enic_main.c

/*
 * Copyright 2008-2010 Cisco Systems, Inc.  All rights reserved.
 * Copyright 2007 Nuova Systems, Inc.  All rights reserved.
 *
 * This program is free software; you may redistribute it and/or modify
 * it under the terms of the GNU General Public License as published by
 * the Free Software Foundation; version 2 of the License.
 *
 * 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/module.h>
#include <linux/kernel.h>
#include <linux/string.h>
#include <linux/errno.h>
#include <linux/types.h>
#include <linux/init.h>
#include <linux/interrupt.h>
#include <linux/workqueue.h>
#include <linux/pci.h>
#include <linux/netdevice.h>
#include <linux/etherdevice.h>
#include <linux/if_ether.h>
#include <linux/if_vlan.h>
#include <linux/ethtool.h>
#include <linux/in.h>
#include <linux/ip.h>
#include <linux/ipv6.h>
#include <linux/tcp.h>
#include <linux/rtnetlink.h>
#include <linux/prefetch.h>
#include <net/ip6_checksum.h>

#include "cq_enet_desc.h"
#include "vnic_dev.h"
#include "vnic_intr.h"
#include "vnic_stats.h"
#include "vnic_vic.h"
#include "enic_res.h"
#include "enic.h"
#include "enic_dev.h"
#include "enic_pp.h"

#define ENIC_NOTIFY_TIMER_PERIOD	(2 * HZ)
#define WQ_ENET_MAX_DESC_LEN		(1 << WQ_ENET_LEN_BITS)
#define MAX_TSO				(1 << 16)
#define ENIC_DESC_MAX_SPLITS		(MAX_TSO / WQ_ENET_MAX_DESC_LEN + 1)

#define PCI_DEVICE_ID_CISCO_VIC_ENET         0x0043  /* ethernet vnic */
#define PCI_DEVICE_ID_CISCO_VIC_ENET_DYN     0x0044  /* enet dynamic vnic */

/* Supported devices */
static DEFINE_PCI_DEVICE_TABLE(enic_id_table) = {
	{ PCI_VDEVICE(CISCO, PCI_DEVICE_ID_CISCO_VIC_ENET) },
	{ PCI_VDEVICE(CISCO, PCI_DEVICE_ID_CISCO_VIC_ENET_DYN) },
	{ 0, }	/* end of table */
};

MODULE_DESCRIPTION(DRV_DESCRIPTION);
MODULE_AUTHOR("Scott Feldman <scofeldm@cisco.com>");
MODULE_LICENSE("GPL");
MODULE_VERSION(DRV_VERSION);
MODULE_DEVICE_TABLE(pci, enic_id_table);

struct enic_stat {
	char name[ETH_GSTRING_LEN];
	unsigned int offset;
};

#define ENIC_TX_STAT(stat)	\
	{ .name = #stat, .offset = offsetof(struct vnic_tx_stats, stat) / 8 }
#define ENIC_RX_STAT(stat)	\
	{ .name = #stat, .offset = offsetof(struct vnic_rx_stats, stat) / 8 }

static const struct enic_stat enic_tx_stats[] = {
	ENIC_TX_STAT(tx_frames_ok),
	ENIC_TX_STAT(tx_unicast_frames_ok),
	ENIC_TX_STAT(tx_multicast_frames_ok),
	ENIC_TX_STAT(tx_broadcast_frames_ok),
	ENIC_TX_STAT(tx_bytes_ok),
	ENIC_TX_STAT(tx_unicast_bytes_ok),
	ENIC_TX_STAT(tx_multicast_bytes_ok),
	ENIC_TX_STAT(tx_broadcast_bytes_ok),
	ENIC_TX_STAT(tx_drops),
	ENIC_TX_STAT(tx_errors),
	ENIC_TX_STAT(tx_tso),
};

static const struct enic_stat enic_rx_stats[] = {
	ENIC_RX_STAT(rx_frames_ok),
	ENIC_RX_STAT(rx_frames_total),
	ENIC_RX_STAT(rx_unicast_frames_ok),
	ENIC_RX_STAT(rx_multicast_frames_ok),
	ENIC_RX_STAT(rx_broadcast_frames_ok),
	ENIC_RX_STAT(rx_bytes_ok),
	ENIC_RX_STAT(rx_unicast_bytes_ok),
	ENIC_RX_STAT(rx_multicast_bytes_ok),
	ENIC_RX_STAT(rx_broadcast_bytes_ok),
	ENIC_RX_STAT(rx_drop),
	ENIC_RX_STAT(rx_no_bufs),
	ENIC_RX_STAT(rx_errors),
	ENIC_RX_STAT(rx_rss),
	ENIC_RX_STAT(rx_crc_errors),
	ENIC_RX_STAT(rx_frames_64),
	ENIC_RX_STAT(rx_frames_127),
	ENIC_RX_STAT(rx_frames_255),
	ENIC_RX_STAT(rx_frames_511),
	ENIC_RX_STAT(rx_frames_1023),
	ENIC_RX_STAT(rx_frames_1518),
	ENIC_RX_STAT(rx_frames_to_max),
};

static const unsigned int enic_n_tx_stats = ARRAY_SIZE(enic_tx_stats);
static const unsigned int enic_n_rx_stats = ARRAY_SIZE(enic_rx_stats);

static int enic_is_dynamic(struct enic *enic)
{
	return enic->pdev->device == PCI_DEVICE_ID_CISCO_VIC_ENET_DYN;
}

static inline unsigned int enic_cq_rq(struct enic *enic, unsigned int rq)
{
	return rq;
}

static inline unsigned int enic_cq_wq(struct enic *enic, unsigned int wq)
{
	return enic->rq_count + wq;
}

static inline unsigned int enic_legacy_io_intr(void)
{
	return 0;
}

static inline unsigned int enic_legacy_err_intr(void)
{
	return 1;
}

static inline unsigned int enic_legacy_notify_intr(void)
{
	return 2;
}

static inline unsigned int enic_msix_rq_intr(struct enic *enic, unsigned int rq)
{
	return enic->cq[enic_cq_rq(enic, rq)].interrupt_offset;
}

static inline unsigned int enic_msix_wq_intr(struct enic *enic, unsigned int wq)
{
	return enic->cq[enic_cq_wq(enic, wq)].interrupt_offset;
}

static inline unsigned int enic_msix_err_intr(struct enic *enic)
{
	return enic->rq_count + enic->wq_count;
}

static inline unsigned int enic_msix_notify_intr(struct enic *enic)
{
	return enic->rq_count + enic->wq_count + 1;
}

static int enic_get_settings(struct net_device *netdev,
	struct ethtool_cmd *ecmd)
{
	struct enic *enic = netdev_priv(netdev);

	ecmd->supported = (SUPPORTED_10000baseT_Full | SUPPORTED_FIBRE);
	ecmd->advertising = (ADVERTISED_10000baseT_Full | ADVERTISED_FIBRE);
	ecmd->port = PORT_FIBRE;
	ecmd->transceiver = XCVR_EXTERNAL;

	if (netif_carrier_ok(netdev)) {
		ethtool_cmd_speed_set(ecmd, vnic_dev_port_speed(enic->vdev));
		ecmd->duplex = DUPLEX_FULL;
	} else {
		ethtool_cmd_speed_set(ecmd, -1);
		ecmd->duplex = -1;
	}

	ecmd->autoneg = AUTONEG_DISABLE;

	return 0;
}

static void enic_get_drvinfo(struct net_device *netdev,
	struct ethtool_drvinfo *drvinfo)
{
	struct enic *enic = netdev_priv(netdev);
	struct vnic_devcmd_fw_info *fw_info;

	enic_dev_fw_info(enic, &fw_info);

	strncpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
	strncpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
	strncpy(drvinfo->fw_version, fw_info->fw_version,
		sizeof(drvinfo->fw_version));
	strncpy(drvinfo->bus_info, pci_name(enic->pdev),
		sizeof(drvinfo->bus_info));
}

static void enic_get_strings(struct net_device *netdev, u32 stringset, u8 *data)
{
	unsigned int i;

	switch (stringset) {
	case ETH_SS_STATS:
		for (i = 0; i < enic_n_tx_stats; i++) {
			memcpy(data, enic_tx_stats[i].name, ETH_GSTRING_LEN);
			data += ETH_GSTRING_LEN;
		}
		for (i = 0; i < enic_n_rx_stats; i++) {
			memcpy(data, enic_rx_stats[i].name, ETH_GSTRING_LEN);
			data += ETH_GSTRING_LEN;
		}
		break;
	}
}

static int enic_get_sset_count(struct net_device *netdev, int sset)
{
	switch (sset) {
	case ETH_SS_STATS:
		return enic_n_tx_stats + enic_n_rx_stats;
	default:
		return -EOPNOTSUPP;
	}
}

static void enic_get_ethtool_stats(struct net_device *netdev,
	struct ethtool_stats *stats, u64 *data)
{
	struct enic *enic = netdev_priv(netdev);
	struct vnic_stats *vstats;
	unsigned int i;

	enic_dev_stats_dump(enic, &vstats);

	for (i = 0; i < enic_n_tx_stats; i++)
		*(data++) = ((u64 *)&vstats->tx)[enic_tx_stats[i].offset];
	for (i = 0; i < enic_n_rx_stats; i++)
		*(data++) = ((u64 *)&vstats->rx)[enic_rx_stats[i].offset];
}

static u32 enic_get_msglevel(struct net_device *netdev)
{
	struct enic *enic = netdev_priv(netdev);
	return enic->msg_enable;
}

static void enic_set_msglevel(struct net_device *netdev, u32 value)
{
	struct enic *enic = netdev_priv(netdev);
	enic->msg_enable = value;
}

static int enic_get_coalesce(struct net_device *netdev,
	struct ethtool_coalesce *ecmd)
{
	struct enic *enic = netdev_priv(netdev);

	ecmd->tx_coalesce_usecs = enic->tx_coalesce_usecs;
	ecmd->rx_coalesce_usecs = enic->rx_coalesce_usecs;

	return 0;
}

static int enic_set_coalesce(struct net_device *netdev,
	struct ethtool_coalesce *ecmd)
{
	struct enic *enic = netdev_priv(netdev);
	u32 tx_coalesce_usecs;
	u32 rx_coalesce_usecs;
	unsigned int i, intr;

	tx_coalesce_usecs = min_t(u32, ecmd->tx_coalesce_usecs,
		vnic_dev_get_intr_coal_timer_max(enic->vdev));
	rx_coalesce_usecs = min_t(u32, ecmd->rx_coalesce_usecs,
		vnic_dev_get_intr_coal_timer_max(enic->vdev));

	switch (vnic_dev_get_intr_mode(enic->vdev)) {
	case VNIC_DEV_INTR_MODE_INTX:
		if (tx_coalesce_usecs != rx_coalesce_usecs)
			return -EINVAL;

		intr = enic_legacy_io_intr();
		vnic_intr_coalescing_timer_set(&enic->intr[intr],
			tx_coalesce_usecs);
		break;
	case VNIC_DEV_INTR_MODE_MSI:
		if (tx_coalesce_usecs != rx_coalesce_usecs)
			return -EINVAL;

		vnic_intr_coalescing_timer_set(&enic->intr[0],
			tx_coalesce_usecs);
		break;
	case VNIC_DEV_INTR_MODE_MSIX:
		for (i = 0; i < enic->wq_count; i++) {
			intr = enic_msix_wq_intr(enic, i);
			vnic_intr_coalescing_timer_set(&enic->intr[intr],
				tx_coalesce_usecs);
		}

		for (i = 0; i < enic->rq_count; i++) {
			intr = enic_msix_rq_intr(enic, i);
			vnic_intr_coalescing_timer_set(&enic->intr[intr],
				rx_coalesce_usecs);
		}

		break;
	default:
		break;
	}

	enic->tx_coalesce_usecs = tx_coalesce_usecs;
	enic->rx_coalesce_usecs = rx_coalesce_usecs;

	return 0;
}

static const struct ethtool_ops enic_ethtool_ops = {
	.get_settings = enic_get_settings,
	.get_drvinfo = enic_get_drvinfo,
	.get_msglevel = enic_get_msglevel,
	.set_msglevel = enic_set_msglevel,
	.get_link = ethtool_op_get_link,
	.get_strings = enic_get_strings,
	.get_sset_count = enic_get_sset_count,
	.get_ethtool_stats = enic_get_ethtool_stats,
	.get_coalesce = enic_get_coalesce,
	.set_coalesce = enic_set_coalesce,
};

static void enic_free_wq_buf(struct vnic_wq *wq, struct vnic_wq_buf *buf)
{
	struct enic *enic = vnic_dev_priv(wq->vdev);

	if (buf->sop)
		pci_unmap_single(enic->pdev, buf->dma_addr,
			buf->len, PCI_DMA_TODEVICE);
	else
		pci_unmap_page(enic->pdev, buf->dma_addr,
			buf->len, PCI_DMA_TODEVICE);

	if (buf->os_buf)
		dev_kfree_skb_any(buf->os_buf);
}

static void enic_wq_free_buf(struct vnic_wq *wq,
	struct cq_desc *cq_desc, struct vnic_wq_buf *buf, void *opaque)
{
	enic_free_wq_buf(wq, buf);
}

static int enic_wq_service(struct vnic_dev *vdev, struct cq_desc *cq_desc,
	u8 type, u16 q_number, u16 completed_index, void *opaque)
{
	struct enic *enic = vnic_dev_priv(vdev);

	spin_lock(&enic->wq_lock[q_number]);

	vnic_wq_service(&enic->wq[q_number], cq_desc,
		completed_index, enic_wq_free_buf,
		opaque);

	if (netif_queue_stopped(enic->netdev) &&
	    vnic_wq_desc_avail(&enic->wq[q_number]) >=
	    (MAX_SKB_FRAGS + ENIC_DESC_MAX_SPLITS))
		netif_wake_queue(enic->netdev);

	spin_unlock(&enic->wq_lock[q_number]);

	return 0;
}

static void enic_log_q_error(struct enic *enic)
{
	unsigned int i;
	u32 error_status;

	for (i = 0; i < enic->wq_count; i++) {
		error_status = vnic_wq_error_status(&enic->wq[i]);
		if (error_status)
			netdev_err(enic->netdev, "WQ[%d] error_status %d\n",
				i, error_status);
	}

	for (i = 0; i < enic->rq_count; i++) {
		error_status = vnic_rq_error_status(&enic->rq[i]);
		if (error_status)
			netdev_err(enic->netdev, "RQ[%d] error_status %d\n",
				i, error_status);
	}
}

static void enic_msglvl_check(struct enic *enic)
{
	u32 msg_enable = vnic_dev_msg_lvl(enic->vdev);

	if (msg_enable != enic->msg_enable) {
		netdev_info(enic->netdev, "msg lvl changed from 0x%x to 0x%x\n",
			enic->msg_enable, msg_enable);
		enic->msg_enable = msg_enable;
	}
}

static void enic_mtu_check(struct enic *enic)
{
	u32 mtu = vnic_dev_mtu(enic->vdev);
	struct net_device *netdev = enic->netdev;

	if (mtu && mtu != enic->port_mtu) {
		enic->port_mtu = mtu;
		if (enic_is_dynamic(enic)) {
			mtu = max_t(int, ENIC_MIN_MTU,
				min_t(int, ENIC_MAX_MTU, mtu));
			if (mtu != netdev->mtu)
				schedule_work(&enic->change_mtu_work);
		} else {
			if (mtu < netdev->mtu)
				netdev_warn(netdev,
					"interface MTU (%d) set higher "
					"than switch port MTU (%d)\n",
					netdev->mtu, mtu);
		}
	}
}

static void enic_link_check(struct enic *enic)
{
	int link_status = vnic_dev_link_status(enic->vdev);
	int carrier_ok = netif_carrier_ok(enic->netdev);

	if (link_status && !carrier_ok) {
		netdev_info(enic->netdev, "Link UP\n");
		netif_carrier_on(enic->netdev);
	} else if (!link_status && carrier_ok) {
		netdev_info(enic->netdev, "Link DOWN\n");
		netif_carrier_off(enic->netdev);
	}
}

static void enic_notify_check(struct enic *enic)
{
	enic_msglvl_check(enic);
	enic_mtu_check(enic);
	enic_link_check(enic);
}

#define ENIC_TEST_INTR(pba, i) (pba & (1 << i))

static irqreturn_t enic_isr_legacy(int irq, void *data)
{
	struct net_device *netdev = data;
	struct enic *enic = netdev_priv(netdev);
	unsigned int io_intr = enic_legacy_io_intr();
	unsigned int err_intr = enic_legacy_err_intr();
	unsigned int notify_intr = enic_legacy_notify_intr();
	u32 pba;

	vnic_intr_mask(&enic->intr[io_intr]);

	pba = vnic_intr_legacy_pba(enic->legacy_pba);
	if (!pba) {
		vnic_intr_unmask(&enic->intr[io_intr]);
		return IRQ_NONE;	/* not our interrupt */
	}

	if (ENIC_TEST_INTR(pba, notify_intr)) {
		vnic_intr_return_all_credits(&enic->intr[notify_intr]);
		enic_notify_check(enic);
	}

	if (ENIC_TEST_INTR(pba, err_intr)) {
		vnic_intr_return_all_credits(&enic->intr[err_intr]);
		enic_log_q_error(enic);
		/* schedule recovery from WQ/RQ error */
		schedule_work(&enic->reset);
		return IRQ_HANDLED;
	}

	if (ENIC_TEST_INTR(pba, io_intr)) {
		if (napi_schedule_prep(&enic->napi[0]))
			__napi_schedule(&enic->napi[0]);
	} else {
		vnic_intr_unmask(&enic->intr[io_intr]);
	}

	return IRQ_HANDLED;
}

static irqreturn_t enic_isr_msi(int irq, void *data)
{
	struct enic *enic = data;

	/* With MSI, there is no sharing of interrupts, so this is
	 * our interrupt and there is no need to ack it.  The device
	 * is not providing per-vector masking, so the OS will not
	 * write to PCI config space to mask/unmask the interrupt.
	 * We're using mask_on_assertion for MSI, so the device
	 * automatically masks the interrupt when the interrupt is
	 * generated.  Later, when exiting polling, the interrupt
	 * will be unmasked (see enic_poll).
	 *
	 * Also, the device uses the same PCIe Traffic Class (TC)
	 * for Memory Write data and MSI, so there are no ordering
	 * issues; the MSI will always arrive at the Root Complex
	 * _after_ corresponding Memory Writes (i.e. descriptor
	 * writes).
	 */

	napi_schedule(&enic->napi[0]);

	return IRQ_HANDLED;
}

static irqreturn_t enic_isr_msix_rq(int irq, void *data)
{
	struct napi_struct *napi = data;

	/* schedule NAPI polling for RQ cleanup */
	napi_schedule(napi);

	return IRQ_HANDLED;
}

static irqreturn_t enic_isr_msix_wq(int irq, void *data)
{
	struct enic *enic = data;
	unsigned int cq = enic_cq_wq(enic, 0);
	unsigned int intr = enic_msix_wq_intr(enic, 0);
	unsigned int wq_work_to_do = -1; /* no limit */
	unsigned int wq_work_done;

	wq_work_done = vnic_cq_service(&enic->cq[cq],
		wq_work_to_do, enic_wq_service, NULL);

	vnic_intr_return_credits(&enic->intr[intr],
		wq_work_done,
		1 /* unmask intr */,
		1 /* reset intr timer */);

	return IRQ_HANDLED;
}

static irqreturn_t enic_isr_msix_err(int irq, void *data)
{
	struct enic *enic = data;
	unsigned int intr = enic_msix_err_intr(enic);

	vnic_intr_return_all_credits(&enic->intr[intr]);

	enic_log_q_error(enic);

	/* schedule recovery from WQ/RQ error */
	schedule_work(&enic->reset);

	return IRQ_HANDLED;
}

static irqreturn_t enic_isr_msix_notify(int irq, void *data)
{
	struct enic *enic = data;
	unsigned int intr = enic_msix_notify_intr(enic);

	vnic_intr_return_all_credits(&enic->intr[intr]);
	enic_notify_check(enic);

	return IRQ_HANDLED;
}

static inline void enic_queue_wq_skb_cont(struct enic *enic,
	struct vnic_wq *wq, struct sk_buff *skb,
	unsigned int len_left, int loopback)
{
	skb_frag_t *frag;

	/* Queue additional data fragments */
	for (frag = skb_shinfo(skb)->frags; len_left; frag++) {
		len_left -= frag->size;
		enic_queue_wq_desc_cont(wq, skb,
			pci_map_page(enic->pdev, frag->page,
				frag->page_offset, frag->size,
				PCI_DMA_TODEVICE),
			frag->size,
			(len_left == 0),	/* EOP? */
			loopback);
	}
}

static inline void enic_queue_wq_skb_vlan(struct enic *enic,
	struct vnic_wq *wq, struct sk_buff *skb,
	int vlan_tag_insert, unsigned int vlan_tag, int loopback)
{
	unsigned int head_len = skb_headlen(skb);
	unsigned int len_left = skb->len - head_len;
	int eop = (len_left == 0);

	/* Queue the main skb fragment. The fragments are no larger
	 * than max MTU(9000)+ETH_HDR_LEN(14) bytes, which is less
	 * than WQ_ENET_MAX_DESC_LEN length. So only one descriptor
	 * per fragment is queued.
	 */
	enic_queue_wq_desc(wq, skb,
		pci_map_single(enic->pdev, skb->data,
			head_len, PCI_DMA_TODEVICE),
		head_len,
		vlan_tag_insert, vlan_tag,
		eop, loopback);

	if (!eop)
		enic_queue_wq_skb_cont(enic, wq, skb, len_left, loopback);
}

static inline void enic_queue_wq_skb_csum_l4(struct enic *enic,
	struct vnic_wq *wq, struct sk_buff *skb,
	int vlan_tag_insert, unsigned int vlan_tag, int loopback)
{
	unsigned int head_len = skb_headlen(skb);
	unsigned int len_left = skb->len - head_len;
	unsigned int hdr_len = skb_checksum_start_offset(skb);
	unsigned int csum_offset = hdr_len + skb->csum_offset;
	int eop = (len_left == 0);

	/* Queue the main skb fragment. The fragments are no larger
	 * than max MTU(9000)+ETH_HDR_LEN(14) bytes, which is less
	 * than WQ_ENET_MAX_DESC_LEN length. So only one descriptor
	 * per fragment is queued.
	 */
	enic_queue_wq_desc_csum_l4(wq, skb,
		pci_map_single(enic->pdev, skb->data,
			head_len, PCI_DMA_TODEVICE),
		head_len,
		csum_offset,
		hdr_len,
		vlan_tag_insert, vlan_tag,
		eop, loopback);

	if (!eop)
		enic_queue_wq_skb_cont(enic, wq, skb, len_left, loopback);
}

static inline void enic_queue_wq_skb_tso(struct enic *enic,
	struct vnic_wq *wq, struct sk_buff *skb, unsigned int mss,
	int vlan_tag_insert, unsigned int vlan_tag, int loopback)
{
	unsigned int frag_len_left = skb_headlen(skb);
	unsigned int len_left = skb->len - frag_len_left;
	unsigned int hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
	int eop = (len_left == 0);
	unsigned int len;
	dma_addr_t dma_addr;
	unsigned int offset = 0;
	skb_frag_t *frag;

	/* Preload TCP csum field with IP pseudo hdr calculated
	 * with IP length set to zero.  HW will later add in length
	 * to each TCP segment resulting from the TSO.
	 */

	if (skb->protocol == cpu_to_be16(ETH_P_IP)) {
		ip_hdr(skb)->check = 0;
		tcp_hdr(skb)->check = ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
			ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
	} else if (skb->protocol == cpu_to_be16(ETH_P_IPV6)) {
		tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
			&ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
	}

	/* Queue WQ_ENET_MAX_DESC_LEN length descriptors
	 * for the main skb fragment
	 */
	while (frag_len_left) {
		len = min(frag_len_left, (unsigned int)WQ_ENET_MAX_DESC_LEN);
		dma_addr = pci_map_single(enic->pdev, skb->data + offset,
				len, PCI_DMA_TODEVICE);
		enic_queue_wq_desc_tso(wq, skb,
			dma_addr,
			len,
			mss, hdr_len,
			vlan_tag_insert, vlan_tag,
			eop && (len == frag_len_left), loopback);
		frag_len_left -= len;
		offset += len;
	}

	if (eop)
		return;

	/* Queue WQ_ENET_MAX_DESC_LEN length descriptors
	 * for additional data fragments
	 */
	for (frag = skb_shinfo(skb)->frags; len_left; frag++) {
		len_left -= frag->size;
		frag_len_left = frag->size;
		offset = frag->page_offset;

		while (frag_len_left) {
			len = min(frag_len_left,
				(unsigned int)WQ_ENET_MAX_DESC_LEN);
			dma_addr = pci_map_page(enic->pdev, frag->page,
				offset, len,
				PCI_DMA_TODEVICE);
			enic_queue_wq_desc_cont(wq, skb,
				dma_addr,
				len,
				(len_left == 0) &&
				(len == frag_len_left),		/* EOP? */
				loopback);
			frag_len_left -= len;
			offset += len;
		}
	}
}

static inline void enic_queue_wq_skb(struct enic *enic,
	struct vnic_wq *wq, struct sk_buff *skb)
{
	unsigned int mss = skb_shinfo(skb)->gso_size;
	unsigned int vlan_tag = 0;
	int vlan_tag_insert = 0;
	int loopback = 0;

	if (vlan_tx_tag_present(skb)) {
		/* VLAN tag from trunking driver */
		vlan_tag_insert = 1;
		vlan_tag = vlan_tx_tag_get(skb);
	} else if (enic->loop_enable) {
		vlan_tag = enic->loop_tag;
		loopback = 1;
	}

	if (mss)
		enic_queue_wq_skb_tso(enic, wq, skb, mss,
			vlan_tag_insert, vlan_tag, loopback);
	else if	(skb->ip_summed == CHECKSUM_PARTIAL)
		enic_queue_wq_skb_csum_l4(enic, wq, skb,
			vlan_tag_insert, vlan_tag, loopback);
	else
		enic_queue_wq_skb_vlan(enic, wq, skb,
			vlan_tag_insert, vlan_tag, loopback);
}

/* netif_tx_lock held, process context with BHs disabled, or BH */
static netdev_tx_t enic_hard_start_xmit(struct sk_buff *skb,
	struct net_device *netdev)
{
	struct enic *enic = netdev_priv(netdev);
	struct vnic_wq *wq = &enic->wq[0];
	unsigned long flags;

	if (skb->len <= 0) {
		dev_kfree_skb(skb);
		return NETDEV_TX_OK;
	}

	/* Non-TSO sends must fit within ENIC_NON_TSO_MAX_DESC descs,
	 * which is very likely.  In the off chance it's going to take
	 * more than * ENIC_NON_TSO_MAX_DESC, linearize the skb.
	 */

	if (skb_shinfo(skb)->gso_size == 0 &&
	    skb_shinfo(skb)->nr_frags + 1 > ENIC_NON_TSO_MAX_DESC &&
	    skb_linearize(skb)) {
		dev_kfree_skb(skb);
		return NETDEV_TX_OK;
	}

	spin_lock_irqsave(&enic->wq_lock[0], flags);

	if (vnic_wq_desc_avail(wq) <
	    skb_shinfo(skb)->nr_frags + ENIC_DESC_MAX_SPLITS) {
		netif_stop_queue(netdev);
		/* This is a hard error, log it */
		netdev_err(netdev, "BUG! Tx ring full when queue awake!\n");
		spin_unlock_irqrestore(&enic->wq_lock[0], flags);
		return NETDEV_TX_BUSY;
	}

	enic_queue_wq_skb(enic, wq, skb);

	if (vnic_wq_desc_avail(wq) < MAX_SKB_FRAGS + ENIC_DESC_MAX_SPLITS)
		netif_stop_queue(netdev);

	spin_unlock_irqrestore(&enic->wq_lock[0], flags);

	return NETDEV_TX_OK;
}

/* dev_base_lock rwlock held, nominally process context */
static struct rtnl_link_stats64 *enic_get_stats(struct net_device *netdev,
						struct rtnl_link_stats64 *net_stats)
{
	struct enic *enic = netdev_priv(netdev);
	struct vnic_stats *stats;

	enic_dev_stats_dump(enic, &stats);

	net_stats->tx_packets = stats->tx.tx_frames_ok;
	net_stats->tx_bytes = stats->tx.tx_bytes_ok;
	net_stats->tx_errors = stats->tx.tx_errors;
	net_stats->tx_dropped = stats->tx.tx_drops;

	net_stats->rx_packets = stats->rx.rx_frames_ok;
	net_stats->rx_bytes = stats->rx.rx_bytes_ok;
	net_stats->rx_errors = stats->rx.rx_errors;
	net_stats->multicast = stats->rx.rx_multicast_frames_ok;
	net_stats->rx_over_errors = enic->rq_truncated_pkts;
	net_stats->rx_crc_errors = enic->rq_bad_fcs;
	net_stats->rx_dropped = stats->rx.rx_no_bufs + stats->rx.rx_drop;

	return net_stats;
}

void enic_reset_addr_lists(struct enic *enic)
{
	enic->mc_count = 0;
	enic->uc_count = 0;
	enic->flags = 0;
}

static int enic_set_mac_addr(struct net_device *netdev, char *addr)
{
	struct enic *enic = netdev_priv(netdev);

	if (enic_is_dynamic(enic)) {
		if (!is_valid_ether_addr(addr) && !is_zero_ether_addr(addr))
			return -EADDRNOTAVAIL;
	} else {
		if (!is_valid_ether_addr(addr))
			return -EADDRNOTAVAIL;
	}

	memcpy(netdev->dev_addr, addr, netdev->addr_len);

	return 0;
}

static int enic_set_mac_address_dynamic(struct net_device *netdev, void *p)
{
	struct enic *enic = netdev_priv(netdev);
	struct sockaddr *saddr = p;
	char *addr = saddr->sa_data;
	int err;

	if (netif_running(enic->netdev)) {
		err = enic_dev_del_station_addr(enic);
		if (err)
			return err;
	}

	err = enic_set_mac_addr(netdev, addr);
	if (err)
		return err;

	if (netif_running(enic->netdev)) {
		err = enic_dev_add_station_addr(enic);
		if (err)
			return err;
	}

	return err;
}

static int enic_set_mac_address(struct net_device *netdev, void *p)
{
	struct sockaddr *saddr = p;
	char *addr = saddr->sa_data;
	struct enic *enic = netdev_priv(netdev);
	int err;

	err = enic_dev_del_station_addr(enic);
	if (err)
		return err;

	err = enic_set_mac_addr(netdev, addr);
	if (err)
		return err;

	return enic_dev_add_station_addr(enic);
}

static void enic_update_multicast_addr_list(struct enic *enic)
{
	struct net_device *netdev = enic->netdev;
	struct netdev_hw_addr *ha;
	unsigned int mc_count = netdev_mc_count(netdev);
	u8 mc_addr[ENIC_MULTICAST_PERFECT_FILTERS][ETH_ALEN];
	unsigned int i, j;

	if (mc_count > ENIC_MULTICAST_PERFECT_FILTERS) {
		netdev_warn(netdev, "Registering only %d out of %d "
			"multicast addresses\n",
			ENIC_MULTICAST_PERFECT_FILTERS, mc_count);
		mc_count = ENIC_MULTICAST_PERFECT_FILTERS;
	}

	/* Is there an easier way?  Trying to minimize to
	 * calls to add/del multicast addrs.  We keep the
	 * addrs from the last call in enic->mc_addr and
	 * look for changes to add/del.
	 */

	i = 0;
	netdev_for_each_mc_addr(ha, netdev) {
		if (i == mc_count)
			break;
		memcpy(mc_addr[i++], ha->addr, ETH_ALEN);
	}

	for (i = 0; i < enic->mc_count; i++) {
		for (j = 0; j < mc_count; j++)
			if (compare_ether_addr(enic->mc_addr[i],
				mc_addr[j]) == 0)
				break;
		if (j == mc_count)
			enic_dev_del_addr(enic, enic->mc_addr[i]);
	}

	for (i = 0; i < mc_count; i++) {
		for (j = 0; j < enic->mc_count; j++)
			if (compare_ether_addr(mc_addr[i],
				enic->mc_addr[j]) == 0)
				break;
		if (j == enic->mc_count)
			enic_dev_add_addr(enic, mc_addr[i]);
	}

	/* Save the list to compare against next time
	 */

	for (i = 0; i < mc_count; i++)
		memcpy(enic->mc_addr[i], mc_addr[i], ETH_ALEN);

	enic->mc_count = mc_count;
}

static void enic_update_unicast_addr_list(struct enic *enic)
{
	struct net_device *netdev = enic->netdev;
	struct netdev_hw_addr *ha;
	unsigned int uc_count = netdev_uc_count(netdev);
	u8 uc_addr[ENIC_UNICAST_PERFECT_FILTERS][ETH_ALEN];
	unsigned int i, j;

	if (uc_count > ENIC_UNICAST_PERFECT_FILTERS) {
		netdev_warn(netdev, "Registering only %d out of %d "
			"unicast addresses\n",
			ENIC_UNICAST_PERFECT_FILTERS, uc_count);
		uc_count = ENIC_UNICAST_PERFECT_FILTERS;
	}

	/* Is there an easier way?  Trying to minimize to
	 * calls to add/del unicast addrs.  We keep the
	 * addrs from the last call in enic->uc_addr and
	 * look for changes to add/del.
	 */

	i = 0;
	netdev_for_each_uc_addr(ha, netdev) {
		if (i == uc_count)
			break;
		memcpy(uc_addr[i++], ha->addr, ETH_ALEN);
	}

	for (i = 0; i < enic->uc_count; i++) {
		for (j = 0; j < uc_count; j++)
			if (compare_ether_addr(enic->uc_addr[i],
				uc_addr[j]) == 0)
				break;
		if (j == uc_count)
			enic_dev_del_addr(enic, enic->uc_addr[i]);
	}

	for (i = 0; i < uc_count; i++) {
		for (j = 0; j < enic->uc_count; j++)
			if (compare_ether_addr(uc_addr[i],
				enic->uc_addr[j]) == 0)
				break;
		if (j == enic->uc_count)
			enic_dev_add_addr(enic, uc_addr[i]);
	}

	/* Save the list to compare against next time
	 */

	for (i = 0; i < uc_count; i++)
		memcpy(enic->uc_addr[i], uc_addr[i], ETH_ALEN);