aboutsummaryrefslogtreecommitdiffstats
path: root/drivers/net/sfc/tx.c
blob: d6681edb7014d799e92ab3978abe2e38c6fd664a (plain) (blame)
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
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
/****************************************************************************
 * Driver for Solarflare Solarstorm network controllers and boards
 * Copyright 2005-2006 Fen Systems Ltd.
 * Copyright 2005-2008 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/if_ether.h>
#include <linux/highmem.h>
#include "net_driver.h"
#include "tx.h"
#include "efx.h"
#include "falcon.h"
#include "workarounds.h"

/*
 * TX descriptor ring full threshold
 *
 * The tx_queue descriptor ring fill-level must fall below this value
 * before we restart the netif queue
 */
#define EFX_NETDEV_TX_THRESHOLD(_tx_queue)	\
	(_tx_queue->efx->type->txd_ring_mask / 2u)

/* We want to be able to nest calls to netif_stop_queue(), since each
 * channel can have an individual stop on the queue.
 */
void efx_stop_queue(struct efx_nic *efx)
{
	spin_lock_bh(&efx->netif_stop_lock);
	EFX_TRACE(efx, "stop TX queue\n");

	atomic_inc(&efx->netif_stop_count);
	netif_stop_queue(efx->net_dev);

	spin_unlock_bh(&efx->netif_stop_lock);
}

/* Wake netif's TX queue
 * We want to be able to nest calls to netif_stop_queue(), since each
 * channel can have an individual stop on the queue.
 */
void efx_wake_queue(struct efx_nic *efx)
{
	local_bh_disable();
	if (atomic_dec_and_lock(&efx->netif_stop_count,
				&efx->netif_stop_lock)) {
		EFX_TRACE(efx, "waking TX queue\n");
		netif_wake_queue(efx->net_dev);
		spin_unlock(&efx->netif_stop_lock);
	}
	local_bh_enable();
}

static void efx_dequeue_buffer(struct efx_tx_queue *tx_queue,
			       struct efx_tx_buffer *buffer)
{
	if (buffer->unmap_len) {
		struct pci_dev *pci_dev = tx_queue->efx->pci_dev;
		dma_addr_t unmap_addr = (buffer->dma_addr + buffer->len -
					 buffer->unmap_len);
		if (buffer->unmap_single)
			pci_unmap_single(pci_dev, unmap_addr, buffer->unmap_len,
					 PCI_DMA_TODEVICE);
		else
			pci_unmap_page(pci_dev, unmap_addr, buffer->unmap_len,
				       PCI_DMA_TODEVICE);
		buffer->unmap_len = 0;
		buffer->unmap_single = false;
	}

	if (buffer->skb) {
		dev_kfree_skb_any((struct sk_buff *) buffer->skb);
		buffer->skb = NULL;
		EFX_TRACE(tx_queue->efx, "TX queue %d transmission id %x "
			  "complete\n", tx_queue->queue, read_ptr);
	}
}

/**
 * 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->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->tsoh = 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.
 *
 * Returns NETDEV_TX_OK or NETDEV_TX_BUSY
 * You must hold netif_tx_lock() to call this function.
 */
static int efx_enqueue_skb(struct efx_tx_queue *tx_queue,
			   struct sk_buff *skb)
{
	struct efx_nic *efx = tx_queue->efx;
	struct pci_dev *pci_dev = efx->pci_dev;
	struct efx_tx_buffer *buffer;
	skb_frag_t *fragment;
	struct page *page;
	int page_offset;
	unsigned int len, unmap_len = 0, fill_level, insert_ptr, misalign;
	dma_addr_t dma_addr, unmap_addr = 0;
	unsigned int dma_len;
	bool unmap_single;
	int q_space, i = 0;
	int rc = NETDEV_TX_OK;

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

	if (skb_shinfo((struct sk_buff *)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;
	}

	fill_level = tx_queue->insert_count - tx_queue->old_read_count;
	q_space = efx->type->txd_ring_mask - 1 - fill_level;

	/* Map for DMA.  Use pci_map_single rather than pci_map_page
	 * since this is more efficient on machines with sparse
	 * memory.
	 */
	unmap_single = true;
	dma_addr = pci_map_single(pci_dev, skb->data, len, PCI_DMA_TODEVICE);

	/* Process all fragments */
	while (1) {
		if (unlikely(pci_dma_mapping_error(pci_dev, dma_addr)))
			goto pci_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 {
			if (unlikely(q_space-- <= 0)) {
				/* It might be that completions have
				 * happened since the xmit path last
				 * checked.  Update the xmit path's
				 * copy of read_count.
				 */
				++tx_queue->stopped;
				/* This memory barrier protects the
				 * change of stopped from the access
				 * of read_count. */
				smp_mb();
				tx_queue->old_read_count =
					*(volatile unsigned *)
					&tx_queue->read_count;
				fill_level = (tx_queue->insert_count
					      - tx_queue->old_read_count);
				q_space = (efx->type->txd_ring_mask - 1 -
					   fill_level);
				if (unlikely(q_space-- <= 0))
					goto stop;
				smp_mb();
				--tx_queue->stopped;
			}

			insert_ptr = (tx_queue->insert_count &
				      efx->type->txd_ring_mask);
			buffer = &tx_queue->buffer[insert_ptr];
			efx_tsoh_free(tx_queue, buffer);
			EFX_BUG_ON_PARANOID(buffer->tsoh);
			EFX_BUG_ON_PARANOID(buffer->skb);
			EFX_BUG_ON_PARANOID(buffer->len);
			EFX_BUG_ON_PARANOID(!buffer->continuation);
			EFX_BUG_ON_PARANOID(buffer->unmap_len);

			dma_len = (((~dma_addr) & efx->type->tx_dma_mask) + 1);
			if (likely(dma_len > len))
				dma_len = len;

			misalign = (unsigned)dma_addr & efx->type->bug5391_mask;
			if (misalign && dma_len + misalign > 512)
				dma_len = 512 - misalign;

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

		/* Transfer ownership of the unmapping to the final buffer */
		buffer->unmap_single = unmap_single;
		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 = fragment->size;
		page = fragment->page;
		page_offset = fragment->page_offset;
		i++;
		/* Map for DMA */
		unmap_single = false;
		dma_addr = pci_map_page(pci_dev, page, page_offset, len,
					PCI_DMA_TODEVICE);
	}

	/* Transfer ownership of the skb to the final buffer */
	buffer->skb = skb;
	buffer->continuation = false;

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

	return NETDEV_TX_OK;

 pci_err:
	EFX_ERR_RL(efx, " 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((struct sk_buff *)skb);
	goto unwind;

 stop:
	rc = NETDEV_TX_BUSY;

	if (tx_queue->stopped == 1)
		efx_stop_queue(efx);

 unwind:
	/* Work backwards until we hit the original insert pointer value */
	while (tx_queue->insert_count != tx_queue->write_count) {
		--tx_queue->insert_count;
		insert_ptr = tx_queue->insert_count & efx->type->txd_ring_mask;
		buffer = &tx_queue->buffer[insert_ptr];
		efx_dequeue_buffer(tx_queue, buffer);
		buffer->len = 0;
	}

	/* Free the fragment we were mid-way through pushing */
	if (unmap_len) {
		if (unmap_single)
			pci_unmap_single(pci_dev, unmap_addr, unmap_len,
					 PCI_DMA_TODEVICE);
		else
			pci_unmap_page(pci_dev, unmap_addr, unmap_len,
				       PCI_DMA_TODEVICE);
	}

	return rc;
}

/* 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)
{
	struct efx_nic *efx = tx_queue->efx;
	unsigned int stop_index, read_ptr;
	unsigned int mask = tx_queue->efx->type->txd_ring_mask;

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

	while (read_ptr != stop_index) {
		struct efx_tx_buffer *buffer = &tx_queue->buffer[read_ptr];
		if (unlikely(buffer->len == 0)) {
			EFX_ERR(tx_queue->efx, "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);
		buffer->continuation = true;
		buffer->len = 0;

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

/* Initiate a packet transmission on the specified TX queue.
 * Note that returning anything other than NETDEV_TX_OK will cause the
 * OS to free the skb.
 *
 * This function is split out from efx_hard_start_xmit to allow the
 * loopback test to direct packets via specific TX queues.  It is
 * therefore a non-static inline, so as not to penalise performance
 * for non-loopback transmissions.
 *
 * Context: netif_tx_lock held
 */
inline int efx_xmit(struct efx_nic *efx,
		    struct efx_tx_queue *tx_queue, struct sk_buff *skb)
{
	int rc;

	/* Map fragments for DMA and add to TX queue */
	rc = efx_enqueue_skb(tx_queue, skb);
	if (unlikely(rc != NETDEV_TX_OK))
		goto out;

	/* Update last TX timer */
	efx->net_dev->trans_start = jiffies;

 out:
	return rc;
}

/* 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.
 */
int 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;

	if (unlikely(efx->port_inhibited))
		return NETDEV_TX_BUSY;

	if (likely(skb->ip_summed == CHECKSUM_PARTIAL))
		tx_queue = &efx->tx_queue[EFX_TX_QUEUE_OFFLOAD_CSUM];
	else
		tx_queue = &efx->tx_queue[EFX_TX_QUEUE_NO_CSUM];

	return efx_xmit(efx, tx_queue, skb);
}

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

	EFX_BUG_ON_PARANOID(index > efx->type->txd_ring_mask);

	efx_dequeue_buffers(tx_queue, index);

	/* See if we need to restart the netif queue.  This barrier
	 * separates the update of read_count from the test of
	 * stopped. */
	smp_mb();
	if (unlikely(tx_queue->stopped) && likely(efx->port_enabled)) {
		fill_level = tx_queue->insert_count - tx_queue->read_count;
		if (fill_level < EFX_NETDEV_TX_THRESHOLD(tx_queue)) {
			EFX_BUG_ON_PARANOID(!efx_dev_registered(efx));

			/* Do this under netif_tx_lock(), to avoid racing
			 * with efx_xmit(). */
			netif_tx_lock(efx->net_dev);
			if (tx_queue->stopped) {
				tx_queue->stopped = 0;
				efx_wake_queue(efx);
			}
			netif_tx_unlock(efx->net_dev);
		}
	}
}

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

	EFX_LOG(efx, "creating TX queue %d\n", tx_queue->queue);

	/* Allocate software ring */
	txq_size = (efx->type->txd_ring_mask + 1) * sizeof(*tx_queue->buffer);
	tx_queue->buffer = kzalloc(txq_size, GFP_KERNEL);
	if (!tx_queue->buffer)
		return -ENOMEM;
	for (i = 0; i <= efx->type->txd_ring_mask; ++i)
		tx_queue->buffer[i].continuation = true;

	/* Allocate hardware ring */
	rc = falcon_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)
{
	EFX_LOG(tx_queue->efx, "initialising TX queue %d\n", tx_queue->queue);

	tx_queue->insert_count = 0;
	tx_queue->write_count = 0;
	tx_queue->read_count = 0;
	tx_queue->old_read_count = 0;
	BUG_ON(tx_queue->stopped);

	/* Set up TX descriptor ring */
	falcon_init_tx(tx_queue);
}

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) {
		buffer = &tx_queue->buffer[tx_queue->read_count &
					   tx_queue->efx->type->txd_ring_mask];
		efx_dequeue_buffer(tx_queue, buffer);
		buffer->continuation = true;
		buffer->len = 0;

		++tx_queue->read_count;
	}
}

void efx_fini_tx_queue(struct efx_tx_queue *tx_queue)
{
	EFX_LOG(tx_queue->efx, "shutting down TX queue %d\n", tx_queue->queue);

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

	efx_release_tx_buffers(tx_queue);

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

	/* Release queue's stop on port, if any */
	if (tx_queue->stopped) {
		tx_queue->stopped = 0;
		efx_wake_queue(tx_queue->efx);
	}
}

void efx_remove_tx_queue(struct efx_tx_queue *tx_queue)
{
	EFX_LOG(tx_queue->efx, "destroying TX queue %d\n", tx_queue->queue);
	falcon_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)

/**
 * 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
 * @unmap_single: DMA single vs page mapping flag
 * @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;
	bool unmap_single;

	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.
 */
static void efx_tso_check_safe(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)) {
		/* Find the encapsulated protocol; reset network header
		 * and transport header based on that. */
		struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
		protocol = veh->h_vlan_encapsulated_proto;
		skb_set_network_header(skb, sizeof(*veh));
		if (protocol == htons(ETH_P_IP))
			skb_set_transport_header(skb, sizeof(*veh) +
						 4 * ip_hdr(skb)->ihl);
	}

	EFX_BUG_ON_PARANOID(protocol != htons(ETH_P_IP));
	EFX_BUG_ON_PARANOID(ip_hdr(skb)->protocol != IPPROTO_TCP);
	EFX_BUG_ON_PARANOID((PTR_DIFF(tcp_hdr(skb), skb->data)
			     + (tcp_hdr(skb)->doff << 2u)) >
			    skb_headlen(skb));
}


/*
 * 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 pci_dev *pci_dev = tx_queue->efx->pci_dev;
	struct efx_tso_header *tsoh;
	dma_addr_t dma_addr;
	u8 *base_kva, *kva;

	base_kva = pci_alloc_consistent(pci_dev, PAGE_SIZE, &dma_addr);
	if (base_kva == NULL) {
		EFX_ERR(tx_queue->efx, "Unable to allocate page for TSO"
			" headers\n");
		return -ENOMEM;
	}

	/* pci_alloc_consistent() 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 pci_dev *pci_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;
	}

	pci_free_consistent(pci_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 = pci_map_single(tx_queue->efx->pci_dev,
					TSOH_BUFFER(tsoh), header_len,
					PCI_DMA_TODEVICE);
	if (unlikely(pci_dma_mapping_error(tx_queue->efx->pci_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)
{
	pci_unmap_single(tx_queue->efx->pci_dev,
			 tsoh->dma_addr, tsoh->unmap_len,
			 PCI_DMA_TODEVICE);
	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.  Return 0 on success or 1 if
 * @tx_queue full.
 */
static int 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, fill_level, insert_ptr, misalign;
	int q_space;

	EFX_BUG_ON_PARANOID(len <= 0);

	fill_level = tx_queue->insert_count - tx_queue->old_read_count;
	/* -1 as there is no way to represent all descriptors used */
	q_space = efx->type->txd_ring_mask - 1 - fill_level;

	while (1) {
		if (unlikely(q_space-- <= 0)) {
			/* It might be that completions have happened
			 * since the xmit path last checked.  Update
			 * the xmit path's copy of read_count.
			 */
			++tx_queue->stopped;
			/* This memory barrier protects the change of
			 * stopped from the access of read_count. */
			smp_mb();
			tx_queue->old_read_count =
				*(volatile unsigned *)&tx_queue->read_count;
			fill_level = (tx_queue->insert_count
				      - tx_queue->old_read_count);
			q_space = efx->type->txd_ring_mask - 1 - fill_level;
			if (unlikely(q_space-- <= 0)) {
				*final_buffer = NULL;
				return 1;
			}
			smp_mb();
			--tx_queue->stopped;
		}

		insert_ptr = tx_queue->insert_count & efx->type->txd_ring_mask;
		buffer = &tx_queue->buffer[insert_ptr];
		++tx_queue->insert_count;

		EFX_BUG_ON_PARANOID(tx_queue->insert_count -
				    tx_queue->read_count >
				    efx->type->txd_ring_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->skb);
		EFX_BUG_ON_PARANOID(!buffer->continuation);
		EFX_BUG_ON_PARANOID(buffer->tsoh);

		buffer->dma_addr = dma_addr;

		/* Ensure we do not cross a boundary unsupported by H/W */
		dma_len = (~dma_addr & efx->type->tx_dma_mask) + 1;

		misalign = (unsigned)dma_addr & efx->type->bug5391_mask;
		if (misalign && dma_len + misalign > 512)
			dma_len = 512 - misalign;

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

		buffer->len = dma_len; /* Don't set the other members */
		dma_addr += dma_len;
		len -= dma_len;
	}

	EFX_BUG_ON_PARANOID(!len);
	buffer->len = len;
	*final_buffer = buffer;
	return 0;
}


/*
 * 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->efx->type->txd_ring_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->skb);
	EFX_BUG_ON_PARANOID(!buffer->continuation);
	EFX_BUG_ON_PARANOID(buffer->tsoh);
	buffer->len = len;
	buffer->dma_addr = tsoh->dma_addr;
	buffer->tsoh = tsoh;

	++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->efx->type->txd_ring_mask];
		efx_tsoh_free(tx_queue, buffer);
		EFX_BUG_ON_PARANOID(buffer->skb);
		buffer->len = 0;
		buffer->continuation = true;
		if (buffer->unmap_len) {
			unmap_addr = (buffer->dma_addr + buffer->len -
				      buffer->unmap_len);
			if (buffer->unmap_single)
				pci_unmap_single(tx_queue->efx->pci_dev,
						 unmap_addr, buffer->unmap_len,
						 PCI_DMA_TODEVICE);
			else
				pci_unmap_page(tx_queue->efx->pci_dev,
					       unmap_addr, buffer->unmap_len,
					       PCI_DMA_TODEVICE);
			buffer->unmap_len = 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;

	st->ipv4_id = ntohs(ip_hdr(skb)->id);
	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->packet_space = st->full_packet_size;
	st->out_len = skb->len - st->header_len;
	st->unmap_len = 0;
	st->unmap_single = false;
}

static int tso_get_fragment(struct tso_state *st, struct efx_nic *efx,
			    skb_frag_t *frag)
{
	st->unmap_addr = pci_map_page(efx->pci_dev, frag->page,
				      frag->page_offset, frag->size,
				      PCI_DMA_TODEVICE);
	if (likely(!pci_dma_mapping_error(efx->pci_dev, st->unmap_addr))) {
		st->unmap_single = false;
		st->unmap_len = frag->size;
		st->in_len = frag->size;
		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 = pci_map_single(efx->pci_dev, skb->data + hl,
					len, PCI_DMA_TODEVICE);
	if (likely(!pci_dma_mapping_error(efx->pci_dev, st->unmap_addr))) {
		st->unmap_single = true;
		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.  Return 0 on success, 1 if not enough
 * space in @tx_queue.
 */
static int 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, end_of_packet, rc;

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

	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;

	rc = efx_tx_queue_insert(tx_queue, st->dma_addr, n, &buffer);
	if (likely(rc == 0)) {
		if (st->out_len == 0)
			/* Transfer ownership of the skb */
			buffer->skb = skb;

		end_of_packet = st->out_len == 0 || st->packet_space == 0;
		buffer->continuation = !end_of_packet;

		if (st->in_len == 0) {
			/* Transfer ownership of the pci mapping */
			buffer->unmap_len = st->unmap_len;
			buffer->unmap_single = st->unmap_single;
			st->unmap_len = 0;
		}
	}

	st->dma_addr += n;
	return rc;
}


/**
 * 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 iphdr *tsoh_iph;
	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));
	tsoh_iph = (struct iphdr *)(header + SKB_IPV4_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;
	}
	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++;

	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 or %NETDEV_TX_BUSY.
 */
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, rc2 = NETDEV_TX_OK;
	struct tso_state state;

	/* Verify TSO is safe - these checks should never fail. */
	efx_tso_check_safe(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) {
		rc = tso_fill_packet_with_fragment(tx_queue, skb, &state);
		if (unlikely(rc))
			goto stop;

		/* 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;
	}

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

	tx_queue->tso_bursts++;
	return NETDEV_TX_OK;

 mem_err:
	EFX_ERR(efx, "Out of memory for TSO headers, or PCI mapping error\n");
	dev_kfree_skb_any((struct sk_buff *)skb);
	goto unwind;

 stop:
	rc2 = NETDEV_TX_BUSY;

	/* Stop the queue if it wasn't stopped before. */
	if (tx_queue->stopped == 1)
		efx_stop_queue(efx);

 unwind:
	/* Free the DMA mapping we were in the process of writing out */
	if (state.unmap_len) {
		if (state.unmap_single)
			pci_unmap_single(efx->pci_dev, state.unmap_addr,
					 state.unmap_len, PCI_DMA_TODEVICE);
		else
			pci_unmap_page(efx->pci_dev, state.unmap_addr,
				       state.unmap_len, PCI_DMA_TODEVICE);
	}

	efx_enqueue_unwind(tx_queue);
	return rc2;
}


/*
 * 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->efx->type->txd_ring_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);
}