drivers/net/ethernet/sfc/tx.c

/****************************************************************************
 * Driver for Solarflare Solarstorm network controllers and boards
 * Copyright 2005-2006 Fen Systems Ltd.
 * Copyright 2005-2010 Solarflare Communications Inc.
 *
 * This program is free software; you can redistribute it and/or modify it
 * under the terms of the GNU General Public License version 2 as published
 * by the Free Software Foundation, incorporated herein by reference.
 */

#include <linux/pci.h>
#include <linux/tcp.h>
#include <linux/ip.h>
#include <linux/in.h>
#include <linux/ipv6.h>
#include <linux/slab.h>
#include <net/ipv6.h>
#include <linux/if_ether.h>
#include <linux/highmem.h>
#include "net_driver.h"
#include "efx.h"
#include "nic.h"
#include "workarounds.h"

static void efx_dequeue_buffer(struct efx_tx_queue *tx_queue,
			       struct efx_tx_buffer *buffer,
			       unsigned int *pkts_compl,
			       unsigned int *bytes_compl)
{
	if (buffer->unmap_len) {
		struct device *dma_dev = &tx_queue->efx->pci_dev->dev;
		dma_addr_t unmap_addr = (buffer->dma_addr + buffer->len -
					 buffer->unmap_len);
		if (buffer->flags & EFX_TX_BUF_MAP_SINGLE)
			dma_unmap_single(dma_dev, unmap_addr, buffer->unmap_len,
					 DMA_TO_DEVICE);
		else
			dma_unmap_page(dma_dev, unmap_addr, buffer->unmap_len,
				       DMA_TO_DEVICE);
		buffer->unmap_len = 0;
	}

	if (buffer->flags & EFX_TX_BUF_SKB) {
		(*pkts_compl)++;
		(*bytes_compl) += buffer->skb->len;
		dev_kfree_skb_any((struct sk_buff *) buffer->skb);
		netif_vdbg(tx_queue->efx, tx_done, tx_queue->efx->net_dev,
			   "TX queue %d transmission id %x complete\n",
			   tx_queue->queue, tx_queue->read_count);
	}

	buffer->flags &= EFX_TX_BUF_TSOH;
}

/**
 * struct efx_tso_header - a DMA mapped buffer for packet headers
 * @next: Linked list of free ones.
 *	The list is protected by the TX queue lock.
 * @dma_unmap_len: Length to unmap for an oversize buffer, or 0.
 * @dma_addr: The DMA address of the header below.
 *
 * This controls the memory used for a TSO header.  Use TSOH_DATA()
 * to find the packet header data.  Use TSOH_SIZE() to calculate the
 * total size required for a given packet header length.  TSO headers
 * in the free list are exactly %TSOH_STD_SIZE bytes in size.
 */
struct efx_tso_header {
	union {
		struct efx_tso_header *next;
		size_t unmap_len;
	};
	dma_addr_t dma_addr;
};

static int efx_enqueue_skb_tso(struct efx_tx_queue *tx_queue,
			       struct sk_buff *skb);
static void efx_fini_tso(struct efx_tx_queue *tx_queue);
static void efx_tsoh_heap_free(struct efx_tx_queue *tx_queue,
			       struct efx_tso_header *tsoh);

static void efx_tsoh_free(struct efx_tx_queue *tx_queue,
			  struct efx_tx_buffer *buffer)
{
	if (buffer->flags & EFX_TX_BUF_TSOH) {
		if (likely(!buffer->tsoh->unmap_len)) {
			buffer->tsoh->next = tx_queue->tso_headers_free;
			tx_queue->tso_headers_free = buffer->tsoh;
		} else {
			efx_tsoh_heap_free(tx_queue, buffer->tsoh);
		}
		buffer->flags &= ~EFX_TX_BUF_TSOH;
	}
}


static inline unsigned
efx_max_tx_len(struct efx_nic *efx, dma_addr_t dma_addr)
{
	/* Depending on the NIC revision, we can use descriptor
	 * lengths up to 8K or 8K-1.  However, since PCI Express
	 * devices must split read requests at 4K boundaries, there is
	 * little benefit from using descriptors that cross those
	 * boundaries and we keep things simple by not doing so.
	 */
	unsigned len = (~dma_addr & (EFX_PAGE_SIZE - 1)) + 1;

	/* Work around hardware bug for unaligned buffers. */
	if (EFX_WORKAROUND_5391(efx) && (dma_addr & 0xf))
		len = min_t(unsigned, len, 512 - (dma_addr & 0xf));

	return len;
}

unsigned int efx_tx_max_skb_descs(struct efx_nic *efx)
{
	/* Header and payload descriptor for each output segment, plus
	 * one for every input fragment boundary within a segment
	 */
	unsigned int max_descs = EFX_TSO_MAX_SEGS * 2 + MAX_SKB_FRAGS;

	/* Possibly one more per segment for the alignment workaround */
	if (EFX_WORKAROUND_5391(efx))
		max_descs += EFX_TSO_MAX_SEGS;

	/* Possibly more for PCIe page boundaries within input fragments */
	if (PAGE_SIZE > EFX_PAGE_SIZE)
		max_descs += max_t(unsigned int, MAX_SKB_FRAGS,
				   DIV_ROUND_UP(GSO_MAX_SIZE, EFX_PAGE_SIZE));

	return max_descs;
}

/* Get partner of a TX queue, seen as part of the same net core queue */
static struct efx_tx_queue *efx_tx_queue_partner(struct efx_tx_queue *tx_queue)
{
	if (tx_queue->queue & EFX_TXQ_TYPE_OFFLOAD)
		return tx_queue - EFX_TXQ_TYPE_OFFLOAD;
	else
		return tx_queue + EFX_TXQ_TYPE_OFFLOAD;
}

static void efx_tx_maybe_stop_queue(struct efx_tx_queue *txq1)
{
	/* We need to consider both queues that the net core sees as one */
	struct efx_tx_queue *txq2 = efx_tx_queue_partner(txq1);
	struct efx_nic *efx = txq1->efx;
	unsigned int fill_level;

	fill_level = max(txq1->insert_count - txq1->old_read_count,
			 txq2->insert_count - txq2->old_read_count);
	if (likely(fill_level < efx->txq_stop_thresh))
		return;

	/* We used the stale old_read_count above, which gives us a
	 * pessimistic estimate of the fill level (which may even
	 * validly be >= efx->txq_entries).  Now try again using
	 * read_count (more likely to be a cache miss).
	 *
	 * If we read read_count and then conditionally stop the
	 * queue, it is possible for the completion path to race with
	 * us and complete all outstanding descriptors in the middle,
	 * after which there will be no more completions to wake it.
	 * Therefore we stop the queue first, then read read_count
	 * (with a memory barrier to ensure the ordering), then
	 * restart the queue if the fill level turns out to be low
	 * enough.
	 */
	netif_tx_stop_queue(txq1->core_txq);
	smp_mb();
	txq1->old_read_count = ACCESS_ONCE(txq1->read_count);
	txq2->old_read_count = ACCESS_ONCE(txq2->read_count);

	fill_level = max(txq1->insert_count - txq1->old_read_count,
			 txq2->insert_count - txq2->old_read_count);
	EFX_BUG_ON_PARANOID(fill_level >= efx->txq_entries);
	if (likely(fill_level < efx->txq_stop_thresh)) {
		smp_mb();
		if (likely(!efx->loopback_selftest))
			netif_tx_start_queue(txq1->core_txq);
	}
}

/*
 * Add a socket buffer to a TX queue
 *
 * This maps all fragments of a socket buffer for DMA and adds them to
 * the TX queue.  The queue's insert pointer will be incremented by
 * the number of fragments in the socket buffer.
 *
 * If any DMA mapping fails, any mapped fragments will be unmapped,
 * the queue's insert pointer will be restored to its original value.
 *
 * This function is split out from efx_hard_start_xmit to allow the
 * loopback test to direct packets via specific TX queues.
 *
 * Returns NETDEV_TX_OK.
 * You must hold netif_tx_lock() to call this function.
 */
netdev_tx_t efx_enqueue_skb(struct efx_tx_queue *tx_queue, struct sk_buff *skb)
{
	struct efx_nic *efx = tx_queue->efx;
	struct device *dma_dev = &efx->pci_dev->dev;
	struct efx_tx_buffer *buffer;
	skb_frag_t *fragment;
	unsigned int len, unmap_len = 0, insert_ptr;
	dma_addr_t dma_addr, unmap_addr = 0;
	unsigned int dma_len;
	unsigned short dma_flags;
	int i = 0;

	EFX_BUG_ON_PARANOID(tx_queue->write_count != tx_queue->insert_count);

	if (skb_shinfo(skb)->gso_size)
		return efx_enqueue_skb_tso(tx_queue, skb);

	/* Get size of the initial fragment */
	len = skb_headlen(skb);

	/* Pad if necessary */
	if (EFX_WORKAROUND_15592(efx) && skb->len <= 32) {
		EFX_BUG_ON_PARANOID(skb->data_len);
		len = 32 + 1;
		if (skb_pad(skb, len - skb->len))
			return NETDEV_TX_OK;
	}

	/* Map for DMA.  Use dma_map_single rather than dma_map_page
	 * since this is more efficient on machines with sparse
	 * memory.
	 */
	dma_flags = EFX_TX_BUF_MAP_SINGLE;
	dma_addr = dma_map_single(dma_dev, skb->data, len, PCI_DMA_TODEVICE);

	/* Process all fragments */
	while (1) {
		if (unlikely(dma_mapping_error(dma_dev, dma_addr)))
			goto dma_err;

		/* Store fields for marking in the per-fragment final
		 * descriptor */
		unmap_len = len;
		unmap_addr = dma_addr;

		/* Add to TX queue, splitting across DMA boundaries */
		do {
			insert_ptr = tx_queue->insert_count & tx_queue->ptr_mask;
			buffer = &tx_queue->buffer[insert_ptr];
			efx_tsoh_free(tx_queue, buffer);
			EFX_BUG_ON_PARANOID(buffer->flags);
			EFX_BUG_ON_PARANOID(buffer->len);
			EFX_BUG_ON_PARANOID(buffer->unmap_len);

			dma_len = efx_max_tx_len(efx, dma_addr);
			if (likely(dma_len >= len))
				dma_len = len;

			/* Fill out per descriptor fields */
			buffer->len = dma_len;
			buffer->dma_addr = dma_addr;
			buffer->flags = EFX_TX_BUF_CONT;
			len -= dma_len;
			dma_addr += dma_len;
			++tx_queue->insert_count;
		} while (len);

		/* Transfer ownership of the unmapping to the final buffer */
		buffer->flags = EFX_TX_BUF_CONT | dma_flags;
		buffer->unmap_len = unmap_len;
		unmap_len = 0;

		/* Get address and size of next fragment */
		if (i >= skb_shinfo(skb)->nr_frags)
			break;
		fragment = &skb_shinfo(skb)->frags[i];
		len = skb_frag_size(fragment);
		i++;
		/* Map for DMA */
		dma_flags = 0;
		dma_addr = skb_frag_dma_map(dma_dev, fragment, 0, len,
					    DMA_TO_DEVICE);
	}

	/* Transfer ownership of the skb to the final buffer */
	buffer->skb = skb;
	buffer->flags = EFX_TX_BUF_SKB | dma_flags;

	netdev_tx_sent_queue(tx_queue->core_txq, skb->len);

	/* Pass off to hardware */
	efx_nic_push_buffers(tx_queue);

	efx_tx_maybe_stop_queue(tx_queue);

	return NETDEV_TX_OK;

 dma_err:
	netif_err(efx, tx_err, efx->net_dev,
		  " TX queue %d could not map skb with %d bytes %d "
		  "fragments for DMA\n", tx_queue->queue, skb->len,
		  skb_shinfo(skb)->nr_frags + 1);

	/* Mark the packet as transmitted, and free the SKB ourselves */
	dev_kfree_skb_any(skb);

	/* Work backwards until we hit the original insert pointer value */
	while (tx_queue->insert_count != tx_queue->write_count) {
		unsigned int pkts_compl = 0, bytes_compl = 0;
		--tx_queue->insert_count;
		insert_ptr = tx_queue->insert_count & tx_queue->ptr_mask;
		buffer = &tx_queue->buffer[insert_ptr];
		efx_dequeue_buffer(tx_queue, buffer, &pkts_compl, &bytes_compl);
		buffer->len = 0;
	}

	/* Free the fragment we were mid-way through pushing */
	if (unmap_len) {
		if (dma_flags & EFX_TX_BUF_MAP_SINGLE)
			dma_unmap_single(dma_dev, unmap_addr, unmap_len,
					 DMA_TO_DEVICE);
		else
			dma_unmap_page(dma_dev, unmap_addr, unmap_len,
				       DMA_TO_DEVICE);
	}

	return NETDEV_TX_OK;
}

/* Remove packets from the TX queue
 *
 * This removes packets from the TX queue, up to and including the
 * specified index.
 */
static void efx_dequeue_buffers(struct efx_tx_queue *tx_queue,
				unsigned int index,
				unsigned int *pkts_compl,
				unsigned int *bytes_compl)
{
	struct efx_nic *efx = tx_queue->efx;
	unsigned int stop_index, read_ptr;

	stop_index = (index + 1) & tx_queue->ptr_mask;
	read_ptr = tx_queue->read_count & tx_queue->ptr_mask;

	while (read_ptr != stop_index) {
		struct efx_tx_buffer *buffer = &tx_queue->buffer[read_ptr];
		if (unlikely(buffer->len == 0)) {
			netif_err(efx, tx_err, efx->net_dev,
				  "TX queue %d spurious TX completion id %x\n",
				  tx_queue->queue, read_ptr);
			efx_schedule_reset(efx, RESET_TYPE_TX_SKIP);
			return;
		}

		efx_dequeue_buffer(tx_queue, buffer, pkts_compl, bytes_compl);
		buffer->len = 0;

		++tx_queue->read_count;
		read_ptr = tx_queue->read_count & tx_queue->ptr_mask;
	}
}

/* Initiate a packet transmission.  We use one channel per CPU
 * (sharing when we have more CPUs than channels).  On Falcon, the TX
 * completion events will be directed back to the CPU that transmitted
 * the packet, which should be cache-efficient.
 *
 * Context: non-blocking.
 * Note that returning anything other than NETDEV_TX_OK will cause the
 * OS to free the skb.
 */
netdev_tx_t efx_hard_start_xmit(struct sk_buff *skb,
				struct net_device *net_dev)
{
	struct efx_nic *efx = netdev_priv(net_dev);
	struct efx_tx_queue *tx_queue;
	unsigned index, type;

	EFX_WARN_ON_PARANOID(!netif_device_present(net_dev));

	index = skb_get_queue_mapping(skb);
	type = skb->ip_summed == CHECKSUM_PARTIAL ? EFX_TXQ_TYPE_OFFLOAD : 0;
	if (index >= efx->n_tx_channels) {
		index -= efx->n_tx_channels;
		type |= EFX_TXQ_TYPE_HIGHPRI;
	}
	tx_queue = efx_get_tx_queue(efx, index, type);

	return efx_enqueue_skb(tx_queue, skb);
}

void efx_init_tx_queue_core_txq(struct efx_tx_queue *tx_queue)
{
	struct efx_nic *efx = tx_queue->efx;

	/* Must be inverse of queue lookup in efx_hard_start_xmit() */
	tx_queue->core_txq =
		netdev_get_tx_queue(efx->net_dev,
				    tx_queue->queue / EFX_TXQ_TYPES +
				    ((tx_queue->queue & EFX_TXQ_TYPE_HIGHPRI) ?
				     efx->n_tx_channels : 0));
}

int efx_setup_tc(struct net_device *net_dev, u8 num_tc)
{
	struct efx_nic *efx = netdev_priv(net_dev);
	struct efx_channel *channel;
	struct efx_tx_queue *tx_queue;
	unsigned tc;
	int rc;

	if (efx_nic_rev(efx) < EFX_REV_FALCON_B0 || num_tc > EFX_MAX_TX_TC)
		return -EINVAL;

	if (num_tc == net_dev->num_tc)
		return 0;

	for (tc = 0; tc < num_tc; tc++) {
		net_dev->tc_to_txq[tc].offset = tc * efx->n_tx_channels;
		net_dev->tc_to_txq[tc].count = efx->n_tx_channels;
	}

	if (num_tc > net_dev->num_tc) {
		/* Initialise high-priority queues as necessary */
		efx_for_each_channel(channel, efx) {
			efx_for_each_possible_channel_tx_queue(tx_queue,
							       channel) {
				if (!(tx_queue->queue & EFX_TXQ_TYPE_HIGHPRI))
					continue;
				if (!tx_queue->buffer) {
					rc = efx_probe_tx_queue(tx_queue);
					if (rc)
						return rc;
				}
				if (!tx_queue->initialised)
					efx_init_tx_queue(tx_queue);
				efx_init_tx_queue_core_txq(tx_queue);
			}
		}
	} else {
		/* Reduce number of classes before number of queues */
		net_dev->num_tc = num_tc;
	}

	rc = netif_set_real_num_tx_queues(net_dev,
					  max_t(int, num_tc, 1) *
					  efx->n_tx_channels);
	if (rc)
		return rc;

	/* Do not destroy high-priority queues when they become
	 * unused.  We would have to flush them first, and it is
	 * fairly difficult to flush a subset of TX queues.  Leave
	 * it to efx_fini_channels().
	 */

	net_dev->num_tc = num_tc;
	return 0;
}

void efx_xmit_done(struct efx_tx_queue *tx_queue, unsigned int index)
{
	unsigned fill_level;
	struct efx_nic *efx = tx_queue->efx;
	struct efx_tx_queue *txq2;
	unsigned int pkts_compl = 0, bytes_compl = 0;

	EFX_BUG_ON_PARANOID(index > tx_queue->ptr_mask);

	efx_dequeue_buffers(tx_queue, index, &pkts_compl, &bytes_compl);
	netdev_tx_completed_queue(tx_queue->core_txq, pkts_compl, bytes_compl);

	/* See if we need to restart the netif queue.  This memory
	 * barrier ensures that we write read_count (inside
	 * efx_dequeue_buffers()) before reading the queue status.
	 */
	smp_mb();
	if (unlikely(netif_tx_queue_stopped(tx_queue->core_txq)) &&
	    likely(efx->port_enabled) &&
	    likely(netif_device_present(efx->net_dev))) {
		txq2 = efx_tx_queue_partner(tx_queue);
		fill_level = max(tx_queue->insert_count - tx_queue->read_count,
				 txq2->insert_count - txq2->read_count);
		if (fill_level <= efx->txq_wake_thresh)
			netif_tx_wake_queue(tx_queue->core_txq);
	}

	/* Check whether the hardware queue is now empty */
	if ((int)(tx_queue->read_count - tx_queue->old_write_count) >= 0) {
		tx_queue->old_write_count = ACCESS_ONCE(tx_queue->write_count);
		if (tx_queue->read_count == tx_queue->old_write_count) {
			smp_mb();
			tx_queue->empty_read_count =
				tx_queue->read_count | EFX_EMPTY_COUNT_VALID;
		}
	}
}

int efx_probe_tx_queue(struct efx_tx_queue *tx_queue)
{
	struct efx_nic *efx = tx_queue->efx;
	unsigned int entries;
	int rc;

	/* Create the smallest power-of-two aligned ring */
	entries = max(roundup_pow_of_two(efx->txq_entries), EFX_MIN_DMAQ_SIZE);
	EFX_BUG_ON_PARANOID(entries > EFX_MAX_DMAQ_SIZE);
	tx_queue->ptr_mask = entries - 1;

	netif_dbg(efx, probe, efx->net_dev,
		  "creating TX queue %d size %#x mask %#x\n",
		  tx_queue->queue, efx->txq_entries, tx_queue->ptr_mask);

	/* Allocate software ring */
	tx_queue->buffer = kcalloc(entries, sizeof(*tx_queue->buffer),
				   GFP_KERNEL);
	if (!tx_queue->buffer)
		return -ENOMEM;

	/* Allocate hardware ring */
	rc = efx_nic_probe_tx(tx_queue);
	if (rc)
		goto fail;

	return 0;

 fail:
	kfree(tx_queue->buffer);
	tx_queue->buffer = NULL;
	return rc;
}

void efx_init_tx_queue(struct efx_tx_queue *tx_queue)
{
	netif_dbg(tx_queue->efx, drv, tx_queue->efx->net_dev,
		  "initialising TX queue %d\n", tx_queue->queue);

	tx_queue->insert_count = 0;
	tx_queue->write_count = 0;
	tx_queue->old_write_count = 0;
	tx_queue->read_count = 0;
	tx_queue->old_read_count = 0;
	tx_queue->empty_read_count = 0 | EFX_EMPTY_COUNT_VALID;

	/* Set up TX descriptor ring */
	efx_nic_init_tx(tx_queue);

	tx_queue->initialised = true;
}

void efx_release_tx_buffers(struct efx_tx_queue *tx_queue)
{
	struct efx_tx_buffer *buffer;

	if (!tx_queue->buffer)
		return;

	/* Free any buffers left in the ring */
	while (tx_queue->read_count != tx_queue->write_count) {
		unsigned int pkts_compl = 0, bytes_compl = 0;
		buffer = &tx_queue->buffer[tx_queue->read_count & tx_queue->ptr_mask];
		efx_dequeue_buffer(tx_queue, buffer, &pkts_compl, &bytes_compl);
		buffer->len = 0;

		++tx_queue->read_count;
	}
	netdev_tx_reset_queue(tx_queue->core_txq);
}

void efx_fini_tx_queue(struct efx_tx_queue *tx_queue)
{
	if (!tx_queue->initialised)
		return;

	netif_dbg(tx_queue->efx, drv, tx_queue->efx->net_dev,
		  "shutting down TX queue %d\n", tx_queue->queue);

	tx_queue->initialised = false;

	/* Flush TX queue, remove descriptor ring */
	efx_nic_fini_tx(tx_queue);

	efx_release_tx_buffers(tx_queue);

	/* Free up TSO header cache */
	efx_fini_tso(tx_queue);
}

void efx_remove_tx_queue(struct efx_tx_queue *tx_queue)
{
	if (!tx_queue->buffer)
		return;

	netif_dbg(tx_queue->efx, drv, tx_queue->efx->net_dev,
		  "destroying TX queue %d\n", tx_queue->queue);
	efx_nic_remove_tx(tx_queue);

	kfree(tx_queue->buffer);
	tx_queue->buffer = NULL;
}


/* Efx TCP segmentation acceleration.
 *
 * Why?  Because by doing it here in the driver we can go significantly
 * faster than the GSO.
 *
 * Requires TX checksum offload support.
 */

/* Number of bytes inserted at the start of a TSO header buffer,
 * similar to NET_IP_ALIGN.
 */
#ifdef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
#define TSOH_OFFSET	0
#else
#define TSOH_OFFSET	NET_IP_ALIGN
#endif

#define TSOH_BUFFER(tsoh)	((u8 *)(tsoh + 1) + TSOH_OFFSET)

/* Total size of struct efx_tso_header, buffer and padding */
#define TSOH_SIZE(hdr_len)					\
	(sizeof(struct efx_tso_header) + TSOH_OFFSET + hdr_len)

/* Size of blocks on free list.  Larger blocks must be allocated from
 * the heap.
 */
#define TSOH_STD_SIZE		128

#define PTR_DIFF(p1, p2)  ((u8 *)(p1) - (u8 *)(p2))
#define ETH_HDR_LEN(skb)  (skb_network_header(skb) - (skb)->data)
#define SKB_TCP_OFF(skb)  PTR_DIFF(tcp_hdr(skb), (skb)->data)
#define SKB_IPV4_OFF(skb) PTR_DIFF(ip_hdr(skb), (skb)->data)
#define SKB_IPV6_OFF(skb) PTR_DIFF(ipv6_hdr(skb), (skb)->data)

/**
 * struct tso_state - TSO state for an SKB
 * @out_len: Remaining length in current segment
 * @seqnum: Current sequence number
 * @ipv4_id: Current IPv4 ID, host endian
 * @packet_space: Remaining space in current packet
 * @dma_addr: DMA address of current position
 * @in_len: Remaining length in current SKB fragment
 * @unmap_len: Length of SKB fragment
 * @unmap_addr: DMA address of SKB fragment
 * @dma_flags: TX buffer flags for DMA mapping - %EFX_TX_BUF_MAP_SINGLE or 0
 * @protocol: Network protocol (after any VLAN header)
 * @header_len: Number of bytes of header
 * @full_packet_size: Number of bytes to put in each outgoing segment
 *
 * The state used during segmentation.  It is put into this data structure
 * just to make it easy to pass into inline functions.
 */
struct tso_state {
	/* Output position */
	unsigned out_len;
	unsigned seqnum;
	unsigned ipv4_id;
	unsigned packet_space;

	/* Input position */
	dma_addr_t dma_addr;
	unsigned in_len;
	unsigned unmap_len;
	dma_addr_t unmap_addr;
	unsigned short dma_flags;

	__be16 protocol;
	unsigned header_len;
	int full_packet_size;
};


/*
 * Verify that our various assumptions about sk_buffs and the conditions
 * under which TSO will be attempted hold true.  Return the protocol number.
 */
static __be16 efx_tso_check_protocol(struct sk_buff *skb)
{
	__be16 protocol = skb->protocol;

	EFX_BUG_ON_PARANOID(((struct ethhdr *)skb->data)->h_proto !=
			    protocol);
	if (protocol == htons(ETH_P_8021Q)) {
		struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
		protocol = veh->h_vlan_encapsulated_proto;
	}

	if (protocol == htons(ETH_P_IP)) {
		EFX_BUG_ON_PARANOID(ip_hdr(skb)->protocol != IPPROTO_TCP);
	} else {
		EFX_BUG_ON_PARANOID(protocol != htons(ETH_P_IPV6));
		EFX_BUG_ON_PARANOID(ipv6_hdr(skb)->nexthdr != NEXTHDR_TCP);
	}
	EFX_BUG_ON_PARANOID((PTR_DIFF(tcp_hdr(skb), skb->data)
			     + (tcp_hdr(skb)->doff << 2u)) >
			    skb_headlen(skb));

	return protocol;
}


/*
 * Allocate a page worth of efx_tso_header structures, and string them
 * into the tx_queue->tso_headers_free linked list. Return 0 or -ENOMEM.
 */
static int efx_tsoh_block_alloc(struct efx_tx_queue *tx_queue)
{
	struct device *dma_dev = &tx_queue->efx->pci_dev->dev;
	struct efx_tso_header *tsoh;
	dma_addr_t dma_addr;
	u8 *base_kva, *kva;

	base_kva = dma_alloc_coherent(dma_dev, PAGE_SIZE, &dma_addr, GFP_ATOMIC);
	if (base_kva == NULL) {
		netif_err(tx_queue->efx, tx_err, tx_queue->efx->net_dev,
			  "Unable to allocate page for TSO headers\n");
		return -ENOMEM;
	}

	/* dma_alloc_coherent() allocates pages. */
	EFX_BUG_ON_PARANOID(dma_addr & (PAGE_SIZE - 1u));

	for (kva = base_kva; kva < base_kva + PAGE_SIZE; kva += TSOH_STD_SIZE) {
		tsoh = (struct efx_tso_header *)kva;
		tsoh->dma_addr = dma_addr + (TSOH_BUFFER(tsoh) - base_kva);
		tsoh->next = tx_queue->tso_headers_free;
		tx_queue->tso_headers_free = tsoh;
	}

	return 0;
}


/* Free up a TSO header, and all others in the same page. */
static void efx_tsoh_block_free(struct efx_tx_queue *tx_queue,
				struct efx_tso_header *tsoh,
				struct device *dma_dev)
{
	struct efx_tso_header **p;
	unsigned long base_kva;
	dma_addr_t base_dma;

	base_kva = (unsigned long)tsoh & PAGE_MASK;
	base_dma = tsoh->dma_addr & PAGE_MASK;

	p = &tx_queue->tso_headers_free;
	while (*p != NULL) {
		if (((unsigned long)*p & PAGE_MASK) == base_kva)
			*p = (*p)->next;
		else
			p = &(*p)->next;
	}

	dma_free_coherent(dma_dev, PAGE_SIZE, (void *)base_kva, base_dma);
}

static struct efx_tso_header *
efx_tsoh_heap_alloc(struct efx_tx_queue *tx_queue, size_t header_len)
{
	struct efx_tso_header *tsoh;

	tsoh = kmalloc(TSOH_SIZE(header_len), GFP_ATOMIC | GFP_DMA);
	if (unlikely(!tsoh))
		return NULL;

	tsoh->dma_addr = dma_map_single(&tx_queue->efx->pci_dev->dev,
					TSOH_BUFFER(tsoh), header_len,
					DMA_TO_DEVICE);
	if (unlikely(dma_mapping_error(&tx_queue->efx->pci_dev->dev,
				       tsoh->dma_addr))) {
		kfree(tsoh);
		return NULL;
	}

	tsoh->unmap_len = header_len;
	return tsoh;
}

static void
efx_tsoh_heap_free(struct efx_tx_queue *tx_queue, struct efx_tso_header *tsoh)
{
	dma_unmap_single(&tx_queue->efx->pci_dev->dev,
			 tsoh->dma_addr, tsoh->unmap_len,
			 DMA_TO_DEVICE);
	kfree(tsoh);
}

/**
 * efx_tx_queue_insert - push descriptors onto the TX queue
 * @tx_queue:		Efx TX queue
 * @dma_addr:		DMA address of fragment
 * @len:		Length of fragment
 * @final_buffer:	The final buffer inserted into the queue
 *
 * Push descriptors onto the TX queue.
 */
static void efx_tx_queue_insert(struct efx_tx_queue *tx_queue,
				dma_addr_t dma_addr, unsigned len,
				struct efx_tx_buffer **final_buffer)
{
	struct efx_tx_buffer *buffer;
	struct efx_nic *efx = tx_queue->efx;
	unsigned dma_len, insert_ptr;

	EFX_BUG_ON_PARANOID(len <= 0);

	while (1) {
		insert_ptr = tx_queue->insert_count & tx_queue->ptr_mask;
		buffer = &tx_queue->buffer[insert_ptr];
		++tx_queue->insert_count;

		EFX_BUG_ON_PARANOID(tx_queue->insert_count -
				    tx_queue->read_count >=
				    efx->txq_entries);

		efx_tsoh_free(tx_queue, buffer);
		EFX_BUG_ON_PARANOID(buffer->len);
		EFX_BUG_ON_PARANOID(buffer->unmap_len);
		EFX_BUG_ON_PARANOID(buffer->flags);

		buffer->dma_addr = dma_addr;

		dma_len = efx_max_tx_len(efx, dma_addr);

		/* If there is enough space to send then do so */
		if (dma_len >= len)
			break;

		buffer->len = dma_len;
		buffer->flags = EFX_TX_BUF_CONT;
		dma_addr += dma_len;
		len -= dma_len;
	}

	EFX_BUG_ON_PARANOID(!len);
	buffer->len = len;
	*final_buffer = buffer;
}


/*
 * Put a TSO header into the TX queue.
 *
 * This is special-cased because we know that it is small enough to fit in
 * a single fragment, and we know it doesn't cross a page boundary.  It
 * also allows us to not worry about end-of-packet etc.
 */
static void efx_tso_put_header(struct efx_tx_queue *tx_queue,
			       struct efx_tso_header *tsoh, unsigned len)
{
	struct efx_tx_buffer *buffer;

	buffer = &tx_queue->buffer[tx_queue->insert_count & tx_queue->ptr_mask];
	efx_tsoh_free(tx_queue, buffer);
	EFX_BUG_ON_PARANOID(buffer->len);
	EFX_BUG_ON_PARANOID(buffer->unmap_len);
	EFX_BUG_ON_PARANOID(buffer->flags);
	buffer->len = len;
	buffer->dma_addr = tsoh->dma_addr;
	buffer->tsoh = tsoh;
	buffer->flags = EFX_TX_BUF_TSOH | EFX_TX_BUF_CONT;

	++tx_queue->insert_count;
}


/* Remove descriptors put into a tx_queue. */
static void efx_enqueue_unwind(struct efx_tx_queue *tx_queue)
{
	struct efx_tx_buffer *buffer;
	dma_addr_t unmap_addr;

	/* Work backwards until we hit the original insert pointer value */
	while (tx_queue->insert_count != tx_queue->write_count) {
		--tx_queue->insert_count;
		buffer = &tx_queue->buffer[tx_queue->insert_count &
					   tx_queue->ptr_mask];
		efx_tsoh_free(tx_queue, buffer);
		EFX_BUG_ON_PARANOID(buffer->flags & EFX_TX_BUF_SKB);
		if (buffer->unmap_len) {
			unmap_addr = (buffer->dma_addr + buffer->len -
				      buffer->unmap_len);
			if (buffer->flags & EFX_TX_BUF_MAP_SINGLE)
				dma_unmap_single(&tx_queue->efx->pci_dev->dev,
						 unmap_addr, buffer->unmap_len,
						 DMA_TO_DEVICE);
			else
				dma_unmap_page(&tx_queue->efx->pci_dev->dev,
					       unmap_addr, buffer->unmap_len,
					       DMA_TO_DEVICE);
			buffer->unmap_len = 0;
		}
		buffer->len = 0;
		buffer->flags = 0;
	}
}


/* Parse the SKB header and initialise state. */
static void tso_start(struct tso_state *st, const struct sk_buff *skb)
{
	/* All ethernet/IP/TCP headers combined size is TCP header size
	 * plus offset of TCP header relative to start of packet.
	 */
	st->header_len = ((tcp_hdr(skb)->doff << 2u)
			  + PTR_DIFF(tcp_hdr(skb), skb->data));
	st->full_packet_size = st->header_len + skb_shinfo(skb)->gso_size;

	if (st->protocol == htons(ETH_P_IP))
		st->ipv4_id = ntohs(ip_hdr(skb)->id);
	else
		st->ipv4_id = 0;
	st->seqnum = ntohl(tcp_hdr(skb)->seq);

	EFX_BUG_ON_PARANOID(tcp_hdr(skb)->urg);
	EFX_BUG_ON_PARANOID(tcp_hdr(skb)->syn);
	EFX_BUG_ON_PARANOID(tcp_hdr(skb)->rst);

	st->out_len = skb->len - st->header_len;
	st->unmap_len = 0;
	st->dma_flags = 0;
}

static int tso_get_fragment(struct tso_state *st, struct efx_nic *efx,
			    skb_frag_t *frag)
{
	st->unmap_addr = skb_frag_dma_map(&efx->pci_dev->dev, frag, 0,
					  skb_frag_size(frag), DMA_TO_DEVICE);
	if (likely(!dma_mapping_error(&efx->pci_dev->dev, st->unmap_addr))) {
		st->dma_flags = 0;
		st->unmap_len = skb_frag_size(frag);
		st->in_len = skb_frag_size(frag);
		st->dma_addr = st->unmap_addr;
		return 0;
	}
	return -ENOMEM;
}

static int tso_get_head_fragment(struct tso_state *st, struct efx_nic *efx,
				 const struct sk_buff *skb)
{
	int hl = st->header_len;
	int len = skb_headlen(skb) - hl;

	st->unmap_addr = dma_map_single(&efx->pci_dev->dev, skb->data + hl,
					len, DMA_TO_DEVICE);
	if (likely(!dma_mapping_error(&efx->pci_dev->dev, st->unmap_addr))) {
		st->dma_flags = EFX_TX_BUF_MAP_SINGLE;
		st->unmap_len = len;
		st->in_len = len;
		st->dma_addr = st->unmap_addr;
		return 0;
	}
	return -ENOMEM;
}


/**
 * tso_fill_packet_with_fragment - form descriptors for the current fragment
 * @tx_queue:		Efx TX queue
 * @skb:		Socket buffer
 * @st:			TSO state
 *
 * Form descriptors for the current fragment, until we reach the end
 * of fragment or end-of-packet.
 */
static void tso_fill_packet_with_fragment(struct efx_tx_queue *tx_queue,
					  const struct sk_buff *skb,
					  struct tso_state *st)
{
	struct efx_tx_buffer *buffer;
	int n;

	if (st->in_len == 0)
		return;
	if (st->packet_space == 0)
		return;

	EFX_BUG_ON_PARANOID(st->in_len <= 0);
	EFX_BUG_ON_PARANOID(st->packet_space <= 0);

	n = min(st->in_len, st->packet_space);

	st->packet_space -= n;
	st->out_len -= n;
	st->in_len -= n;

	efx_tx_queue_insert(tx_queue, st->dma_addr, n, &buffer);

	if (st->out_len == 0) {
		/* Transfer ownership of the skb */
		buffer->skb = skb;
		buffer->flags = EFX_TX_BUF_SKB;
	} else if (st->packet_space != 0) {
		buffer->flags = EFX_TX_BUF_CONT;
	}

	if (st->in_len == 0) {
		/* Transfer ownership of the DMA mapping */
		buffer->unmap_len = st->unmap_len;
		buffer->flags |= st->dma_flags;
		st->unmap_len = 0;
	}

	st->dma_addr += n;
}


/**
 * tso_start_new_packet - generate a new header and prepare for the new packet
 * @tx_queue:		Efx TX queue
 * @skb:		Socket buffer
 * @st:			TSO state
 *
 * Generate a new header and prepare for the new packet.  Return 0 on
 * success, or -1 if failed to alloc header.
 */
static int tso_start_new_packet(struct efx_tx_queue *tx_queue,
				const struct sk_buff *skb,
				struct tso_state *st)
{
	struct efx_tso_header *tsoh;
	struct tcphdr *tsoh_th;
	unsigned ip_length;
	u8 *header;

	/* Allocate a DMA-mapped header buffer. */
	if (likely(TSOH_SIZE(st->header_len) <= TSOH_STD_SIZE)) {
		if (tx_queue->tso_headers_free == NULL) {
			if (efx_tsoh_block_alloc(tx_queue))
				return -1;
		}
		EFX_BUG_ON_PARANOID(!tx_queue->tso_headers_free);
		tsoh = tx_queue->tso_headers_free;
		tx_queue->tso_headers_free = tsoh->next;
		tsoh->unmap_len = 0;
	} else {
		tx_queue->tso_long_headers++;
		tsoh = efx_tsoh_heap_alloc(tx_queue, st->header_len);
		if (unlikely(!tsoh))
			return -1;
	}

	header = TSOH_BUFFER(tsoh);
	tsoh_th = (struct tcphdr *)(header + SKB_TCP_OFF(skb));

	/* Copy and update the headers. */
	memcpy(header, skb->data, st->header_len);

	tsoh_th->seq = htonl(st->seqnum);
	st->seqnum += skb_shinfo(skb)->gso_size;
	if (st->out_len > skb_shinfo(skb)->gso_size) {
		/* This packet will not finish the TSO burst. */
		ip_length = st->full_packet_size - ETH_HDR_LEN(skb);
		tsoh_th->fin = 0;
		tsoh_th->psh = 0;
	} else {
		/* This packet will be the last in the TSO burst. */
		ip_length = st->header_len - ETH_HDR_LEN(skb) + st->out_len;
		tsoh_th->fin = tcp_hdr(skb)->fin;
		tsoh_th->psh = tcp_hdr(skb)->psh;
	}

	if (st->protocol == htons(ETH_P_IP)) {
		struct iphdr *tsoh_iph =
			(struct iphdr *)(header + SKB_IPV4_OFF(skb));

		tsoh_iph->tot_len = htons(ip_length);

		/* Linux leaves suitable gaps in the IP ID space for us to fill. */
		tsoh_iph->id = htons(st->ipv4_id);
		st->ipv4_id++;
	} else {
		struct ipv6hdr *tsoh_iph =
			(struct ipv6hdr *)(header + SKB_IPV6_OFF(skb));

		tsoh_iph->payload_len = htons(ip_length - sizeof(*tsoh_iph));
	}

	st->packet_space = skb_shinfo(skb)->gso_size;
	++tx_queue->tso_packets;

	/* Form a descriptor for this header. */
	efx_tso_put_header(tx_queue, tsoh, st->header_len);

	return 0;
}


/**
 * efx_enqueue_skb_tso - segment and transmit a TSO socket buffer
 * @tx_queue:		Efx TX queue
 * @skb:		Socket buffer
 *
 * Context: You must hold netif_tx_lock() to call this function.
 *
 * Add socket buffer @skb to @tx_queue, doing TSO or return != 0 if
 * @skb was not enqueued.  In all cases @skb is consumed.  Return
 * %NETDEV_TX_OK.
 */
static int efx_enqueue_skb_tso(struct efx_tx_queue *tx_queue,
			       struct sk_buff *skb)
{
	struct efx_nic *efx = tx_queue->efx;
	int frag_i, rc;
	struct tso_state state;

	/* Find the packet protocol and sanity-check it */
	state.protocol = efx_tso_check_protocol(skb);

	EFX_BUG_ON_PARANOID(tx_queue->write_count != tx_queue->insert_count);

	tso_start(&state, skb);

	/* Assume that skb header area contains exactly the headers, and
	 * all payload is in the frag list.
	 */
	if (skb_headlen(skb) == state.header_len) {
		/* Grab the first payload fragment. */
		EFX_BUG_ON_PARANOID(skb_shinfo(skb)->nr_frags < 1);
		frag_i = 0;
		rc = tso_get_fragment(&state, efx,
				      skb_shinfo(skb)->frags + frag_i);
		if (rc)
			goto mem_err;
	} else {
		rc = tso_get_head_fragment(&state, efx, skb);
		if (rc)
			goto mem_err;
		frag_i = -1;
	}

	if (tso_start_new_packet(tx_queue, skb, &state) < 0)
		goto mem_err;

	while (1) {
		tso_fill_packet_with_fragment(tx_queue, skb, &state);

		/* Move onto the next fragment? */
		if (state.in_len == 0) {
			if (++frag_i >= skb_shinfo(skb)->nr_frags)
				/* End of payload reached. */
				break;
			rc = tso_get_fragment(&state, efx,
					      skb_shinfo(skb)->frags + frag_i);
			if (rc)
				goto mem_err;
		}

		/* Start at new packet? */
		if (state.packet_space == 0 &&
		    tso_start_new_packet(tx_queue, skb, &state) < 0)
			goto mem_err;
	}

	netdev_tx_sent_queue(tx_queue->core_txq, skb->len);

	/* Pass off to hardware */
	efx_nic_push_buffers(tx_queue);

	efx_tx_maybe_stop_queue(tx_queue);

	tx_queue->tso_bursts++;
	return NETDEV_TX_OK;

 mem_err:
	netif_err(efx, tx_err, efx->net_dev,
		  "Out of memory for TSO headers, or DMA mapping error\n");
	dev_kfree_skb_any(skb);

	/* Free the DMA mapping we were in the process of writing out */
	if (state.unmap_len) {
		if (state.dma_flags & EFX_TX_BUF_MAP_SINGLE)
			dma_unmap_single(&efx->pci_dev->dev, state.unmap_addr,
					 state.unmap_len, DMA_TO_DEVICE);
		else
			dma_unmap_page(&efx->pci_dev->dev, state.unmap_addr,
				       state.unmap_len, DMA_TO_DEVICE);
	}

	efx_enqueue_unwind(tx_queue);
	return NETDEV_TX_OK;
}


/*
 * Free up all TSO datastructures associated with tx_queue. This
 * routine should be called only once the tx_queue is both empty and
 * will no longer be used.
 */
static void efx_fini_tso(struct efx_tx_queue *tx_queue)
{
	unsigned i;

	if (tx_queue->buffer) {
		for (i = 0; i <= tx_queue->ptr_mask; ++i)
			efx_tsoh_free(tx_queue, &tx_queue->buffer[i]);
	}

	while (tx_queue->tso_headers_free != NULL)
		efx_tsoh_block_free(tx_queue, tx_queue->tso_headers_free,
				    &tx_queue->efx->pci_dev->dev);
}