __be64 T511_prio_6;
+       __be64 T511_prio_7;
+       __be64 T511_novlan;
+       __be64 T511_loopbk;
+       /* Transmit frames with a length of 512 to 1023 octets */
+       __be64 T1023_prio_0;
+       __be64 T1023_prio_1;
+       __be64 T1023_prio_2;
+       __be64 T1023_prio_3;
+       __be64 T1023_prio_4;
+       __be64 T1023_prio_5;
+       __be64 T1023_prio_6;
+       __be64 T1023_prio_7;
+       __be64 T1023_novlan;
+       __be64 T1023_loopbk;
+       /* Transmit frames with a length of 1024 to 1518 octets */
+       __be64 T1518_prio_0;
+       __be64 T1518_prio_1;
+       __be64 T1518_prio_2;
+       __be64 T1518_prio_3;
+       __be64 T1518_prio_4;
+       __be64 T1518_prio_5;
+       __be64 T1518_prio_6;
+       __be64 T1518_prio_7;
+       __be64 T1518_novlan;
+       __be64 T1518_loopbk;
+       /* Counts transmit frames with a length of 1519 to 1522 bytes */
+       __be64 T1522_prio_0;
+       __be64 T1522_prio_1;
+       __be64 T1522_prio_2;
+       __be64 T1522_prio_3;
+       __be64 T1522_prio_4;
+       __be64 T1522_prio_5;
+       __be64 T1522_prio_6;
+       __be64 T1522_prio_7;
+       __be64 T1522_novlan;
+       __be64 T1522_loopbk;
+       /* Transmit frames with a length of 1523 to 1548 octets */
+       __be64 T1548_prio_0;
+       __be64 T1548_prio_1;
+       __be64 T1548_prio_2;
+       __be64 T1548_prio_3;
+       __be64 T1548_prio_4;
+       __be64 T1548_prio_5;
+       __be64 T1548_prio_6;
+       __be64 T1548_prio_7;
+       __be64 T1548_novlan;
+       __be64 T1548_loopbk;
+       /* Counts transmit frames with a length of 1549 to MTU bytes */
+       __be64 T2MTU_prio_0;
+       __be64 T2MTU_prio_1;
+       __be64 T2MTU_prio_2;
+       __be64 T2MTU_prio_3;
+       __be64 T2MTU_prio_4;
+       __be64 T2MTU_prio_5;
+       __be64 T2MTU_prio_6;
+       __be64 T2MTU_prio_7;
+       __be64 T2MTU_novlan;
+       __be64 T2MTU_loopbk;
+       /* Transmit frames with a length greater than MTU octets and a good CRC. */
+       __be64 TGIANT_prio_0;
+       __be64 TGIANT_prio_1;
+       __be64 TGIANT_prio_2;
+       __be64 TGIANT_prio_3;
+       __be64 TGIANT_prio_4;
+       __be64 TGIANT_prio_5;
+       __be64 TGIANT_prio_6;
+       __be64 TGIANT_prio_7;
+       __be64 TGIANT_novlan;
+       __be64 TGIANT_loopbk;
+       /* Transmit broadcast frames with a good CRC */
+       __be64 TBCAST_prio_0;
+       __be64 TBCAST_prio_1;
+       __be64 TBCAST_prio_2;
+       __be64 TBCAST_prio_3;
+       __be64 TBCAST_prio_4;
+       __be64 TBCAST_prio_5;
+       __be64 TBCAST_prio_6;
+       __be64 TBCAST_prio_7;
+       __be64 TBCAST_novlan;
+       __be64 TBCAST_loopbk;
+       /* Transmit multicast frames with a good CRC */
+       __be64 TMCAST_prio_0;
+       __be64 TMCAST_prio_1;
+       __be64 TMCAST_prio_2;
+       __be64 TMCAST_prio_3;
+       __be64 TMCAST_prio_4;
+       __be64 TMCAST_prio_5;
+       __be64 TMCAST_prio_6;
+       __be64 TMCAST_prio_7;
+       __be64 TMCAST_novlan;
+       __be64 TMCAST_loopbk;
+       /* Transmit good frames that are neither broadcast nor multicast */
+       __be64 TTOTG_prio_0;
+       __be64 TTOTG_prio_1;
+       __be64 TTOTG_prio_2;
+       __be64 TTOTG_prio_3;
+       __be64 TTOTG_prio_4;
+       __be64 TTOTG_prio_5;
+       __be64 TTOTG_prio_6;
+       __be64 TTOTG_prio_7;
+       __be64 TTOTG_novlan;
+       __be64 TTOTG_loopbk;
+
+       /* total octets of transmitted frames, including framing characters */
+       __be64 TTTLOCT_prio_0;
+       /* total octets of transmitted frames, not including framing characters */
+       __be64 TTTLOCT_NOFRM_prio_0;
+       /* ifOutOctets */
+       __be64 TOCT_prio_0;
+
+       __be64 TTTLOCT_prio_1;
+       __be64 TTTLOCT_NOFRM_prio_1;
+       __be64 TOCT_prio_1;
+
+       __be64 TTTLOCT_prio_2;
+       __be64 TTTLOCT_NOFRM_prio_2;
+       __be64 TOCT_prio_2;
+
+       __be64 TTTLOCT_prio_3;
+       __be64 TTTLOCT_NOFRM_prio_3;
+       __be64 TOCT_prio_3;
+
+       __be64 TTTLOCT_prio_4;
+       __be64 TTTLOCT_NOFRM_prio_4;
+       __be64 TOCT_prio_4;
+
+       __be64 TTTLOCT_prio_5;
+       __be64 TTTLOCT_NOFRM_prio_5;
+       __be64 TOCT_prio_5;
+
+       __be64 TTTLOCT_prio_6;
+       __be64 TTTLOCT_NOFRM_prio_6;
+       __be64 TOCT_prio_6;
+
+       __be64 TTTLOCT_prio_7;
+       __be64 TTTLOCT_NOFRM_prio_7;
+       __be64 TOCT_prio_7;
+
+       __be64 TTTLOCT_novlan;
+       __be64 TTTLOCT_NOFRM_novlan;
+       __be64 TOCT_novlan;
+
+       __be64 TTTLOCT_loopbk;
+       __be64 TTTLOCT_NOFRM_loopbk;
+       __be64 TOCT_loopbk;
+
+       /* Total frames transmitted with a good CRC that are not aborted  */
+       __be64 TTOT_prio_0;
+       /* Total number of frames transmitted with 802.1Q encapsulation */
+       __be64 T1Q_prio_0;
+       __be64 reserved13;
+
+       __be64 TTOT_prio_1;
+       __be64 T1Q_prio_1;
+       __be64 reserved14;
+
+       __be64 TTOT_prio_2;
+       __be64 T1Q_prio_2;
+       __be64 reserved15;
+
+       __be64 TTOT_prio_3;
+       __be64 T1Q_prio_3;
+       __be64 reserved16;
+
+       __be64 TTOT_prio_4;
+       __be64 T1Q_prio_4;
+       __be64 reserved17;
+
+       __be64 TTOT_prio_5;
+       __be64 T1Q_prio_5;
+       __be64 reserved18;
+
+       __be64 TTOT_prio_6;
+       __be64 T1Q_prio_6;
+       __be64 reserved19;
+
+       __be64 TTOT_prio_7;
+       __be64 T1Q_prio_7;
+       __be64 reserved20;
+
+       __be64 TTOT_novlan;
+       __be64 T1Q_novlan;
+       __be64 reserved21;
+
+       __be64 TTOT_loopbk;
+       __be64 T1Q_loopbk;
+       __be64 reserved22;
+
+       /* Received frames with a length greater than MTU octets and a bad CRC */
+       __be32 RJBBR;
+       /* Received frames with a bad CRC that are not runts, jabbers,
+          or alignment errors */
+       __be32 RCRC;
+       /* Received frames with SFD with a length of less than 64 octets and a
+          bad CRC */
+       __be32 RRUNT;
+       /* Received frames with a length less than 64 octets and a good CRC */
+       __be32 RSHORT;
+       /* Total Number of Received Packets Dropped */
+       __be32 RDROP;
+       /* Drop due to overflow  */
+       __be32 RdropOvflw;
+       /* Drop due to overflow */
+       __be32 RdropLength;
+       /* Total of good frames. Does not include frames received with
+          frame-too-long, FCS, or length errors */
+       __be32 RTOTFRMS;
+       /* Total dropped Xmited packets */
+       __be32 TDROP;
+};
+
+
+#endif
diff --git a/drivers/net/mlx4/en_resources.c b/drivers/net/mlx4/en_resources.c
new file mode 100644 (file)
index 0000000..a054520
--- /dev/null
@@ -0,0 +1,96 @@
+/*
+ * 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/vmalloc.h>
+#include <linux/mlx4/qp.h>
+
+#include "mlx4_en.h"
+
+void mlx4_en_fill_qp_context(struct mlx4_en_priv *priv, int size, int stride,
+                            int is_tx, int rss, int qpn, int cqn, int srqn,
+                            struct mlx4_qp_context *context)
+{
+       struct mlx4_en_dev *mdev = priv->mdev;
+
+       memset(context, 0, sizeof *context);
+       context->flags = cpu_to_be32(7 << 16 | rss << 13);
+       context->pd = cpu_to_be32(mdev->priv_pdn);
+       context->mtu_msgmax = 0xff;
+       context->rq_size_stride = 0;
+       if (is_tx)
+               context->sq_size_stride = ilog2(size) << 3 | (ilog2(stride) - 4);
+       else
+               context->sq_size_stride = 1;
+       context->usr_page = cpu_to_be32(mdev->priv_uar.index);
+       context->local_qpn = cpu_to_be32(qpn);
+       context->pri_path.ackto = 1 & 0x07;
+       context->pri_path.sched_queue = 0x83 | (priv->port - 1) << 6;
+       context->pri_path.counter_index = 0xff;
+       context->cqn_send = cpu_to_be32(cqn);
+       context->cqn_recv = cpu_to_be32(cqn);
+       context->db_rec_addr = cpu_to_be64(priv->res.db.dma << 2);
+       if (!rss)
+               context->srqn = cpu_to_be32(MLX4_EN_USE_SRQ | srqn);
+}
+
+
+int mlx4_en_map_buffer(struct mlx4_buf *buf)
+{
+       struct page **pages;
+       int i;
+
+       if (BITS_PER_LONG == 64 || buf->nbufs == 1)
+               return 0;
+
+       pages = kmalloc(sizeof *pages * buf->nbufs, GFP_KERNEL);
+       if (!pages)
+               return -ENOMEM;
+
+       for (i = 0; i < buf->nbufs; ++i)
+               pages[i] = virt_to_page(buf->page_list[i].buf);
+
+       buf->direct.buf = vmap(pages, buf->nbufs, VM_MAP, PAGE_KERNEL);
+       kfree(pages);
+       if (!buf->direct.buf)
+               return -ENOMEM;
+
+       return 0;
+}
+
+void mlx4_en_unmap_buffer(struct mlx4_buf *buf)
+{
+       if (BITS_PER_LONG == 64 || buf->nbufs == 1)
+               return;
+
+       vunmap(buf->direct.buf);
+}
diff --git a/drivers/net/mlx4/en_rx.c b/drivers/net/mlx4/en_rx.c
new file mode 100644 (file)
index 0000000..6232227
--- /dev/null
@@ -0,0 +1,1080 @@
+/*
+ * 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/skbuff.h>
+#include <linux/if_ether.h>
+#include <linux/if_vlan.h>
+#include <linux/vmalloc.h>
+
+#include "mlx4_en.h"
+
+static void *get_wqe(struct mlx4_en_rx_ring *ring, int n)
+{
+       int offset = n << ring->srq.wqe_shift;
+       return ring->buf + offset;
+}
+
+static void mlx4_en_srq_event(struct mlx4_srq *srq, enum mlx4_event type)
+{
+       return;
+}
+
+static int mlx4_en_get_frag_header(struct skb_frag_struct *frags, void **mac_hdr,
+                                  void **ip_hdr, void **tcpudp_hdr,
+                                  u64 *hdr_flags, void *priv)
+{
+       *mac_hdr = page_address(frags->page) + frags->page_offset;
+       *ip_hdr = *mac_hdr + ETH_HLEN;
+       *tcpudp_hdr = (struct tcphdr *)(*ip_hdr + sizeof(struct iphdr));
+       *hdr_flags = LRO_IPV4 | LRO_TCP;
+
+       return 0;
+}
+
+static int mlx4_en_alloc_frag(struct mlx4_en_priv *priv,
+                             struct mlx4_en_rx_desc *rx_desc,
+                             struct skb_frag_struct *skb_frags,
+                             struct mlx4_en_rx_alloc *ring_alloc,
+                             int i)
+{
+       struct mlx4_en_dev *mdev = priv->mdev;
+       struct mlx4_en_frag_info *frag_info = &priv->frag_info[i];
+       struct mlx4_en_rx_alloc *page_alloc = &ring_alloc[i];
+       struct page *page;
+       dma_addr_t dma;
+
+       if (page_alloc->offset == frag_info->last_offset) {
+               /* Allocate new page */
+               page = alloc_pages(GFP_ATOMIC | __GFP_COMP, MLX4_EN_ALLOC_ORDER);
+               if (!page)
+                       return -ENOMEM;
+
+               skb_frags[i].page = page_alloc->page;
+               skb_frags[i].page_offset = page_alloc->offset;
+               page_alloc->page = page;
+               page_alloc->offset = frag_info->frag_align;
+       } else {
+               page = page_alloc->page;
+               get_page(page);
+
+               skb_frags[i].page = page;
+               skb_frags[i].page_offset = page_alloc->offset;
+               page_alloc->offset += frag_info->frag_stride;
+       }
+       dma = pci_map_single(mdev->pdev, page_address(skb_frags[i].page) +
+                            skb_frags[i].page_offset, frag_info->frag_size,
+                            PCI_DMA_FROMDEVICE);
+       rx_desc->data[i].addr = cpu_to_be64(dma);
+       return 0;
+}
+
+static int mlx4_en_init_allocator(struct mlx4_en_priv *priv,
+                                 struct mlx4_en_rx_ring *ring)
+{
+       struct mlx4_en_rx_alloc *page_alloc;
+       int i;
+
+       for (i = 0; i < priv->num_frags; i++) {
+               page_alloc = &ring->page_alloc[i];
+               page_alloc->page = alloc_pages(GFP_ATOMIC | __GFP_COMP,
+                                              MLX4_EN_ALLOC_ORDER);
+               if (!page_alloc->page)
+                       goto out;
+
+               page_alloc->offset = priv->frag_info[i].frag_align;
+               mlx4_dbg(DRV, priv, "Initialized allocator:%d with page:%p\n",
+                        i, page_alloc->page);
+       }
+       return 0;
+
+out:
+       while (i--) {
+               page_alloc = &ring->page_alloc[i];
+               put_page(page_alloc->page);
+               page_alloc->page = NULL;
+       }
+       return -ENOMEM;
+}
+
+static void mlx4_en_destroy_allocator(struct mlx4_en_priv *priv,
+                                     struct mlx4_en_rx_ring *ring)
+{
+       struct mlx4_en_rx_alloc *page_alloc;
+       int i;
+
+       for (i = 0; i < priv->num_frags; i++) {
+               page_alloc = &ring->page_alloc[i];
+               mlx4_dbg(DRV, priv, "Freeing allocator:%d count:%d\n",
+                        i, page_count(page_alloc->page));
+
+               put_page(page_alloc->page);
+               page_alloc->page = NULL;
+       }
+}
+
+
+static void mlx4_en_init_rx_desc(struct mlx4_en_priv *priv,
+                                struct mlx4_en_rx_ring *ring, int index)
+{
+       struct mlx4_en_rx_desc *rx_desc = ring->buf + ring->stride * index;
+       struct skb_frag_struct *skb_frags = ring->rx_info +
+                                           (index << priv->log_rx_info);
+       int possible_frags;
+       int i;
+
+       /* Pre-link descriptor */
+       rx_desc->next.next_wqe_index = cpu_to_be16((index + 1) & ring->size_mask);
+
+       /* Set size and memtype fields */
+       for (i = 0; i < priv->num_frags; i++) {
+               skb_frags[i].size = priv->frag_info[i].frag_size;
+               rx_desc->data[i].byte_count =
+                       cpu_to_be32(priv->frag_info[i].frag_size);
+               rx_desc->data[i].lkey = cpu_to_be32(priv->mdev->mr.key);
+       }
+
+       /* If the number of used fragments does not fill up the ring stride,
+        * remaining (unused) fragments must be padded with null address/size
+        * and a special memory key */
+       possible_frags = (ring->stride - sizeof(struct mlx4_en_rx_desc)) / DS_SIZE;
+       for (i = priv->num_frags; i < possible_frags; i++) {
+               rx_desc->data[i].byte_count = 0;
+               rx_desc->data[i].lkey = cpu_to_be32(MLX4_EN_MEMTYPE_PAD);
+               rx_desc->data[i].addr = 0;
+       }
+}
+
+
+static int mlx4_en_prepare_rx_desc(struct mlx4_en_priv *priv,
+                                  struct mlx4_en_rx_ring *ring, int index)
+{
+       struct mlx4_en_rx_desc *rx_desc = ring->buf + (index * ring->stride);
+       struct skb_frag_struct *skb_frags = ring->rx_info +
+                                           (index << priv->log_rx_info);
+       int i;
+
+       for (i = 0; i < priv->num_frags; i++)
+               if (mlx4_en_alloc_frag(priv, rx_desc, skb_frags, ring->page_alloc, i))
+                       goto err;
+
+       return 0;
+
+err:
+       while (i--)
+               put_page(skb_frags[i].page);
+       return -ENOMEM;
+}
+
+static inline void mlx4_en_update_rx_prod_db(struct mlx4_en_rx_ring *ring)
+{
+       *ring->wqres.db.db = cpu_to_be32(ring->prod & 0xffff);
+}
+
+static int mlx4_en_fill_rx_buffers(struct mlx4_en_priv *priv)
+{
+       struct mlx4_en_dev *mdev = priv->mdev;
+       struct mlx4_en_rx_ring *ring;
+       int ring_ind;
+       int buf_ind;
+
+       for (buf_ind = 0; buf_ind < priv->prof->rx_ring_size; buf_ind++) {
+               for (ring_ind = 0; ring_ind < priv->rx_ring_num; ring_ind++) {
+                       ring = &priv->rx_ring[ring_ind];
+
+                       if (mlx4_en_prepare_rx_desc(priv, ring,
+                                                   ring->actual_size)) {
+                               if (ring->actual_size < MLX4_EN_MIN_RX_SIZE) {
+                                       mlx4_err(mdev, "Failed to allocate "
+                                                      "enough rx buffers\n");
+                                       return -ENOMEM;
+                               } else {
+                                       if (netif_msg_rx_err(priv))
+                                               mlx4_warn(mdev,
+                                                         "Only %d buffers allocated\n",
+                                                         ring->actual_size);
+                                       goto out;
+                               }
+                       }
+                       ring->actual_size++;
+                       ring->prod++;
+               }
+       }
+out:
+       return 0;
+}
+
+static int mlx4_en_fill_rx_buf(struct net_device *dev,
+                              struct mlx4_en_rx_ring *ring)
+{
+       struct mlx4_en_priv *priv = netdev_priv(dev);
+       int num = 0;
+       int err;
+
+       while ((u32) (ring->prod - ring->cons) < ring->actual_size) {
+               err = mlx4_en_prepare_rx_desc(priv, ring, ring->prod &
+                                             ring->size_mask);
+               if (err) {
+                       if (netif_msg_rx_err(priv))
+                               mlx4_warn(priv->mdev,
+                                         "Failed preparing rx descriptor\n");
+                       priv->port_stats.rx_alloc_failed++;
+                       break;
+               }
+               ++num;
+               ++ring->prod;
+       }
+       if ((u32) (ring->prod - ring->cons) == ring->size)
+               ring->full = 1;
+
+       return num;
+}
+
+static void mlx4_en_free_rx_buf(struct mlx4_en_priv *priv,
+                               struct mlx4_en_rx_ring *ring)
+{
+       struct mlx4_en_dev *mdev = priv->mdev;
+       struct skb_frag_struct *skb_frags;
+       struct mlx4_en_rx_desc *rx_desc;
+       dma_addr_t dma;
+       int index;
+       int nr;
+
+       mlx4_dbg(DRV, priv, "Freeing Rx buf - cons:%d prod:%d\n",
+                       ring->cons, ring->prod);
+
+       /* Unmap and free Rx buffers */
+       BUG_ON((u32) (ring->prod - ring->cons) > ring->size);
+       while (ring->cons != ring->prod) {
+               index = ring->cons & ring->size_mask;
+               rx_desc = ring->buf + (index << ring->log_stride);
+               skb_frags = ring->rx_info + (index << priv->log_rx_info);
+               mlx4_dbg(DRV, priv, "Processing descriptor:%d\n", index);
+
+               for (nr = 0; nr < priv->num_frags; nr++) {
+                       mlx4_dbg(DRV, priv, "Freeing fragment:%d\n", nr);
+                       dma = be64_to_cpu(rx_desc->data[nr].addr);
+
+                       mlx4_dbg(DRV, priv, "Unmaping buffer at dma:0x%llx\n", (u64) dma);
+                       pci_unmap_single(mdev->pdev, dma, skb_frags[nr].size,
+                                        PCI_DMA_FROMDEVICE);
+                       put_page(skb_frags[nr].page);
+               }
+               ++ring->cons;
+       }
+}
+
+
+void mlx4_en_rx_refill(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,
+                                                refill_task);
+       struct mlx4_en_dev *mdev = priv->mdev;
+       struct net_device *dev = priv->dev;
+       struct mlx4_en_rx_ring *ring;
+       int need_refill = 0;
+       int i;
+
+       mutex_lock(&mdev->state_lock);
+       if (!mdev->device_up || !priv->port_up)
+               goto out;
+
+       /* We only get here if there are no receive buffers, so we can't race
+        * with Rx interrupts while filling buffers */
+       for (i = 0; i < priv->rx_ring_num; i++) {
+               ring = &priv->rx_ring[i];
+               if (ring->need_refill) {
+                       if (mlx4_en_fill_rx_buf(dev, ring)) {
+                               ring->need_refill = 0;
+                               mlx4_en_update_rx_prod_db(ring);
+                       } else
+                               need_refill = 1;
+               }
+       }
+       if (need_refill)
+               queue_delayed_work(mdev->workqueue, &priv->refill_task, HZ);
+
+out:
+       mutex_unlock(&mdev->state_lock);
+}
+
+
+int mlx4_en_create_rx_ring(struct mlx4_en_priv *priv,
+                          struct mlx4_en_rx_ring *ring, u32 size, u16 stride)
+{
+       struct mlx4_en_dev *mdev = priv->mdev;
+       int err;
+       int tmp;
+
+       /* Sanity check SRQ size before proceeding */
+       if (size >= mdev->dev->caps.max_srq_wqes)
+               return -EINVAL;
+
+       ring->prod = 0;
+       ring->cons = 0;
+       ring->size = size;
+       ring->size_mask = size - 1;
+       ring->stride = stride;
+       ring->log_stride = ffs(ring->stride) - 1;
+       ring->buf_size = ring->size * ring->stride;
+
+       tmp = size * roundup_pow_of_two(MLX4_EN_MAX_RX_FRAGS *
+                                       sizeof(struct skb_frag_struct));
+       ring->rx_info = vmalloc(tmp);
+       if (!ring->rx_info) {
+               mlx4_err(mdev, "Failed allocating rx_info ring\n");
+               return -ENOMEM;
+       }
+       mlx4_dbg(DRV, priv, "Allocated rx_info ring at addr:%p size:%d\n",
+                ring->rx_info, tmp);
+
+       err = mlx4_alloc_hwq_res(mdev->dev, &ring->wqres,
+                                ring->buf_size, 2 * PAGE_SIZE);
+       if (err)
+               goto err_ring;
+
+       err = mlx4_en_map_buffer(&ring->wqres.buf);
+       if (err) {
+               mlx4_err(mdev, "Failed to map RX buffer\n");
+               goto err_hwq;
+       }
+       ring->buf = ring->wqres.buf.direct.buf;
+
+       /* Configure lro mngr */
+       memset(&ring->lro, 0, sizeof(struct net_lro_mgr));
+       ring->lro.dev = priv->dev;
+       ring->lro.features = LRO_F_NAPI;
+       ring->lro.frag_align_pad = NET_IP_ALIGN;
+       ring->lro.ip_summed = CHECKSUM_UNNECESSARY;
+       ring->lro.ip_summed_aggr = CHECKSUM_UNNECESSARY;
+       ring->lro.max_desc = mdev->profile.num_lro;
+       ring->lro.max_aggr = MAX_SKB_FRAGS;
+       ring->lro.lro_arr = kzalloc(mdev->profile.num_lro *
+                                   sizeof(struct net_lro_desc),
+                                   GFP_KERNEL);
+       if (!ring->lro.lro_arr) {
+               mlx4_err(mdev, "Failed to allocate lro array\n");
+               goto err_map;
+       }
+       ring->lro.get_frag_header = mlx4_en_get_frag_header;
+
+       return 0;
+
+err_map:
+       mlx4_en_unmap_buffer(&ring->wqres.buf);
+err_hwq:
+       mlx4_free_hwq_res(mdev->dev, &ring->wqres, ring->buf_size);
+err_ring:
+       vfree(ring->rx_info);
+       ring->rx_info = NULL;
+       return err;
+}
+
+int mlx4_en_activate_rx_rings(struct mlx4_en_priv *priv)
+{
+       struct mlx4_en_dev *mdev = priv->mdev;
+       struct mlx4_wqe_srq_next_seg *next;
+       struct mlx4_en_rx_ring *ring;
+       int i;
+       int ring_ind;
+       int err;
+       int stride = roundup_pow_of_two(sizeof(struct mlx4_en_rx_desc) +
+                                       DS_SIZE * priv->num_frags);
+       int max_gs = (stride - sizeof(struct mlx4_wqe_srq_next_seg)) / DS_SIZE;
+
+       for (ring_ind = 0; ring_ind < priv->rx_ring_num; ring_ind++) {
+               ring = &priv->rx_ring[ring_ind];
+
+               ring->prod = 0;
+               ring->cons = 0;
+               ring->actual_size = 0;
+               ring->cqn = priv->rx_cq[ring_ind].mcq.cqn;
+
+               ring->stride = stride;
+               ring->log_stride = ffs(ring->stride) - 1;
+               ring->buf_size = ring->size * ring->stride;
+
+               memset(ring->buf, 0, ring->buf_size);
+               mlx4_en_update_rx_prod_db(ring);
+
+               /* Initailize all descriptors */
+               for (i = 0; i < ring->size; i++)
+                       mlx4_en_init_rx_desc(priv, ring, i);
+
+               /* Initialize page allocators */
+               err = mlx4_en_init_allocator(priv, ring);
+               if (err) {
+                        mlx4_err(mdev, "Failed initializing ring allocator\n");
+                        goto err_allocator;
+               }
+
+               /* Fill Rx buffers */
+               ring->full = 0;
+       }
+       if (mlx4_en_fill_rx_buffers(priv))
+               goto err_buffers;
+
+       for (ring_ind = 0; ring_ind < priv->rx_ring_num; ring_ind++) {
+               ring = &priv->rx_ring[ring_ind];
+
+               mlx4_en_update_rx_prod_db(ring);
+
+               /* Configure SRQ representing the ring */
+               ring->srq.max    = ring->size;
+               ring->srq.max_gs = max_gs;
+               ring->srq.wqe_shift = ilog2(ring->stride);
+
+               for (i = 0; i < ring->srq.max; ++i) {
+                       next = get_wqe(ring, i);
+                       next->next_wqe_index =
+                       cpu_to_be16((i + 1) & (ring->srq.max - 1));
+               }
+
+               err = mlx4_srq_alloc(mdev->dev, mdev->priv_pdn, &ring->wqres.mtt,
+                                    ring->wqres.db.dma, &ring->srq);
+               if (err){
+                       mlx4_err(mdev, "Failed to allocate srq\n");
+                       goto err_srq;
+               }
+               ring->srq.event = mlx4_en_srq_event;
+       }
+
+       return 0;
+
+err_srq:
+       while (ring_ind >= 0) {
+               ring = &priv->rx_ring[ring_ind];
+               mlx4_srq_free(mdev->dev, &ring->srq);
+               ring_ind--;
+       }
+
+err_buffers:
+       for (ring_ind = 0; ring_ind < priv->rx_ring_num; ring_ind++)
+               mlx4_en_free_rx_buf(priv, &priv->rx_ring[ring_ind]);
+
+       ring_ind = priv->rx_ring_num - 1;
+err_allocator:
+       while (ring_ind >= 0) {
+               mlx4_en_destroy_allocator(priv, &priv->rx_ring[ring_ind]);
+               ring_ind--;
+       }
+       return err;
+}
+
+void mlx4_en_destroy_rx_ring(struct mlx4_en_priv *priv,
+                            struct mlx4_en_rx_ring *ring)
+{
+       struct mlx4_en_dev *mdev = priv->mdev;
+
+       kfree(ring->lro.lro_arr);
+       mlx4_en_unmap_buffer(&ring->wqres.buf);
+       mlx4_free_hwq_res(mdev->dev, &ring->wqres, ring->buf_size);
+       vfree(ring->rx_info);
+       ring->rx_info = NULL;
+}
+
+void mlx4_en_deactivate_rx_ring(struct mlx4_en_priv *priv,
+                               struct mlx4_en_rx_ring *ring)
+{
+       struct mlx4_en_dev *mdev = priv->mdev;
+
+       mlx4_srq_free(mdev->dev, &ring->srq);
+       mlx4_en_free_rx_buf(priv, ring);
+       mlx4_en_destroy_allocator(priv, ring);
+}
+
+
+/* Unmap a completed descriptor and free unused pages */
+static int mlx4_en_complete_rx_desc(struct mlx4_en_priv *priv,
+                                   struct mlx4_en_rx_desc *rx_desc,
+                                   struct skb_frag_struct *skb_frags,
+                                   struct skb_frag_struct *skb_frags_rx,
+                                   struct mlx4_en_rx_alloc *page_alloc,
+                                   int length)
+{
+       struct mlx4_en_dev *mdev = priv->mdev;
+       struct mlx4_en_frag_info *frag_info;
+       int nr;
+       dma_addr_t dma;
+
+       /* Collect used fragments while replacing them in the HW descirptors */
+       for (nr = 0; nr < priv->num_frags; nr++) {
+               frag_info = &priv->frag_info[nr];
+               if (length <= frag_info->frag_prefix_size)
+                       break;
+
+               /* Save page reference in skb */
+               skb_frags_rx[nr].page = skb_frags[nr].page;
+               skb_frags_rx[nr].size = skb_frags[nr].size;
+               skb_frags_rx[nr].page_offset = skb_frags[nr].page_offset;
+               dma = be64_to_cpu(rx_desc->data[nr].addr);
+
+               /* Allocate a replacement page */
+               if (mlx4_en_alloc_frag(priv, rx_desc, skb_frags, page_alloc, nr))
+                       goto fail;
+
+               /* Unmap buffer */
+               pci_unmap_single(mdev->pdev, dma, skb_frags[nr].size,
+                                PCI_DMA_FROMDEVICE);
+       }
+       /* Adjust size of last fragment to match actual length */
+       skb_frags_rx[nr - 1].size = length -
+               priv->frag_info[nr - 1].frag_prefix_size;
+       return nr;
+
+fail:
+       /* Drop all accumulated fragments (which have already been replaced in
+        * the descriptor) of this packet; remaining fragments are reused... */
+       while (nr > 0) {
+               nr--;
+               put_page(skb_frags_rx[nr].page);
+       }
+       return 0;
+}
+
+
+static struct sk_buff *mlx4_en_rx_skb(struct mlx4_en_priv *priv,
+                                     struct mlx4_en_rx_desc *rx_desc,
+                                     struct skb_frag_struct *skb_frags,
+                                     struct mlx4_en_rx_alloc *page_alloc,
+                                     unsigned int length)
+{
+       struct mlx4_en_dev *mdev = priv->mdev;
+       struct sk_buff *skb;
+       void *va;
+       int used_frags;
+       dma_addr_t dma;
+
+       skb = dev_alloc_skb(SMALL_PACKET_SIZE + NET_IP_ALIGN);
+       if (!skb) {
+               mlx4_dbg(RX_ERR, priv, "Failed allocating skb\n");
+               return NULL;
+       }
+       skb->dev = priv->dev;
+       skb_reserve(skb, NET_IP_ALIGN);
+       skb->len = length;
+       skb->truesize = length + sizeof(struct sk_buff);
+
+       /* Get pointer to first fragment so we could copy the headers into the
+        * (linear part of the) skb */
+       va = page_address(skb_frags[0].page) + skb_frags[0].page_offset;
+
+       if (length <= SMALL_PACKET_SIZE) {
+               /* We are copying all relevant data to the skb - temporarily
+                * synch buffers for the copy */
+               dma = be64_to_cpu(rx_desc->data[0].addr);
+               dma_sync_single_range_for_cpu(&mdev->pdev->dev, dma, 0,
+                                             length, DMA_FROM_DEVICE);
+               skb_copy_to_linear_data(skb, va, length);
+               dma_sync_single_range_for_device(&mdev->pdev->dev, dma, 0,
+                                                length, DMA_FROM_DEVICE);
+               skb->tail += length;
+       } else {
+
+               /* Move relevant fragments to skb */
+               used_frags = mlx4_en_complete_rx_desc(priv, rx_desc, skb_frags,
+                                                     skb_shinfo(skb)->frags,
+                                                     page_alloc, length);
+               skb_shinfo(skb)->nr_frags = used_frags;
+
+               /* Copy headers into the skb linear buffer */
+               memcpy(skb->data, va, HEADER_COPY_SIZE);
+               skb->tail += HEADER_COPY_SIZE;
+
+               /* Skip headers in first fragment */
+               skb_shinfo(skb)->frags[0].page_offset += HEADER_COPY_SIZE;
+
+               /* Adjust size of first fragment */
+               skb_shinfo(skb)->frags[0].size -= HEADER_COPY_SIZE;
+               skb->data_len = length - HEADER_COPY_SIZE;
+       }
+       return skb;
+}
+
+static void mlx4_en_copy_desc(struct mlx4_en_priv *priv,
+                             struct mlx4_en_rx_ring *ring,
+                             int from, int to, int num)
+{
+       struct skb_frag_struct *skb_frags_from;
+       struct skb_frag_struct *skb_frags_to;
+       struct mlx4_en_rx_desc *rx_desc_from;
+       struct mlx4_en_rx_desc *rx_desc_to;
+       int from_index, to_index;
+       int nr, i;
+
+       for (i = 0; i < num; i++) {
+               from_index = (from + i) & ring->size_mask;
+               to_index = (to + i) & ring->size_mask;
+               skb_frags_from = ring->rx_info + (from_index << priv->log_rx_info);
+               skb_frags_to = ring->rx_info + (to_index << priv->log_rx_info);
+               rx_desc_from = ring->buf + (from_index << ring->log_stride);
+               rx_desc_to = ring->buf + (to_index << ring->log_stride);
+
+               for (nr = 0; nr < priv->num_frags; nr++) {
+                       skb_frags_to[nr].page = skb_frags_from[nr].page;
+                       skb_frags_to[nr].page_offset = skb_frags_from[nr].page_offset;
+                       rx_desc_to->data[nr].addr = rx_desc_from->data[nr].addr;
+               }
+       }
+}
+
+
+int mlx4_en_process_rx_cq(struct net_device *dev, struct mlx4_en_cq *cq, int budget)
+{
+       struct mlx4_en_priv *priv = netdev_priv(dev);
+       struct mlx4_en_dev *mdev = priv->mdev;
+       struct mlx4_cqe *cqe;
+       struct mlx4_en_rx_ring *ring = &priv->rx_ring[cq->ring];
+       struct skb_frag_struct *skb_frags;
+       struct skb_frag_struct lro_frags[MLX4_EN_MAX_RX_FRAGS];
+       struct mlx4_en_rx_desc *rx_desc;
+       struct sk_buff *skb;
+       int index;
+       int nr;
+       unsigned int length;
+       int polled = 0;
+       int ip_summed;
+
+       if (!priv->port_up)
+               return 0;
+
+       /* We assume a 1:1 mapping between CQEs and Rx descriptors, so Rx
+        * descriptor offset can be deduced from the CQE index instead of
+        * reading 'cqe->index' */
+       index = cq->mcq.cons_index & ring->size_mask;
+       cqe = &cq->buf[index];
+
+       /* Process all completed CQEs */
+       while (XNOR(cqe->owner_sr_opcode & MLX4_CQE_OWNER_MASK,
+                   cq->mcq.cons_index & cq->size)) {
+
+               skb_frags = ring->rx_info + (index << priv->log_rx_info);
+               rx_desc = ring->buf + (index << ring->log_stride);
+
+               /*
+                * make sure we read the CQE after we read the ownership bit
+                */
+               rmb();
+
+               /* Drop packet on bad receive or bad checksum */
+               if (unlikely((cqe->owner_sr_opcode & MLX4_CQE_OPCODE_MASK) ==
+                                               MLX4_CQE_OPCODE_ERROR)) {
+                       mlx4_err(mdev, "CQE completed in error - vendor "
+                                 "syndrom:%d syndrom:%d\n",
+                                 ((struct mlx4_err_cqe *) cqe)->vendor_err_syndrome,
+                                 ((struct mlx4_err_cqe *) cqe)->syndrome);
+                       goto next;
+               }
+               if (unlikely(cqe->badfcs_enc & MLX4_CQE_BAD_FCS)) {
+                       mlx4_dbg(RX_ERR, priv, "Accepted frame with bad FCS\n");
+                       goto next;
+               }
+
+               /*
+                * Packet is OK - process it.
+                */
+               length = be32_to_cpu(cqe->byte_cnt);
+               ring->bytes += length;
+               ring->packets++;
+
+               if (likely(priv->rx_csum)) {
+                       if ((cqe->status & cpu_to_be16(MLX4_CQE_STATUS_IPOK)) &&
+                           (cqe->checksum == cpu_to_be16(0xffff))) {
+                               priv->port_stats.rx_chksum_good++;
+                               /* This packet is eligible for LRO if it is:
+                                * - DIX Ethernet (type interpretation)
+                                * - TCP/IP (v4)
+                                * - without IP options
+                                * - not an IP fragment */
+                               if (mlx4_en_can_lro(cqe->status) &&
+                                   dev->features & NETIF_F_LRO) {
+
+                                       nr = mlx4_en_complete_rx_desc(
+                                               priv, rx_desc,
+                                               skb_frags, lro_frags,
+                                               ring->page_alloc, length);
+                                       if (!nr)
+                                               goto next;
+
+                                       if (priv->vlgrp && (cqe->vlan_my_qpn &
+                                                           cpu_to_be32(MLX4_CQE_VLAN_PRESENT_MASK))) {
+                                               lro_vlan_hwaccel_receive_frags(
+                                                      &ring->lro, lro_frags,
+                                                      length, length,
+                                                      priv->vlgrp,
+                                                      be16_to_cpu(cqe->sl_vid),
+                                                      NULL, 0);
+                                       } else
+                                               lro_receive_frags(&ring->lro,
+                                                                 lro_frags,
+                                                                 length,
+                                                                 length,
+                                                                 NULL, 0);
+
+                                       goto next;
+                               }
+
+                               /* LRO not possible, complete processing here */
+                               ip_summed = CHECKSUM_UNNECESSARY;
+                               INC_PERF_COUNTER(priv->pstats.lro_misses);
+                       } else {
+                               ip_summed = CHECKSUM_NONE;
+                               priv->port_stats.rx_chksum_none++;
+                       }
+               } else {
+                       ip_summed = CHECKSUM_NONE;
+                       priv->port_stats.rx_chksum_none++;
+               }
+
+               skb = mlx4_en_rx_skb(priv, rx_desc, skb_frags,
+                                    ring->page_alloc, length);
+               if (!skb) {
+                       priv->stats.rx_dropped++;
+                       goto next;
+               }
+
+               skb->ip_summed = ip_summed;
+               skb->protocol = eth_type_trans(skb, dev);
+
+               /* Push it up the stack */
+               if (priv->vlgrp && (be32_to_cpu(cqe->vlan_my_qpn) &
+                                   MLX4_CQE_VLAN_PRESENT_MASK)) {
+                       vlan_hwaccel_receive_skb(skb, priv->vlgrp,
+                                               be16_to_cpu(cqe->sl_vid));
+               } else
+                       netif_receive_skb(skb);
+
+               dev->last_rx = jiffies;
+
+next:
+               ++cq->mcq.cons_index;
+               index = (cq->mcq.cons_index) & ring->size_mask;
+               cqe = &cq->buf[index];
+               if (++polled == budget) {
+                       /* We are here because we reached the NAPI budget -
+                        * flush only pending LRO sessions */
+                       lro_flush_all(&ring->lro);
+                       goto out;
+               }
+       }
+
+       /* If CQ is empty flush all LRO sessions unconditionally */
+       lro_flush_all(&ring->lro);
+
+out:
+       AVG_PERF_COUNTER(priv->pstats.rx_coal_avg, polled);
+       mlx4_cq_set_ci(&cq->mcq);
+       wmb(); /* ensure HW sees CQ consumer before we post new buffers */
+       ring->cons = cq->mcq.cons_index;
+       ring->prod += polled; /* Polled descriptors were realocated in place */
+       if (unlikely(!ring->full)) {
+               mlx4_en_copy_desc(priv, ring, ring->cons - polled,
+                                 ring->prod - polled, polled);
+               mlx4_en_fill_rx_buf(dev, ring);
+       }
+       mlx4_en_update_rx_prod_db(ring);
+       return polled;
+}
+
+
+void mlx4_en_rx_irq(struct mlx4_cq *mcq)
+{
+       struct mlx4_en_cq *cq = container_of(mcq, struct mlx4_en_cq, mcq);
+       struct mlx4_en_priv *priv = netdev_priv(cq->dev);
+
+       if (priv->port_up)
+               netif_rx_schedule(cq->dev, &cq->napi);
+       else
+               mlx4_en_arm_cq(priv, cq);
+}
+
+/* Rx CQ polling - called by NAPI */
+int mlx4_en_poll_rx_cq(struct napi_struct *napi, int budget)
+{
+       struct mlx4_en_cq *cq = container_of(napi, struct mlx4_en_cq, napi);
+       struct net_device *dev = cq->dev;
+       struct mlx4_en_priv *priv = netdev_priv(dev);
+       int done;
+
+       done = mlx4_en_process_rx_cq(dev, cq, budget);
+
+       /* If we used up all the quota - we're probably not done yet... */
+       if (done == budget)
+               INC_PERF_COUNTER(priv->pstats.napi_quota);
+       else {
+               /* Done for now */
+               netif_rx_complete(dev, napi);
+               mlx4_en_arm_cq(priv, cq);
+       }
+       return done;
+}
+
+
+/* Calculate the last offset position that accomodates a full fragment
+ * (assuming fagment size = stride-align) */
+static int mlx4_en_last_alloc_offset(struct mlx4_en_priv *priv, u16 stride, u16 align)
+{
+       u16 res = MLX4_EN_ALLOC_SIZE % stride;
+       u16 offset = MLX4_EN_ALLOC_SIZE - stride - res + align;
+
+       mlx4_dbg(DRV, priv, "Calculated last offset for stride:%d align:%d "
+                           "res:%d offset:%d\n", stride, align, res, offset);
+       return offset;
+}
+
+
+static int frag_sizes[] = {
+       FRAG_SZ0,
+       FRAG_SZ1,
+       FRAG_SZ2,
+       FRAG_SZ3
+};
+
+void mlx4_en_calc_rx_buf(struct net_device *dev)
+{
+       struct mlx4_en_priv *priv = netdev_priv(dev);
+       int eff_mtu = dev->mtu + ETH_HLEN + VLAN_HLEN + ETH_LLC_SNAP_SIZE;
+       int buf_size = 0;
+       int i = 0;
+
+       while (buf_size < eff_mtu) {
+               priv->frag_info[i].frag_size =
+                       (eff_mtu > buf_size + frag_sizes[i]) ?
+                               frag_sizes[i] : eff_mtu - buf_size;
+               priv->frag_info[i].frag_prefix_size = buf_size;
+               if (!i) {
+                       priv->frag_info[i].frag_align = NET_IP_ALIGN;
+                       priv->frag_info[i].frag_stride =
+                               ALIGN(frag_sizes[i] + NET_IP_ALIGN, SMP_CACHE_BYTES);
+               } else {
+                       priv->frag_info[i].frag_align = 0;
+                       priv->frag_info[i].frag_stride =
+                               ALIGN(frag_sizes[i], SMP_CACHE_BYTES);
+               }
+               priv->frag_info[i].last_offset = mlx4_en_last_alloc_offset(
+                                               priv, priv->frag_info[i].frag_stride,
+                                               priv->frag_info[i].frag_align);
+               buf_size += priv->frag_info[i].frag_size;
+               i++;
+       }
+
+       priv->num_frags = i;
+       priv->rx_skb_size = eff_mtu;
+       priv->log_rx_info = ROUNDUP_LOG2(i * sizeof(struct skb_frag_struct));
+
+       mlx4_dbg(DRV, priv, "Rx buffer scatter-list (effective-mtu:%d "
+                 "num_frags:%d):\n", eff_mtu, priv->num_frags);
+       for (i = 0; i < priv->num_frags; i++) {
+               mlx4_dbg(DRV, priv, "  frag:%d - size:%d prefix:%d align:%d "
+                               "stride:%d last_offset:%d\n", i,
+                               priv->frag_info[i].frag_size,
+                               priv->frag_info[i].frag_prefix_size,
+                               priv->frag_info[i].frag_align,
+                               priv->frag_info[i].frag_stride,
+                               priv->frag_info[i].last_offset);
+       }
+}
+
+/* RSS related functions */
+
+/* Calculate rss size and map each entry in rss table to rx ring */
+void mlx4_en_set_default_rss_map(struct mlx4_en_priv *priv,
+                                struct mlx4_en_rss_map *rss_map,
+                                int num_entries, int num_rings)
+{
+       int i;
+
+       rss_map->size = roundup_pow_of_two(num_entries);
+       mlx4_dbg(DRV, priv, "Setting default RSS map of %d entires\n",
+                rss_map->size);
+
+       for (i = 0; i < rss_map->size; i++) {
+               rss_map->map[i] = i % num_rings;
+               mlx4_dbg(DRV, priv, "Entry %d ---> ring %d\n", i, rss_map->map[i]);
+       }
+}
+
+static void mlx4_en_sqp_event(struct mlx4_qp *qp, enum mlx4_event event)
+{
+    return;
+}
+
+
+static int mlx4_en_config_rss_qp(struct mlx4_en_priv *priv,
+                                int qpn, int srqn, int cqn,
+                                enum mlx4_qp_state *state,
+                                struct mlx4_qp *qp)
+{
+       struct mlx4_en_dev *mdev = priv->mdev;
+       struct mlx4_qp_context *context;
+       int err = 0;
+
+       context = kmalloc(sizeof *context , GFP_KERNEL);
+       if (!context) {
+               mlx4_err(mdev, "Failed to allocate qp context\n");
+               return -ENOMEM;
+       }
+
+       err = mlx4_qp_alloc(mdev->dev, qpn, qp);
+       if (err) {
+               mlx4_err(mdev, "Failed to allocate qp #%d\n", qpn);
+               goto out;
+               return err;
+       }
+       qp->event = mlx4_en_sqp_event;
+
+       memset(context, 0, sizeof *context);
+       mlx4_en_fill_qp_context(priv, 0, 0, 0, 0, qpn, cqn, srqn, context);
+
+       err = mlx4_qp_to_ready(mdev->dev, &priv->res.mtt, context, qp, state);
+       if (err) {
+               mlx4_qp_remove(mdev->dev, qp);
+               mlx4_qp_free(mdev->dev, qp);
+       }
+out:
+       kfree(context);
+       return err;
+}
+
+/* Allocate rx qp's and configure them according to rss map */
+int mlx4_en_config_rss_steer(struct mlx4_en_priv *priv)
+{
+       struct mlx4_en_dev *mdev = priv->mdev;
+       struct mlx4_en_rss_map *rss_map = &priv->rss_map;
+       struct mlx4_qp_context context;
+       struct mlx4_en_rss_context *rss_context;
+       void *ptr;
+       int rss_xor = mdev->profile.rss_xor;
+       u8 rss_mask = mdev->profile.rss_mask;
+       int i, srqn, qpn, cqn;
+       int err = 0;
+       int good_qps = 0;
+
+       mlx4_dbg(DRV, priv, "Configuring rss steering for port %u\n", priv->port);
+       err = mlx4_qp_reserve_range(mdev->dev, rss_map->size,
+                                   rss_map->size, &rss_map->base_qpn);
+       if (err) {
+               mlx4_err(mdev, "Failed reserving %d qps for port %u\n",
+                        rss_map->size, priv->port);
+               return err;
+       }
+
+       for (i = 0; i < rss_map->size; i++) {
+               cqn = priv->rx_ring[rss_map->map[i]].cqn;
+               srqn = priv->rx_ring[rss_map->map[i]].srq.srqn;
+               qpn = rss_map->base_qpn + i;
+               err = mlx4_en_config_rss_qp(priv, qpn, srqn, cqn,
+                                           &rss_map->state[i],
+                                           &rss_map->qps[i]);
+               if (err)
+                       goto rss_err;
+
+               ++good_qps;
+       }
+
+       /* Configure RSS indirection qp */
+       err = mlx4_qp_reserve_range(mdev->dev, 1, 1, &priv->base_qpn);
+       if (err) {
+               mlx4_err(mdev, "Failed to reserve range for RSS "
+                              "indirection qp\n");
+               goto rss_err;
+       }
+       err = mlx4_qp_alloc(mdev->dev, priv->base_qpn, &rss_map->indir_qp);
+       if (err) {
+               mlx4_err(mdev, "Failed to allocate RSS indirection QP\n");
+               goto reserve_err;
+       }
+       rss_map->indir_qp.event = mlx4_en_sqp_event;
+       mlx4_en_fill_qp_context(priv, 0, 0, 0, 1, priv->base_qpn,
+                               priv->rx_ring[0].cqn, 0, &context);
+
+       ptr = ((void *) &context) + 0x3c;
+       rss_context = (struct mlx4_en_rss_context *) ptr;
+       rss_context->base_qpn = cpu_to_be32(ilog2(rss_map->size) << 24 |
+                                           (rss_map->base_qpn));
+       rss_context->default_qpn = cpu_to_be32(rss_map->base_qpn);
+       rss_context->hash_fn = rss_xor & 0x3;
+       rss_context->flags = rss_mask << 2;
+
+       err = mlx4_qp_to_ready(mdev->dev, &priv->res.mtt, &context,
+                              &rss_map->indir_qp, &rss_map->indir_state);
+       if (err)
+               goto indir_err;
+
+       return 0;
+
+indir_err:
+       mlx4_qp_modify(mdev->dev, NULL, rss_map->indir_state,
+                      MLX4_QP_STATE_RST, NULL, 0, 0, &rss_map->indir_qp);
+       mlx4_qp_remove(mdev->dev, &rss_map->indir_qp);
+       mlx4_qp_free(mdev->dev, &rss_map->indir_qp);
+reserve_err:
+       mlx4_qp_release_range(mdev->dev, priv->base_qpn, 1);
+rss_err:
+       for (i = 0; i < good_qps; i++) {
+               mlx4_qp_modify(mdev->dev, NULL, rss_map->state[i],
+                              MLX4_QP_STATE_RST, NULL, 0, 0, &rss_map->qps[i]);
+               mlx4_qp_remove(mdev->dev, &rss_map->qps[i]);
+               mlx4_qp_free(mdev->dev, &rss_map->qps[i]);
+       }
+       mlx4_qp_release_range(mdev->dev, rss_map->base_qpn, rss_map->size);
+       return err;
+}
+
+void mlx4_en_release_rss_steer(struct mlx4_en_priv *priv)
+{
+       struct mlx4_en_dev *mdev = priv->mdev;
+       struct mlx4_en_rss_map *rss_map = &priv->rss_map;
+       int i;
+
+       mlx4_qp_modify(mdev->dev, NULL, rss_map->indir_state,
+                      MLX4_QP_STATE_RST, NULL, 0, 0, &rss_map->indir_qp);
+       mlx4_qp_remove(mdev->dev, &rss_map->indir_qp);
+       mlx4_qp_free(mdev->dev, &rss_map->indir_qp);
+       mlx4_qp_release_range(mdev->dev, priv->base_qpn, 1);
+
+       for (i = 0; i < rss_map->size; i++) {
+               mlx4_qp_modify(mdev->dev, NULL, rss_map->state[i],
+                              MLX4_QP_STATE_RST, NULL, 0, 0, &rss_map->qps[i]);
+               mlx4_qp_remove(mdev->dev, &rss_map->qps[i]);
+               mlx4_qp_free(mdev->dev, &rss_map->qps[i]);
+       }
+       mlx4_qp_release_range(mdev->dev, rss_map->base_qpn, rss_map->size);
+}
+
+
+
+
+
diff --git a/drivers/net/mlx4/en_tx.c b/drivers/net/mlx4/en_tx.c
new file mode 100644 (file)
index 0000000..8592f8f
--- /dev/null
@@ -0,0 +1,820 @@
+/*
+ * 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 <asm/page.h>
+#include <linux/mlx4/cq.h>
+#include <linux/mlx4/qp.h>
+#include <linux/skbuff.h>
+#include <linux/if_vlan.h>
+#include <linux/vmalloc.h>
+
+#include "mlx4_en.h"
+
+enum {
+       MAX_INLINE = 104, /* 128 - 16 - 4 - 4 */
+};
+
+static int inline_thold __read_mostly = MAX_INLINE;
+
+module_param_named(inline_thold, inline_thold, int, 0444);
+MODULE_PARM_DESC(inline_thold, "treshold for using inline data");
+
+int mlx4_en_create_tx_ring(struct mlx4_en_priv *priv,
+                          struct mlx4_en_tx_ring *ring, u32 size,
+                          u16 stride)
+{
+       struct mlx4_en_dev *mdev = priv->mdev;
+       int tmp;
+       int err;
+
+       ring->size = size;
+       ring->size_mask = size - 1;
+       ring->stride = stride;
+
+       inline_thold = min(inline_thold, MAX_INLINE);
+
+       spin_lock_init(&ring->comp_lock);
+
+       tmp = size * sizeof(struct mlx4_en_tx_info);
+       ring->tx_info = vmalloc(tmp);
+       if (!ring->tx_info) {
+               mlx4_err(mdev, "Failed allocating tx_info ring\n");
+               return -ENOMEM;
+       }
+       mlx4_dbg(DRV, priv, "Allocated tx_info ring at addr:%p size:%d\n",
+                ring->tx_info, tmp);
+
+       ring->bounce_buf = kmalloc(MAX_DESC_SIZE, GFP_KERNEL);
+       if (!ring->bounce_buf) {
+               mlx4_err(mdev, "Failed allocating bounce buffer\n");
+               err = -ENOMEM;
+               goto err_tx;
+       }
+       ring->buf_size = ALIGN(size * ring->stride, MLX4_EN_PAGE_SIZE);
+
+       err = mlx4_alloc_hwq_res(mdev->dev, &ring->wqres, ring->buf_size,
+                                2 * PAGE_SIZE);
+       if (err) {
+               mlx4_err(mdev, "Failed allocating hwq resources\n");
+               goto err_bounce;
+       }
+
+       err = mlx4_en_map_buffer(&ring->wqres.buf);
+       if (err) {
+               mlx4_err(mdev, "Failed to map TX buffer\n");
+               goto err_hwq_res;
+       }
+
+       ring->buf = ring->wqres.buf.direct.buf;
+
+       mlx4_dbg(DRV, priv, "Allocated TX ring (addr:%p) - buf:%p size:%d "
+                "buf_size:%d dma:%llx\n", ring, ring->buf, ring->size,
+                ring->buf_size, (unsigned long long) ring->wqres.buf.direct.map);
+
+       err = mlx4_qp_reserve_range(mdev->dev, 1, 1, &ring->qpn);
+       if (err) {
+               mlx4_err(mdev, "Failed reserving qp for tx ring.\n");
+               goto err_map;
+       }
+
+       err = mlx4_qp_alloc(mdev->dev, ring->qpn, &ring->qp);
+       if (err) {
+               mlx4_err(mdev, "Failed allocating qp %d\n", ring->qpn);
+               goto err_reserve;
+       }
+
+       return 0;
+
+err_reserve:
+       mlx4_qp_release_range(mdev->dev, ring->qpn, 1);
+err_map:
+       mlx4_en_unmap_buffer(&ring->wqres.buf);
+err_hwq_res:
+       mlx4_free_hwq_res(mdev->dev, &ring->wqres, ring->buf_size);
+err_bounce:
+       kfree(ring->bounce_buf);
+       ring->bounce_buf = NULL;
+err_tx:
+       vfree(ring->tx_info);
+       ring->tx_info = NULL;
+       return err;
+}
+
+void mlx4_en_destroy_tx_ring(struct mlx4_en_priv *priv,
+                            struct mlx4_en_tx_ring *ring)
+{
+       struct mlx4_en_dev *mdev = priv->mdev;
+       mlx4_dbg(DRV, priv, "Destroying tx ring, qpn: %d\n", ring->qpn);
+
+       mlx4_qp_remove(mdev->dev, &ring->qp);
+       mlx4_qp_free(mdev->dev, &ring->qp);
+       mlx4_qp_release_range(mdev->dev, ring->qpn, 1);
+       mlx4_en_unmap_buffer(&ring->wqres.buf);
+       mlx4_free_hwq_res(mdev->dev, &ring->wqres, ring->buf_size);
+       kfree(ring->bounce_buf);
+       ring->bounce_buf = NULL;
+       vfree(ring->tx_info);
+       ring->tx_info = NULL;
+}
+
+int mlx4_en_activate_tx_ring(struct mlx4_en_priv *priv,
+                            struct mlx4_en_tx_ring *ring,
+                            int cq, int srqn)
+{
+       struct mlx4_en_dev *mdev = priv->mdev;
+       int err;
+
+       ring->cqn = cq;
+       ring->prod = 0;
+       ring->cons = 0xffffffff;
+       ring->last_nr_txbb = 1;
+       ring->poll_cnt = 0;
+       ring->blocked = 0;
+       memset(ring->tx_info, 0, ring->size * sizeof(struct mlx4_en_tx_info));
+       memset(ring->buf, 0, ring->buf_size);
+
+       ring->qp_state = MLX4_QP_STATE_RST;
+       ring->doorbell_qpn = swab32(ring->qp.qpn << 8);
+
+       mlx4_en_fill_qp_context(priv, ring->size, ring->stride, 1, 0, ring->qpn,
+                               ring->cqn, srqn, &ring->context);
+
+       err = mlx4_qp_to_ready(mdev->dev, &ring->wqres.mtt, &ring->context,
+                              &ring->qp, &ring->qp_state);
+
+       return err;
+}
+
+void mlx4_en_deactivate_tx_ring(struct mlx4_en_priv *priv,
+                               struct mlx4_en_tx_ring *ring)
+{
+       struct mlx4_en_dev *mdev = priv->mdev;
+
+       mlx4_qp_modify(mdev->dev, NULL, ring->qp_state,
+                      MLX4_QP_STATE_RST, NULL, 0, 0, &ring->qp);
+}
+
+
+static u32 mlx4_en_free_tx_desc(struct mlx4_en_priv *priv,
+                               struct mlx4_en_tx_ring *ring,
+                               int index, u8 owner)
+{
+       struct mlx4_en_dev *mdev = priv->mdev;
+       struct mlx4_en_tx_info *tx_info = &ring->tx_info[index];
+       struct mlx4_en_tx_desc *tx_desc = ring->buf + index * TXBB_SIZE;
+       struct mlx4_wqe_data_seg *data = (void *) tx_desc + tx_info->data_offset;
+       struct sk_buff *skb = tx_info->skb;
+       struct skb_frag_struct *frag;
+       void *end = ring->buf + ring->buf_size;
+       int frags = skb_shinfo(skb)->nr_frags;
+       int i;
+       __be32 *ptr = (__be32 *)tx_desc;
+       __be32 stamp = cpu_to_be32(STAMP_VAL | (!!owner << STAMP_SHIFT));
+
+       /* Optimize the common case when there are no wraparounds */
+       if (likely((void *) tx_desc + tx_info->nr_txbb * TXBB_SIZE <= end)) {
+               if (tx_info->linear) {
+                       pci_unmap_single(mdev->pdev,
+                                        (dma_addr_t) be64_to_cpu(data->addr),
+                                        be32_to_cpu(data->byte_count),
+                                        PCI_DMA_TODEVICE);
+                       ++data;
+               }
+
+               for (i = 0; i < frags; i++) {
+                       frag = &skb_shinfo(skb)->frags[i];
+                       pci_unmap_page(mdev->pdev,
+                                      (dma_addr_t) be64_to_cpu(data[i].addr),
+                                      frag->size, PCI_DMA_TODEVICE);
+               }
+               /* Stamp the freed descriptor */
+               for (i = 0; i < tx_info->nr_txbb * TXBB_SIZE; i += STAMP_STRIDE) {
+                       *ptr = stamp;
+                       ptr += STAMP_DWORDS;
+               }
+
+       } else {
+               if ((void *) data >= end) {
+                       data = (struct mlx4_wqe_data_seg *)
+                                       (ring->buf + ((void *) data - end));
+               }
+
+               if (tx_info->linear) {
+                       pci_unmap_single(mdev->pdev,
+                                        (dma_addr_t) be64_to_cpu(data->addr),
+                                        be32_to_cpu(data->byte_count),
+                                        PCI_DMA_TODEVICE);
+                       ++data;
+               }
+
+               for (i = 0; i < frags; i++) {
+                       /* Check for wraparound before unmapping */
+                       if ((void *) data >= end)
+                               data = (struct mlx4_wqe_data_seg *) ring->buf;
+                       frag = &skb_shinfo(skb)->frags[i];
+                       pci_unmap_page(mdev->pdev,
+                                       (dma_addr_t) be64_to_cpu(data->addr),
+                                        frag->size, PCI_DMA_TODEVICE);
+               }
+               /* Stamp the freed descriptor */
+               for (i = 0; i < tx_info->nr_txbb * TXBB_SIZE; i += STAMP_STRIDE) {
+                       *ptr = stamp;
+                       ptr += STAMP_DWORDS;
+                       if ((void *) ptr >= end) {
+                               ptr = ring->buf;
+                               stamp ^= cpu_to_be32(0x80000000);
+                       }
+               }
+
+       }
+       dev_kfree_skb_any(skb);
+       return tx_info->nr_txbb;
+}
+
+
+int mlx4_en_free_tx_buf(struct net_device *dev, struct mlx4_en_tx_ring *ring)
+{
+       struct mlx4_en_priv *priv = netdev_priv(dev);
+       int cnt = 0;
+
+       /* Skip last polled descriptor */
+       ring->cons += ring->last_nr_txbb;
+       mlx4_dbg(DRV, priv, "Freeing Tx buf - cons:0x%x prod:0x%x\n",
+                ring->cons, ring->prod);
+
+       if ((u32) (ring->prod - ring->cons) > ring->size) {
+               if (netif_msg_tx_err(priv))
+                       mlx4_warn(priv->mdev, "Tx consumer passed producer!\n");
+               return 0;
+       }
+
+       while (ring->cons != ring->prod) {
+               ring->last_nr_txbb = mlx4_en_free_tx_desc(priv, ring,
+                                               ring->cons & ring->size_mask,
+                                               !!(ring->cons & ring->size));
+               ring->cons += ring->last_nr_txbb;
+               cnt++;
+       }
+
+       if (cnt)
+               mlx4_dbg(DRV, priv, "Freed %d uncompleted tx descriptors\n", cnt);
+
+       return cnt;
+}
+
+void mlx4_en_set_prio_map(struct mlx4_en_priv *priv, u16 *prio_map, u32 ring_num)
+{
+       int block = 8 / ring_num;
+       int extra = 8 - (block * ring_num);
+       int num = 0;
+       u16 ring = 1;
+       int prio;
+
+       if (ring_num == 1) {
+               for (prio = 0; prio < 8; prio++)
+                       prio_map[prio] = 0;
+               return;
+       }
+
+       for (prio = 0; prio < 8; prio++) {
+               if (extra && (num == block + 1)) {
+                       ring++;
+                       num = 0;
+                       extra--;
+               } else if (!extra && (num == block)) {
+                       ring++;
+                       num = 0;
+               }
+               prio_map[prio] = ring;
+               mlx4_dbg(DRV, priv, " prio:%d --> ring:%d\n", prio, ring);
+               num++;
+       }
+}
+
+static void mlx4_en_process_tx_cq(struct net_device *dev, struct mlx4_en_cq *cq)
+{
+       struct mlx4_en_priv *priv = netdev_priv(dev);
+       struct mlx4_cq *mcq = &cq->mcq;
+       struct mlx4_en_tx_ring *ring = &priv->tx_ring[cq->ring];
+       struct mlx4_cqe *cqe = cq->buf;
+       u16 index;
+       u16 new_index;
+       u32 txbbs_skipped = 0;
+       u32 cq_last_sav;
+
+       /* index always points to the first TXBB of the last polled descriptor */
+       index = ring->cons & ring->size_mask;
+       new_index = be16_to_cpu(cqe->wqe_index) & ring->size_mask;
+       if (index == new_index)
+               return;
+
+       if (!priv->port_up)
+               return;
+
+       /*
+        * We use a two-stage loop:
+        * - the first samples the HW-updated CQE
+        * - the second frees TXBBs until the last sample
+        * This lets us amortize CQE cache misses, while still polling the CQ
+        * until is quiescent.
+        */
+       cq_last_sav = mcq->cons_index;
+       do {
+               do {
+                       /* Skip over last polled CQE */
+                       index = (index + ring->last_nr_txbb) & ring->size_mask;
+                       txbbs_skipped += ring->last_nr_txbb;
+
+                       /* Poll next CQE */
+                       ring->last_nr_txbb = mlx4_en_free_tx_desc(
+                                               priv, ring, index,
+                                               !!((ring->cons + txbbs_skipped) &
+                                                  ring->size));
+                       ++mcq->cons_index;
+
+               } while (index != new_index);
+
+               new_index = be16_to_cpu(cqe->wqe_index) & ring->size_mask;
+       } while (index != new_index);
+       AVG_PERF_COUNTER(priv->pstats.tx_coal_avg,
+                        (u32) (mcq->cons_index - cq_last_sav));
+
+       /*
+        * To prevent CQ overflow we first update CQ consumer and only then
+        * the ring consumer.
+        */
+       mlx4_cq_set_ci(mcq);
+       wmb();
+       ring->cons += txbbs_skipped;
+
+       /* Wakeup Tx queue if this ring stopped it */
+       if (unlikely(ring->blocked)) {
+               if (((u32) (ring->prod - ring->cons) <=
+                    ring->size - HEADROOM - MAX_DESC_TXBBS) && !cq->armed) {
+
+                       /* TODO: support multiqueue netdevs. Currently, we block
+                        * when *any* ring is full. Note that:
+                        * - 2 Tx rings can unblock at the same time and call
+                        *   netif_wake_queue(), which is OK since this
+                        *   operation is idempotent.
+                        * - We might wake the queue just after another ring
+                        *   stopped it. This is no big deal because the next
+                        *   transmission on that ring would stop the queue.
+                        */
+                       ring->blocked = 0;
+                       netif_wake_queue(dev);
+                       priv->port_stats.wake_queue++;
+               }
+       }
+}
+
+void mlx4_en_tx_irq(struct mlx4_cq *mcq)
+{
+       struct mlx4_en_cq *cq = container_of(mcq, struct mlx4_en_cq, mcq);
+       struct mlx4_en_priv *priv = netdev_priv(cq->dev);
+       struct mlx4_en_tx_ring *ring = &priv->tx_ring[cq->ring];
+
+       spin_lock_irq(&ring->comp_lock);
+       cq->armed = 0;
+       mlx4_en_process_tx_cq(cq->dev, cq);
+       if (ring->blocked)
+               mlx4_en_arm_cq(priv, cq);
+       else
+               mod_timer(&cq->timer, jiffies + 1);
+       spin_unlock_irq(&ring->comp_lock);
+}
+
+
+void mlx4_en_poll_tx_cq(unsigned long data)
+{
+       struct mlx4_en_cq *cq = (struct mlx4_en_cq *) data;
+       struct mlx4_en_priv *priv = netdev_priv(cq->dev);
+       struct mlx4_en_tx_ring *ring = &priv->tx_ring[cq->ring];
+       u32 inflight;
+
+       INC_PERF_COUNTER(priv->pstats.tx_poll);
+
+       netif_tx_lock(priv->dev);
+       spin_lock_irq(&ring->comp_lock);
+       mlx4_en_process_tx_cq(cq->dev, cq);
+       inflight = (u32) (ring->prod - ring->cons - ring->last_nr_txbb);
+
+       /* If there are still packets in flight and the timer has not already
+        * been scheduled by the Tx routine then schedule it here to guarantee
+        * completion processing of these packets */
+       if (inflight && priv->port_up)
+               mod_timer(&cq->timer, jiffies + MLX4_EN_TX_POLL_TIMEOUT);
+
+       spin_unlock_irq(&ring->comp_lock);
+       netif_tx_unlock(priv->dev);
+}
+
+static struct mlx4_en_tx_desc *mlx4_en_bounce_to_desc(struct mlx4_en_priv *priv,
+                                                     struct mlx4_en_tx_ring *ring,
+                                                     u32 index,
+                                                     unsigned int desc_size)
+{
+       u32 copy = (ring->size - index) * TXBB_SIZE;
+       int i;
+
+       for (i = desc_size - copy - 4; i >= 0; i -= 4) {
+               if ((i & (TXBB_SIZE - 1)) == 0)
+                       wmb();
+
+               *((u32 *) (ring->buf + i)) =
+                       *((u32 *) (ring->bounce_buf + copy + i));
+       }
+
+       for (i = copy - 4; i >= 4 ; i -= 4) {
+               if ((i & (TXBB_SIZE - 1)) == 0)
+                       wmb();
+
+               *((u32 *) (ring->buf + index * TXBB_SIZE + i)) =
+                       *((u32 *) (ring->bounce_buf + i));
+       }
+
+       /* Return real descriptor location */
+       return ring->buf + index * TXBB_SIZE;
+}
+
+static inline void mlx4_en_xmit_poll(struct mlx4_en_priv *priv, int tx_ind)
+{
+       struct mlx4_en_cq *cq = &priv->tx_cq[tx_ind];
+       struct mlx4_en_tx_ring *ring = &priv->tx_ring[tx_ind];
+
+       /* If we don't have a pending timer, set one up to catch our recent
+          post in case the interface becomes idle */
+       if (!timer_pending(&cq->timer))
+               mod_timer(&cq->timer, jiffies + MLX4_EN_TX_POLL_TIMEOUT);
+
+       /* Poll the CQ every mlx4_en_TX_MODER_POLL packets */
+       if ((++ring->poll_cnt & (MLX4_EN_TX_POLL_MODER - 1)) == 0)
+               mlx4_en_process_tx_cq(priv->dev, cq);
+}
+
+static void *get_frag_ptr(struct sk_buff *skb)
+{
+       struct skb_frag_struct *frag =  &skb_shinfo(skb)->frags[0];
+       struct page *page = frag->page;
+       void *ptr;
+
+       ptr = page_address(page);
+       if (unlikely(!ptr))
+               return NULL;
+
+       return ptr + frag->page_offset;
+}
+
+static int is_inline(struct sk_buff *skb, void **pfrag)
+{
+       void *ptr;
+
+       if (inline_thold && !skb_is_gso(skb) && skb->len <= inline_thold) {
+               if (skb_shinfo(skb)->nr_frags == 1) {
+                       ptr = get_frag_ptr(skb);
+                       if (unlikely(!ptr))
+                               return 0;
+
+                       if (pfrag)
+                               *pfrag = ptr;
+
+                       return 1;
+               } else if (unlikely(skb_shinfo(skb)->nr_frags))
+                       return 0;
+               else
+                       return 1;
+       }
+
+       return 0;
+}
+
+static int inline_size(struct sk_buff *skb)
+{
+       if (skb->len + CTRL_SIZE + sizeof(struct mlx4_wqe_inline_seg)
+           <= MLX4_INLINE_ALIGN)
+               return ALIGN(skb->len + CTRL_SIZE +
+                            sizeof(struct mlx4_wqe_inline_seg), 16);
+       else
+               return ALIGN(skb->len + CTRL_SIZE + 2 *
+                            sizeof(struct mlx4_wqe_inline_seg), 16);
+}
+
+static int get_real_size(struct sk_buff *skb, struct net_device *dev,
+                        int *lso_header_size)
+{
+       struct mlx4_en_priv *priv = netdev_priv(dev);
+       struct mlx4_en_dev *mdev = priv->mdev;
+       int real_size;
+
+       if (skb_is_gso(skb)) {
+               *lso_header_size = skb_transport_offset(skb) + tcp_hdrlen(skb);
+               real_size = CTRL_SIZE + skb_shinfo(skb)->nr_frags * DS_SIZE +
+                       ALIGN(*lso_header_size + 4, DS_SIZE);
+               if (unlikely(*lso_header_size != skb_headlen(skb))) {
+                       /* We add a segment for the skb linear buffer only if
+                        * it contains data */
+                       if (*lso_header_size < skb_headlen(skb))
+                               real_size += DS_SIZE;
+                       else {
+                               if (netif_msg_tx_err(priv))
+                                       mlx4_warn(mdev, "Non-linear headers\n");
+                               dev_kfree_skb_any(skb);
+                               return 0;
+                       }
+               }
+               if (unlikely(*lso_header_size > MAX_LSO_HDR_SIZE)) {
+                       if (netif_msg_tx_err(priv))
+                               mlx4_warn(mdev, "LSO header size too big\n");
+                       dev_kfree_skb_any(skb);
+                       return 0;
+               }
+       } else {
+               *lso_header_size = 0;
+               if (!is_inline(skb, NULL))
+                       real_size = CTRL_SIZE + (skb_shinfo(skb)->nr_frags + 1) * DS_SIZE;
+               else
+                       real_size = inline_size(skb);
+       }
+
+       return real_size;
+}
+
+static void build_inline_wqe(struct mlx4_en_tx_desc *tx_desc, struct sk_buff *skb,
+                            int real_size, u16 *vlan_tag, int tx_ind, void *fragptr)
+{
+       struct mlx4_wqe_inline_seg *inl = &tx_desc->inl;
+       int spc = MLX4_INLINE_ALIGN - CTRL_SIZE - sizeof *inl;
+
+       if (skb->len <= spc) {
+               inl->byte_count = cpu_to_be32(1 << 31 | skb->len);
+               skb_copy_from_linear_data(skb, inl + 1, skb_headlen(skb));
+               if (skb_shinfo(skb)->nr_frags)
+                       memcpy(((void *)(inl + 1)) + skb_headlen(skb), fragptr,
+                              skb_shinfo(skb)->frags[0].size);
+
+       } else {
+               inl->byte_count = cpu_to_be32(1 << 31 | spc);
+               if (skb_headlen(skb) <= spc) {
+                       skb_copy_from_linear_data(skb, inl + 1, skb_headlen(skb));
+                       if (skb_headlen(skb) < spc) {
+                               memcpy(((void *)(inl + 1)) + skb_headlen(skb),
+                                       fragptr, spc - skb_headlen(skb));
+                               fragptr +=  spc - skb_headlen(skb);
+                       }
+                       inl = (void *) (inl + 1) + spc;
+                       memcpy(((void *)(inl + 1)), fragptr, skb->len - spc);
+               } else {
+                       skb_copy_from_linear_data(skb, inl + 1, spc);
+                       inl = (void *) (inl + 1) + spc;
+                       skb_copy_from_linear_data_offset(skb, spc, inl + 1,
+                                       skb_headlen(skb) - spc);
+                       if (skb_shinfo(skb)->nr_frags)
+                               memcpy(((void *)(inl + 1)) + skb_headlen(skb) - spc,
+                                       fragptr, skb_shinfo(skb)->frags[0].size);
+               }
+
+               wmb();
+               inl->byte_count = cpu_to_be32(1 << 31 | (skb->len - spc));
+       }
+       tx_desc->ctrl.vlan_tag = cpu_to_be16(*vlan_tag);
+       tx_desc->ctrl.ins_vlan = MLX4_WQE_CTRL_INS_VLAN * !!(*vlan_tag);
+       tx_desc->ctrl.fence_size = (real_size / 16) & 0x3f;
+}
+
+static int get_vlan_info(struct mlx4_en_priv *priv, struct sk_buff *skb,
+                        u16 *vlan_tag)
+{
+       int tx_ind;
+
+       /* Obtain VLAN information if present */
+       if (priv->vlgrp && vlan_tx_tag_present(skb)) {
+               *vlan_tag = vlan_tx_tag_get(skb);
+               /* Set the Tx ring to use according to vlan priority */
+               tx_ind = priv->tx_prio_map[*vlan_tag >> 13];
+       } else {
+               *vlan_tag = 0;
+               tx_ind = 0;
+       }
+       return tx_ind;
+}
+
+int mlx4_en_xmit(struct sk_buff *skb, struct net_device *dev)
+{
+       struct mlx4_en_priv *priv = netdev_priv(dev);
+       struct mlx4_en_dev *mdev = priv->mdev;
+       struct mlx4_en_tx_ring *ring;
+       struct mlx4_en_cq *cq;
+       struct mlx4_en_tx_desc *tx_desc;
+       struct mlx4_wqe_data_seg *data;
+       struct skb_frag_struct *frag;
+       struct mlx4_en_tx_info *tx_info;
+       int tx_ind = 0;
+       int nr_txbb;
+       int desc_size;
+       int real_size;
+       dma_addr_t dma;
+       u32 index;
+       __be32 op_own;
+       u16 vlan_tag;
+       int i;
+       int lso_header_size;
+       void *fragptr;
+
+       if (unlikely(!skb->len)) {
+               dev_kfree_skb_any(skb);
+               return NETDEV_TX_OK;
+       }
+       real_size = get_real_size(skb, dev, &lso_header_size);
+       if (unlikely(!real_size))
+               return NETDEV_TX_OK;
+
+       /* Allign descriptor to TXBB size */
+       desc_size = ALIGN(real_size, TXBB_SIZE);
+       nr_txbb = desc_size / TXBB_SIZE;
+       if (unlikely(nr_txbb > MAX_DESC_TXBBS)) {
+               if (netif_msg_tx_err(priv))
+                       mlx4_warn(mdev, "Oversized header or SG list\n");
+               dev_kfree_skb_any(skb);
+               return NETDEV_TX_OK;
+       }
+
+       tx_ind = get_vlan_info(priv, skb, &vlan_tag);
+       ring = &priv->tx_ring[tx_ind];
+
+       /* Check available TXBBs And 2K spare for prefetch */
+       if (unlikely(((int)(ring->prod - ring->cons)) >
+                    ring->size - HEADROOM - MAX_DESC_TXBBS)) {
+               /* every full Tx ring stops queue.
+                * TODO: implement multi-queue support (per-queue stop) */
+               netif_stop_queue(dev);
+               ring->blocked = 1;
+               priv->port_stats.queue_stopped++;
+
+               /* Use interrupts to find out when queue opened */
+               cq = &priv->tx_cq[tx_ind];
+               mlx4_en_arm_cq(priv, cq);
+               return NETDEV_TX_BUSY;
+       }
+
+       /* Now that we know what Tx ring to use */
+       if (unlikely(!priv->port_up)) {
+               if (netif_msg_tx_err(priv))
+                       mlx4_warn(mdev, "xmit: port down!\n");
+               dev_kfree_skb_any(skb);
+               return NETDEV_TX_OK;
+       }
+
+       /* Track current inflight packets for performance analysis */
+       AVG_PERF_COUNTER(priv->pstats.inflight_avg,
+                        (u32) (ring->prod - ring->cons - 1));
+
+       /* Packet is good - grab an index and transmit it */
+       index = ring->prod & ring->size_mask;
+
+       /* See if we have enough space for whole descriptor TXBB for setting
+        * SW ownership on next descriptor; if not, use a bounce buffer. */
+       if (likely(index + nr_txbb <= ring->size))
+               tx_desc = ring->buf + index * TXBB_SIZE;
+       else
+               tx_desc = (struct mlx4_en_tx_desc *) ring->bounce_buf;
+
+       /* Save skb in tx_info ring */
+       tx_info = &ring->tx_info[index];
+       tx_info->skb = skb;
+       tx_info->nr_txbb = nr_txbb;
+
+       /* Prepare ctrl segement apart opcode+ownership, which depends on
+        * whether LSO is used */
+       tx_desc->ctrl.vlan_tag = cpu_to_be16(vlan_tag);
+       tx_desc->ctrl.ins_vlan = MLX4_WQE_CTRL_INS_VLAN * !!vlan_tag;
+       tx_desc->ctrl.fence_size = (real_size / 16) & 0x3f;
+       tx_desc->ctrl.srcrb_flags = cpu_to_be32(MLX4_WQE_CTRL_CQ_UPDATE |
+                                               MLX4_WQE_CTRL_SOLICITED);
+       if (likely(skb->ip_summed == CHECKSUM_PARTIAL)) {
+               tx_desc->ctrl.srcrb_flags |= cpu_to_be32(MLX4_WQE_CTRL_IP_CSUM |
+                                                        MLX4_WQE_CTRL_TCP_UDP_CSUM);
+               priv->port_stats.tx_chksum_offload++;
+       }
+
+       /* Handle LSO (TSO) packets */
+       if (lso_header_size) {
+               /* Mark opcode as LSO */
+               op_own = cpu_to_be32(MLX4_OPCODE_LSO | (1 << 6)) |
+                       ((ring->prod & ring->size) ?
+                               cpu_to_be32(MLX4_EN_BIT_DESC_OWN) : 0);
+
+               /* Fill in the LSO prefix */
+               tx_desc->lso.mss_hdr_size = cpu_to_be32(
+                       skb_shinfo(skb)->gso_size << 16 | lso_header_size);
+
+               /* Copy headers;
+                * note that we already verified that it is linear */
+               memcpy(tx_desc->lso.header, skb->data, lso_header_size);
+               data = ((void *) &tx_desc->lso +
+                       ALIGN(lso_header_size + 4, DS_SIZE));
+
+               priv->port_stats.tso_packets++;
+               i = ((skb->len - lso_header_size) / skb_shinfo(skb)->gso_size) +
+                       !!((skb->len - lso_header_size) % skb_shinfo(skb)->gso_size);
+               ring->bytes += skb->len + (i - 1) * lso_header_size;
+               ring->packets += i;
+       } else {
+               /* Normal (Non LSO) packet */
+               op_own = cpu_to_be32(MLX4_OPCODE_SEND) |
+                       ((ring->prod & ring->size) ?
+                        cpu_to_be32(MLX4_EN_BIT_DESC_OWN) : 0);
+               data = &tx_desc->data;
+               ring->bytes += max(skb->len, (unsigned int) ETH_ZLEN);
+               ring->packets++;
+
+       }
+       AVG_PERF_COUNTER(priv->pstats.tx_pktsz_avg, skb->len);
+
+
+       /* valid only for none inline segments */
+       tx_info->data_offset = (void *) data - (void *) tx_desc;
+
+       tx_info->linear = (lso_header_size < skb_headlen(skb) && !is_inline(skb, NULL)) ? 1 : 0;
+       data += skb_shinfo(skb)->nr_frags + tx_info->linear - 1;
+
+       if (!is_inline(skb, &fragptr)) {
+               /* Map fragments */
+               for (i = skb_shinfo(skb)->nr_frags - 1; i >= 0; i--) {
+                       frag = &skb_shinfo(skb)->frags[i];
+                       dma = pci_map_page(mdev->dev->pdev, frag->page, frag->page_offset,
+                                          frag->size, PCI_DMA_TODEVICE);
+                       data->addr = cpu_to_be64(dma);
+                       data->lkey = cpu_to_be32(mdev->mr.key);
+                       wmb();
+                       data->byte_count = cpu_to_be32(frag->size);
+                       --data;
+               }
+
+               /* Map linear part */
+               if (tx_info->linear) {
+                       dma = pci_map_single(mdev->dev->pdev, skb->data + lso_header_size,
+                                            skb_headlen(skb) - lso_header_size, PCI_DMA_TODEVICE);
+                       data->addr = cpu_to_be64(dma);
+                       data->lkey = cpu_to_be32(mdev->mr.key);
+                       wmb();
+                       data->byte_count = cpu_to_be32(skb_headlen(skb) - lso_header_size);
+               }
+       } else
+               build_inline_wqe(tx_desc, skb, real_size, &vlan_tag, tx_ind, fragptr);
+
+       ring->prod += nr_txbb;
+
+       /* If we used a bounce buffer then copy descriptor back into place */
+       if (tx_desc == (struct mlx4_en_tx_desc *) ring->bounce_buf)
+               tx_desc = mlx4_en_bounce_to_desc(priv, ring, index, desc_size);
+
+       /* Run destructor before passing skb to HW */
+       if (likely(!skb_shared(skb)))
+               skb_orphan(skb);
+
+       /* Ensure new descirptor hits memory
+        * before setting ownership of this descriptor to HW */
+       wmb();
+       tx_desc->ctrl.owner_opcode = op_own;
+
+       /* Ring doorbell! */
+       wmb();
+       writel(ring->doorbell_qpn, mdev->uar_map + MLX4_SEND_DOORBELL);
+       dev->trans_start = jiffies;
+
+       /* Poll CQ here */
+       mlx4_en_xmit_poll(priv, tx_ind);
+
+       return 0;
+}
+
diff --git a/drivers/net/mlx4/mlx4_en.h b/drivers/net/mlx4/mlx4_en.h
new file mode 100644 (file)
index 0000000..11fb17c
--- /dev/null
@@ -0,0 +1,561 @@
+/*
+ * 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.
+ *
+ */
+
+#ifndef _MLX4_EN_H_
+#define _MLX4_EN_H_
+
+#include <linux/compiler.h>
+#include <linux/list.h>
+#include <linux/mutex.h>
+#include <linux/netdevice.h>
+#include <linux/inet_lro.h>
+
+#include <linux/mlx4/device.h>
+#include <linux/mlx4/qp.h>
+#include <linux/mlx4/cq.h>
+#include <linux/mlx4/srq.h>
+#include <linux/mlx4/doorbell.h>
+
+#include "en_port.h"
+
+#define DRV_NAME       "mlx4_en"
+#define DRV_VERSION    "1.4.0"
+#define DRV_RELDATE    "Sep 2008"
+
+
+#define MLX4_EN_MSG_LEVEL      (NETIF_MSG_LINK | NETIF_MSG_IFDOWN)
+
+#define mlx4_dbg(mlevel, priv, format, arg...) \
+       if (NETIF_MSG_##mlevel & priv->msg_enable) \
+       printk(KERN_DEBUG "%s %s: " format , DRV_NAME ,\
+               (&priv->mdev->pdev->dev)->bus_id , ## arg)
+
+#define mlx4_err(mdev, format, arg...) \
+       printk(KERN_ERR "%s %s: " format , DRV_NAME ,\
+               (&mdev->pdev->dev)->bus_id , ## arg)
+#define mlx4_info(mdev, format, arg...) \
+       printk(KERN_INFO "%s %s: " format , DRV_NAME ,\
+               (&mdev->pdev->dev)->bus_id , ## arg)
+#define mlx4_warn(mdev, format, arg...) \
+       printk(KERN_WARNING "%s %s: " format , DRV_NAME ,\
+               (&mdev->pdev->dev)->bus_id , ## arg)
+
+/*
+ * Device constants
+ */
+
+
+#define MLX4_EN_PAGE_SHIFT     12
+#define MLX4_EN_PAGE_SIZE      (1 << MLX4_EN_PAGE_SHIFT)
+#define MAX_TX_RINGS           16
+#define MAX_RX_RINGS           16
+#define MAX_RSS_MAP_SIZE       64
+#define RSS_FACTOR             2
+#define TXBB_SIZE              64
+#define HEADROOM               (2048 / TXBB_SIZE + 1)
+#define MAX_LSO_HDR_SIZE       92
+#define STAMP_STRIDE           64
+#define STAMP_DWORDS           (STAMP_STRIDE / 4)
+#define STAMP_SHIFT            31
+#define STAMP_VAL              0x7fffffff
+#define STATS_DELAY            (HZ / 4)
+
+/* Typical TSO descriptor with 16 gather entries is 352 bytes... */
+#define MAX_DESC_SIZE          512
+#define MAX_DESC_TXBBS         (MAX_DESC_SIZE / TXBB_SIZE)
+
+/*
+ * OS related constants and tunables
+ */
+
+#define MLX4_EN_WATCHDOG_TIMEOUT       (15 * HZ)
+
+#define MLX4_EN_ALLOC_ORDER    2
+#define MLX4_EN_ALLOC_SIZE     (PAGE_SIZE << MLX4_EN_ALLOC_ORDER)
+
+#define MLX4_EN_MAX_LRO_DESCRIPTORS    32
+
+/* Receive fragment sizes; we use at most 4 fragments (for 9600 byte MTU
+ * and 4K allocations) */
+enum {
+       FRAG_SZ0 = 512 - NET_IP_ALIGN,
+       FRAG_SZ1 = 1024,
+       FRAG_SZ2 = 4096,
+       FRAG_SZ3 = MLX4_EN_ALLOC_SIZE
+};
+#define MLX4_EN_MAX_RX_FRAGS   4
+
+/* Minimum ring size for our page-allocation sceme to work */
+#define MLX4_EN_MIN_RX_SIZE    (MLX4_EN_ALLOC_SIZE / SMP_CACHE_BYTES)
+#define MLX4_EN_MIN_TX_SIZE    (4096 / TXBB_SIZE)
+
+#define MLX4_EN_TX_RING_NUM            9
+#define MLX4_EN_DEF_TX_RING_SIZE       1024
+#define MLX4_EN_DEF_RX_RING_SIZE       1024
+
+/* Target number of bytes to coalesce with interrupt moderation */
+#define MLX4_EN_RX_COAL_TARGET 0x20000
+#define MLX4_EN_RX_COAL_TIME   0x10
+
+#define MLX4_EN_TX_COAL_PKTS   5
+#define MLX4_EN_TX_COAL_TIME   0x80
+
+#define MLX4_EN_RX_RATE_LOW            400000
+#define MLX4_EN_RX_COAL_TIME_LOW       0
+#define MLX4_EN_RX_RATE_HIGH           450000
+#define MLX4_EN_RX_COAL_TIME_HIGH      128
+#define MLX4_EN_RX_SIZE_THRESH         1024
+#define MLX4_EN_RX_RATE_THRESH         (1000000 / MLX4_EN_RX_COAL_TIME_HIGH)
+#define MLX4_EN_SAMPLE_INTERVAL                0
+
+#define MLX4_EN_AUTO_CONF      0xffff
+
+#define MLX4_EN_DEF_RX_PAUSE   1
+#define MLX4_EN_DEF_TX_PAUSE   1
+
+/* Interval between sucessive polls in the Tx routine when polling is used
+   instead of interrupts (in per-core Tx rings) - should be power of 2 */
+#define MLX4_EN_TX_POLL_MODER  16
+#define MLX4_EN_TX_POLL_TIMEOUT        (HZ / 4)
+
+#define ETH_LLC_SNAP_SIZE      8
+
+#define SMALL_PACKET_SIZE      (256 - NET_IP_ALIGN)
+#define HEADER_COPY_SIZE       (128 - NET_IP_ALIGN)
+
+#define MLX4_EN_MIN_MTU                46
+#define ETH_BCAST              0xffffffffffffULL
+
+#ifdef MLX4_EN_PERF_STAT
+/* Number of samples to 'average' */
+#define AVG_SIZE                       128
+#define AVG_FACTOR                     1024
+#define NUM_PERF_STATS                 NUM_PERF_COUNTERS
+
+#define INC_PERF_COUNTER(cnt)          (++(cnt))
+#define ADD_PERF_COUNTER(cnt, add)     ((cnt) += (add))
+#define AVG_PERF_COUNTER(cnt, sample) \
+       ((cnt) = ((cnt) * (AVG_SIZE - 1) + (sample) * AVG_FACTOR) / AVG_SIZE)
+#define GET_PERF_COUNTER(cnt)          (cnt)
+#define GET_AVG_PERF_COUNTER(cnt)      ((cnt) / AVG_FACTOR)
+
+#else
+
+#define NUM_PERF_STATS                 0
+#define INC_PERF_COUNTER(cnt)          do {} while (0)
+#define ADD_PERF_COUNTER(cnt, add)     do {} while (0)
+#define AVG_PERF_COUNTER(cnt, sample)  do {} while (0)
+#define GET_PERF_COUNTER(cnt)          (0)
+#define GET_AVG_PERF_COUNTER(cnt)      (0)
+#endif /* MLX4_EN_PERF_STAT */
+
+/*
+ * Configurables
+ */
+
+enum cq_type {
+       RX = 0,
+       TX = 1,
+};
+
+
+/*
+ * Useful macros
+ */
+#define ROUNDUP_LOG2(x)                ilog2(roundup_pow_of_two(x))
+#define XNOR(x, y)             (!(x) == !(y))
+#define ILLEGAL_MAC(addr)      (addr == 0xffffffffffffULL || addr == 0x0)
+
+
+struct mlx4_en_tx_info {
+       struct sk_buff *skb;
+       u32 nr_txbb;
+       u8 linear;
+       u8 data_offset;
+};
+
+
+#define MLX4_EN_BIT_DESC_OWN   0x80000000
+#define CTRL_SIZE      sizeof(struct mlx4_wqe_ctrl_seg)
+#define MLX4_EN_MEMTYPE_PAD    0x100
+#define DS_SIZE                sizeof(struct mlx4_wqe_data_seg)
+
+
+struct mlx4_en_tx_desc {
+       struct mlx4_wqe_ctrl_seg ctrl;
+       union {
+               struct mlx4_wqe_data_seg data; /* at least one data segment */
+               struct mlx4_wqe_lso_seg lso;
+               struct mlx4_wqe_inline_seg inl;
+       };
+};
+
+#define MLX4_EN_USE_SRQ                0x01000000
+
+struct mlx4_en_rx_alloc {
+       struct page *page;
+       u16 offset;
+};
+
+struct mlx4_en_tx_ring {
+       struct mlx4_hwq_resources wqres;
+       u32 size ; /* number of TXBBs */
+       u32 size_mask;
+       u16 stride;
+       u16 cqn;        /* index of port CQ associated with this ring */
+       u32 prod;
+       u32 cons;
+       u32 buf_size;
+       u32 doorbell_qpn;
+       void *buf;
+       u16 poll_cnt;
+       int blocked;
+       struct mlx4_en_tx_info *tx_info;
+       u8 *bounce_buf;
+       u32 last_nr_txbb;
+       struct mlx4_qp qp;
+       struct mlx4_qp_context context;
+       int qpn;
+       enum mlx4_qp_state qp_state;
+       struct mlx4_srq dummy;
+       unsigned long bytes;
+       unsigned long packets;
+       spinlock_t comp_lock;
+};
+
+struct mlx4_en_rx_desc {
+       struct mlx4_wqe_srq_next_seg next;
+       /* actual number of entries depends on rx ring stride */
+       struct mlx4_wqe_data_seg data[0];
+};
+
+struct mlx4_en_rx_ring {
+       struct mlx4_srq srq;
+       struct mlx4_hwq_resources wqres;
+       struct mlx4_en_rx_alloc page_alloc[MLX4_EN_MAX_RX_FRAGS];
+       struct net_lro_mgr lro;
+       u32 size ;      /* number of Rx descs*/
+       u32 actual_size;
+       u32 size_mask;
+       u16 stride;
+       u16 log_stride;
+       u16 cqn;        /* index of port CQ associated with this ring */
+       u32 prod;
+       u32 cons;
+       u32 buf_size;
+       int need_refill;
+       int full;
+       void *buf;
+       void *rx_info;
+       unsigned long bytes;
+       unsigned long packets;
+};
+
+
+static inline int mlx4_en_can_lro(__be16 status)
+{
+       return (status & cpu_to_be16(MLX4_CQE_STATUS_IPV4       |
+                                    MLX4_CQE_STATUS_IPV4F      |
+                                    MLX4_CQE_STATUS_IPV6       |
+                                    MLX4_CQE_STATUS_IPV4OPT    |
+                                    MLX4_CQE_STATUS_TCP        |
+                                    MLX4_CQE_STATUS_UDP        |
+                                    MLX4_CQE_STATUS_IPOK)) ==
+               cpu_to_be16(MLX4_CQE_STATUS_IPV4 |
+                           MLX4_CQE_STATUS_IPOK |
+                           MLX4_CQE_STATUS_TCP);
+}
+
+struct mlx4_en_cq {
+       struct mlx4_cq          mcq;
+       struct mlx4_hwq_resources wqres;
+       int                     ring;
+       spinlock_t              lock;
+       struct net_device      *dev;
+       struct napi_struct      napi;
+       /* Per-core Tx cq processing support */
+       struct timer_list timer;
+       int size;
+       int buf_size;
+       unsigned vector;
+       enum cq_type is_tx;
+       u16 moder_time;
+       u16 moder_cnt;
+       int armed;
+       struct mlx4_cqe *buf;
+#define MLX4_EN_OPCODE_ERROR   0x1e
+};
+
+struct mlx4_en_port_profile {
+       u32 flags;
+       u32 tx_ring_num;
+       u32 rx_ring_num;
+       u32 tx_ring_size;
+       u32 rx_ring_size;
+};
+
+struct mlx4_en_profile {
+       int rss_xor;
+       int num_lro;
+       u8 rss_mask;
+       u32 active_ports;
+       u32 small_pkt_int;
+       int rx_moder_cnt;
+       int rx_moder_time;
+       int auto_moder;
+       u8 rx_pause;
+       u8 rx_ppp;
+       u8 tx_pause;
+       u8 tx_ppp;
+       u8 no_reset;
+       struct mlx4_en_port_profile prof[MLX4_MAX_PORTS + 1];
+};
+
+struct mlx4_en_dev {
+       struct mlx4_dev         *dev;
+       struct pci_dev          *pdev;
+       struct mutex            state_lock;
+       struct net_device       *pndev[MLX4_MAX_PORTS + 1];
+       u32                     port_cnt;
+       bool                    device_up;
+       struct mlx4_en_profile  profile;
+       u32                     LSO_support;
+       struct workqueue_struct *workqueue;
+       struct device           *dma_device;
+       void __iomem            *uar_map;
+       struct mlx4_uar         priv_uar;
+       struct mlx4_mr          mr;
+       u32                     priv_pdn;
+       spinlock_t              uar_lock;
+};
+
+
+struct mlx4_en_rss_map {
+       int size;
+       int base_qpn;
+       u16 map[MAX_RSS_MAP_SIZE];
+       struct mlx4_qp qps[MAX_RSS_MAP_SIZE];
+       enum mlx4_qp_state state[MAX_RSS_MAP_SIZE];
+       struct mlx4_qp indir_qp;
+       enum mlx4_qp_state indir_state;
+};
+
+struct mlx4_en_rss_context {
+       __be32 base_qpn;
+       __be32 default_qpn;
+       u16 reserved;
+       u8 hash_fn;
+       u8 flags;
+       __be32 rss_key[10];
+};
+
+struct mlx4_en_pkt_stats {
+       unsigned long broadcast;
+       unsigned long rx_prio[8];
+       unsigned long tx_prio[8];
+#define NUM_PKT_STATS          17
+};
+
+struct mlx4_en_port_stats {
+       unsigned long lro_aggregated;
+       unsigned long lro_flushed;
+       unsigned long lro_no_desc;
+       unsigned long tso_packets;
+       unsigned long queue_stopped;
+       unsigned long wake_queue;
+       unsigned long tx_timeout;
+       unsigned long rx_alloc_failed;
+       unsigned long rx_chksum_good;
+       unsigned long rx_chksum_none;
+       unsigned long tx_chksum_offload;
+#define NUM_PORT_STATS         11
+};
+
+struct mlx4_en_perf_stats {
+       u32 tx_poll;
+       u64 tx_pktsz_avg;
+       u32 inflight_avg;
+       u16 tx_coal_avg;
+       u16 rx_coal_avg;
+       u32 napi_quota;
+#define NUM_PERF_COUNTERS              6
+};
+
+struct mlx4_en_frag_info {
+       u16 frag_size;
+       u16 frag_prefix_size;
+       u16 frag_stride;
+       u16 frag_align;
+       u16 last_offset;
+
+};
+
+struct mlx4_en_priv {
+       struct mlx4_en_dev *mdev;
+       struct mlx4_en_port_profile *prof;
+       struct net_device *dev;
+       struct vlan_group *vlgrp;
+       struct net_device_stats stats;
+       struct net_device_stats ret_stats;
+       spinlock_t stats_lock;
+
+       unsigned long last_moder_packets;
+       unsigned long last_moder_tx_packets;
+       unsigned long last_moder_bytes;
+       unsigned long last_moder_jiffies;
+       int last_moder_time;
+       u16 rx_usecs;
+       u16 rx_frames;
+       u16 tx_usecs;
+       u16 tx_frames;
+       u32 pkt_rate_low;
+       u16 rx_usecs_low;
+       u32 pkt_rate_high;
+       u16 rx_usecs_high;
+       u16 sample_interval;
+       u16 adaptive_rx_coal;
+       u32 msg_enable;
+
+       struct mlx4_hwq_resources res;
+       int link_state;
+       int last_link_state;
+       bool port_up;
+       int port;
+       int registered;
+       int allocated;
+       int stride;
+       int rx_csum;
+       u64 mac;
+       int mac_index;
+       unsigned max_mtu;
+       int base_qpn;
+
+       struct mlx4_en_rss_map rss_map;
+       u16 tx_prio_map[8];
+       u32 flags;
+#define MLX4_EN_FLAG_PROMISC   0x1
+       u32 tx_ring_num;
+       u32 rx_ring_num;
+       u32 rx_skb_size;
+       struct mlx4_en_frag_info frag_info[MLX4_EN_MAX_RX_FRAGS];
+       u16 num_frags;
+       u16 log_rx_info;
+
+       struct mlx4_en_tx_ring tx_ring[MAX_TX_RINGS];
+       struct mlx4_en_rx_ring rx_ring[MAX_RX_RINGS];
+       struct mlx4_en_cq tx_cq[MAX_TX_RINGS];
+       struct mlx4_en_cq rx_cq[MAX_RX_RINGS];
+       struct work_struct mcast_task;
+       struct work_struct mac_task;
+       struct delayed_work refill_task;
+       struct work_struct watchdog_task;
+       struct work_struct linkstate_task;
+       struct delayed_work stats_task;
+       struct mlx4_en_perf_stats pstats;
+       struct mlx4_en_pkt_stats pkstats;
+       struct mlx4_en_port_stats port_stats;
+       struct dev_mc_list *mc_list;
+       struct mlx4_en_stat_out_mbox hw_stats;
+};
+
+
+void mlx4_en_destroy_netdev(struct net_device *dev);
+int mlx4_en_init_netdev(struct mlx4_en_dev *mdev, int port,
+                       struct mlx4_en_port_profile *prof);
+
+int mlx4_en_get_profile(struct mlx4_en_dev *mdev);
+
+int mlx4_en_create_cq(struct mlx4_en_priv *priv, struct mlx4_en_cq *cq,
+                     int entries, int ring, enum cq_type mode);
+void mlx4_en_destroy_cq(struct mlx4_en_priv *priv, struct mlx4_en_cq *cq);
+int mlx4_en_activate_cq(struct mlx4_en_priv *priv, struct mlx4_en_cq *cq);
+void mlx4_en_deactivate_cq(struct mlx4_en_priv *priv, struct mlx4_en_cq *cq);
+int mlx4_en_set_cq_moder(struct mlx4_en_priv *priv, struct mlx4_en_cq *cq);
+int mlx4_en_arm_cq(struct mlx4_en_priv *priv, struct mlx4_en_cq *cq);
+
+void mlx4_en_poll_tx_cq(unsigned long data);
+void mlx4_en_tx_irq(struct mlx4_cq *mcq);
+int mlx4_en_xmit(struct sk_buff *skb, struct net_device *dev);
+
+int mlx4_en_create_tx_ring(struct mlx4_en_priv *priv, struct mlx4_en_tx_ring *ring,
+                          u32 size, u16 stride);
+void mlx4_en_destroy_tx_ring(struct mlx4_en_priv *priv, struct mlx4_en_tx_ring *ring);
+int mlx4_en_activate_tx_ring(struct mlx4_en_priv *priv,
+                            struct mlx4_en_tx_ring *ring,
+                            int cq, int srqn);
+void mlx4_en_deactivate_tx_ring(struct mlx4_en_priv *priv,
+                               struct mlx4_en_tx_ring *ring);
+
+int mlx4_en_create_rx_ring(struct mlx4_en_priv *priv,
+                          struct mlx4_en_rx_ring *ring,
+                          u32 size, u16 stride);
+void mlx4_en_destroy_rx_ring(struct mlx4_en_priv *priv,
+                            struct mlx4_en_rx_ring *ring);
+int mlx4_en_activate_rx_rings(struct mlx4_en_priv *priv);
+void mlx4_en_deactivate_rx_ring(struct mlx4_en_priv *priv,
+                               struct mlx4_en_rx_ring *ring);
+int mlx4_en_process_rx_cq(struct net_device *dev,
+                         struct mlx4_en_cq *cq,
+                         int budget);
+int mlx4_en_poll_rx_cq(struct napi_struct *napi, int budget);
+void mlx4_en_fill_qp_context(struct mlx4_en_priv *priv, int size, int stride,
+                            int is_tx, int rss, int qpn, int cqn, int srqn,
+                            struct mlx4_qp_context *context);
+int mlx4_en_map_buffer(struct mlx4_buf *buf);
+void mlx4_en_unmap_buffer(struct mlx4_buf *buf);
+
+void mlx4_en_calc_rx_buf(struct net_device *dev);
+void mlx4_en_set_default_rss_map(struct mlx4_en_priv *priv,
+                                struct mlx4_en_rss_map *rss_map,
+                                int num_entries, int num_rings);
+void mlx4_en_set_prio_map(struct mlx4_en_priv *priv, u16 *prio_map, u32 ring_num);
+int mlx4_en_config_rss_steer(struct mlx4_en_priv *priv);
+void mlx4_en_release_rss_steer(struct mlx4_en_priv *priv);
+int mlx4_en_free_tx_buf(struct net_device *dev, struct mlx4_en_tx_ring *ring);
+void mlx4_en_rx_refill(struct work_struct *work);
+void mlx4_en_rx_irq(struct mlx4_cq *mcq);
+
+int mlx4_SET_MCAST_FLTR(struct mlx4_dev *dev, u8 port, u64 mac, u64 clear, u8 mode);
+int mlx4_SET_VLAN_FLTR(struct mlx4_dev *dev, u8 port, struct vlan_group *grp);
+int mlx4_SET_PORT_general(struct mlx4_dev *dev, u8 port, int mtu,
+                         u8 pptx, u8 pfctx, u8 pprx, u8 pfcrx);
+int mlx4_SET_PORT_qpn_calc(struct mlx4_dev *dev, u8 port, u32 base_qpn,
+                          u8 promisc);
+
+int mlx4_en_DUMP_ETH_STATS(struct mlx4_en_dev *mdev, u8 port, u8 reset);
+
+/*
+ * Globals
+ */
+extern const struct ethtool_ops mlx4_en_ethtool_ops;
+#endif