summaryrefslogtreecommitdiffstats
path: root/drivers/misc/mic/scif/scif_api.c
blob: 8dd0ccedeb94a304bb5b8217a9c834711c60af77 (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
/*
 * Intel MIC Platform Software Stack (MPSS)
 *
 * Copyright(c) 2014 Intel Corporation.
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License, version 2, as
 * published by the Free Software Foundation.
 *
 * 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.
 *
 * Intel SCIF driver.
 *
 */
#include <linux/scif.h>
#include "scif_main.h"
#include "scif_map.h"

static const char * const scif_ep_states[] = {
	"Unbound",
	"Bound",
	"Listening",
	"Connected",
	"Connecting",
	"Mapping",
	"Closing",
	"Close Listening",
	"Disconnected",
	"Zombie"};

enum conn_async_state {
	ASYNC_CONN_IDLE = 1,	/* ep setup for async connect */
	ASYNC_CONN_INPROGRESS,	/* async connect in progress */
	ASYNC_CONN_FLUSH_WORK	/* async work flush in progress  */
};

/*
 * File operations for anonymous inode file associated with a SCIF endpoint,
 * used in kernel mode SCIF poll. Kernel mode SCIF poll calls portions of the
 * poll API in the kernel and these take in a struct file *. Since a struct
 * file is not available to kernel mode SCIF, it uses an anonymous file for
 * this purpose.
 */
const struct file_operations scif_anon_fops = {
	.owner = THIS_MODULE,
};

scif_epd_t scif_open(void)
{
	struct scif_endpt *ep;
	int err;

	might_sleep();
	ep = kzalloc(sizeof(*ep), GFP_KERNEL);
	if (!ep)
		goto err_ep_alloc;

	ep->qp_info.qp = kzalloc(sizeof(*ep->qp_info.qp), GFP_KERNEL);
	if (!ep->qp_info.qp)
		goto err_qp_alloc;

	err = scif_anon_inode_getfile(ep);
	if (err)
		goto err_anon_inode;

	spin_lock_init(&ep->lock);
	mutex_init(&ep->sendlock);
	mutex_init(&ep->recvlock);

	scif_rma_ep_init(ep);
	ep->state = SCIFEP_UNBOUND;
	dev_dbg(scif_info.mdev.this_device,
		"SCIFAPI open: ep %p success\n", ep);
	return ep;

err_anon_inode:
	kfree(ep->qp_info.qp);
err_qp_alloc:
	kfree(ep);
err_ep_alloc:
	return NULL;
}
EXPORT_SYMBOL_GPL(scif_open);

/*
 * scif_disconnect_ep - Disconnects the endpoint if found
 * @epd: The end point returned from scif_open()
 */
static struct scif_endpt *scif_disconnect_ep(struct scif_endpt *ep)
{
	struct scifmsg msg;
	struct scif_endpt *fep = NULL;
	struct scif_endpt *tmpep;
	struct list_head *pos, *tmpq;
	int err;

	/*
	 * Wake up any threads blocked in send()/recv() before closing
	 * out the connection. Grabbing and releasing the send/recv lock
	 * will ensure that any blocked senders/receivers have exited for
	 * Ring 0 endpoints. It is a Ring 0 bug to call send/recv after
	 * close. Ring 3 endpoints are not affected since close will not
	 * be called while there are IOCTLs executing.
	 */
	wake_up_interruptible(&ep->sendwq);
	wake_up_interruptible(&ep->recvwq);
	mutex_lock(&ep->sendlock);
	mutex_unlock(&ep->sendlock);
	mutex_lock(&ep->recvlock);
	mutex_unlock(&ep->recvlock);

	/* Remove from the connected list */
	mutex_lock(&scif_info.connlock);
	list_for_each_safe(pos, tmpq, &scif_info.connected) {
		tmpep = list_entry(pos, struct scif_endpt, list);
		if (tmpep == ep) {
			list_del(pos);
			fep = tmpep;
			spin_lock(&ep->lock);
			break;
		}
	}

	if (!fep) {
		/*
		 * The other side has completed the disconnect before
		 * the end point can be removed from the list. Therefore
		 * the ep lock is not locked, traverse the disconnected
		 * list to find the endpoint and release the conn lock.
		 */
		list_for_each_safe(pos, tmpq, &scif_info.disconnected) {
			tmpep = list_entry(pos, struct scif_endpt, list);
			if (tmpep == ep) {
				list_del(pos);
				break;
			}
		}
		mutex_unlock(&scif_info.connlock);
		return NULL;
	}

	init_completion(&ep->discon);
	msg.uop = SCIF_DISCNCT;
	msg.src = ep->port;
	msg.dst = ep->peer;
	msg.payload[0] = (u64)ep;
	msg.payload[1] = ep->remote_ep;

	err = scif_nodeqp_send(ep->remote_dev, &msg);
	spin_unlock(&ep->lock);
	mutex_unlock(&scif_info.connlock);

	if (!err)
		/* Wait for the remote node to respond with SCIF_DISCNT_ACK */
		wait_for_completion_timeout(&ep->discon,
					    SCIF_NODE_ALIVE_TIMEOUT);
	return ep;
}

int scif_close(scif_epd_t epd)
{
	struct scif_endpt *ep = (struct scif_endpt *)epd;
	struct scif_endpt *tmpep;
	struct list_head *pos, *tmpq;
	enum scif_epd_state oldstate;
	bool flush_conn;

	dev_dbg(scif_info.mdev.this_device, "SCIFAPI close: ep %p %s\n",
		ep, scif_ep_states[ep->state]);
	might_sleep();
	spin_lock(&ep->lock);
	flush_conn = (ep->conn_async_state == ASYNC_CONN_INPROGRESS);
	spin_unlock(&ep->lock);

	if (flush_conn)
		flush_work(&scif_info.conn_work);

	spin_lock(&ep->lock);
	oldstate = ep->state;

	ep->state = SCIFEP_CLOSING;

	switch (oldstate) {
	case SCIFEP_ZOMBIE:
		dev_err(scif_info.mdev.this_device,
			"SCIFAPI close: zombie state unexpected\n");
		/* fall through */
	case SCIFEP_DISCONNECTED:
		spin_unlock(&ep->lock);
		scif_unregister_all_windows(epd);
		/* Remove from the disconnected list */
		mutex_lock(&scif_info.connlock);
		list_for_each_safe(pos, tmpq, &scif_info.disconnected) {
			tmpep = list_entry(pos, struct scif_endpt, list);
			if (tmpep == ep) {
				list_del(pos);
				break;
			}
		}
		mutex_unlock(&scif_info.connlock);
		break;
	case SCIFEP_UNBOUND:
	case SCIFEP_BOUND:
	case SCIFEP_CONNECTING:
		spin_unlock(&ep->lock);
		break;
	case SCIFEP_MAPPING:
	case SCIFEP_CONNECTED:
	case SCIFEP_CLOSING:
	{
		spin_unlock(&ep->lock);
		scif_unregister_all_windows(epd);
		scif_disconnect_ep(ep);
		break;
	}
	case SCIFEP_LISTENING:
	case SCIFEP_CLLISTEN:
	{
		struct scif_conreq *conreq;
		struct scifmsg msg;
		struct scif_endpt *aep;

		spin_unlock(&ep->lock);
		mutex_lock(&scif_info.eplock);

		/* remove from listen list */
		list_for_each_safe(pos, tmpq, &scif_info.listen) {
			tmpep = list_entry(pos, struct scif_endpt, list);
			if (tmpep == ep)
				list_del(pos);
		}
		/* Remove any dangling accepts */
		while (ep->acceptcnt) {
			aep = list_first_entry(&ep->li_accept,
					       struct scif_endpt, liacceptlist);
			list_del(&aep->liacceptlist);
			scif_put_port(aep->port.port);
			list_for_each_safe(pos, tmpq, &scif_info.uaccept) {
				tmpep = list_entry(pos, struct scif_endpt,
						   miacceptlist);
				if (tmpep == aep) {
					list_del(pos);
					break;
				}
			}
			mutex_unlock(&scif_info.eplock);
			mutex_lock(&scif_info.connlock);
			list_for_each_safe(pos, tmpq, &scif_info.connected) {
				tmpep = list_entry(pos,
						   struct scif_endpt, list);
				if (tmpep == aep) {
					list_del(pos);
					break;
				}
			}
			list_for_each_safe(pos, tmpq, &scif_info.disconnected) {
				tmpep = list_entry(pos,
						   struct scif_endpt, list);
				if (tmpep == aep) {
					list_del(pos);
					break;
				}
			}
			mutex_unlock(&scif_info.connlock);
			scif_teardown_ep(aep);
			mutex_lock(&scif_info.eplock);
			scif_add_epd_to_zombie_list(aep, SCIF_EPLOCK_HELD);
			ep->acceptcnt--;
		}

		spin_lock(&ep->lock);
		mutex_unlock(&scif_info.eplock);

		/* Remove and reject any pending connection requests. */
		while (ep->conreqcnt) {
			conreq = list_first_entry(&ep->conlist,
						  struct scif_conreq, list);
			list_del(&conreq->list);

			msg.uop = SCIF_CNCT_REJ;
			msg.dst.node = conreq->msg.src.node;
			msg.dst.port = conreq->msg.src.port;
			msg.payload[0] = conreq->msg.payload[0];
			msg.payload[1] = conreq->msg.payload[1];
			/*
			 * No Error Handling on purpose for scif_nodeqp_send().
			 * If the remote node is lost we still want free the
			 * connection requests on the self node.
			 */
			scif_nodeqp_send(&scif_dev[conreq->msg.src.node],
					 &msg);
			ep->conreqcnt--;
			kfree(conreq);
		}

		spin_unlock(&ep->lock);
		/* If a kSCIF accept is waiting wake it up */
		wake_up_interruptible(&ep->conwq);
		break;
	}
	}
	scif_put_port(ep->port.port);
	scif_anon_inode_fput(ep);
	scif_teardown_ep(ep);
	scif_add_epd_to_zombie_list(ep, !SCIF_EPLOCK_HELD);
	return 0;
}
EXPORT_SYMBOL_GPL(scif_close);

/**
 * scif_flush() - Wakes up any blocking accepts. The endpoint will no longer
 *			accept new connections.
 * @epd: The end point returned from scif_open()
 */
int __scif_flush(scif_epd_t epd)
{
	struct scif_endpt *ep = (struct scif_endpt *)epd;

	switch (ep->state) {
	case SCIFEP_LISTENING:
	{
		ep->state = SCIFEP_CLLISTEN;

		/* If an accept is waiting wake it up */
		wake_up_interruptible(&ep->conwq);
		break;
	}
	default:
		break;
	}
	return 0;
}

int scif_bind(scif_epd_t epd, u16 pn)
{
	struct scif_endpt *ep = (struct scif_endpt *)epd;
	int ret = 0;
	int tmp;

	dev_dbg(scif_info.mdev.this_device,
		"SCIFAPI bind: ep %p %s requested port number %d\n",
		ep, scif_ep_states[ep->state], pn);
	if (pn) {
		/*
		 * Similar to IETF RFC 1700, SCIF ports below
		 * SCIF_ADMIN_PORT_END can only be bound by system (or root)
		 * processes or by processes executed by privileged users.
		 */
		if (pn < SCIF_ADMIN_PORT_END && !capable(CAP_SYS_ADMIN)) {
			ret = -EACCES;
			goto scif_bind_admin_exit;
		}
	}

	spin_lock(&ep->lock);
	if (ep->state == SCIFEP_BOUND) {
		ret = -EINVAL;
		goto scif_bind_exit;
	} else if (ep->state != SCIFEP_UNBOUND) {
		ret = -EISCONN;
		goto scif_bind_exit;
	}

	if (pn) {
		tmp = scif_rsrv_port(pn);
		if (tmp != pn) {
			ret = -EINVAL;
			goto scif_bind_exit;
		}
	} else {
		ret = scif_get_new_port();
		if (ret < 0)
			goto scif_bind_exit;
		pn = ret;
	}

	ep->state = SCIFEP_BOUND;
	ep->port.node = scif_info.nodeid;
	ep->port.port = pn;
	ep->conn_async_state = ASYNC_CONN_IDLE;
	ret = pn;
	dev_dbg(scif_info.mdev.this_device,
		"SCIFAPI bind: bound to port number %d\n", pn);
scif_bind_exit:
	spin_unlock(&ep->lock);
scif_bind_admin_exit:
	return ret;
}
EXPORT_SYMBOL_GPL(scif_bind);

int scif_listen(scif_epd_t epd, int backlog)
{
	struct scif_endpt *ep = (struct scif_endpt *)epd;

	dev_dbg(scif_info.mdev.this_device,
		"SCIFAPI listen: ep %p %s\n", ep, scif_ep_states[ep->state]);
	spin_lock(&ep->lock);
	switch (ep->state) {
	case SCIFEP_ZOMBIE:
	case SCIFEP_CLOSING:
	case SCIFEP_CLLISTEN:
	case SCIFEP_UNBOUND:
	case SCIFEP_DISCONNECTED:
		spin_unlock(&ep->lock);
		return -EINVAL;
	case SCIFEP_LISTENING:
	case SCIFEP_CONNECTED:
	case SCIFEP_CONNECTING:
	case SCIFEP_MAPPING:
		spin_unlock(&ep->lock);
		return -EISCONN;
	case SCIFEP_BOUND:
		break;
	}

	ep->state = SCIFEP_LISTENING;
	ep->backlog = backlog;

	ep->conreqcnt = 0;
	ep->acceptcnt = 0;
	INIT_LIST_HEAD(&ep->conlist);
	init_waitqueue_head(&ep->conwq);
	INIT_LIST_HEAD(&ep->li_accept);
	spin_unlock(&ep->lock);

	/*
	 * Listen status is complete so delete the qp information not needed
	 * on a listen before placing on the list of listening ep's
	 */
	scif_teardown_ep(ep);
	ep->qp_info.qp = NULL;

	mutex_lock(&scif_info.eplock);
	list_add_tail(&ep->list, &scif_info.listen);
	mutex_unlock(&scif_info.eplock);
	return 0;
}
EXPORT_SYMBOL_GPL(scif_listen);

/*
 ************************************************************************
 * SCIF connection flow:
 *
 * 1) A SCIF listening endpoint can call scif_accept(..) to wait for SCIF
 *	connections via a SCIF_CNCT_REQ message
 * 2) A SCIF endpoint can initiate a SCIF connection by calling
 *	scif_connect(..) which calls scif_setup_qp_connect(..) which
 *	allocates the local qp for the endpoint ring buffer and then sends
 *	a SCIF_CNCT_REQ to the remote node and waits for a SCIF_CNCT_GNT or
 *	a SCIF_CNCT_REJ message
 * 3) The peer node handles a SCIF_CNCT_REQ via scif_cnctreq_resp(..) which
 *	wakes up any threads blocked in step 1 or sends a SCIF_CNCT_REJ
 *	message otherwise
 * 4) A thread blocked waiting for incoming connections allocates its local
 *	endpoint QP and ring buffer following which it sends a SCIF_CNCT_GNT
 *	and waits for a SCIF_CNCT_GNT(N)ACK. If the allocation fails then
 *	the node sends a SCIF_CNCT_REJ message
 * 5) Upon receipt of a SCIF_CNCT_GNT or a SCIF_CNCT_REJ message the
 *	connecting endpoint is woken up as part of handling
 *	scif_cnctgnt_resp(..) following which it maps the remote endpoints'
 *	QP, updates its outbound QP and sends a SCIF_CNCT_GNTACK message on
 *	success or a SCIF_CNCT_GNTNACK message on failure and completes
 *	the scif_connect(..) API
 * 6) Upon receipt of a SCIF_CNCT_GNT(N)ACK the accepting endpoint blocked
 *	in step 4 is woken up and completes the scif_accept(..) API
 * 7) The SCIF connection is now established between the two SCIF endpoints.
 */
static int scif_conn_func(struct scif_endpt *ep)
{
	int err = 0;
	struct scifmsg msg;
	struct device *spdev;

	err = scif_reserve_dma_chan(ep);
	if (err) {
		dev_err(&ep->remote_dev->sdev->dev,
			"%s %d err %d\n", __func__, __LINE__, err);
		ep->state = SCIFEP_BOUND;
		goto connect_error_simple;
	}
	/* Initiate the first part of the endpoint QP setup */
	err = scif_setup_qp_connect(ep->qp_info.qp, &ep->qp_info.qp_offset,
				    SCIF_ENDPT_QP_SIZE, ep->remote_dev);
	if (err) {
		dev_err(&ep->remote_dev->sdev->dev,
			"%s err %d qp_offset 0x%llx\n",
			__func__, err, ep->qp_info.qp_offset);
		ep->state = SCIFEP_BOUND;
		goto connect_error_simple;
	}

	spdev = scif_get_peer_dev(ep->remote_dev);
	if (IS_ERR(spdev)) {
		err = PTR_ERR(spdev);
		goto cleanup_qp;
	}
	/* Format connect message and send it */
	msg.src = ep->port;
	msg.dst = ep->conn_port;
	msg.uop = SCIF_CNCT_REQ;
	msg.payload[0] = (u64)ep;
	msg.payload[1] = ep->qp_info.qp_offset;
	err = _scif_nodeqp_send(ep->remote_dev, &msg);
	if (err)
		goto connect_error_dec;
	scif_put_peer_dev(spdev);
	/*
	 * Wait for the remote node to respond with SCIF_CNCT_GNT or
	 * SCIF_CNCT_REJ message.
	 */
	err = wait_event_timeout(ep->conwq, ep->state != SCIFEP_CONNECTING,
				 SCIF_NODE_ALIVE_TIMEOUT);
	if (!err) {
		dev_err(&ep->remote_dev->sdev->dev,
			"%s %d timeout\n", __func__, __LINE__);
		ep->state = SCIFEP_BOUND;
	}
	spdev = scif_get_peer_dev(ep->remote_dev);
	if (IS_ERR(spdev)) {
		err = PTR_ERR(spdev);
		goto cleanup_qp;
	}
	if (ep->state == SCIFEP_MAPPING) {
		err = scif_setup_qp_connect_response(ep->remote_dev,
						     ep->qp_info.qp,
						     ep->qp_info.gnt_pld);
		/*
		 * If the resource to map the queue are not available then
		 * we need to tell the other side to terminate the accept
		 */
		if (err) {
			dev_err(&ep->remote_dev->sdev->dev,
				"%s %d err %d\n", __func__, __LINE__, err);
			msg.uop = SCIF_CNCT_GNTNACK;
			msg.payload[0] = ep->remote_ep;
			_scif_nodeqp_send(ep->remote_dev, &msg);
			ep->state = SCIFEP_BOUND;
			goto connect_error_dec;
		}

		msg.uop = SCIF_CNCT_GNTACK;
		msg.payload[0] = ep->remote_ep;
		err = _scif_nodeqp_send(ep->remote_dev, &msg);
		if (err) {
			ep->state = SCIFEP_BOUND;
			goto connect_error_dec;
		}
		ep->state = SCIFEP_CONNECTED;
		mutex_lock(&scif_info.connlock);
		list_add_tail(&ep->list, &scif_info.connected);
		mutex_unlock(&scif_info.connlock);
		dev_dbg(&ep->remote_dev->sdev->dev,
			"SCIFAPI connect: ep %p connected\n", ep);
	} else if (ep->state == SCIFEP_BOUND) {
		dev_dbg(&ep->remote_dev->sdev->dev,
			"SCIFAPI connect: ep %p connection refused\n", ep);
		err = -ECONNREFUSED;
		goto connect_error_dec;
	}
	scif_put_peer_dev(spdev);
	return err;
connect_error_dec:
	scif_put_peer_dev(spdev);
cleanup_qp:
	scif_cleanup_ep_qp(ep);
connect_error_simple:
	return err;
}

/*
 * scif_conn_handler:
 *
 * Workqueue handler for servicing non-blocking SCIF connect
 *
 */
void scif_conn_handler(struct work_struct *work)
{
	struct scif_endpt *ep;

	do {
		ep = NULL;
		spin_lock(&scif_info.nb_connect_lock);
		if (!list_empty(&scif_info.nb_connect_list)) {
			ep = list_first_entry(&scif_info.nb_connect_list,
					      struct scif_endpt, conn_list);
			list_del(&ep->conn_list);
		}
		spin_unlock(&scif_info.nb_connect_lock);
		if (ep) {
			ep->conn_err = scif_conn_func(ep);
			wake_up_interruptible(&ep->conn_pend_wq);
		}
	} while (ep);
}

int __scif_connect(scif_epd_t epd, struct scif_port_id *dst, bool non_block)
{
	struct scif_endpt *ep = (struct scif_endpt *)epd;
	int err = 0;
	struct scif_dev *remote_dev;
	struct device *spdev;

	dev_dbg(scif_info.mdev.this_device, "SCIFAPI connect: ep %p %s\n", ep,
		scif_ep_states[ep->state]);

	if (!scif_dev || dst->node > scif_info.maxid)
		return -ENODEV;

	might_sleep();

	remote_dev = &scif_dev[dst->node];
	spdev = scif_get_peer_dev(remote_dev);
	if (IS_ERR(spdev)) {
		err = PTR_ERR(spdev);
		return err;
	}

	spin_lock(&ep->lock);
	switch (ep->state) {
	case SCIFEP_ZOMBIE:
	case SCIFEP_CLOSING:
		err = -EINVAL;
		break;
	case SCIFEP_DISCONNECTED:
		if (ep->conn_async_state == ASYNC_CONN_INPROGRESS)
			ep->conn_async_state = ASYNC_CONN_FLUSH_WORK;
		else
			err = -EINVAL;
		break;
	case SCIFEP_LISTENING:
	case SCIFEP_CLLISTEN:
		err = -EOPNOTSUPP;
		break;
	case SCIFEP_CONNECTING:
	case SCIFEP_MAPPING:
		if (ep->conn_async_state == ASYNC_CONN_INPROGRESS)
			err = -EINPROGRESS;
		else
			err = -EISCONN;
		break;
	case SCIFEP_CONNECTED:
		if (ep->conn_async_state == ASYNC_CONN_INPROGRESS)
			ep->conn_async_state = ASYNC_CONN_FLUSH_WORK;
		else
			err = -EISCONN;
		break;
	case SCIFEP_UNBOUND:
		err = scif_get_new_port();
		if (err < 0)
			break;
		ep->port.port = err;
		ep->port.node = scif_info.nodeid;
		ep->conn_async_state = ASYNC_CONN_IDLE;
		/* Fall through */
	case SCIFEP_BOUND:
		/*
		 * If a non-blocking connect has been already initiated
		 * (conn_async_state is either ASYNC_CONN_INPROGRESS or
		 * ASYNC_CONN_FLUSH_WORK), the end point could end up in
		 * SCIF_BOUND due an error in the connection process
		 * (e.g., connection refused) If conn_async_state is
		 * ASYNC_CONN_INPROGRESS - transition to ASYNC_CONN_FLUSH_WORK
		 * so that the error status can be collected. If the state is
		 * already ASYNC_CONN_FLUSH_WORK - then set the error to
		 * EINPROGRESS since some other thread is waiting to collect
		 * error status.
		 */
		if (ep->conn_async_state == ASYNC_CONN_INPROGRESS) {
			ep->conn_async_state = ASYNC_CONN_FLUSH_WORK;
		} else if (ep->conn_async_state == ASYNC_CONN_FLUSH_WORK) {
			err = -EINPROGRESS;
		} else {
			ep->conn_port = *dst;
			init_waitqueue_head(&ep->sendwq);
			init_waitqueue_head(&ep->recvwq);
			init_waitqueue_head(&ep->conwq);
			ep->conn_async_state = 0;

			if (unlikely(non_block))
				ep->conn_async_state = ASYNC_CONN_INPROGRESS;
		}
		break;
	}

	if (err || ep->conn_async_state == ASYNC_CONN_FLUSH_WORK)
			goto connect_simple_unlock1;

	ep->state = SCIFEP_CONNECTING;
	ep->remote_dev = &scif_dev[dst->node];
	ep->qp_info.qp->magic = SCIFEP_MAGIC;
	if (ep->conn_async_state == ASYNC_CONN_INPROGRESS) {
		init_waitqueue_head(&ep->conn_pend_wq);
		spin_lock(&scif_info.nb_connect_lock);
		list_add_tail(&ep->conn_list, &scif_info.nb_connect_list);
		spin_unlock(&scif_info.nb_connect_lock);
		err = -EINPROGRESS;
		schedule_work(&scif_info.conn_work);
	}
connect_simple_unlock1:
	spin_unlock(&ep->lock);
	scif_put_peer_dev(spdev);
	if (err) {
		return err;
	} else if (ep->conn_async_state == ASYNC_CONN_FLUSH_WORK) {
		flush_work(&scif_info.conn_work);
		err = ep->conn_err;
		spin_lock(&ep->lock);
		ep->conn_async_state = ASYNC_CONN_IDLE;
		spin_unlock(&ep->lock);
	} else {
		err = scif_conn_func(ep);
	}
	return err;
}

int scif_connect(scif_epd_t epd, struct scif_port_id *dst)
{
	return __scif_connect(epd, dst, false);
}
EXPORT_SYMBOL_GPL(scif_connect);

/**
 * scif_accept() - Accept a connection request from the remote node
 *
 * The function accepts a connection request from the remote node.  Successful
 * complete is indicate by a new end point being created and passed back
 * to the caller for future reference.
 *
 * Upon successful complete a zero will be returned and the peer information
 * will be filled in.
 *
 * If the end point is not in the listening state -EINVAL will be returned.
 *
 * If during the connection sequence resource allocation fails the -ENOMEM
 * will be returned.
 *
 * If the function is called with the ASYNC flag set and no connection requests
 * are pending it will return -EAGAIN.
 *
 * If the remote side is not sending any connection requests the caller may
 * terminate this function with a signal.  If so a -EINTR will be returned.
 */
int scif_accept(scif_epd_t epd, struct scif_port_id *peer,
		scif_epd_t *newepd, int flags)
{
	struct scif_endpt *lep = (struct scif_endpt *)epd;
	struct scif_endpt *cep;
	struct scif_conreq *conreq;
	struct scifmsg msg;
	int err;
	struct device *spdev;

	dev_dbg(scif_info.mdev.this_device,
		"SCIFAPI accept: ep %p %s\n", lep, scif_ep_states[lep->state]);

	if (flags & ~SCIF_ACCEPT_SYNC)
		return -EINVAL;

	if (!peer || !newepd)
		return -EINVAL;

	might_sleep();
	spin_lock(&lep->lock);
	if (lep->state != SCIFEP_LISTENING) {
		spin_unlock(&lep->lock);
		return -EINVAL;
	}

	if (!lep->conreqcnt && !(flags & SCIF_ACCEPT_SYNC)) {
		/* No connection request present and we do not want to wait */
		spin_unlock(&lep->lock);
		return -EAGAIN;
	}

	lep->files = current->files;
retry_connection:
	spin_unlock(&lep->lock);
	/* Wait for the remote node to send us a SCIF_CNCT_REQ */
	err = wait_event_interruptible(lep->conwq,
				       (lep->conreqcnt ||
				       (lep->state != SCIFEP_LISTENING)));
	if (err)
		return err;

	if (lep->state != SCIFEP_LISTENING)
		return -EINTR;

	spin_lock(&lep->lock);

	if (!lep->conreqcnt)
		goto retry_connection;

	/* Get the first connect request off the list */
	conreq = list_first_entry(&lep->conlist, struct scif_conreq, list);
	list_del(&conreq->list);
	lep->conreqcnt--;
	spin_unlock(&lep->lock);

	/* Fill in the peer information */
	peer->node = conreq->msg.src.node;
	peer->port = conreq->msg.src.port;

	cep = kzalloc(sizeof(*cep), GFP_KERNEL);
	if (!cep) {
		err = -ENOMEM;
		goto scif_accept_error_epalloc;
	}
	spin_lock_init(&cep->lock);
	mutex_init(&cep->sendlock);
	mutex_init(&cep->recvlock);
	cep->state = SCIFEP_CONNECTING;
	cep->remote_dev = &scif_dev[peer->node];
	cep->remote_ep = conreq->msg.payload[0];

	scif_rma_ep_init(cep);

	err = scif_reserve_dma_chan(cep);
	if (err) {
		dev_err(scif_info.mdev.this_device,
			"%s %d err %d\n", __func__, __LINE__, err);
		goto scif_accept_error_qpalloc;
	}

	cep->qp_info.qp = kzalloc(sizeof(*cep->qp_info.qp), GFP_KERNEL);
	if (!cep->qp_info.qp) {
		err = -ENOMEM;
		goto scif_accept_error_qpalloc;
	}

	err = scif_anon_inode_getfile(cep);
	if (err)
		goto scif_accept_error_anon_inode;

	cep->qp_info.qp->magic = SCIFEP_MAGIC;
	spdev = scif_get_peer_dev(cep->remote_dev);
	if (IS_ERR(spdev)) {
		err = PTR_ERR(spdev);
		goto scif_accept_error_map;
	}
	err = scif_setup_qp_accept(cep->qp_info.qp, &cep->qp_info.qp_offset,
				   conreq->msg.payload[1], SCIF_ENDPT_QP_SIZE,
				   cep->remote_dev);
	if (err) {
		dev_dbg(&cep->remote_dev->sdev->dev,
			"SCIFAPI accept: ep %p new %p scif_setup_qp_accept %d qp_offset 0x%llx\n",
			lep, cep, err, cep->qp_info.qp_offset);
		scif_put_peer_dev(spdev);
		goto scif_accept_error_map;
	}

	cep->port.node = lep->port.node;
	cep->port.port = lep->port.port;
	cep->peer.node = peer->node;
	cep->peer.port = peer->port;
	init_waitqueue_head(&cep->sendwq);
	init_waitqueue_head(&cep->recvwq);
	init_waitqueue_head(&cep->conwq);

	msg.uop = SCIF_CNCT_GNT;
	msg.src = cep->port;
	msg.payload[0] = cep->remote_ep;
	msg.payload[1] = cep->qp_info.qp_offset;
	msg.payload[2] = (u64)cep;

	err = _scif_nodeqp_send(cep->remote_dev, &msg);
	scif_put_peer_dev(spdev);
	if (err)
		goto scif_accept_error_map;
retry:
	/* Wait for the remote node to respond with SCIF_CNCT_GNT(N)ACK */
	err = wait_event_timeout(cep->conwq, cep->state != SCIFEP_CONNECTING,
				 SCIF_NODE_ACCEPT_TIMEOUT);
	if (!err && scifdev_alive(cep))
		goto retry;
	err = !err ? -ENODEV : 0;
	if (err)
		goto scif_accept_error_map;
	kfree(conreq);

	spin_lock(&cep->lock);

	if (cep->state == SCIFEP_CLOSING) {
		/*
		 * Remote failed to allocate resources and NAKed the grant.
		 * There is at this point nothing referencing the new end point.
		 */
		spin_unlock(&cep->lock);
		scif_teardown_ep(cep);
		kfree(cep);

		/* If call with sync flag then go back and wait. */
		if (flags & SCIF_ACCEPT_SYNC) {
			spin_lock(&lep->lock);
			goto retry_connection;
		}
		return -EAGAIN;
	}

	scif_get_port(cep->port.port);
	*newepd = (scif_epd_t)cep;
	spin_unlock(&cep->lock);
	return 0;
scif_accept_error_map:
	scif_anon_inode_fput(cep);
scif_accept_error_anon_inode:
	scif_teardown_ep(cep);
scif_accept_error_qpalloc:
	kfree(cep);
scif_accept_error_epalloc:
	msg.uop = SCIF_CNCT_REJ;
	msg.dst.node = conreq->msg.src.node;
	msg.dst.port = conreq->msg.src.port;
	msg.payload[0] = conreq->msg.payload[0];
	msg.payload[1] = conreq->msg.payload[1];
	scif_nodeqp_send(&scif_dev[conreq->msg.src.node], &msg);
	kfree(conreq);
	return err;
}
EXPORT_SYMBOL_GPL(scif_accept);

/*
 * scif_msg_param_check:
 * @epd: The end point returned from scif_open()
 * @len: Length to receive
 * @flags: blocking or non blocking
 *
 * Validate parameters for messaging APIs scif_send(..)/scif_recv(..).
 */
static inline int scif_msg_param_check(scif_epd_t epd, int len, int flags)
{
	int ret = -EINVAL;

	if (len < 0)
		goto err_ret;
	if (flags && (!(flags & SCIF_RECV_BLOCK)))
		goto err_ret;
	ret = 0;
err_ret:
	return ret;
}

static int _scif_send(scif_epd_t epd, void *msg, int len, int flags)
{
	struct scif_endpt *ep = (struct scif_endpt *)epd;
	struct scifmsg notif_msg;
	int curr_xfer_len = 0, sent_len = 0, write_count;
	int ret = 0;
	struct scif_qp *qp = ep->qp_info.qp;

	if (flags & SCIF_SEND_BLOCK)
		might_sleep();

	spin_lock(&ep->lock);
	while (sent_len != len && SCIFEP_CONNECTED == ep->state) {
		write_count = scif_rb_space(&qp->outbound_q);
		if (write_count) {
			/* Best effort to send as much data as possible */
			curr_xfer_len = min(len - sent_len, write_count);
			ret = scif_rb_write(&qp->outbound_q, msg,
					    curr_xfer_len);
			if (ret < 0)
				break;
			/* Success. Update write pointer */
			scif_rb_commit(&qp->outbound_q);
			/*
			 * Send a notification to the peer about the
			 * produced data message.
			 */
			notif_msg.src = ep->port;
			notif_msg.uop = SCIF_CLIENT_SENT;
			notif_msg.payload[0] = ep->remote_ep;
			ret = _scif_nodeqp_send(ep->remote_dev, &notif_msg);
			if (ret)
				break;
			sent_len += curr_xfer_len;
			msg = msg + curr_xfer_len;
			continue;
		}
		curr_xfer_len = min(len - sent_len, SCIF_ENDPT_QP_SIZE - 1);
		/* Not enough RB space. return for the Non Blocking case */
		if (!(flags & SCIF_SEND_BLOCK))
			break;

		spin_unlock(&ep->lock);
		/* Wait for a SCIF_CLIENT_RCVD message in the Blocking case */
		ret =
		wait_event_interruptible(ep->sendwq,
					 (SCIFEP_CONNECTED != ep->state) ||
					 (scif_rb_space(&qp->outbound_q) >=
					 curr_xfer_len));
		spin_lock(&ep->lock);
		if (ret)
			break;
	}
	if (sent_len)
		ret = sent_len;
	else if (!ret && SCIFEP_CONNECTED != ep->state)
		ret = SCIFEP_DISCONNECTED == ep->state ?
			-ECONNRESET : -ENOTCONN;
	spin_unlock(&ep->lock);
	return ret;
}

static int _scif_recv(scif_epd_t epd, void *msg, int len, int flags)
{
	int read_size;
	struct scif_endpt *ep = (struct scif_endpt *)epd;
	struct scifmsg notif_msg;
	int curr_recv_len = 0, remaining_len = len, read_count;
	int ret = 0;
	struct scif_qp *qp = ep->qp_info.qp;

	if (flags & SCIF_RECV_BLOCK)
		might_sleep();
	spin_lock(&ep->lock);
	while (remaining_len && (SCIFEP_CONNECTED == ep->state ||
				 SCIFEP_DISCONNECTED == ep->state)) {
		read_count = scif_rb_count(&qp->inbound_q, remaining_len);
		if (read_count) {
			/*
			 * Best effort to recv as much data as there
			 * are bytes to read in the RB particularly
			 * important for the Non Blocking case.
			 */
			curr_recv_len = min(remaining_len, read_count);
			read_size = scif_rb_get_next(&qp->inbound_q,
						     msg, curr_recv_len);
			if (ep->state == SCIFEP_CONNECTED) {
				/*
				 * Update the read pointer only if the endpoint
				 * is still connected else the read pointer
				 * might no longer exist since the peer has
				 * freed resources!
				 */
				scif_rb_update_read_ptr(&qp->inbound_q);
				/*
				 * Send a notification to the peer about the
				 * consumed data message only if the EP is in
				 * SCIFEP_CONNECTED state.
				 */
				notif_msg.src = ep->port;
				notif_msg.uop = SCIF_CLIENT_RCVD;
				notif_msg.payload[0] = ep->remote_ep;
				ret = _scif_nodeqp_send(ep->remote_dev,
							&notif_msg);
				if (ret)
					break;
			}
			remaining_len -= curr_recv_len;
			msg = msg + curr_recv_len;
			continue;
		}
		/*
		 * Bail out now if the EP is in SCIFEP_DISCONNECTED state else
		 * we will keep looping forever.
		 */
		if (ep->state == SCIFEP_DISCONNECTED)
			break;
		/*
		 * Return in the Non Blocking case if there is no data
		 * to read in this iteration.
		 */
		if (!(flags & SCIF_RECV_BLOCK))
			break;
		curr_recv_len = min(remaining_len, SCIF_ENDPT_QP_SIZE - 1);
		spin_unlock(&ep->lock);
		/*
		 * Wait for a SCIF_CLIENT_SEND message in the blocking case
		 * or until other side disconnects.
		 */
		ret =
		wait_event_interruptible(ep->recvwq,
					 SCIFEP_CONNECTED != ep->state ||
					 scif_rb_count(&qp->inbound_q,
						       curr_recv_len)
					 >= curr_recv_len);
		spin_lock(&ep->lock);
		if (ret)
			break;
	}
	if (len - remaining_len)
		ret = len - remaining_len;
	else if (!ret && ep->state != SCIFEP_CONNECTED)
		ret = ep->state == SCIFEP_DISCONNECTED ?
			-ECONNRESET : -ENOTCONN;
	spin_unlock(&ep->lock);
	return ret;
}

/**
 * scif_user_send() - Send data to connection queue
 * @epd: The end point returned from scif_open()
 * @msg: Address to place data
 * @len: Length to receive
 * @flags: blocking or non blocking
 *
 * This function is called from the driver IOCTL entry point
 * only and is a wrapper for _scif_send().
 */
int scif_user_send(scif_epd_t epd, void __user *msg, int len, int flags)
{
	struct scif_endpt *ep = (struct scif_endpt *)epd;
	int err = 0;
	int sent_len = 0;
	char *tmp;
	int loop_len;
	int chunk_len = min(len, (1 << (MAX_ORDER + PAGE_SHIFT - 1)));

	dev_dbg(scif_info.mdev.this_device,
		"SCIFAPI send (U): ep %p %s\n", ep, scif_ep_states[ep->state]);
	if (!len)
		return 0;

	err = scif_msg_param_check(epd, len, flags);
	if (err)
		goto send_err;

	tmp = kmalloc(chunk_len, GFP_KERNEL);
	if (!tmp) {
		err = -ENOMEM;
		goto send_err;
	}
	/*
	 * Grabbing the lock before breaking up the transfer in
	 * multiple chunks is required to ensure that messages do
	 * not get fragmented and reordered.
	 */
	mutex_lock(&ep->sendlock);
	while (sent_len != len) {
		loop_len = len - sent_len;
		loop_len = min(chunk_len, loop_len);
		if (copy_from_user(tmp, msg, loop_len)) {
			err = -EFAULT;
			goto send_free_err;
		}
		err = _scif_send(epd, tmp, loop_len, flags);
		if (err < 0)
			goto send_free_err;
		sent_len += err;
		msg += err;
		if (err != loop_len)
			goto send_free_err;
	}
send_free_err:
	mutex_unlock(&ep->sendlock);
	kfree(tmp);
send_err:
	return err < 0 ? err : sent_len;
}

/**
 * scif_user_recv() - Receive data from connection queue
 * @epd: The end point returned from scif_open()
 * @msg: Address to place data
 * @len: Length to receive
 * @flags: blocking or non blocking
 *
 * This function is called from the driver IOCTL entry point
 * only and is a wrapper for _scif_recv().
 */
int scif_user_recv(scif_epd_t epd, void __user *msg, int len, int flags)
{
	struct scif_endpt *ep = (struct scif_endpt *)epd;
	int err = 0;
	int recv_len = 0;
	char *tmp;
	int loop_len;
	int chunk_len = min(len, (1 << (MAX_ORDER + PAGE_SHIFT - 1)));

	dev_dbg(scif_info.mdev.this_device,
		"SCIFAPI recv (U): ep %p %s\n", ep, scif_ep_states[ep->state]);
	if (!len)
		return 0;

	err = scif_msg_param_check(epd, len, flags);
	if (err)
		goto recv_err;

	tmp = kmalloc(chunk_len, GFP_KERNEL);
	if (!tmp) {
		err = -ENOMEM;
		goto recv_err;
	}
	/*
	 * Grabbing the lock before breaking up the transfer in
	 * multiple chunks is required to ensure that messages do
	 * not get fragmented and reordered.
	 */
	mutex_lock(&ep->recvlock);
	while (recv_len != len) {
		loop_len = len - recv_len;
		loop_len = min(chunk_len, loop_len);
		err = _scif_recv(epd, tmp, loop_len, flags);
		if (err < 0)
			goto recv_free_err;
		if (copy_to_user(msg, tmp, err)) {
			err = -EFAULT;
			goto recv_free_err;
		}
		recv_len += err;
		msg += err;
		if (err != loop_len)
			goto recv_free_err;
	}
recv_free_err:
	mutex_unlock(&ep->recvlock);
	kfree(tmp);
recv_err:
	return err < 0 ? err : recv_len;
}

/**
 * scif_send() - Send data to connection queue
 * @epd: The end point returned from scif_open()
 * @msg: Address to place data
 * @len: Length to receive
 * @flags: blocking or non blocking
 *
 * This function is called from the kernel mode only and is
 * a wrapper for _scif_send().
 */
int scif_send(scif_epd_t epd, void *msg, int len, int flags)
{
	struct scif_endpt *ep = (struct scif_endpt *)epd;
	int ret;

	dev_dbg(scif_info.mdev.this_device,
		"SCIFAPI send (K): ep %p %s\n", ep, scif_ep_states[ep->state]);
	if (!len)
		return 0;

	ret = scif_msg_param_check(epd, len, flags);
	if (ret)
		return ret;
	if (!ep->remote_dev)
		return -ENOTCONN;
	/*
	 * Grab the mutex lock in the blocking case only
	 * to ensure messages do not get fragmented/reordered.
	 * The non blocking mode is protected using spin locks
	 * in _scif_send().
	 */
	if (flags & SCIF_SEND_BLOCK)
		mutex_lock(&ep->sendlock);

	ret = _scif_send(epd, msg, len, flags);

	if (flags & SCIF_SEND_BLOCK)
		mutex_unlock(&ep->sendlock);
	return ret;
}
EXPORT_SYMBOL_GPL(scif_send);

/**
 * scif_recv() - Receive data from connection queue
 * @epd: The end point returned from scif_open()
 * @msg: Address to place data
 * @len: Length to receive
 * @flags: blocking or non blocking
 *
 * This function is called from the kernel mode only and is
 * a wrapper for _scif_recv().
 */
int scif_recv(scif_epd_t epd, void *msg, int len, int flags)
{
	struct scif_endpt *ep = (struct scif_endpt *)epd;
	int ret;

	dev_dbg(scif_info.mdev.this_device,
		"SCIFAPI recv (K): ep %p %s\n", ep, scif_ep_states[ep->state]);
	if (!len)
		return 0;

	ret = scif_msg_param_check(epd, len, flags);
	if (ret)
		return ret;
	/*
	 * Grab the mutex lock in the blocking case only
	 * to ensure messages do not get fragmented/reordered.
	 * The non blocking mode is protected using spin locks
	 * in _scif_send().
	 */
	if (flags & SCIF_RECV_BLOCK)
		mutex_lock(&ep->recvlock);

	ret = _scif_recv(epd, msg, len, flags);

	if (flags & SCIF_RECV_BLOCK)
		mutex_unlock(&ep->recvlock);

	return ret;
}
EXPORT_SYMBOL_GPL(scif_recv);

static inline void _scif_poll_wait(struct file *f, wait_queue_head_t *wq,
				   poll_table *p, struct scif_endpt *ep)
{
	/*
	 * Because poll_wait makes a GFP_KERNEL allocation, give up the lock
	 * and regrab it afterwards. Because the endpoint state might have
	 * changed while the lock was given up, the state must be checked
	 * again after re-acquiring the lock. The code in __scif_pollfd(..)
	 * does this.
	 */
	spin_unlock(&ep->lock);
	poll_wait(f, wq, p);
	spin_lock(&ep->lock);
}

__poll_t
__scif_pollfd(struct file *f, poll_table *wait, struct scif_endpt *ep)
{
	__poll_t mask = 0;

	dev_dbg(scif_info.mdev.this_device,
		"SCIFAPI pollfd: ep %p %s\n", ep, scif_ep_states[ep->state]);

	spin_lock(&ep->lock);

	/* Endpoint is waiting for a non-blocking connect to complete */
	if (ep->conn_async_state == ASYNC_CONN_INPROGRESS) {
		_scif_poll_wait(f, &ep->conn_pend_wq, wait, ep);
		if (ep->conn_async_state == ASYNC_CONN_INPROGRESS) {
			if (ep->state == SCIFEP_CONNECTED ||
			    ep->state == SCIFEP_DISCONNECTED ||
			    ep->conn_err)
				mask |= EPOLLOUT;
			goto exit;
		}
	}

	/* Endpoint is listening for incoming connection requests */
	if (ep->state == SCIFEP_LISTENING) {
		_scif_poll_wait(f, &ep->conwq, wait, ep);
		if (ep->state == SCIFEP_LISTENING) {
			if (ep->conreqcnt)
				mask |= EPOLLIN;
			goto exit;
		}
	}

	/* Endpoint is connected or disconnected */
	if (ep->state == SCIFEP_CONNECTED || ep->state == SCIFEP_DISCONNECTED) {
		if (poll_requested_events(wait) & EPOLLIN)
			_scif_poll_wait(f, &ep->recvwq, wait, ep);
		if (poll_requested_events(wait) & EPOLLOUT)
			_scif_poll_wait(f, &ep->sendwq, wait, ep);
		if (ep->state == SCIFEP_CONNECTED ||
		    ep->state == SCIFEP_DISCONNECTED) {
			/* Data can be read without blocking */
			if (scif_rb_count(&ep->qp_info.qp->inbound_q, 1))
				mask |= EPOLLIN;
			/* Data can be written without blocking */
			if (scif_rb_space(&ep->qp_info.qp->outbound_q))
				mask |= EPOLLOUT;
			/* Return EPOLLHUP if endpoint is disconnected */
			if (ep->state == SCIFEP_DISCONNECTED)
				mask |= EPOLLHUP;
			goto exit;
		}
	}

	/* Return EPOLLERR if the endpoint is in none of the above states */
	mask |= EPOLLERR;
exit:
	spin_unlock(&ep->lock);
	return mask;
}

/**
 * scif_poll() - Kernel mode SCIF poll
 * @ufds: Array of scif_pollepd structures containing the end points
 *	  and events to poll on
 * @nfds: Size of the ufds array
 * @timeout_msecs: Timeout in msecs, -ve implies infinite timeout
 *
 * The code flow in this function is based on do_poll(..) in select.c
 *
 * Returns the number of endpoints which have pending events or 0 in
 * the event of a timeout. If a signal is used for wake up, -EINTR is
 * returned.
 */
int
scif_poll(struct scif_pollepd *ufds, unsigned int nfds, long timeout_msecs)
{
	struct poll_wqueues table;
	poll_table *pt;
	int i, count = 0, timed_out = timeout_msecs == 0;
	__poll_t mask;
	u64 timeout = timeout_msecs < 0 ? MAX_SCHEDULE_TIMEOUT
		: msecs_to_jiffies(timeout_msecs);

	poll_initwait(&table);
	pt = &table.pt;
	while (1) {
		for (i = 0; i < nfds; i++) {
			pt->_key = ufds[i].events | EPOLLERR | EPOLLHUP;
			mask = __scif_pollfd(ufds[i].epd->anon,
					     pt, ufds[i].epd);
			mask &= ufds[i].events | EPOLLERR | EPOLLHUP;
			if (mask) {
				count++;
				pt->_qproc = NULL;
			}
			ufds[i].revents = mask;
		}
		pt->_qproc = NULL;
		if (!count) {
			count = table.error;
			if (signal_pending(current))
				count = -EINTR;
		}
		if (count || timed_out)
			break;

		if (!schedule_timeout_interruptible(timeout))
			timed_out = 1;
	}
	poll_freewait(&table);
	return count;
}
EXPORT_SYMBOL_GPL(scif_poll);

int scif_get_node_ids(u16 *nodes, int len, u16 *self)
{
	int online = 0;
	int offset = 0;
	int node;

	if (!scif_is_mgmt_node())
		scif_get_node_info();

	*self = scif_info.nodeid;
	mutex_lock(&scif_info.conflock);
	len = min_t(int, len, scif_info.total);
	for (node = 0; node <= scif_info.maxid; node++) {
		if (_scifdev_alive(&scif_dev[node])) {
			online++;
			if (offset < len)
				nodes[offset++] = node;
		}
	}
	dev_dbg(scif_info.mdev.this_device,
		"SCIFAPI get_node_ids total %d online %d filled in %d nodes\n",
		scif_info.total, online, offset);
	mutex_unlock(&scif_info.conflock);

	return online;
}
EXPORT_SYMBOL_GPL(scif_get_node_ids);

static int scif_add_client_dev(struct device *dev, struct subsys_interface *si)
{
	struct scif_client *client =
		container_of(si, struct scif_client, si);
	struct scif_peer_dev *spdev =
		container_of(dev, struct scif_peer_dev, dev);

	if (client->probe)
		client->probe(spdev);
	return 0;
}

static void scif_remove_client_dev(struct device *dev,
				   struct subsys_interface *si)
{
	struct scif_client *client =
		container_of(si, struct scif_client, si);
	struct scif_peer_dev *spdev =
		container_of(dev, struct scif_peer_dev, dev);

	if (client->remove)
		client->remove(spdev);
}

void scif_client_unregister(struct scif_client *client)
{
	subsys_interface_unregister(&client->si);
}
EXPORT_SYMBOL_GPL(scif_client_unregister);

int scif_client_register(struct scif_client *client)
{
	struct subsys_interface *si = &client->si;

	si->name = client->name;
	si->subsys = &scif_peer_bus;
	si->add_dev = scif_add_client_dev;
	si->remove_dev = scif_remove_client_dev;

	return subsys_interface_register(&client->si);
}
EXPORT_SYMBOL_GPL(scif_client_register);