summaryrefslogtreecommitdiff
path: root/drivers/infiniband/hw/qib/qib_user_sdma.c
blob: d0a0ea0c14d6a965afe2bcf8d359052fdb946e26 (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
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
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
/*
 * Copyright (c) 2007, 2008, 2009 QLogic Corporation. All rights reserved.
 *
 * This software is available to you under a choice of one of two
 * licenses.  You may choose to be licensed under the terms of the GNU
 * General Public License (GPL) Version 2, available from the file
 * COPYING in the main directory of this source tree, or the
 * OpenIB.org BSD license below:
 *
 *     Redistribution and use in source and binary forms, with or
 *     without modification, are permitted provided that the following
 *     conditions are met:
 *
 *      - Redistributions of source code must retain the above
 *        copyright notice, this list of conditions and the following
 *        disclaimer.
 *
 *      - Redistributions in binary form must reproduce the above
 *        copyright notice, this list of conditions and the following
 *        disclaimer in the documentation and/or other materials
 *        provided with the distribution.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
 * SOFTWARE.
 */
#include <linux/mm.h>
#include <linux/types.h>
#include <linux/device.h>
#include <linux/dmapool.h>
#include <linux/slab.h>
#include <linux/list.h>
#include <linux/highmem.h>
#include <linux/io.h>
#include <linux/uio.h>
#include <linux/rbtree.h>
#include <linux/spinlock.h>
#include <linux/delay.h>

#include "qib.h"
#include "qib_user_sdma.h"

/* minimum size of header */
#define QIB_USER_SDMA_MIN_HEADER_LENGTH 64
/* expected size of headers (for dma_pool) */
#define QIB_USER_SDMA_EXP_HEADER_LENGTH 64
/* attempt to drain the queue for 5secs */
#define QIB_USER_SDMA_DRAIN_TIMEOUT 500

struct qib_user_sdma_pkt {
	struct list_head list;  /* list element */

	u8  tiddma;		/* if this is NEW tid-sdma */
	u8  largepkt;		/* this is large pkt from kmalloc */
	u16 frag_size;		/* frag size used by PSM */
	u16 index;              /* last header index or push index */
	u16 naddr;              /* dimension of addr (1..3) ... */
	u16 addrlimit;		/* addr array size */
	u16 tidsmidx;		/* current tidsm index */
	u16 tidsmcount;		/* tidsm array item count */
	u16 payload_size;	/* payload size so far for header */
	u32 bytes_togo;		/* bytes for processing */
	u32 counter;            /* sdma pkts queued counter for this entry */
	struct qib_tid_session_member *tidsm;	/* tid session member array */
	struct qib_user_sdma_queue *pq;	/* which pq this pkt belongs to */
	u64 added;              /* global descq number of entries */

	struct {
		u16 offset;                     /* offset for kvaddr, addr */
		u16 length;                     /* length in page */
		u16 first_desc;			/* first desc */
		u16 last_desc;			/* last desc */
		u16 put_page;                   /* should we put_page? */
		u16 dma_mapped;                 /* is page dma_mapped? */
		u16 dma_length;			/* for dma_unmap_page() */
		u16 padding;
		struct page *page;              /* may be NULL (coherent mem) */
		void *kvaddr;                   /* FIXME: only for pio hack */
		dma_addr_t addr;
	} addr[4];   /* max pages, any more and we coalesce */
};

struct qib_user_sdma_queue {
	/*
	 * pkts sent to dma engine are queued on this
	 * list head.  the type of the elements of this
	 * list are struct qib_user_sdma_pkt...
	 */
	struct list_head sent;

	/*
	 * Because above list will be accessed by both process and
	 * signal handler, we need a spinlock for it.
	 */
	spinlock_t sent_lock ____cacheline_aligned_in_smp;

	/* headers with expected length are allocated from here... */
	char header_cache_name[64];
	struct dma_pool *header_cache;

	/* packets are allocated from the slab cache... */
	char pkt_slab_name[64];
	struct kmem_cache *pkt_slab;

	/* as packets go on the queued queue, they are counted... */
	u32 counter;
	u32 sent_counter;
	/* pending packets, not sending yet */
	u32 num_pending;
	/* sending packets, not complete yet */
	u32 num_sending;
	/* global descq number of entry of last sending packet */
	u64 added;

	/* dma page table */
	struct rb_root dma_pages_root;

	/* protect everything above... */
	struct mutex lock;
};

struct qib_user_sdma_queue *
qib_user_sdma_queue_create(struct device *dev, int unit, int ctxt, int sctxt)
{
	struct qib_user_sdma_queue *pq =
		kmalloc(sizeof(struct qib_user_sdma_queue), GFP_KERNEL);

	if (!pq)
		goto done;

	pq->counter = 0;
	pq->sent_counter = 0;
	pq->num_pending = 0;
	pq->num_sending = 0;
	pq->added = 0;

	INIT_LIST_HEAD(&pq->sent);
	spin_lock_init(&pq->sent_lock);
	mutex_init(&pq->lock);

	snprintf(pq->pkt_slab_name, sizeof(pq->pkt_slab_name),
		 "qib-user-sdma-pkts-%u-%02u.%02u", unit, ctxt, sctxt);
	pq->pkt_slab = kmem_cache_create(pq->pkt_slab_name,
					 sizeof(struct qib_user_sdma_pkt),
					 0, 0, NULL);

	if (!pq->pkt_slab)
		goto err_kfree;

	snprintf(pq->header_cache_name, sizeof(pq->header_cache_name),
		 "qib-user-sdma-headers-%u-%02u.%02u", unit, ctxt, sctxt);
	pq->header_cache = dma_pool_create(pq->header_cache_name,
					   dev,
					   QIB_USER_SDMA_EXP_HEADER_LENGTH,
					   4, 0);
	if (!pq->header_cache)
		goto err_slab;

	pq->dma_pages_root = RB_ROOT;

	goto done;

err_slab:
	kmem_cache_destroy(pq->pkt_slab);
err_kfree:
	kfree(pq);
	pq = NULL;

done:
	return pq;
}

static void qib_user_sdma_init_frag(struct qib_user_sdma_pkt *pkt,
				    int i, u16 offset, u16 len,
				    u16 first_desc, u16 last_desc,
				    u16 put_page, u16 dma_mapped,
				    struct page *page, void *kvaddr,
				    dma_addr_t dma_addr, u16 dma_length)
{
	pkt->addr[i].offset = offset;
	pkt->addr[i].length = len;
	pkt->addr[i].first_desc = first_desc;
	pkt->addr[i].last_desc = last_desc;
	pkt->addr[i].put_page = put_page;
	pkt->addr[i].dma_mapped = dma_mapped;
	pkt->addr[i].page = page;
	pkt->addr[i].kvaddr = kvaddr;
	pkt->addr[i].addr = dma_addr;
	pkt->addr[i].dma_length = dma_length;
}

static void *qib_user_sdma_alloc_header(struct qib_user_sdma_queue *pq,
				size_t len, dma_addr_t *dma_addr)
{
	void *hdr;

	if (len == QIB_USER_SDMA_EXP_HEADER_LENGTH)
		hdr = dma_pool_alloc(pq->header_cache, GFP_KERNEL,
					     dma_addr);
	else
		hdr = NULL;

	if (!hdr) {
		hdr = kmalloc(len, GFP_KERNEL);
		if (!hdr)
			return NULL;

		*dma_addr = 0;
	}

	return hdr;
}

static int qib_user_sdma_page_to_frags(const struct qib_devdata *dd,
				       struct qib_user_sdma_queue *pq,
				       struct qib_user_sdma_pkt *pkt,
				       struct page *page, u16 put,
				       u16 offset, u16 len, void *kvaddr)
{
	__le16 *pbc16;
	void *pbcvaddr;
	struct qib_message_header *hdr;
	u16 newlen, pbclen, lastdesc, dma_mapped;
	u32 vcto;
	union qib_seqnum seqnum;
	dma_addr_t pbcdaddr;
	dma_addr_t dma_addr =
		dma_map_page(&dd->pcidev->dev,
			page, offset, len, DMA_TO_DEVICE);
	int ret = 0;

	if (dma_mapping_error(&dd->pcidev->dev, dma_addr)) {
		/*
		 * dma mapping error, pkt has not managed
		 * this page yet, return the page here so
		 * the caller can ignore this page.
		 */
		if (put) {
			put_page(page);
		} else {
			/* coalesce case */
			kunmap(page);
			__free_page(page);
		}
		ret = -ENOMEM;
		goto done;
	}
	offset = 0;
	dma_mapped = 1;


next_fragment:

	/*
	 * In tid-sdma, the transfer length is restricted by
	 * receiver side current tid page length.
	 */
	if (pkt->tiddma && len > pkt->tidsm[pkt->tidsmidx].length)
		newlen = pkt->tidsm[pkt->tidsmidx].length;
	else
		newlen = len;

	/*
	 * Then the transfer length is restricted by MTU.
	 * the last descriptor flag is determined by:
	 * 1. the current packet is at frag size length.
	 * 2. the current tid page is done if tid-sdma.
	 * 3. there is no more byte togo if sdma.
	 */
	lastdesc = 0;
	if ((pkt->payload_size + newlen) >= pkt->frag_size) {
		newlen = pkt->frag_size - pkt->payload_size;
		lastdesc = 1;
	} else if (pkt->tiddma) {
		if (newlen == pkt->tidsm[pkt->tidsmidx].length)
			lastdesc = 1;
	} else {
		if (newlen == pkt->bytes_togo)
			lastdesc = 1;
	}

	/* fill the next fragment in this page */
	qib_user_sdma_init_frag(pkt, pkt->naddr, /* index */
		offset, newlen,		/* offset, len */
		0, lastdesc,		/* first last desc */
		put, dma_mapped,	/* put page, dma mapped */
		page, kvaddr,		/* struct page, virt addr */
		dma_addr, len);		/* dma addr, dma length */
	pkt->bytes_togo -= newlen;
	pkt->payload_size += newlen;
	pkt->naddr++;
	if (pkt->naddr == pkt->addrlimit) {
		ret = -EFAULT;
		goto done;
	}

	/* If there is no more byte togo. (lastdesc==1) */
	if (pkt->bytes_togo == 0) {
		/* The packet is done, header is not dma mapped yet.
		 * it should be from kmalloc */
		if (!pkt->addr[pkt->index].addr) {
			pkt->addr[pkt->index].addr =
				dma_map_single(&dd->pcidev->dev,
					pkt->addr[pkt->index].kvaddr,
					pkt->addr[pkt->index].dma_length,
					DMA_TO_DEVICE);
			if (dma_mapping_error(&dd->pcidev->dev,
					pkt->addr[pkt->index].addr)) {
				ret = -ENOMEM;
				goto done;
			}
			pkt->addr[pkt->index].dma_mapped = 1;
		}

		goto done;
	}

	/* If tid-sdma, advance tid info. */
	if (pkt->tiddma) {
		pkt->tidsm[pkt->tidsmidx].length -= newlen;
		if (pkt->tidsm[pkt->tidsmidx].length) {
			pkt->tidsm[pkt->tidsmidx].offset += newlen;
		} else {
			pkt->tidsmidx++;
			if (pkt->tidsmidx == pkt->tidsmcount) {
				ret = -EFAULT;
				goto done;
			}
		}
	}

	/*
	 * If this is NOT the last descriptor. (newlen==len)
	 * the current packet is not done yet, but the current
	 * send side page is done.
	 */
	if (lastdesc == 0)
		goto done;

	/*
	 * If running this driver under PSM with message size
	 * fitting into one transfer unit, it is not possible
	 * to pass this line. otherwise, it is a buggggg.
	 */

	/*
	 * Since the current packet is done, and there are more
	 * bytes togo, we need to create a new sdma header, copying
	 * from previous sdma header and modify both.
	 */
	pbclen = pkt->addr[pkt->index].length;
	pbcvaddr = qib_user_sdma_alloc_header(pq, pbclen, &pbcdaddr);
	if (!pbcvaddr) {
		ret = -ENOMEM;
		goto done;
	}
	/* Copy the previous sdma header to new sdma header */
	pbc16 = (__le16 *)pkt->addr[pkt->index].kvaddr;
	memcpy(pbcvaddr, pbc16, pbclen);

	/* Modify the previous sdma header */
	hdr = (struct qib_message_header *)&pbc16[4];

	/* New pbc length */
	pbc16[0] = cpu_to_le16(le16_to_cpu(pbc16[0])-(pkt->bytes_togo>>2));

	/* New packet length */
	hdr->lrh[2] = cpu_to_be16(le16_to_cpu(pbc16[0]));

	if (pkt->tiddma) {
		/* turn on the header suppression */
		hdr->iph.pkt_flags =
			cpu_to_le16(le16_to_cpu(hdr->iph.pkt_flags)|0x2);
		/* turn off ACK_REQ: 0x04 and EXPECTED_DONE: 0x20 */
		hdr->flags &= ~(0x04|0x20);
	} else {
		/* turn off extra bytes: 20-21 bits */
		hdr->bth[0] = cpu_to_be32(be32_to_cpu(hdr->bth[0])&0xFFCFFFFF);
		/* turn off ACK_REQ: 0x04 */
		hdr->flags &= ~(0x04);
	}

	/* New kdeth checksum */
	vcto = le32_to_cpu(hdr->iph.ver_ctxt_tid_offset);
	hdr->iph.chksum = cpu_to_le16(QIB_LRH_BTH +
		be16_to_cpu(hdr->lrh[2]) -
		((vcto>>16)&0xFFFF) - (vcto&0xFFFF) -
		le16_to_cpu(hdr->iph.pkt_flags));

	/* The packet is done, header is not dma mapped yet.
	 * it should be from kmalloc */
	if (!pkt->addr[pkt->index].addr) {
		pkt->addr[pkt->index].addr =
			dma_map_single(&dd->pcidev->dev,
				pkt->addr[pkt->index].kvaddr,
				pkt->addr[pkt->index].dma_length,
				DMA_TO_DEVICE);
		if (dma_mapping_error(&dd->pcidev->dev,
				pkt->addr[pkt->index].addr)) {
			ret = -ENOMEM;
			goto done;
		}
		pkt->addr[pkt->index].dma_mapped = 1;
	}

	/* Modify the new sdma header */
	pbc16 = (__le16 *)pbcvaddr;
	hdr = (struct qib_message_header *)&pbc16[4];

	/* New pbc length */
	pbc16[0] = cpu_to_le16(le16_to_cpu(pbc16[0])-(pkt->payload_size>>2));

	/* New packet length */
	hdr->lrh[2] = cpu_to_be16(le16_to_cpu(pbc16[0]));

	if (pkt->tiddma) {
		/* Set new tid and offset for new sdma header */
		hdr->iph.ver_ctxt_tid_offset = cpu_to_le32(
			(le32_to_cpu(hdr->iph.ver_ctxt_tid_offset)&0xFF000000) +
			(pkt->tidsm[pkt->tidsmidx].tid<<QLOGIC_IB_I_TID_SHIFT) +
			(pkt->tidsm[pkt->tidsmidx].offset>>2));
	} else {
		/* Middle protocol new packet offset */
		hdr->uwords[2] += pkt->payload_size;
	}

	/* New kdeth checksum */
	vcto = le32_to_cpu(hdr->iph.ver_ctxt_tid_offset);
	hdr->iph.chksum = cpu_to_le16(QIB_LRH_BTH +
		be16_to_cpu(hdr->lrh[2]) -
		((vcto>>16)&0xFFFF) - (vcto&0xFFFF) -
		le16_to_cpu(hdr->iph.pkt_flags));

	/* Next sequence number in new sdma header */
	seqnum.val = be32_to_cpu(hdr->bth[2]);
	if (pkt->tiddma)
		seqnum.seq++;
	else
		seqnum.pkt++;
	hdr->bth[2] = cpu_to_be32(seqnum.val);

	/* Init new sdma header. */
	qib_user_sdma_init_frag(pkt, pkt->naddr, /* index */
		0, pbclen,		/* offset, len */
		1, 0,			/* first last desc */
		0, 0,			/* put page, dma mapped */
		NULL, pbcvaddr,		/* struct page, virt addr */
		pbcdaddr, pbclen);	/* dma addr, dma length */
	pkt->index = pkt->naddr;
	pkt->payload_size = 0;
	pkt->naddr++;
	if (pkt->naddr == pkt->addrlimit) {
		ret = -EFAULT;
		goto done;
	}

	/* Prepare for next fragment in this page */
	if (newlen != len) {
		if (dma_mapped) {
			put = 0;
			dma_mapped = 0;
			page = NULL;
			kvaddr = NULL;
		}
		len -= newlen;
		offset += newlen;

		goto next_fragment;
	}

done:
	return ret;
}

/* we've too many pages in the iovec, coalesce to a single page */
static int qib_user_sdma_coalesce(const struct qib_devdata *dd,
				  struct qib_user_sdma_queue *pq,
				  struct qib_user_sdma_pkt *pkt,
				  const struct iovec *iov,
				  unsigned long niov)
{
	int ret = 0;
	struct page *page = alloc_page(GFP_KERNEL);
	void *mpage_save;
	char *mpage;
	int i;
	int len = 0;

	if (!page) {
		ret = -ENOMEM;
		goto done;
	}

	mpage = kmap(page);
	mpage_save = mpage;
	for (i = 0; i < niov; i++) {
		int cfur;

		cfur = copy_from_user(mpage,
				      iov[i].iov_base, iov[i].iov_len);
		if (cfur) {
			ret = -EFAULT;
			goto free_unmap;
		}

		mpage += iov[i].iov_len;
		len += iov[i].iov_len;
	}

	ret = qib_user_sdma_page_to_frags(dd, pq, pkt,
			page, 0, 0, len, mpage_save);
	goto done;

free_unmap:
	kunmap(page);
	__free_page(page);
done:
	return ret;
}

/*
 * How many pages in this iovec element?
 */
static int qib_user_sdma_num_pages(const struct iovec *iov)
{
	const unsigned long addr  = (unsigned long) iov->iov_base;
	const unsigned long  len  = iov->iov_len;
	const unsigned long spage = addr & PAGE_MASK;
	const unsigned long epage = (addr + len - 1) & PAGE_MASK;

	return 1 + ((epage - spage) >> PAGE_SHIFT);
}

static void qib_user_sdma_free_pkt_frag(struct device *dev,
					struct qib_user_sdma_queue *pq,
					struct qib_user_sdma_pkt *pkt,
					int frag)
{
	const int i = frag;

	if (pkt->addr[i].page) {
		/* only user data has page */
		if (pkt->addr[i].dma_mapped)
			dma_unmap_page(dev,
				       pkt->addr[i].addr,
				       pkt->addr[i].dma_length,
				       DMA_TO_DEVICE);

		if (pkt->addr[i].kvaddr)
			kunmap(pkt->addr[i].page);

		if (pkt->addr[i].put_page)
			put_page(pkt->addr[i].page);
		else
			__free_page(pkt->addr[i].page);
	} else if (pkt->addr[i].kvaddr) {
		/* for headers */
		if (pkt->addr[i].dma_mapped) {
			/* from kmalloc & dma mapped */
			dma_unmap_single(dev,
				       pkt->addr[i].addr,
				       pkt->addr[i].dma_length,
				       DMA_TO_DEVICE);
			kfree(pkt->addr[i].kvaddr);
		} else if (pkt->addr[i].addr) {
			/* free coherent mem from cache... */
			dma_pool_free(pq->header_cache,
			      pkt->addr[i].kvaddr, pkt->addr[i].addr);
		} else {
			/* from kmalloc but not dma mapped */
			kfree(pkt->addr[i].kvaddr);
		}
	}
}

/* return number of pages pinned... */
static int qib_user_sdma_pin_pages(const struct qib_devdata *dd,
				   struct qib_user_sdma_queue *pq,
				   struct qib_user_sdma_pkt *pkt,
				   unsigned long addr, int tlen, int npages)
{
	struct page *pages[8];
	int i, j;
	int ret = 0;

	while (npages) {
		if (npages > 8)
			j = 8;
		else
			j = npages;

		ret = get_user_pages(current, current->mm, addr,
			     j, 0, 1, pages, NULL);
		if (ret != j) {
			i = 0;
			j = ret;
			ret = -ENOMEM;
			goto free_pages;
		}

		for (i = 0; i < j; i++) {
			/* map the pages... */
			unsigned long fofs = addr & ~PAGE_MASK;
			int flen = ((fofs + tlen) > PAGE_SIZE) ?
				(PAGE_SIZE - fofs) : tlen;

			ret = qib_user_sdma_page_to_frags(dd, pq, pkt,
				pages[i], 1, fofs, flen, NULL);
			if (ret < 0) {
				/* current page has beed taken
				 * care of inside above call.
				 */
				i++;
				goto free_pages;
			}

			addr += flen;
			tlen -= flen;
		}

		npages -= j;
	}

	goto done;

	/* if error, return all pages not managed by pkt */
free_pages:
	while (i < j)
		put_page(pages[i++]);

done:
	return ret;
}

static int qib_user_sdma_pin_pkt(const struct qib_devdata *dd,
				 struct qib_user_sdma_queue *pq,
				 struct qib_user_sdma_pkt *pkt,
				 const struct iovec *iov,
				 unsigned long niov)
{
	int ret = 0;
	unsigned long idx;

	for (idx = 0; idx < niov; idx++) {
		const int npages = qib_user_sdma_num_pages(iov + idx);
		const unsigned long addr = (unsigned long) iov[idx].iov_base;

		ret = qib_user_sdma_pin_pages(dd, pq, pkt, addr,
					      iov[idx].iov_len, npages);
		if (ret < 0)
			goto free_pkt;
	}

	goto done;

free_pkt:
	/* we need to ignore the first entry here */
	for (idx = 1; idx < pkt->naddr; idx++)
		qib_user_sdma_free_pkt_frag(&dd->pcidev->dev, pq, pkt, idx);

	/* need to dma unmap the first entry, this is to restore to
	 * the original state so that caller can free the memory in
	 * error condition. Caller does not know if dma mapped or not*/
	if (pkt->addr[0].dma_mapped) {
		dma_unmap_single(&dd->pcidev->dev,
		       pkt->addr[0].addr,
		       pkt->addr[0].dma_length,
		       DMA_TO_DEVICE);
		pkt->addr[0].addr = 0;
		pkt->addr[0].dma_mapped = 0;
	}

done:
	return ret;
}

static int qib_user_sdma_init_payload(const struct qib_devdata *dd,
				      struct qib_user_sdma_queue *pq,
				      struct qib_user_sdma_pkt *pkt,
				      const struct iovec *iov,
				      unsigned long niov, int npages)
{
	int ret = 0;

	if (pkt->frag_size == pkt->bytes_togo &&
			npages >= ARRAY_SIZE(pkt->addr))
		ret = qib_user_sdma_coalesce(dd, pq, pkt, iov, niov);
	else
		ret = qib_user_sdma_pin_pkt(dd, pq, pkt, iov, niov);

	return ret;
}

/* free a packet list -- return counter value of last packet */
static void qib_user_sdma_free_pkt_list(struct device *dev,
					struct qib_user_sdma_queue *pq,
					struct list_head *list)
{
	struct qib_user_sdma_pkt *pkt, *pkt_next;

	list_for_each_entry_safe(pkt, pkt_next, list, list) {
		int i;

		for (i = 0; i < pkt->naddr; i++)
			qib_user_sdma_free_pkt_frag(dev, pq, pkt, i);

		if (pkt->largepkt)
			kfree(pkt);
		else
			kmem_cache_free(pq->pkt_slab, pkt);
	}
	INIT_LIST_HEAD(list);
}

/*
 * copy headers, coalesce etc -- pq->lock must be held
 *
 * we queue all the packets to list, returning the
 * number of bytes total.  list must be empty initially,
 * as, if there is an error we clean it...
 */
static int qib_user_sdma_queue_pkts(const struct qib_devdata *dd,
				    struct qib_pportdata *ppd,
				    struct qib_user_sdma_queue *pq,
				    const struct iovec *iov,
				    unsigned long niov,
				    struct list_head *list,
				    int *maxpkts, int *ndesc)
{
	unsigned long idx = 0;
	int ret = 0;
	int npkts = 0;
	__le32 *pbc;
	dma_addr_t dma_addr;
	struct qib_user_sdma_pkt *pkt = NULL;
	size_t len;
	size_t nw;
	u32 counter = pq->counter;
	u16 frag_size;

	while (idx < niov && npkts < *maxpkts) {
		const unsigned long addr = (unsigned long) iov[idx].iov_base;
		const unsigned long idx_save = idx;
		unsigned pktnw;
		unsigned pktnwc;
		int nfrags = 0;
		int npages = 0;
		int bytes_togo = 0;
		int tiddma = 0;
		int cfur;

		len = iov[idx].iov_len;
		nw = len >> 2;

		if (len < QIB_USER_SDMA_MIN_HEADER_LENGTH ||
		    len > PAGE_SIZE || len & 3 || addr & 3) {
			ret = -EINVAL;
			goto free_list;
		}

		pbc = qib_user_sdma_alloc_header(pq, len, &dma_addr);
		if (!pbc) {
			ret = -ENOMEM;
			goto free_list;
		}

		cfur = copy_from_user(pbc, iov[idx].iov_base, len);
		if (cfur) {
			ret = -EFAULT;
			goto free_pbc;
		}

		/*
		 * This assignment is a bit strange.  it's because the
		 * the pbc counts the number of 32 bit words in the full
		 * packet _except_ the first word of the pbc itself...
		 */
		pktnwc = nw - 1;

		/*
		 * pktnw computation yields the number of 32 bit words
		 * that the caller has indicated in the PBC.  note that
		 * this is one less than the total number of words that
		 * goes to the send DMA engine as the first 32 bit word
		 * of the PBC itself is not counted.  Armed with this count,
		 * we can verify that the packet is consistent with the
		 * iovec lengths.
		 */
		pktnw = le32_to_cpu(*pbc) & 0xFFFF;
		if (pktnw < pktnwc) {
			ret = -EINVAL;
			goto free_pbc;
		}

		idx++;
		while (pktnwc < pktnw && idx < niov) {
			const size_t slen = iov[idx].iov_len;
			const unsigned long faddr =
				(unsigned long) iov[idx].iov_base;

			if (slen & 3 || faddr & 3 || !slen) {
				ret = -EINVAL;
				goto free_pbc;
			}

			npages += qib_user_sdma_num_pages(&iov[idx]);

			bytes_togo += slen;
			pktnwc += slen >> 2;
			idx++;
			nfrags++;
		}

		if (pktnwc != pktnw) {
			ret = -EINVAL;
			goto free_pbc;
		}

		frag_size = ((le32_to_cpu(*pbc))>>16) & 0xFFFF;
		if (((frag_size ? frag_size : bytes_togo) + len) >
						ppd->ibmaxlen) {
			ret = -EINVAL;
			goto free_pbc;
		}

		if (frag_size) {
			int pktsize, tidsmsize, n;

			n = npages*((2*PAGE_SIZE/frag_size)+1);
			pktsize = sizeof(*pkt) + sizeof(pkt->addr[0])*n;

			/*
			 * Determine if this is tid-sdma or just sdma.
			 */
			tiddma = (((le32_to_cpu(pbc[7])>>
				QLOGIC_IB_I_TID_SHIFT)&
				QLOGIC_IB_I_TID_MASK) !=
				QLOGIC_IB_I_TID_MASK);

			if (tiddma)
				tidsmsize = iov[idx].iov_len;
			else
				tidsmsize = 0;

			pkt = kmalloc(pktsize+tidsmsize, GFP_KERNEL);
			if (!pkt) {
				ret = -ENOMEM;
				goto free_pbc;
			}
			pkt->largepkt = 1;
			pkt->frag_size = frag_size;
			pkt->addrlimit = n + ARRAY_SIZE(pkt->addr);

			if (tiddma) {
				char *tidsm = (char *)pkt + pktsize;
				cfur = copy_from_user(tidsm,
					iov[idx].iov_base, tidsmsize);
				if (cfur) {
					ret = -EFAULT;
					goto free_pkt;
				}
				pkt->tidsm =
					(struct qib_tid_session_member *)tidsm;
				pkt->tidsmcount = tidsmsize/
					sizeof(struct qib_tid_session_member);
				pkt->tidsmidx = 0;
				idx++;
			}

			/*
			 * pbc 'fill1' field is borrowed to pass frag size,
			 * we need to clear it after picking frag size, the
			 * hardware requires this field to be zero.
			 */
			*pbc = cpu_to_le32(le32_to_cpu(*pbc) & 0x0000FFFF);
		} else {
			pkt = kmem_cache_alloc(pq->pkt_slab, GFP_KERNEL);
			if (!pkt) {
				ret = -ENOMEM;
				goto free_pbc;
			}
			pkt->largepkt = 0;
			pkt->frag_size = bytes_togo;
			pkt->addrlimit = ARRAY_SIZE(pkt->addr);
		}
		pkt->bytes_togo = bytes_togo;
		pkt->payload_size = 0;
		pkt->counter = counter;
		pkt->tiddma = tiddma;

		/* setup the first header */
		qib_user_sdma_init_frag(pkt, 0, /* index */
			0, len,		/* offset, len */
			1, 0,		/* first last desc */
			0, 0,		/* put page, dma mapped */
			NULL, pbc,	/* struct page, virt addr */
			dma_addr, len);	/* dma addr, dma length */
		pkt->index = 0;
		pkt->naddr = 1;

		if (nfrags) {
			ret = qib_user_sdma_init_payload(dd, pq, pkt,
							 iov + idx_save + 1,
							 nfrags, npages);
			if (ret < 0)
				goto free_pkt;
		} else {
			/* since there is no payload, mark the
			 * header as the last desc. */
			pkt->addr[0].last_desc = 1;

			if (dma_addr == 0) {
				/*
				 * the header is not dma mapped yet.
				 * it should be from kmalloc.
				 */
				dma_addr = dma_map_single(&dd->pcidev->dev,
					pbc, len, DMA_TO_DEVICE);
				if (dma_mapping_error(&dd->pcidev->dev,
								dma_addr)) {
					ret = -ENOMEM;
					goto free_pkt;
				}
				pkt->addr[0].addr = dma_addr;
				pkt->addr[0].dma_mapped = 1;
			}
		}

		counter++;
		npkts++;
		pkt->pq = pq;
		pkt->index = 0; /* reset index for push on hw */
		*ndesc += pkt->naddr;

		list_add_tail(&pkt->list, list);
	}

	*maxpkts = npkts;
	ret = idx;
	goto done;

free_pkt:
	if (pkt->largepkt)
		kfree(pkt);
	else
		kmem_cache_free(pq->pkt_slab, pkt);
free_pbc:
	if (dma_addr)
		dma_pool_free(pq->header_cache, pbc, dma_addr);
	else
		kfree(pbc);
free_list:
	qib_user_sdma_free_pkt_list(&dd->pcidev->dev, pq, list);
done:
	return ret;
}

static void qib_user_sdma_set_complete_counter(struct qib_user_sdma_queue *pq,
					       u32 c)
{
	pq->sent_counter = c;
}

/* try to clean out queue -- needs pq->lock */
static int qib_user_sdma_queue_clean(struct qib_pportdata *ppd,
				     struct qib_user_sdma_queue *pq)
{
	struct qib_devdata *dd = ppd->dd;
	struct list_head free_list;
	struct qib_user_sdma_pkt *pkt;
	struct qib_user_sdma_pkt *pkt_prev;
	unsigned long flags;
	int ret = 0;

	if (!pq->num_sending)
		return 0;

	INIT_LIST_HEAD(&free_list);

	/*
	 * We need this spin lock here because interrupt handler
	 * might modify this list in qib_user_sdma_send_desc(), also
	 * we can not get interrupted, otherwise it is a deadlock.
	 */
	spin_lock_irqsave(&pq->sent_lock, flags);
	list_for_each_entry_safe(pkt, pkt_prev, &pq->sent, list) {
		s64 descd = ppd->sdma_descq_removed - pkt->added;

		if (descd < 0)
			break;

		list_move_tail(&pkt->list, &free_list);

		/* one more packet cleaned */
		ret++;
		pq->num_sending--;
	}
	spin_unlock_irqrestore(&pq->sent_lock, flags);

	if (!list_empty(&free_list)) {
		u32 counter;

		pkt = list_entry(free_list.prev,
				 struct qib_user_sdma_pkt, list);
		counter = pkt->counter;

		qib_user_sdma_free_pkt_list(&dd->pcidev->dev, pq, &free_list);
		qib_user_sdma_set_complete_counter(pq, counter);
	}

	return ret;
}

void qib_user_sdma_queue_destroy(struct qib_user_sdma_queue *pq)
{
	if (!pq)
		return;

	kmem_cache_destroy(pq->pkt_slab);
	dma_pool_destroy(pq->header_cache);
	kfree(pq);
}

/* clean descriptor queue, returns > 0 if some elements cleaned */
static int qib_user_sdma_hwqueue_clean(struct qib_pportdata *ppd)
{
	int ret;
	unsigned long flags;

	spin_lock_irqsave(&ppd->sdma_lock, flags);
	ret = qib_sdma_make_progress(ppd);
	spin_unlock_irqrestore(&ppd->sdma_lock, flags);

	return ret;
}

/* we're in close, drain packets so that we can cleanup successfully... */
void qib_user_sdma_queue_drain(struct qib_pportdata *ppd,
			       struct qib_user_sdma_queue *pq)
{
	struct qib_devdata *dd = ppd->dd;
	unsigned long flags;
	int i;

	if (!pq)
		return;

	for (i = 0; i < QIB_USER_SDMA_DRAIN_TIMEOUT; i++) {
		mutex_lock(&pq->lock);
		if (!pq->num_pending && !pq->num_sending) {
			mutex_unlock(&pq->lock);
			break;
		}
		qib_user_sdma_hwqueue_clean(ppd);
		qib_user_sdma_queue_clean(ppd, pq);
		mutex_unlock(&pq->lock);
		msleep(10);
	}

	if (pq->num_pending || pq->num_sending) {
		struct qib_user_sdma_pkt *pkt;
		struct qib_user_sdma_pkt *pkt_prev;
		struct list_head free_list;

		mutex_lock(&pq->lock);
		spin_lock_irqsave(&ppd->sdma_lock, flags);
		/*
		 * Since we hold sdma_lock, it is safe without sent_lock.
		 */
		if (pq->num_pending) {
			list_for_each_entry_safe(pkt, pkt_prev,
					&ppd->sdma_userpending, list) {
				if (pkt->pq == pq) {
					list_move_tail(&pkt->list, &pq->sent);
					pq->num_pending--;
					pq->num_sending++;
				}
			}
		}
		spin_unlock_irqrestore(&ppd->sdma_lock, flags);

		qib_dev_err(dd, "user sdma lists not empty: forcing!\n");
		INIT_LIST_HEAD(&free_list);
		list_splice_init(&pq->sent, &free_list);
		pq->num_sending = 0;
		qib_user_sdma_free_pkt_list(&dd->pcidev->dev, pq, &free_list);
		mutex_unlock(&pq->lock);
	}
}

static inline __le64 qib_sdma_make_desc0(u8 gen,
					 u64 addr, u64 dwlen, u64 dwoffset)
{
	return cpu_to_le64(/* SDmaPhyAddr[31:0] */
			   ((addr & 0xfffffffcULL) << 32) |
			   /* SDmaGeneration[1:0] */
			   ((gen & 3ULL) << 30) |
			   /* SDmaDwordCount[10:0] */
			   ((dwlen & 0x7ffULL) << 16) |
			   /* SDmaBufOffset[12:2] */
			   (dwoffset & 0x7ffULL));
}

static inline __le64 qib_sdma_make_first_desc0(__le64 descq)
{
	return descq | cpu_to_le64(1ULL << 12);
}

static inline __le64 qib_sdma_make_last_desc0(__le64 descq)
{
					      /* last */  /* dma head */
	return descq | cpu_to_le64(1ULL << 11 | 1ULL << 13);
}

static inline __le64 qib_sdma_make_desc1(u64 addr)
{
	/* SDmaPhyAddr[47:32] */
	return cpu_to_le64(addr >> 32);
}

static void qib_user_sdma_send_frag(struct qib_pportdata *ppd,
				    struct qib_user_sdma_pkt *pkt, int idx,
				    unsigned ofs, u16 tail, u8 gen)
{
	const u64 addr = (u64) pkt->addr[idx].addr +
		(u64) pkt->addr[idx].offset;
	const u64 dwlen = (u64) pkt->addr[idx].length / 4;
	__le64 *descqp;
	__le64 descq0;

	descqp = &ppd->sdma_descq[tail].qw[0];

	descq0 = qib_sdma_make_desc0(gen, addr, dwlen, ofs);
	if (pkt->addr[idx].first_desc)
		descq0 = qib_sdma_make_first_desc0(descq0);
	if (pkt->addr[idx].last_desc) {
		descq0 = qib_sdma_make_last_desc0(descq0);
		if (ppd->sdma_intrequest) {
			descq0 |= cpu_to_le64(1ULL << 15);
			ppd->sdma_intrequest = 0;
		}
	}

	descqp[0] = descq0;
	descqp[1] = qib_sdma_make_desc1(addr);
}

void qib_user_sdma_send_desc(struct qib_pportdata *ppd,
				struct list_head *pktlist)
{
	struct qib_devdata *dd = ppd->dd;
	u16 nfree, nsent;
	u16 tail, tail_c;
	u8 gen, gen_c;

	nfree = qib_sdma_descq_freecnt(ppd);
	if (!nfree)
		return;

retry:
	nsent = 0;
	tail_c = tail = ppd->sdma_descq_tail;
	gen_c = gen = ppd->sdma_generation;
	while (!list_empty(pktlist)) {
		struct qib_user_sdma_pkt *pkt =
			list_entry(pktlist->next, struct qib_user_sdma_pkt,
				   list);
		int i, j, c = 0;
		unsigned ofs = 0;
		u16 dtail = tail;

		for (i = pkt->index; i < pkt->naddr && nfree; i++) {
			qib_user_sdma_send_frag(ppd, pkt, i, ofs, tail, gen);
			ofs += pkt->addr[i].length >> 2;

			if (++tail == ppd->sdma_descq_cnt) {
				tail = 0;
				++gen;
				ppd->sdma_intrequest = 1;
			} else if (tail == (ppd->sdma_descq_cnt>>1)) {
				ppd->sdma_intrequest = 1;
			}
			nfree--;
			if (pkt->addr[i].last_desc == 0)
				continue;

			/*
			 * If the packet is >= 2KB mtu equivalent, we
			 * have to use the large buffers, and have to
			 * mark each descriptor as part of a large
			 * buffer packet.
			 */
			if (ofs > dd->piosize2kmax_dwords) {
				for (j = pkt->index; j <= i; j++) {
					ppd->sdma_descq[dtail].qw[0] |=
						cpu_to_le64(1ULL << 14);
					if (++dtail == ppd->sdma_descq_cnt)
						dtail = 0;
				}
			}
			c += i + 1 - pkt->index;
			pkt->index = i + 1; /* index for next first */
			tail_c = dtail = tail;
			gen_c = gen;
			ofs = 0;  /* reset for next packet */
		}

		ppd->sdma_descq_added += c;
		nsent += c;
		if (pkt->index == pkt->naddr) {
			pkt->added = ppd->sdma_descq_added;
			pkt->pq->added = pkt->added;
			pkt->pq->num_pending--;
			spin_lock(&pkt->pq->sent_lock);
			pkt->pq->num_sending++;
			list_move_tail(&pkt->list, &pkt->pq->sent);
			spin_unlock(&pkt->pq->sent_lock);
		}
		if (!nfree || (nsent<<2) > ppd->sdma_descq_cnt)
			break;
	}

	/* advance the tail on the chip if necessary */
	if (ppd->sdma_descq_tail != tail_c) {
		ppd->sdma_generation = gen_c;
		dd->f_sdma_update_tail(ppd, tail_c);
	}

	if (nfree && !list_empty(pktlist))
		goto retry;

	return;
}

/* pq->lock must be held, get packets on the wire... */
static int qib_user_sdma_push_pkts(struct qib_pportdata *ppd,
				 struct qib_user_sdma_queue *pq,
				 struct list_head *pktlist, int count)
{
	int ret = 0;
	unsigned long flags;

	if (unlikely(!(ppd->lflags & QIBL_LINKACTIVE)))
		return -ECOMM;

	spin_lock_irqsave(&ppd->sdma_lock, flags);

	if (unlikely(!__qib_sdma_running(ppd))) {
		ret = -ECOMM;
		goto unlock;
	}

	pq->num_pending += count;
	list_splice_tail_init(pktlist, &ppd->sdma_userpending);
	qib_user_sdma_send_desc(ppd, &ppd->sdma_userpending);

unlock:
	spin_unlock_irqrestore(&ppd->sdma_lock, flags);
	return ret;
}

int qib_user_sdma_writev(struct qib_ctxtdata *rcd,
			 struct qib_user_sdma_queue *pq,
			 const struct iovec *iov,
			 unsigned long dim)
{
	struct qib_devdata *dd = rcd->dd;
	struct qib_pportdata *ppd = rcd->ppd;
	int ret = 0;
	struct list_head list;
	int npkts = 0;

	INIT_LIST_HEAD(&list);

	mutex_lock(&pq->lock);

	/* why not -ECOMM like qib_user_sdma_push_pkts() below? */
	if (!qib_sdma_running(ppd))
		goto done_unlock;

	/* if I have packets not complete yet */
	if (pq->added > ppd->sdma_descq_removed)
		qib_user_sdma_hwqueue_clean(ppd);
	/* if I have complete packets to be freed */
	if (pq->num_sending)
		qib_user_sdma_queue_clean(ppd, pq);

	while (dim) {
		int mxp = 8;
		int ndesc = 0;

		down_write(&current->mm->mmap_sem);
		ret = qib_user_sdma_queue_pkts(dd, ppd, pq,
				iov, dim, &list, &mxp, &ndesc);
		up_write(&current->mm->mmap_sem);

		if (ret < 0)
			goto done_unlock;
		else {
			dim -= ret;
			iov += ret;
		}

		/* force packets onto the sdma hw queue... */
		if (!list_empty(&list)) {
			/*
			 * Lazily clean hw queue.
			 */
			if (qib_sdma_descq_freecnt(ppd) < ndesc) {
				qib_user_sdma_hwqueue_clean(ppd);
				if (pq->num_sending)
					qib_user_sdma_queue_clean(ppd, pq);
			}

			ret = qib_user_sdma_push_pkts(ppd, pq, &list, mxp);
			if (ret < 0)
				goto done_unlock;
			else {
				npkts += mxp;
				pq->counter += mxp;
			}
		}
	}

done_unlock:
	if (!list_empty(&list))
		qib_user_sdma_free_pkt_list(&dd->pcidev->dev, pq, &list);
	mutex_unlock(&pq->lock);

	return (ret < 0) ? ret : npkts;
}

int qib_user_sdma_make_progress(struct qib_pportdata *ppd,
				struct qib_user_sdma_queue *pq)
{
	int ret = 0;

	mutex_lock(&pq->lock);
	qib_user_sdma_hwqueue_clean(ppd);
	ret = qib_user_sdma_queue_clean(ppd, pq);
	mutex_unlock(&pq->lock);

	return ret;
}

u32 qib_user_sdma_complete_counter(const struct qib_user_sdma_queue *pq)
{
	return pq ? pq->sent_counter : 0;
}

u32 qib_user_sdma_inflight_counter(struct qib_user_sdma_queue *pq)
{
	return pq ? pq->counter : 0;
}