summaryrefslogtreecommitdiff
path: root/drivers/infiniband/hw/hfi1/user_sdma.c
blob: 47ffd273ecbd7745d25067f4f8268301d89dde25 (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
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
/*
 * Copyright(c) 2015, 2016 Intel Corporation.
 *
 * This file is provided under a dual BSD/GPLv2 license.  When using or
 * redistributing this file, you may do so under either license.
 *
 * GPL LICENSE SUMMARY
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of version 2 of the GNU General Public License as
 * published by the Free Software Foundation.
 *
 * This program is distributed in the hope that it will be useful, but
 * WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * General Public License for more details.
 *
 * BSD LICENSE
 *
 * 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.
 *  - Neither the name of Intel Corporation nor the names of its
 *    contributors may be used to endorse or promote products derived
 *    from this software without specific prior written permission.
 *
 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 *
 */
#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 <linux/kthread.h>
#include <linux/mmu_context.h>
#include <linux/module.h>
#include <linux/vmalloc.h>

#include "hfi.h"
#include "sdma.h"
#include "user_sdma.h"
#include "verbs.h"  /* for the headers */
#include "common.h" /* for struct hfi1_tid_info */
#include "trace.h"
#include "mmu_rb.h"

static uint hfi1_sdma_comp_ring_size = 128;
module_param_named(sdma_comp_size, hfi1_sdma_comp_ring_size, uint, S_IRUGO);
MODULE_PARM_DESC(sdma_comp_size, "Size of User SDMA completion ring. Default: 128");

/* The maximum number of Data io vectors per message/request */
#define MAX_VECTORS_PER_REQ 8
/*
 * Maximum number of packet to send from each message/request
 * before moving to the next one.
 */
#define MAX_PKTS_PER_QUEUE 16

#define num_pages(x) (1 + ((((x) - 1) & PAGE_MASK) >> PAGE_SHIFT))

#define req_opcode(x) \
	(((x) >> HFI1_SDMA_REQ_OPCODE_SHIFT) & HFI1_SDMA_REQ_OPCODE_MASK)
#define req_version(x) \
	(((x) >> HFI1_SDMA_REQ_VERSION_SHIFT) & HFI1_SDMA_REQ_OPCODE_MASK)
#define req_iovcnt(x) \
	(((x) >> HFI1_SDMA_REQ_IOVCNT_SHIFT) & HFI1_SDMA_REQ_IOVCNT_MASK)

/* Number of BTH.PSN bits used for sequence number in expected rcvs */
#define BTH_SEQ_MASK 0x7ffull

/*
 * Define fields in the KDETH header so we can update the header
 * template.
 */
#define KDETH_OFFSET_SHIFT        0
#define KDETH_OFFSET_MASK         0x7fff
#define KDETH_OM_SHIFT            15
#define KDETH_OM_MASK             0x1
#define KDETH_TID_SHIFT           16
#define KDETH_TID_MASK            0x3ff
#define KDETH_TIDCTRL_SHIFT       26
#define KDETH_TIDCTRL_MASK        0x3
#define KDETH_INTR_SHIFT          28
#define KDETH_INTR_MASK           0x1
#define KDETH_SH_SHIFT            29
#define KDETH_SH_MASK             0x1
#define KDETH_HCRC_UPPER_SHIFT    16
#define KDETH_HCRC_UPPER_MASK     0xff
#define KDETH_HCRC_LOWER_SHIFT    24
#define KDETH_HCRC_LOWER_MASK     0xff

#define PBC2LRH(x) ((((x) & 0xfff) << 2) - 4)
#define LRH2PBC(x) ((((x) >> 2) + 1) & 0xfff)

#define KDETH_GET(val, field)						\
	(((le32_to_cpu((val))) >> KDETH_##field##_SHIFT) & KDETH_##field##_MASK)
#define KDETH_SET(dw, field, val) do {					\
		u32 dwval = le32_to_cpu(dw);				\
		dwval &= ~(KDETH_##field##_MASK << KDETH_##field##_SHIFT); \
		dwval |= (((val) & KDETH_##field##_MASK) << \
			  KDETH_##field##_SHIFT);			\
		dw = cpu_to_le32(dwval);				\
	} while (0)

#define AHG_HEADER_SET(arr, idx, dw, bit, width, value)			\
	do {								\
		if ((idx) < ARRAY_SIZE((arr)))				\
			(arr)[(idx++)] = sdma_build_ahg_descriptor(	\
				(__force u16)(value), (dw), (bit),	\
							(width));	\
		else							\
			return -ERANGE;					\
	} while (0)

/* KDETH OM multipliers and switch over point */
#define KDETH_OM_SMALL     4
#define KDETH_OM_LARGE     64
#define KDETH_OM_MAX_SIZE  (1 << ((KDETH_OM_LARGE / KDETH_OM_SMALL) + 1))

/* Last packet in the request */
#define TXREQ_FLAGS_REQ_LAST_PKT BIT(0)

#define SDMA_REQ_IN_USE     0
#define SDMA_REQ_FOR_THREAD 1
#define SDMA_REQ_SEND_DONE  2
#define SDMA_REQ_HAVE_AHG   3
#define SDMA_REQ_HAS_ERROR  4
#define SDMA_REQ_DONE_ERROR 5

#define SDMA_PKT_Q_INACTIVE BIT(0)
#define SDMA_PKT_Q_ACTIVE   BIT(1)
#define SDMA_PKT_Q_DEFERRED BIT(2)

/*
 * Maximum retry attempts to submit a TX request
 * before putting the process to sleep.
 */
#define MAX_DEFER_RETRY_COUNT 1

static unsigned initial_pkt_count = 8;

#define SDMA_IOWAIT_TIMEOUT 1000 /* in milliseconds */

struct sdma_mmu_node;

struct user_sdma_iovec {
	struct list_head list;
	struct iovec iov;
	/* number of pages in this vector */
	unsigned npages;
	/* array of pinned pages for this vector */
	struct page **pages;
	/*
	 * offset into the virtual address space of the vector at
	 * which we last left off.
	 */
	u64 offset;
	struct sdma_mmu_node *node;
};

#define SDMA_CACHE_NODE_EVICT 0

struct sdma_mmu_node {
	struct mmu_rb_node rb;
	struct list_head list;
	struct hfi1_user_sdma_pkt_q *pq;
	atomic_t refcount;
	struct page **pages;
	unsigned npages;
	unsigned long flags;
};

struct user_sdma_request {
	struct sdma_req_info info;
	struct hfi1_user_sdma_pkt_q *pq;
	struct hfi1_user_sdma_comp_q *cq;
	/* This is the original header from user space */
	struct hfi1_pkt_header hdr;
	/*
	 * Pointer to the SDMA engine for this request.
	 * Since different request could be on different VLs,
	 * each request will need it's own engine pointer.
	 */
	struct sdma_engine *sde;
	u8 ahg_idx;
	u32 ahg[9];
	/*
	 * KDETH.Offset (Eager) field
	 * We need to remember the initial value so the headers
	 * can be updated properly.
	 */
	u32 koffset;
	/*
	 * KDETH.OFFSET (TID) field
	 * The offset can cover multiple packets, depending on the
	 * size of the TID entry.
	 */
	u32 tidoffset;
	/*
	 * KDETH.OM
	 * Remember this because the header template always sets it
	 * to 0.
	 */
	u8 omfactor;
	/*
	 * We copy the iovs for this request (based on
	 * info.iovcnt). These are only the data vectors
	 */
	unsigned data_iovs;
	/* total length of the data in the request */
	u32 data_len;
	/* progress index moving along the iovs array */
	unsigned iov_idx;
	struct user_sdma_iovec iovs[MAX_VECTORS_PER_REQ];
	/* number of elements copied to the tids array */
	u16 n_tids;
	/* TID array values copied from the tid_iov vector */
	u32 *tids;
	u16 tididx;
	u32 sent;
	u64 seqnum;
	u64 seqcomp;
	u64 seqsubmitted;
	struct list_head txps;
	unsigned long flags;
	/* status of the last txreq completed */
	int status;
};

/*
 * A single txreq could span up to 3 physical pages when the MTU
 * is sufficiently large (> 4K). Each of the IOV pointers also
 * needs it's own set of flags so the vector has been handled
 * independently of each other.
 */
struct user_sdma_txreq {
	/* Packet header for the txreq */
	struct hfi1_pkt_header hdr;
	struct sdma_txreq txreq;
	struct list_head list;
	struct user_sdma_request *req;
	u16 flags;
	unsigned busycount;
	u64 seqnum;
};

#define SDMA_DBG(req, fmt, ...)				     \
	hfi1_cdbg(SDMA, "[%u:%u:%u:%u] " fmt, (req)->pq->dd->unit, \
		 (req)->pq->ctxt, (req)->pq->subctxt, (req)->info.comp_idx, \
		 ##__VA_ARGS__)
#define SDMA_Q_DBG(pq, fmt, ...)			 \
	hfi1_cdbg(SDMA, "[%u:%u:%u] " fmt, (pq)->dd->unit, (pq)->ctxt, \
		 (pq)->subctxt, ##__VA_ARGS__)

static int user_sdma_send_pkts(struct user_sdma_request *, unsigned);
static int num_user_pages(const struct iovec *);
static void user_sdma_txreq_cb(struct sdma_txreq *, int);
static inline void pq_update(struct hfi1_user_sdma_pkt_q *);
static void user_sdma_free_request(struct user_sdma_request *, bool);
static int pin_vector_pages(struct user_sdma_request *,
			    struct user_sdma_iovec *);
static void unpin_vector_pages(struct mm_struct *, struct page **, unsigned,
			       unsigned);
static int check_header_template(struct user_sdma_request *,
				 struct hfi1_pkt_header *, u32, u32);
static int set_txreq_header(struct user_sdma_request *,
			    struct user_sdma_txreq *, u32);
static int set_txreq_header_ahg(struct user_sdma_request *,
				struct user_sdma_txreq *, u32);
static inline void set_comp_state(struct hfi1_user_sdma_pkt_q *,
				  struct hfi1_user_sdma_comp_q *,
				  u16, enum hfi1_sdma_comp_state, int);
static inline u32 set_pkt_bth_psn(__be32, u8, u32);
static inline u32 get_lrh_len(struct hfi1_pkt_header, u32 len);

static int defer_packet_queue(
	struct sdma_engine *,
	struct iowait *,
	struct sdma_txreq *,
	unsigned seq);
static void activate_packet_queue(struct iowait *, int);
static bool sdma_rb_filter(struct mmu_rb_node *, unsigned long, unsigned long);
static int sdma_rb_insert(struct rb_root *, struct mmu_rb_node *);
static void sdma_rb_remove(struct rb_root *, struct mmu_rb_node *,
			   struct mm_struct *);
static int sdma_rb_invalidate(struct rb_root *, struct mmu_rb_node *);

static struct mmu_rb_ops sdma_rb_ops = {
	.filter = sdma_rb_filter,
	.insert = sdma_rb_insert,
	.remove = sdma_rb_remove,
	.invalidate = sdma_rb_invalidate
};

static int defer_packet_queue(
	struct sdma_engine *sde,
	struct iowait *wait,
	struct sdma_txreq *txreq,
	unsigned seq)
{
	struct hfi1_user_sdma_pkt_q *pq =
		container_of(wait, struct hfi1_user_sdma_pkt_q, busy);
	struct hfi1_ibdev *dev = &pq->dd->verbs_dev;
	struct user_sdma_txreq *tx =
		container_of(txreq, struct user_sdma_txreq, txreq);

	if (sdma_progress(sde, seq, txreq)) {
		if (tx->busycount++ < MAX_DEFER_RETRY_COUNT)
			goto eagain;
	}
	/*
	 * We are assuming that if the list is enqueued somewhere, it
	 * is to the dmawait list since that is the only place where
	 * it is supposed to be enqueued.
	 */
	xchg(&pq->state, SDMA_PKT_Q_DEFERRED);
	write_seqlock(&dev->iowait_lock);
	if (list_empty(&pq->busy.list))
		list_add_tail(&pq->busy.list, &sde->dmawait);
	write_sequnlock(&dev->iowait_lock);
	return -EBUSY;
eagain:
	return -EAGAIN;
}

static void activate_packet_queue(struct iowait *wait, int reason)
{
	struct hfi1_user_sdma_pkt_q *pq =
		container_of(wait, struct hfi1_user_sdma_pkt_q, busy);
	xchg(&pq->state, SDMA_PKT_Q_ACTIVE);
	wake_up(&wait->wait_dma);
};

static void sdma_kmem_cache_ctor(void *obj)
{
	struct user_sdma_txreq *tx = obj;

	memset(tx, 0, sizeof(*tx));
}

int hfi1_user_sdma_alloc_queues(struct hfi1_ctxtdata *uctxt, struct file *fp)
{
	struct hfi1_filedata *fd;
	int ret = 0;
	unsigned memsize;
	char buf[64];
	struct hfi1_devdata *dd;
	struct hfi1_user_sdma_comp_q *cq;
	struct hfi1_user_sdma_pkt_q *pq;
	unsigned long flags;

	if (!uctxt || !fp) {
		ret = -EBADF;
		goto done;
	}

	fd = fp->private_data;

	if (!hfi1_sdma_comp_ring_size) {
		ret = -EINVAL;
		goto done;
	}

	dd = uctxt->dd;

	pq = kzalloc(sizeof(*pq), GFP_KERNEL);
	if (!pq)
		goto pq_nomem;

	memsize = sizeof(*pq->reqs) * hfi1_sdma_comp_ring_size;
	pq->reqs = kzalloc(memsize, GFP_KERNEL);
	if (!pq->reqs)
		goto pq_reqs_nomem;

	INIT_LIST_HEAD(&pq->list);
	pq->dd = dd;
	pq->ctxt = uctxt->ctxt;
	pq->subctxt = fd->subctxt;
	pq->n_max_reqs = hfi1_sdma_comp_ring_size;
	pq->state = SDMA_PKT_Q_INACTIVE;
	atomic_set(&pq->n_reqs, 0);
	init_waitqueue_head(&pq->wait);
	pq->sdma_rb_root = RB_ROOT;
	INIT_LIST_HEAD(&pq->evict);
	spin_lock_init(&pq->evict_lock);

	iowait_init(&pq->busy, 0, NULL, defer_packet_queue,
		    activate_packet_queue, NULL);
	pq->reqidx = 0;
	snprintf(buf, 64, "txreq-kmem-cache-%u-%u-%u", dd->unit, uctxt->ctxt,
		 fd->subctxt);
	pq->txreq_cache = kmem_cache_create(buf,
			       sizeof(struct user_sdma_txreq),
					    L1_CACHE_BYTES,
					    SLAB_HWCACHE_ALIGN,
					    sdma_kmem_cache_ctor);
	if (!pq->txreq_cache) {
		dd_dev_err(dd, "[%u] Failed to allocate TxReq cache\n",
			   uctxt->ctxt);
		goto pq_txreq_nomem;
	}
	fd->pq = pq;
	cq = kzalloc(sizeof(*cq), GFP_KERNEL);
	if (!cq)
		goto cq_nomem;

	memsize = PAGE_ALIGN(sizeof(*cq->comps) * hfi1_sdma_comp_ring_size);
	cq->comps = vmalloc_user(memsize);
	if (!cq->comps)
		goto cq_comps_nomem;

	cq->nentries = hfi1_sdma_comp_ring_size;
	fd->cq = cq;

	ret = hfi1_mmu_rb_register(&pq->sdma_rb_root, &sdma_rb_ops);
	if (ret) {
		dd_dev_err(dd, "Failed to register with MMU %d", ret);
		goto done;
	}

	spin_lock_irqsave(&uctxt->sdma_qlock, flags);
	list_add(&pq->list, &uctxt->sdma_queues);
	spin_unlock_irqrestore(&uctxt->sdma_qlock, flags);
	goto done;

cq_comps_nomem:
	kfree(cq);
cq_nomem:
	kmem_cache_destroy(pq->txreq_cache);
pq_txreq_nomem:
	kfree(pq->reqs);
pq_reqs_nomem:
	kfree(pq);
	fd->pq = NULL;
pq_nomem:
	ret = -ENOMEM;
done:
	return ret;
}

int hfi1_user_sdma_free_queues(struct hfi1_filedata *fd)
{
	struct hfi1_ctxtdata *uctxt = fd->uctxt;
	struct hfi1_user_sdma_pkt_q *pq;
	unsigned long flags;

	hfi1_cdbg(SDMA, "[%u:%u:%u] Freeing user SDMA queues", uctxt->dd->unit,
		  uctxt->ctxt, fd->subctxt);
	pq = fd->pq;
	hfi1_mmu_rb_unregister(&pq->sdma_rb_root);
	if (pq) {
		spin_lock_irqsave(&uctxt->sdma_qlock, flags);
		if (!list_empty(&pq->list))
			list_del_init(&pq->list);
		spin_unlock_irqrestore(&uctxt->sdma_qlock, flags);
		iowait_sdma_drain(&pq->busy);
		/* Wait until all requests have been freed. */
		wait_event_interruptible(
			pq->wait,
			(ACCESS_ONCE(pq->state) == SDMA_PKT_Q_INACTIVE));
		kfree(pq->reqs);
		kmem_cache_destroy(pq->txreq_cache);
		kfree(pq);
		fd->pq = NULL;
	}
	if (fd->cq) {
		vfree(fd->cq->comps);
		kfree(fd->cq);
		fd->cq = NULL;
	}
	return 0;
}

int hfi1_user_sdma_process_request(struct file *fp, struct iovec *iovec,
				   unsigned long dim, unsigned long *count)
{
	int ret = 0, i = 0;
	struct hfi1_filedata *fd = fp->private_data;
	struct hfi1_ctxtdata *uctxt = fd->uctxt;
	struct hfi1_user_sdma_pkt_q *pq = fd->pq;
	struct hfi1_user_sdma_comp_q *cq = fd->cq;
	struct hfi1_devdata *dd = pq->dd;
	unsigned long idx = 0;
	u8 pcount = initial_pkt_count;
	struct sdma_req_info info;
	struct user_sdma_request *req;
	u8 opcode, sc, vl;
	int req_queued = 0;

	if (iovec[idx].iov_len < sizeof(info) + sizeof(req->hdr)) {
		hfi1_cdbg(
		   SDMA,
		   "[%u:%u:%u] First vector not big enough for header %lu/%lu",
		   dd->unit, uctxt->ctxt, fd->subctxt,
		   iovec[idx].iov_len, sizeof(info) + sizeof(req->hdr));
		return -EINVAL;
	}
	ret = copy_from_user(&info, iovec[idx].iov_base, sizeof(info));
	if (ret) {
		hfi1_cdbg(SDMA, "[%u:%u:%u] Failed to copy info QW (%d)",
			  dd->unit, uctxt->ctxt, fd->subctxt, ret);
		return -EFAULT;
	}

	trace_hfi1_sdma_user_reqinfo(dd, uctxt->ctxt, fd->subctxt,
				     (u16 *)&info);
	if (cq->comps[info.comp_idx].status == QUEUED ||
	    test_bit(SDMA_REQ_IN_USE, &pq->reqs[info.comp_idx].flags)) {
		hfi1_cdbg(SDMA, "[%u:%u:%u] Entry %u is in QUEUED state",
			  dd->unit, uctxt->ctxt, fd->subctxt,
			  info.comp_idx);
		return -EBADSLT;
	}
	if (!info.fragsize) {
		hfi1_cdbg(SDMA,
			  "[%u:%u:%u:%u] Request does not specify fragsize",
			  dd->unit, uctxt->ctxt, fd->subctxt, info.comp_idx);
		return -EINVAL;
	}
	/*
	 * We've done all the safety checks that we can up to this point,
	 * "allocate" the request entry.
	 */
	hfi1_cdbg(SDMA, "[%u:%u:%u] Using req/comp entry %u\n", dd->unit,
		  uctxt->ctxt, fd->subctxt, info.comp_idx);
	req = pq->reqs + info.comp_idx;
	memset(req, 0, sizeof(*req));
	/* Mark the request as IN_USE before we start filling it in. */
	set_bit(SDMA_REQ_IN_USE, &req->flags);
	req->data_iovs = req_iovcnt(info.ctrl) - 1;
	req->pq = pq;
	req->cq = cq;
	req->status = -1;
	INIT_LIST_HEAD(&req->txps);

	memcpy(&req->info, &info, sizeof(info));

	if (req_opcode(info.ctrl) == EXPECTED)
		req->data_iovs--;

	if (!info.npkts || req->data_iovs > MAX_VECTORS_PER_REQ) {
		SDMA_DBG(req, "Too many vectors (%u/%u)", req->data_iovs,
			 MAX_VECTORS_PER_REQ);
		return -EINVAL;
	}
	/* Copy the header from the user buffer */
	ret = copy_from_user(&req->hdr, iovec[idx].iov_base + sizeof(info),
			     sizeof(req->hdr));
	if (ret) {
		SDMA_DBG(req, "Failed to copy header template (%d)", ret);
		ret = -EFAULT;
		goto free_req;
	}

	/* If Static rate control is not enabled, sanitize the header. */
	if (!HFI1_CAP_IS_USET(STATIC_RATE_CTRL))
		req->hdr.pbc[2] = 0;

	/* Validate the opcode. Do not trust packets from user space blindly. */
	opcode = (be32_to_cpu(req->hdr.bth[0]) >> 24) & 0xff;
	if ((opcode & USER_OPCODE_CHECK_MASK) !=
	     USER_OPCODE_CHECK_VAL) {
		SDMA_DBG(req, "Invalid opcode (%d)", opcode);
		ret = -EINVAL;
		goto free_req;
	}
	/*
	 * Validate the vl. Do not trust packets from user space blindly.
	 * VL comes from PBC, SC comes from LRH, and the VL needs to
	 * match the SC look up.
	 */
	vl = (le16_to_cpu(req->hdr.pbc[0]) >> 12) & 0xF;
	sc = (((be16_to_cpu(req->hdr.lrh[0]) >> 12) & 0xF) |
	      (((le16_to_cpu(req->hdr.pbc[1]) >> 14) & 0x1) << 4));
	if (vl >= dd->pport->vls_operational ||
	    vl != sc_to_vlt(dd, sc)) {
		SDMA_DBG(req, "Invalid SC(%u)/VL(%u)", sc, vl);
		ret = -EINVAL;
		goto free_req;
	}

	/* Checking P_KEY for requests from user-space */
	if (egress_pkey_check(dd->pport, req->hdr.lrh, req->hdr.bth, sc,
			      PKEY_CHECK_INVALID)) {
		ret = -EINVAL;
		goto free_req;
	}

	/*
	 * Also should check the BTH.lnh. If it says the next header is GRH then
	 * the RXE parsing will be off and will land in the middle of the KDETH
	 * or miss it entirely.
	 */
	if ((be16_to_cpu(req->hdr.lrh[0]) & 0x3) == HFI1_LRH_GRH) {
		SDMA_DBG(req, "User tried to pass in a GRH");
		ret = -EINVAL;
		goto free_req;
	}

	req->koffset = le32_to_cpu(req->hdr.kdeth.swdata[6]);
	/*
	 * Calculate the initial TID offset based on the values of
	 * KDETH.OFFSET and KDETH.OM that are passed in.
	 */
	req->tidoffset = KDETH_GET(req->hdr.kdeth.ver_tid_offset, OFFSET) *
		(KDETH_GET(req->hdr.kdeth.ver_tid_offset, OM) ?
		 KDETH_OM_LARGE : KDETH_OM_SMALL);
	SDMA_DBG(req, "Initial TID offset %u", req->tidoffset);
	idx++;

	/* Save all the IO vector structures */
	while (i < req->data_iovs) {
		INIT_LIST_HEAD(&req->iovs[i].list);
		memcpy(&req->iovs[i].iov, iovec + idx++, sizeof(struct iovec));
		ret = pin_vector_pages(req, &req->iovs[i]);
		if (ret) {
			req->status = ret;
			goto free_req;
		}
		req->data_len += req->iovs[i++].iov.iov_len;
	}
	SDMA_DBG(req, "total data length %u", req->data_len);

	if (pcount > req->info.npkts)
		pcount = req->info.npkts;
	/*
	 * Copy any TID info
	 * User space will provide the TID info only when the
	 * request type is EXPECTED. This is true even if there is
	 * only one packet in the request and the header is already
	 * setup. The reason for the singular TID case is that the
	 * driver needs to perform safety checks.
	 */
	if (req_opcode(req->info.ctrl) == EXPECTED) {
		u16 ntids = iovec[idx].iov_len / sizeof(*req->tids);

		if (!ntids || ntids > MAX_TID_PAIR_ENTRIES) {
			ret = -EINVAL;
			goto free_req;
		}
		req->tids = kcalloc(ntids, sizeof(*req->tids), GFP_KERNEL);
		if (!req->tids) {
			ret = -ENOMEM;
			goto free_req;
		}
		/*
		 * We have to copy all of the tids because they may vary
		 * in size and, therefore, the TID count might not be
		 * equal to the pkt count. However, there is no way to
		 * tell at this point.
		 */
		ret = copy_from_user(req->tids, iovec[idx].iov_base,
				     ntids * sizeof(*req->tids));
		if (ret) {
			SDMA_DBG(req, "Failed to copy %d TIDs (%d)",
				 ntids, ret);
			ret = -EFAULT;
			goto free_req;
		}
		req->n_tids = ntids;
		idx++;
	}

	/* Have to select the engine */
	req->sde = sdma_select_engine_vl(dd,
					 (u32)(uctxt->ctxt + fd->subctxt),
					 vl);
	if (!req->sde || !sdma_running(req->sde)) {
		ret = -ECOMM;
		goto free_req;
	}

	/* We don't need an AHG entry if the request contains only one packet */
	if (req->info.npkts > 1 && HFI1_CAP_IS_USET(SDMA_AHG)) {
		int ahg = sdma_ahg_alloc(req->sde);

		if (likely(ahg >= 0)) {
			req->ahg_idx = (u8)ahg;
			set_bit(SDMA_REQ_HAVE_AHG, &req->flags);
		}
	}

	set_comp_state(pq, cq, info.comp_idx, QUEUED, 0);
	atomic_inc(&pq->n_reqs);
	req_queued = 1;
	/* Send the first N packets in the request to buy us some time */
	ret = user_sdma_send_pkts(req, pcount);
	if (unlikely(ret < 0 && ret != -EBUSY)) {
		req->status = ret;
		goto free_req;
	}

	/*
	 * It is possible that the SDMA engine would have processed all the
	 * submitted packets by the time we get here. Therefore, only set
	 * packet queue state to ACTIVE if there are still uncompleted
	 * requests.
	 */
	if (atomic_read(&pq->n_reqs))
		xchg(&pq->state, SDMA_PKT_Q_ACTIVE);

	/*
	 * This is a somewhat blocking send implementation.
	 * The driver will block the caller until all packets of the
	 * request have been submitted to the SDMA engine. However, it
	 * will not wait for send completions.
	 */
	while (!test_bit(SDMA_REQ_SEND_DONE, &req->flags)) {
		ret = user_sdma_send_pkts(req, pcount);
		if (ret < 0) {
			if (ret != -EBUSY) {
				req->status = ret;
				set_bit(SDMA_REQ_DONE_ERROR, &req->flags);
				if (ACCESS_ONCE(req->seqcomp) ==
				    req->seqsubmitted - 1)
					goto free_req;
				return ret;
			}
			wait_event_interruptible_timeout(
				pq->busy.wait_dma,
				(pq->state == SDMA_PKT_Q_ACTIVE),
				msecs_to_jiffies(
					SDMA_IOWAIT_TIMEOUT));
		}
	}
	*count += idx;
	return 0;
free_req:
	user_sdma_free_request(req, true);
	if (req_queued)
		pq_update(pq);
	set_comp_state(pq, cq, info.comp_idx, ERROR, req->status);
	return ret;
}

static inline u32 compute_data_length(struct user_sdma_request *req,
				      struct user_sdma_txreq *tx)
{
	/*
	 * Determine the proper size of the packet data.
	 * The size of the data of the first packet is in the header
	 * template. However, it includes the header and ICRC, which need
	 * to be subtracted.
	 * The size of the remaining packets is the minimum of the frag
	 * size (MTU) or remaining data in the request.
	 */
	u32 len;

	if (!req->seqnum) {
		len = ((be16_to_cpu(req->hdr.lrh[2]) << 2) -
		       (sizeof(tx->hdr) - 4));
	} else if (req_opcode(req->info.ctrl) == EXPECTED) {
		u32 tidlen = EXP_TID_GET(req->tids[req->tididx], LEN) *
			PAGE_SIZE;
		/*
		 * Get the data length based on the remaining space in the
		 * TID pair.
		 */
		len = min(tidlen - req->tidoffset, (u32)req->info.fragsize);
		/* If we've filled up the TID pair, move to the next one. */
		if (unlikely(!len) && ++req->tididx < req->n_tids &&
		    req->tids[req->tididx]) {
			tidlen = EXP_TID_GET(req->tids[req->tididx],
					     LEN) * PAGE_SIZE;
			req->tidoffset = 0;
			len = min_t(u32, tidlen, req->info.fragsize);
		}
		/*
		 * Since the TID pairs map entire pages, make sure that we
		 * are not going to try to send more data that we have
		 * remaining.
		 */
		len = min(len, req->data_len - req->sent);
	} else {
		len = min(req->data_len - req->sent, (u32)req->info.fragsize);
	}
	SDMA_DBG(req, "Data Length = %u", len);
	return len;
}

static inline u32 get_lrh_len(struct hfi1_pkt_header hdr, u32 len)
{
	/* (Size of complete header - size of PBC) + 4B ICRC + data length */
	return ((sizeof(hdr) - sizeof(hdr.pbc)) + 4 + len);
}

static int user_sdma_send_pkts(struct user_sdma_request *req, unsigned maxpkts)
{
	int ret = 0;
	unsigned npkts = 0;
	struct user_sdma_txreq *tx = NULL;
	struct hfi1_user_sdma_pkt_q *pq = NULL;
	struct user_sdma_iovec *iovec = NULL;

	if (!req->pq)
		return -EINVAL;

	pq = req->pq;

	/* If tx completion has reported an error, we are done. */
	if (test_bit(SDMA_REQ_HAS_ERROR, &req->flags)) {
		set_bit(SDMA_REQ_DONE_ERROR, &req->flags);
		return -EFAULT;
	}

	/*
	 * Check if we might have sent the entire request already
	 */
	if (unlikely(req->seqnum == req->info.npkts)) {
		if (!list_empty(&req->txps))
			goto dosend;
		return ret;
	}

	if (!maxpkts || maxpkts > req->info.npkts - req->seqnum)
		maxpkts = req->info.npkts - req->seqnum;

	while (npkts < maxpkts) {
		u32 datalen = 0, queued = 0, data_sent = 0;
		u64 iov_offset = 0;

		/*
		 * Check whether any of the completions have come back
		 * with errors. If so, we are not going to process any
		 * more packets from this request.
		 */
		if (test_bit(SDMA_REQ_HAS_ERROR, &req->flags)) {
			set_bit(SDMA_REQ_DONE_ERROR, &req->flags);
			return -EFAULT;
		}

		tx = kmem_cache_alloc(pq->txreq_cache, GFP_KERNEL);
		if (!tx)
			return -ENOMEM;

		tx->flags = 0;
		tx->req = req;
		tx->busycount = 0;
		INIT_LIST_HEAD(&tx->list);

		if (req->seqnum == req->info.npkts - 1)
			tx->flags |= TXREQ_FLAGS_REQ_LAST_PKT;

		/*
		 * Calculate the payload size - this is min of the fragment
		 * (MTU) size or the remaining bytes in the request but only
		 * if we have payload data.
		 */
		if (req->data_len) {
			iovec = &req->iovs[req->iov_idx];
			if (ACCESS_ONCE(iovec->offset) == iovec->iov.iov_len) {
				if (++req->iov_idx == req->data_iovs) {
					ret = -EFAULT;
					goto free_txreq;
				}
				iovec = &req->iovs[req->iov_idx];
				WARN_ON(iovec->offset);
			}

			datalen = compute_data_length(req, tx);
			if (!datalen) {
				SDMA_DBG(req,
					 "Request has data but pkt len is 0");
				ret = -EFAULT;
				goto free_tx;
			}
		}

		if (test_bit(SDMA_REQ_HAVE_AHG, &req->flags)) {
			if (!req->seqnum) {
				u16 pbclen = le16_to_cpu(req->hdr.pbc[0]);
				u32 lrhlen = get_lrh_len(req->hdr, datalen);
				/*
				 * Copy the request header into the tx header
				 * because the HW needs a cacheline-aligned
				 * address.
				 * This copy can be optimized out if the hdr
				 * member of user_sdma_request were also
				 * cacheline aligned.
				 */
				memcpy(&tx->hdr, &req->hdr, sizeof(tx->hdr));
				if (PBC2LRH(pbclen) != lrhlen) {
					pbclen = (pbclen & 0xf000) |
						LRH2PBC(lrhlen);
					tx->hdr.pbc[0] = cpu_to_le16(pbclen);
				}
				ret = sdma_txinit_ahg(&tx->txreq,
						      SDMA_TXREQ_F_AHG_COPY,
						      sizeof(tx->hdr) + datalen,
						      req->ahg_idx, 0, NULL, 0,
						      user_sdma_txreq_cb);
				if (ret)
					goto free_tx;
				ret = sdma_txadd_kvaddr(pq->dd, &tx->txreq,
							&tx->hdr,
							sizeof(tx->hdr));
				if (ret)
					goto free_txreq;
			} else {
				int changes;

				changes = set_txreq_header_ahg(req, tx,
							       datalen);
				if (changes < 0)
					goto free_tx;
				sdma_txinit_ahg(&tx->txreq,
						SDMA_TXREQ_F_USE_AHG,
						datalen, req->ahg_idx, changes,
						req->ahg, sizeof(req->hdr),
						user_sdma_txreq_cb);
			}
		} else {
			ret = sdma_txinit(&tx->txreq, 0, sizeof(req->hdr) +
					  datalen, user_sdma_txreq_cb);
			if (ret)
				goto free_tx;
			/*
			 * Modify the header for this packet. This only needs
			 * to be done if we are not going to use AHG. Otherwise,
			 * the HW will do it based on the changes we gave it
			 * during sdma_txinit_ahg().
			 */
			ret = set_txreq_header(req, tx, datalen);
			if (ret)
				goto free_txreq;
		}

		/*
		 * If the request contains any data vectors, add up to
		 * fragsize bytes to the descriptor.
		 */
		while (queued < datalen &&
		       (req->sent + data_sent) < req->data_len) {
			unsigned long base, offset;
			unsigned pageidx, len;

			base = (unsigned long)iovec->iov.iov_base;
			offset = offset_in_page(base + iovec->offset +
						iov_offset);
			pageidx = (((iovec->offset + iov_offset +
				     base) - (base & PAGE_MASK)) >> PAGE_SHIFT);
			len = offset + req->info.fragsize > PAGE_SIZE ?
				PAGE_SIZE - offset : req->info.fragsize;
			len = min((datalen - queued), len);
			ret = sdma_txadd_page(pq->dd, &tx->txreq,
					      iovec->pages[pageidx],
					      offset, len);
			if (ret) {
				SDMA_DBG(req, "SDMA txreq add page failed %d\n",
					 ret);
				goto free_txreq;
			}
			iov_offset += len;
			queued += len;
			data_sent += len;
			if (unlikely(queued < datalen &&
				     pageidx == iovec->npages &&
				     req->iov_idx < req->data_iovs - 1)) {
				iovec->offset += iov_offset;
				iovec = &req->iovs[++req->iov_idx];
				iov_offset = 0;
			}
		}
		/*
		 * The txreq was submitted successfully so we can update
		 * the counters.
		 */
		req->koffset += datalen;
		if (req_opcode(req->info.ctrl) == EXPECTED)
			req->tidoffset += datalen;
		req->sent += data_sent;
		if (req->data_len)
			iovec->offset += iov_offset;
		list_add_tail(&tx->txreq.list, &req->txps);
		/*
		 * It is important to increment this here as it is used to
		 * generate the BTH.PSN and, therefore, can't be bulk-updated
		 * outside of the loop.
		 */
		tx->seqnum = req->seqnum++;
		npkts++;
	}
dosend:
	ret = sdma_send_txlist(req->sde, &pq->busy, &req->txps);
	if (list_empty(&req->txps)) {
		req->seqsubmitted = req->seqnum;
		if (req->seqnum == req->info.npkts) {
			set_bit(SDMA_REQ_SEND_DONE, &req->flags);
			/*
			 * The txreq has already been submitted to the HW queue
			 * so we can free the AHG entry now. Corruption will not
			 * happen due to the sequential manner in which
			 * descriptors are processed.
			 */
			if (test_bit(SDMA_REQ_HAVE_AHG, &req->flags))
				sdma_ahg_free(req->sde, req->ahg_idx);
		}
	} else if (ret > 0) {
		req->seqsubmitted += ret;
		ret = 0;
	}
	return ret;

free_txreq:
	sdma_txclean(pq->dd, &tx->txreq);
free_tx:
	kmem_cache_free(pq->txreq_cache, tx);
	return ret;
}

/*
 * How many pages in this iovec element?
 */
static inline int num_user_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 u32 sdma_cache_evict(struct hfi1_user_sdma_pkt_q *pq, u32 npages)
{
	u32 cleared = 0;
	struct sdma_mmu_node *node, *ptr;
	struct list_head to_evict = LIST_HEAD_INIT(to_evict);

	spin_lock(&pq->evict_lock);
	list_for_each_entry_safe_reverse(node, ptr, &pq->evict, list) {
		/* Make sure that no one is still using the node. */
		if (!atomic_read(&node->refcount)) {
			set_bit(SDMA_CACHE_NODE_EVICT, &node->flags);
			list_del_init(&node->list);
			list_add(&node->list, &to_evict);
			cleared += node->npages;
			if (cleared >= npages)
				break;
		}
	}
	spin_unlock(&pq->evict_lock);

	list_for_each_entry_safe(node, ptr, &to_evict, list)
		hfi1_mmu_rb_remove(&pq->sdma_rb_root, &node->rb);

	return cleared;
}

static int pin_vector_pages(struct user_sdma_request *req,
			    struct user_sdma_iovec *iovec) {
	int ret = 0, pinned, npages, cleared;
	struct page **pages;
	struct hfi1_user_sdma_pkt_q *pq = req->pq;
	struct sdma_mmu_node *node = NULL;
	struct mmu_rb_node *rb_node;

	rb_node = hfi1_mmu_rb_extract(&pq->sdma_rb_root,
				      (unsigned long)iovec->iov.iov_base,
				      iovec->iov.iov_len);
	if (rb_node && !IS_ERR(rb_node))
		node = container_of(rb_node, struct sdma_mmu_node, rb);
	else
		rb_node = NULL;

	if (!node) {
		node = kzalloc(sizeof(*node), GFP_KERNEL);
		if (!node)
			return -ENOMEM;

		node->rb.addr = (unsigned long)iovec->iov.iov_base;
		node->pq = pq;
		atomic_set(&node->refcount, 0);
		INIT_LIST_HEAD(&node->list);
	}

	npages = num_user_pages(&iovec->iov);
	if (node->npages < npages) {
		pages = kcalloc(npages, sizeof(*pages), GFP_KERNEL);
		if (!pages) {
			SDMA_DBG(req, "Failed page array alloc");
			ret = -ENOMEM;
			goto bail;
		}
		memcpy(pages, node->pages, node->npages * sizeof(*pages));

		npages -= node->npages;

		/*
		 * If rb_node is NULL, it means that this is brand new node
		 * and, therefore not on the eviction list.
		 * If, however, the rb_node is non-NULL, it means that the
		 * node is already in RB tree and, therefore on the eviction
		 * list (nodes are unconditionally inserted in the eviction
		 * list). In that case, we have to remove the node prior to
		 * calling the eviction function in order to prevent it from
		 * freeing this node.
		 */
		if (rb_node) {
			spin_lock(&pq->evict_lock);
			list_del_init(&node->list);
			spin_unlock(&pq->evict_lock);
		}
retry:
		if (!hfi1_can_pin_pages(pq->dd, pq->n_locked, npages)) {
			cleared = sdma_cache_evict(pq, npages);
			if (cleared >= npages)
				goto retry;
		}
		pinned = hfi1_acquire_user_pages(
			((unsigned long)iovec->iov.iov_base +
			 (node->npages * PAGE_SIZE)), npages, 0,
			pages + node->npages);
		if (pinned < 0) {
			kfree(pages);
			ret = pinned;
			goto bail;
		}
		if (pinned != npages) {
			unpin_vector_pages(current->mm, pages, node->npages,
					   pinned);
			ret = -EFAULT;
			goto bail;
		}
		kfree(node->pages);
		node->rb.len = iovec->iov.iov_len;
		node->pages = pages;
		node->npages += pinned;
		npages = node->npages;
		spin_lock(&pq->evict_lock);
		list_add(&node->list, &pq->evict);
		pq->n_locked += pinned;
		spin_unlock(&pq->evict_lock);
	}
	iovec->pages = node->pages;
	iovec->npages = npages;
	iovec->node = node;

	ret = hfi1_mmu_rb_insert(&req->pq->sdma_rb_root, &node->rb);
	if (ret) {
		spin_lock(&pq->evict_lock);
		if (!list_empty(&node->list))
			list_del(&node->list);
		pq->n_locked -= node->npages;
		spin_unlock(&pq->evict_lock);
		goto bail;
	}
	return 0;
bail:
	if (rb_node)
		unpin_vector_pages(current->mm, node->pages, 0, node->npages);
	kfree(node);
	return ret;
}

static void unpin_vector_pages(struct mm_struct *mm, struct page **pages,
			       unsigned start, unsigned npages)
{
	hfi1_release_user_pages(mm, pages + start, npages, 0);
	kfree(pages);
}

static int check_header_template(struct user_sdma_request *req,
				 struct hfi1_pkt_header *hdr, u32 lrhlen,
				 u32 datalen)
{
	/*
	 * Perform safety checks for any type of packet:
	 *    - transfer size is multiple of 64bytes
	 *    - packet length is multiple of 4bytes
	 *    - entire request length is multiple of 4bytes
	 *    - packet length is not larger than MTU size
	 *
	 * These checks are only done for the first packet of the
	 * transfer since the header is "given" to us by user space.
	 * For the remainder of the packets we compute the values.
	 */
	if (req->info.fragsize % PIO_BLOCK_SIZE ||
	    lrhlen & 0x3 || req->data_len & 0x3  ||
	    lrhlen > get_lrh_len(*hdr, req->info.fragsize))
		return -EINVAL;

	if (req_opcode(req->info.ctrl) == EXPECTED) {
		/*
		 * The header is checked only on the first packet. Furthermore,
		 * we ensure that at least one TID entry is copied when the
		 * request is submitted. Therefore, we don't have to verify that
		 * tididx points to something sane.
		 */
		u32 tidval = req->tids[req->tididx],
			tidlen = EXP_TID_GET(tidval, LEN) * PAGE_SIZE,
			tididx = EXP_TID_GET(tidval, IDX),
			tidctrl = EXP_TID_GET(tidval, CTRL),
			tidoff;
		__le32 kval = hdr->kdeth.ver_tid_offset;

		tidoff = KDETH_GET(kval, OFFSET) *
			  (KDETH_GET(req->hdr.kdeth.ver_tid_offset, OM) ?
			   KDETH_OM_LARGE : KDETH_OM_SMALL);
		/*
		 * Expected receive packets have the following
		 * additional checks:
		 *     - offset is not larger than the TID size
		 *     - TIDCtrl values match between header and TID array
		 *     - TID indexes match between header and TID array
		 */
		if ((tidoff + datalen > tidlen) ||
		    KDETH_GET(kval, TIDCTRL) != tidctrl ||
		    KDETH_GET(kval, TID) != tididx)
			return -EINVAL;
	}
	return 0;
}

/*
 * Correctly set the BTH.PSN field based on type of
 * transfer - eager packets can just increment the PSN but
 * expected packets encode generation and sequence in the
 * BTH.PSN field so just incrementing will result in errors.
 */
static inline u32 set_pkt_bth_psn(__be32 bthpsn, u8 expct, u32 frags)
{
	u32 val = be32_to_cpu(bthpsn),
		mask = (HFI1_CAP_IS_KSET(EXTENDED_PSN) ? 0x7fffffffull :
			0xffffffull),
		psn = val & mask;
	if (expct)
		psn = (psn & ~BTH_SEQ_MASK) | ((psn + frags) & BTH_SEQ_MASK);
	else
		psn = psn + frags;
	return psn & mask;
}

static int set_txreq_header(struct user_sdma_request *req,
			    struct user_sdma_txreq *tx, u32 datalen)
{
	struct hfi1_user_sdma_pkt_q *pq = req->pq;
	struct hfi1_pkt_header *hdr = &tx->hdr;
	u16 pbclen;
	int ret;
	u32 tidval = 0, lrhlen = get_lrh_len(*hdr, datalen);

	/* Copy the header template to the request before modification */
	memcpy(hdr, &req->hdr, sizeof(*hdr));

	/*
	 * Check if the PBC and LRH length are mismatched. If so
	 * adjust both in the header.
	 */
	pbclen = le16_to_cpu(hdr->pbc[0]);
	if (PBC2LRH(pbclen) != lrhlen) {
		pbclen = (pbclen & 0xf000) | LRH2PBC(lrhlen);
		hdr->pbc[0] = cpu_to_le16(pbclen);
		hdr->lrh[2] = cpu_to_be16(lrhlen >> 2);
		/*
		 * Third packet
		 * This is the first packet in the sequence that has
		 * a "static" size that can be used for the rest of
		 * the packets (besides the last one).
		 */
		if (unlikely(req->seqnum == 2)) {
			/*
			 * From this point on the lengths in both the
			 * PBC and LRH are the same until the last
			 * packet.
			 * Adjust the template so we don't have to update
			 * every packet
			 */
			req->hdr.pbc[0] = hdr->pbc[0];
			req->hdr.lrh[2] = hdr->lrh[2];
		}
	}
	/*
	 * We only have to modify the header if this is not the
	 * first packet in the request. Otherwise, we use the
	 * header given to us.
	 */
	if (unlikely(!req->seqnum)) {
		ret = check_header_template(req, hdr, lrhlen, datalen);
		if (ret)
			return ret;
		goto done;
	}

	hdr->bth[2] = cpu_to_be32(
		set_pkt_bth_psn(hdr->bth[2],
				(req_opcode(req->info.ctrl) == EXPECTED),
				req->seqnum));

	/* Set ACK request on last packet */
	if (unlikely(tx->flags & TXREQ_FLAGS_REQ_LAST_PKT))
		hdr->bth[2] |= cpu_to_be32(1UL << 31);

	/* Set the new offset */
	hdr->kdeth.swdata[6] = cpu_to_le32(req->koffset);
	/* Expected packets have to fill in the new TID information */
	if (req_opcode(req->info.ctrl) == EXPECTED) {
		tidval = req->tids[req->tididx];
		/*
		 * If the offset puts us at the end of the current TID,
		 * advance everything.
		 */
		if ((req->tidoffset) == (EXP_TID_GET(tidval, LEN) *
					 PAGE_SIZE)) {
			req->tidoffset = 0;
			/*
			 * Since we don't copy all the TIDs, all at once,
			 * we have to check again.
			 */
			if (++req->tididx > req->n_tids - 1 ||
			    !req->tids[req->tididx]) {
				return -EINVAL;
			}
			tidval = req->tids[req->tididx];
		}
		req->omfactor = EXP_TID_GET(tidval, LEN) * PAGE_SIZE >=
			KDETH_OM_MAX_SIZE ? KDETH_OM_LARGE : KDETH_OM_SMALL;
		/* Set KDETH.TIDCtrl based on value for this TID. */
		KDETH_SET(hdr->kdeth.ver_tid_offset, TIDCTRL,
			  EXP_TID_GET(tidval, CTRL));
		/* Set KDETH.TID based on value for this TID */
		KDETH_SET(hdr->kdeth.ver_tid_offset, TID,
			  EXP_TID_GET(tidval, IDX));
		/* Clear KDETH.SH only on the last packet */
		if (unlikely(tx->flags & TXREQ_FLAGS_REQ_LAST_PKT))
			KDETH_SET(hdr->kdeth.ver_tid_offset, SH, 0);
		/*
		 * Set the KDETH.OFFSET and KDETH.OM based on size of
		 * transfer.
		 */
		SDMA_DBG(req, "TID offset %ubytes %uunits om%u",
			 req->tidoffset, req->tidoffset / req->omfactor,
			 req->omfactor != KDETH_OM_SMALL);
		KDETH_SET(hdr->kdeth.ver_tid_offset, OFFSET,
			  req->tidoffset / req->omfactor);
		KDETH_SET(hdr->kdeth.ver_tid_offset, OM,
			  req->omfactor != KDETH_OM_SMALL);
	}
done:
	trace_hfi1_sdma_user_header(pq->dd, pq->ctxt, pq->subctxt,
				    req->info.comp_idx, hdr, tidval);
	return sdma_txadd_kvaddr(pq->dd, &tx->txreq, hdr, sizeof(*hdr));
}

static int set_txreq_header_ahg(struct user_sdma_request *req,
				struct user_sdma_txreq *tx, u32 len)
{
	int diff = 0;
	struct hfi1_user_sdma_pkt_q *pq = req->pq;
	struct hfi1_pkt_header *hdr = &req->hdr;
	u16 pbclen = le16_to_cpu(hdr->pbc[0]);
	u32 val32, tidval = 0, lrhlen = get_lrh_len(*hdr, len);

	if (PBC2LRH(pbclen) != lrhlen) {
		/* PBC.PbcLengthDWs */
		AHG_HEADER_SET(req->ahg, diff, 0, 0, 12,
			       cpu_to_le16(LRH2PBC(lrhlen)));
		/* LRH.PktLen (we need the full 16 bits due to byte swap) */
		AHG_HEADER_SET(req->ahg, diff, 3, 0, 16,
			       cpu_to_be16(lrhlen >> 2));
	}

	/*
	 * Do the common updates
	 */
	/* BTH.PSN and BTH.A */
	val32 = (be32_to_cpu(hdr->bth[2]) + req->seqnum) &
		(HFI1_CAP_IS_KSET(EXTENDED_PSN) ? 0x7fffffff : 0xffffff);
	if (unlikely(tx->flags & TXREQ_FLAGS_REQ_LAST_PKT))
		val32 |= 1UL << 31;
	AHG_HEADER_SET(req->ahg, diff, 6, 0, 16, cpu_to_be16(val32 >> 16));
	AHG_HEADER_SET(req->ahg, diff, 6, 16, 16, cpu_to_be16(val32 & 0xffff));
	/* KDETH.Offset */
	AHG_HEADER_SET(req->ahg, diff, 15, 0, 16,
		       cpu_to_le16(req->koffset & 0xffff));
	AHG_HEADER_SET(req->ahg, diff, 15, 16, 16,
		       cpu_to_le16(req->koffset >> 16));
	if (req_opcode(req->info.ctrl) == EXPECTED) {
		__le16 val;

		tidval = req->tids[req->tididx];

		/*
		 * If the offset puts us at the end of the current TID,
		 * advance everything.
		 */
		if ((req->tidoffset) == (EXP_TID_GET(tidval, LEN) *
					 PAGE_SIZE)) {
			req->tidoffset = 0;
			/*
			 * Since we don't copy all the TIDs, all at once,
			 * we have to check again.
			 */
			if (++req->tididx > req->n_tids - 1 ||
			    !req->tids[req->tididx]) {
				return -EINVAL;
			}
			tidval = req->tids[req->tididx];
		}
		req->omfactor = ((EXP_TID_GET(tidval, LEN) *
				  PAGE_SIZE) >=
				 KDETH_OM_MAX_SIZE) ? KDETH_OM_LARGE :
			KDETH_OM_SMALL;
		/* KDETH.OM and KDETH.OFFSET (TID) */
		AHG_HEADER_SET(req->ahg, diff, 7, 0, 16,
			       ((!!(req->omfactor - KDETH_OM_SMALL)) << 15 |
				((req->tidoffset / req->omfactor) & 0x7fff)));
		/* KDETH.TIDCtrl, KDETH.TID */
		val = cpu_to_le16(((EXP_TID_GET(tidval, CTRL) & 0x3) << 10) |
					(EXP_TID_GET(tidval, IDX) & 0x3ff));
		/* Clear KDETH.SH on last packet */
		if (unlikely(tx->flags & TXREQ_FLAGS_REQ_LAST_PKT)) {
			val |= cpu_to_le16(KDETH_GET(hdr->kdeth.ver_tid_offset,
								INTR) >> 16);
			val &= cpu_to_le16(~(1U << 13));
			AHG_HEADER_SET(req->ahg, diff, 7, 16, 14, val);
		} else {
			AHG_HEADER_SET(req->ahg, diff, 7, 16, 12, val);
		}
	}

	trace_hfi1_sdma_user_header_ahg(pq->dd, pq->ctxt, pq->subctxt,
					req->info.comp_idx, req->sde->this_idx,
					req->ahg_idx, req->ahg, diff, tidval);
	return diff;
}

/*
 * SDMA tx request completion callback. Called when the SDMA progress
 * state machine gets notification that the SDMA descriptors for this
 * tx request have been processed by the DMA engine. Called in
 * interrupt context.
 */
static void user_sdma_txreq_cb(struct sdma_txreq *txreq, int status)
{
	struct user_sdma_txreq *tx =
		container_of(txreq, struct user_sdma_txreq, txreq);
	struct user_sdma_request *req;
	struct hfi1_user_sdma_pkt_q *pq;
	struct hfi1_user_sdma_comp_q *cq;
	u16 idx;

	if (!tx->req)
		return;

	req = tx->req;
	pq = req->pq;
	cq = req->cq;

	if (status != SDMA_TXREQ_S_OK) {
		SDMA_DBG(req, "SDMA completion with error %d",
			 status);
		set_bit(SDMA_REQ_HAS_ERROR, &req->flags);
	}

	req->seqcomp = tx->seqnum;
	kmem_cache_free(pq->txreq_cache, tx);
	tx = NULL;

	idx = req->info.comp_idx;
	if (req->status == -1 && status == SDMA_TXREQ_S_OK) {
		if (req->seqcomp == req->info.npkts - 1) {
			req->status = 0;
			user_sdma_free_request(req, false);
			pq_update(pq);
			set_comp_state(pq, cq, idx, COMPLETE, 0);
		}
	} else {
		if (status != SDMA_TXREQ_S_OK)
			req->status = status;
		if (req->seqcomp == (ACCESS_ONCE(req->seqsubmitted) - 1) &&
		    (test_bit(SDMA_REQ_SEND_DONE, &req->flags) ||
		     test_bit(SDMA_REQ_DONE_ERROR, &req->flags))) {
			user_sdma_free_request(req, false);
			pq_update(pq);
			set_comp_state(pq, cq, idx, ERROR, req->status);
		}
	}
}

static inline void pq_update(struct hfi1_user_sdma_pkt_q *pq)
{
	if (atomic_dec_and_test(&pq->n_reqs)) {
		xchg(&pq->state, SDMA_PKT_Q_INACTIVE);
		wake_up(&pq->wait);
	}
}

static void user_sdma_free_request(struct user_sdma_request *req, bool unpin)
{
	if (!list_empty(&req->txps)) {
		struct sdma_txreq *t, *p;

		list_for_each_entry_safe(t, p, &req->txps, list) {
			struct user_sdma_txreq *tx =
				container_of(t, struct user_sdma_txreq, txreq);
			list_del_init(&t->list);
			sdma_txclean(req->pq->dd, t);
			kmem_cache_free(req->pq->txreq_cache, tx);
		}
	}
	if (req->data_iovs) {
		struct sdma_mmu_node *node;
		int i;

		for (i = 0; i < req->data_iovs; i++) {
			node = req->iovs[i].node;
			if (!node)
				continue;

			if (unpin)
				hfi1_mmu_rb_remove(&req->pq->sdma_rb_root,
						   &node->rb);
			else
				atomic_dec(&node->refcount);
		}
	}
	kfree(req->tids);
	clear_bit(SDMA_REQ_IN_USE, &req->flags);
}

static inline void set_comp_state(struct hfi1_user_sdma_pkt_q *pq,
				  struct hfi1_user_sdma_comp_q *cq,
				  u16 idx, enum hfi1_sdma_comp_state state,
				  int ret)
{
	hfi1_cdbg(SDMA, "[%u:%u:%u:%u] Setting completion status %u %d",
		  pq->dd->unit, pq->ctxt, pq->subctxt, idx, state, ret);
	cq->comps[idx].status = state;
	if (state == ERROR)
		cq->comps[idx].errcode = -ret;
	trace_hfi1_sdma_user_completion(pq->dd, pq->ctxt, pq->subctxt,
					idx, state, ret);
}

static bool sdma_rb_filter(struct mmu_rb_node *node, unsigned long addr,
			   unsigned long len)
{
	return (bool)(node->addr == addr);
}

static int sdma_rb_insert(struct rb_root *root, struct mmu_rb_node *mnode)
{
	struct sdma_mmu_node *node =
		container_of(mnode, struct sdma_mmu_node, rb);

	atomic_inc(&node->refcount);
	return 0;
}

static void sdma_rb_remove(struct rb_root *root, struct mmu_rb_node *mnode,
			   struct mm_struct *mm)
{
	struct sdma_mmu_node *node =
		container_of(mnode, struct sdma_mmu_node, rb);

	spin_lock(&node->pq->evict_lock);
	/*
	 * We've been called by the MMU notifier but this node has been
	 * scheduled for eviction. The eviction function will take care
	 * of freeing this node.
	 * We have to take the above lock first because we are racing
	 * against the setting of the bit in the eviction function.
	 */
	if (mm && test_bit(SDMA_CACHE_NODE_EVICT, &node->flags)) {
		spin_unlock(&node->pq->evict_lock);
		return;
	}

	if (!list_empty(&node->list))
		list_del(&node->list);
	node->pq->n_locked -= node->npages;
	spin_unlock(&node->pq->evict_lock);

	/*
	 * If mm is set, we are being called by the MMU notifier and we
	 * should not pass a mm_struct to unpin_vector_page(). This is to
	 * prevent a deadlock when hfi1_release_user_pages() attempts to
	 * take the mmap_sem, which the MMU notifier has already taken.
	 */
	unpin_vector_pages(mm ? NULL : current->mm, node->pages, 0,
			   node->npages);
	/*
	 * If called by the MMU notifier, we have to adjust the pinned
	 * page count ourselves.
	 */
	if (mm)
		mm->pinned_vm -= node->npages;
	kfree(node);
}

static int sdma_rb_invalidate(struct rb_root *root, struct mmu_rb_node *mnode)
{
	struct sdma_mmu_node *node =
		container_of(mnode, struct sdma_mmu_node, rb);

	if (!atomic_read(&node->refcount))
		return 1;
	return 0;
}