summaryrefslogtreecommitdiff
path: root/drivers/net/ethernet/sfc/falcon/tx.c
blob: f1520a404ac619851d1f6ed5622828b6c7fbe44c (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
/****************************************************************************
 * Driver for Solarflare network controllers and boards
 * Copyright 2005-2006 Fen Systems Ltd.
 * Copyright 2005-2013 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 <linux/cache.h>
#include "net_driver.h"
#include "efx.h"
#include "io.h"
#include "nic.h"
#include "tx.h"
#include "workarounds.h"

static inline u8 *ef4_tx_get_copy_buffer(struct ef4_tx_queue *tx_queue,
					 struct ef4_tx_buffer *buffer)
{
	unsigned int index = ef4_tx_queue_get_insert_index(tx_queue);
	struct ef4_buffer *page_buf =
		&tx_queue->cb_page[index >> (PAGE_SHIFT - EF4_TX_CB_ORDER)];
	unsigned int offset =
		((index << EF4_TX_CB_ORDER) + NET_IP_ALIGN) & (PAGE_SIZE - 1);

	if (unlikely(!page_buf->addr) &&
	    ef4_nic_alloc_buffer(tx_queue->efx, page_buf, PAGE_SIZE,
				 GFP_ATOMIC))
		return NULL;
	buffer->dma_addr = page_buf->dma_addr + offset;
	buffer->unmap_len = 0;
	return (u8 *)page_buf->addr + offset;
}

u8 *ef4_tx_get_copy_buffer_limited(struct ef4_tx_queue *tx_queue,
				   struct ef4_tx_buffer *buffer, size_t len)
{
	if (len > EF4_TX_CB_SIZE)
		return NULL;
	return ef4_tx_get_copy_buffer(tx_queue, buffer);
}

static void ef4_dequeue_buffer(struct ef4_tx_queue *tx_queue,
			       struct ef4_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->dma_offset;
		if (buffer->flags & EF4_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 & EF4_TX_BUF_SKB) {
		(*pkts_compl)++;
		(*bytes_compl) += buffer->skb->len;
		dev_consume_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->len = 0;
	buffer->flags = 0;
}

unsigned int ef4_tx_max_skb_descs(struct ef4_nic *efx)
{
	/* This is probably too much since we don't have any TSO support;
	 * it's a left-over from when we had Software TSO.  But it's safer
	 * to leave it as-is than try to determine a new bound.
	 */
	/* Header and payload descriptor for each output segment, plus
	 * one for every input fragment boundary within a segment
	 */
	unsigned int max_descs = EF4_TSO_MAX_SEGS * 2 + MAX_SKB_FRAGS;

	/* Possibly one more per segment for the alignment workaround,
	 * or for option descriptors
	 */
	if (EF4_WORKAROUND_5391(efx))
		max_descs += EF4_TSO_MAX_SEGS;

	/* Possibly more for PCIe page boundaries within input fragments */
	if (PAGE_SIZE > EF4_PAGE_SIZE)
		max_descs += max_t(unsigned int, MAX_SKB_FRAGS,
				   DIV_ROUND_UP(GSO_MAX_SIZE, EF4_PAGE_SIZE));

	return max_descs;
}

static void ef4_tx_maybe_stop_queue(struct ef4_tx_queue *txq1)
{
	/* We need to consider both queues that the net core sees as one */
	struct ef4_tx_queue *txq2 = ef4_tx_queue_partner(txq1);
	struct ef4_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);
	EF4_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);
	}
}

static int ef4_enqueue_skb_copy(struct ef4_tx_queue *tx_queue,
				struct sk_buff *skb)
{
	unsigned int min_len = tx_queue->tx_min_size;
	unsigned int copy_len = skb->len;
	struct ef4_tx_buffer *buffer;
	u8 *copy_buffer;
	int rc;

	EF4_BUG_ON_PARANOID(copy_len > EF4_TX_CB_SIZE);

	buffer = ef4_tx_queue_get_insert_buffer(tx_queue);

	copy_buffer = ef4_tx_get_copy_buffer(tx_queue, buffer);
	if (unlikely(!copy_buffer))
		return -ENOMEM;

	rc = skb_copy_bits(skb, 0, copy_buffer, copy_len);
	EF4_WARN_ON_PARANOID(rc);
	if (unlikely(copy_len < min_len)) {
		memset(copy_buffer + copy_len, 0, min_len - copy_len);
		buffer->len = min_len;
	} else {
		buffer->len = copy_len;
	}

	buffer->skb = skb;
	buffer->flags = EF4_TX_BUF_SKB;

	++tx_queue->insert_count;
	return rc;
}

static struct ef4_tx_buffer *ef4_tx_map_chunk(struct ef4_tx_queue *tx_queue,
					      dma_addr_t dma_addr,
					      size_t len)
{
	const struct ef4_nic_type *nic_type = tx_queue->efx->type;
	struct ef4_tx_buffer *buffer;
	unsigned int dma_len;

	/* Map the fragment taking account of NIC-dependent DMA limits. */
	do {
		buffer = ef4_tx_queue_get_insert_buffer(tx_queue);
		dma_len = nic_type->tx_limit_len(tx_queue, dma_addr, len);

		buffer->len = dma_len;
		buffer->dma_addr = dma_addr;
		buffer->flags = EF4_TX_BUF_CONT;
		len -= dma_len;
		dma_addr += dma_len;
		++tx_queue->insert_count;
	} while (len);

	return buffer;
}

/* Map all data from an SKB for DMA and create descriptors on the queue.
 */
static int ef4_tx_map_data(struct ef4_tx_queue *tx_queue, struct sk_buff *skb)
{
	struct ef4_nic *efx = tx_queue->efx;
	struct device *dma_dev = &efx->pci_dev->dev;
	unsigned int frag_index, nr_frags;
	dma_addr_t dma_addr, unmap_addr;
	unsigned short dma_flags;
	size_t len, unmap_len;

	nr_frags = skb_shinfo(skb)->nr_frags;
	frag_index = 0;

	/* Map header data. */
	len = skb_headlen(skb);
	dma_addr = dma_map_single(dma_dev, skb->data, len, DMA_TO_DEVICE);
	dma_flags = EF4_TX_BUF_MAP_SINGLE;
	unmap_len = len;
	unmap_addr = dma_addr;

	if (unlikely(dma_mapping_error(dma_dev, dma_addr)))
		return -EIO;

	/* Add descriptors for each fragment. */
	do {
		struct ef4_tx_buffer *buffer;
		skb_frag_t *fragment;

		buffer = ef4_tx_map_chunk(tx_queue, dma_addr, len);

		/* The final descriptor for a fragment is responsible for
		 * unmapping the whole fragment.
		 */
		buffer->flags = EF4_TX_BUF_CONT | dma_flags;
		buffer->unmap_len = unmap_len;
		buffer->dma_offset = buffer->dma_addr - unmap_addr;

		if (frag_index >= nr_frags) {
			/* Store SKB details with the final buffer for
			 * the completion.
			 */
			buffer->skb = skb;
			buffer->flags = EF4_TX_BUF_SKB | dma_flags;
			return 0;
		}

		/* Move on to the next fragment. */
		fragment = &skb_shinfo(skb)->frags[frag_index++];
		len = skb_frag_size(fragment);
		dma_addr = skb_frag_dma_map(dma_dev, fragment,
				0, len, DMA_TO_DEVICE);
		dma_flags = 0;
		unmap_len = len;
		unmap_addr = dma_addr;

		if (unlikely(dma_mapping_error(dma_dev, dma_addr)))
			return -EIO;
	} while (1);
}

/* Remove buffers put into a tx_queue.  None of the buffers must have
 * an skb attached.
 */
static void ef4_enqueue_unwind(struct ef4_tx_queue *tx_queue)
{
	struct ef4_tx_buffer *buffer;

	/* Work backwards until we hit the original insert pointer value */
	while (tx_queue->insert_count != tx_queue->write_count) {
		--tx_queue->insert_count;
		buffer = __ef4_tx_queue_get_insert_buffer(tx_queue);
		ef4_dequeue_buffer(tx_queue, buffer, NULL, NULL);
	}
}

/*
 * 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 ef4_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 ef4_enqueue_skb(struct ef4_tx_queue *tx_queue, struct sk_buff *skb)
{
	bool data_mapped = false;
	unsigned int skb_len;

	skb_len = skb->len;
	EF4_WARN_ON_PARANOID(skb_is_gso(skb));

	if (skb_len < tx_queue->tx_min_size ||
			(skb->data_len && skb_len <= EF4_TX_CB_SIZE)) {
		/* Pad short packets or coalesce short fragmented packets. */
		if (ef4_enqueue_skb_copy(tx_queue, skb))
			goto err;
		tx_queue->cb_packets++;
		data_mapped = true;
	}

	/* Map for DMA and create descriptors if we haven't done so already. */
	if (!data_mapped && (ef4_tx_map_data(tx_queue, skb)))
		goto err;

	/* Update BQL */
	netdev_tx_sent_queue(tx_queue->core_txq, skb_len);

	/* Pass off to hardware */
	if (!skb->xmit_more || netif_xmit_stopped(tx_queue->core_txq)) {
		struct ef4_tx_queue *txq2 = ef4_tx_queue_partner(tx_queue);

		/* There could be packets left on the partner queue if those
		 * SKBs had skb->xmit_more set. If we do not push those they
		 * could be left for a long time and cause a netdev watchdog.
		 */
		if (txq2->xmit_more_available)
			ef4_nic_push_buffers(txq2);

		ef4_nic_push_buffers(tx_queue);
	} else {
		tx_queue->xmit_more_available = skb->xmit_more;
	}

	tx_queue->tx_packets++;

	ef4_tx_maybe_stop_queue(tx_queue);

	return NETDEV_TX_OK;


err:
	ef4_enqueue_unwind(tx_queue);
	dev_kfree_skb_any(skb);
	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 ef4_dequeue_buffers(struct ef4_tx_queue *tx_queue,
				unsigned int index,
				unsigned int *pkts_compl,
				unsigned int *bytes_compl)
{
	struct ef4_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 ef4_tx_buffer *buffer = &tx_queue->buffer[read_ptr];

		if (!(buffer->flags & EF4_TX_BUF_OPTION) &&
		    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);
			ef4_schedule_reset(efx, RESET_TYPE_TX_SKIP);
			return;
		}

		ef4_dequeue_buffer(tx_queue, buffer, pkts_compl, bytes_compl);

		++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 ef4_hard_start_xmit(struct sk_buff *skb,
				struct net_device *net_dev)
{
	struct ef4_nic *efx = netdev_priv(net_dev);
	struct ef4_tx_queue *tx_queue;
	unsigned index, type;

	EF4_WARN_ON_PARANOID(!netif_device_present(net_dev));

	index = skb_get_queue_mapping(skb);
	type = skb->ip_summed == CHECKSUM_PARTIAL ? EF4_TXQ_TYPE_OFFLOAD : 0;
	if (index >= efx->n_tx_channels) {
		index -= efx->n_tx_channels;
		type |= EF4_TXQ_TYPE_HIGHPRI;
	}
	tx_queue = ef4_get_tx_queue(efx, index, type);

	return ef4_enqueue_skb(tx_queue, skb);
}

void ef4_init_tx_queue_core_txq(struct ef4_tx_queue *tx_queue)
{
	struct ef4_nic *efx = tx_queue->efx;

	/* Must be inverse of queue lookup in ef4_hard_start_xmit() */
	tx_queue->core_txq =
		netdev_get_tx_queue(efx->net_dev,
				    tx_queue->queue / EF4_TXQ_TYPES +
				    ((tx_queue->queue & EF4_TXQ_TYPE_HIGHPRI) ?
				     efx->n_tx_channels : 0));
}

int ef4_setup_tc(struct net_device *net_dev, u32 handle, u32 chain_index,
		 __be16 proto, struct tc_to_netdev *ntc)
{
	struct ef4_nic *efx = netdev_priv(net_dev);
	struct ef4_channel *channel;
	struct ef4_tx_queue *tx_queue;
	unsigned tc, num_tc;
	int rc;

	if (ntc->type != TC_SETUP_MQPRIO)
		return -EINVAL;

	num_tc = ntc->mqprio->num_tc;

	if (ef4_nic_rev(efx) < EF4_REV_FALCON_B0 || num_tc > EF4_MAX_TX_TC)
		return -EINVAL;

	ntc->mqprio->hw = TC_MQPRIO_HW_OFFLOAD_TCS;

	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 */
		ef4_for_each_channel(channel, efx) {
			ef4_for_each_possible_channel_tx_queue(tx_queue,
							       channel) {
				if (!(tx_queue->queue & EF4_TXQ_TYPE_HIGHPRI))
					continue;
				if (!tx_queue->buffer) {
					rc = ef4_probe_tx_queue(tx_queue);
					if (rc)
						return rc;
				}
				if (!tx_queue->initialised)
					ef4_init_tx_queue(tx_queue);
				ef4_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 ef4_fini_channels().
	 */

	net_dev->num_tc = num_tc;
	return 0;
}

void ef4_xmit_done(struct ef4_tx_queue *tx_queue, unsigned int index)
{
	unsigned fill_level;
	struct ef4_nic *efx = tx_queue->efx;
	struct ef4_tx_queue *txq2;
	unsigned int pkts_compl = 0, bytes_compl = 0;

	EF4_BUG_ON_PARANOID(index > tx_queue->ptr_mask);

	ef4_dequeue_buffers(tx_queue, index, &pkts_compl, &bytes_compl);
	tx_queue->pkts_compl += pkts_compl;
	tx_queue->bytes_compl += bytes_compl;

	if (pkts_compl > 1)
		++tx_queue->merge_events;

	/* See if we need to restart the netif queue.  This memory
	 * barrier ensures that we write read_count (inside
	 * ef4_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 = ef4_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 | EF4_EMPTY_COUNT_VALID;
		}
	}
}

static unsigned int ef4_tx_cb_page_count(struct ef4_tx_queue *tx_queue)
{
	return DIV_ROUND_UP(tx_queue->ptr_mask + 1, PAGE_SIZE >> EF4_TX_CB_ORDER);
}

int ef4_probe_tx_queue(struct ef4_tx_queue *tx_queue)
{
	struct ef4_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), EF4_MIN_DMAQ_SIZE);
	EF4_BUG_ON_PARANOID(entries > EF4_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;

	tx_queue->cb_page = kcalloc(ef4_tx_cb_page_count(tx_queue),
				    sizeof(tx_queue->cb_page[0]), GFP_KERNEL);
	if (!tx_queue->cb_page) {
		rc = -ENOMEM;
		goto fail1;
	}

	/* Allocate hardware ring */
	rc = ef4_nic_probe_tx(tx_queue);
	if (rc)
		goto fail2;

	return 0;

fail2:
	kfree(tx_queue->cb_page);
	tx_queue->cb_page = NULL;
fail1:
	kfree(tx_queue->buffer);
	tx_queue->buffer = NULL;
	return rc;
}

void ef4_init_tx_queue(struct ef4_tx_queue *tx_queue)
{
	struct ef4_nic *efx = tx_queue->efx;

	netif_dbg(efx, drv, 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 | EF4_EMPTY_COUNT_VALID;
	tx_queue->xmit_more_available = false;

	/* Some older hardware requires Tx writes larger than 32. */
	tx_queue->tx_min_size = EF4_WORKAROUND_15592(efx) ? 33 : 0;

	/* Set up TX descriptor ring */
	ef4_nic_init_tx(tx_queue);

	tx_queue->initialised = true;
}

void ef4_fini_tx_queue(struct ef4_tx_queue *tx_queue)
{
	struct ef4_tx_buffer *buffer;

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

	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];
		ef4_dequeue_buffer(tx_queue, buffer, &pkts_compl, &bytes_compl);

		++tx_queue->read_count;
	}
	tx_queue->xmit_more_available = false;
	netdev_tx_reset_queue(tx_queue->core_txq);
}

void ef4_remove_tx_queue(struct ef4_tx_queue *tx_queue)
{
	int i;

	if (!tx_queue->buffer)
		return;

	netif_dbg(tx_queue->efx, drv, tx_queue->efx->net_dev,
		  "destroying TX queue %d\n", tx_queue->queue);
	ef4_nic_remove_tx(tx_queue);

	if (tx_queue->cb_page) {
		for (i = 0; i < ef4_tx_cb_page_count(tx_queue); i++)
			ef4_nic_free_buffer(tx_queue->efx,
					    &tx_queue->cb_page[i]);
		kfree(tx_queue->cb_page);
		tx_queue->cb_page = NULL;
	}

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