Free Electrons

Embedded Linux Experts

   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
/*
 * Copyright (c) 2012 Mellanox Technologies. All rights reserved.
 *
 * This software is available to you under a choice of one of two
 * licenses.  You may choose to be licensed under the terms of the GNU
 * General Public License (GPL) Version 2, available from the file
 * COPYING in the main directory of this source tree, or the
 * OpenIB.org BSD license below:
 *
 *     Redistribution and use in source and binary forms, with or
 *     without modification, are permitted provided that the following
 *     conditions are met:
 *
 *      - Redistributions of source code must retain the above
 *        copyright notice, this list of conditions and the following
 *        disclaimer.
 *
 *      - Redistributions in binary form must reproduce the above
 *        copyright notice, this list of conditions and the following
 *        disclaimer in the documentation and/or other materials
 *        provided with the distribution.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
 * SOFTWARE.
 */

#include <rdma/ib_mad.h>
#include <rdma/ib_smi.h>
#include <rdma/ib_cache.h>
#include <rdma/ib_sa.h>

#include <linux/mlx4/cmd.h>
#include <linux/rbtree.h>
#include <linux/delay.h>

#include "mlx4_ib.h"

#define MAX_VFS		80
#define MAX_PEND_REQS_PER_FUNC 4
#define MAD_TIMEOUT_MS	2000

#define mcg_warn(fmt, arg...)	pr_warn("MCG WARNING: " fmt, ##arg)
#define mcg_error(fmt, arg...)	pr_err(fmt, ##arg)
#define mcg_warn_group(group, format, arg...) \
	pr_warn("%s-%d: %16s (port %d): WARNING: " format, __func__, __LINE__,\
	(group)->name, group->demux->port, ## arg)

#define mcg_error_group(group, format, arg...) \
	pr_err("  %16s: " format, (group)->name, ## arg)


static union ib_gid mgid0;

static struct workqueue_struct *clean_wq;

enum mcast_state {
	MCAST_NOT_MEMBER = 0,
	MCAST_MEMBER,
};

enum mcast_group_state {
	MCAST_IDLE,
	MCAST_JOIN_SENT,
	MCAST_LEAVE_SENT,
	MCAST_RESP_READY
};

struct mcast_member {
	enum mcast_state state;
	uint8_t			join_state;
	int			num_pend_reqs;
	struct list_head	pending;
};

struct ib_sa_mcmember_data {
	union ib_gid	mgid;
	union ib_gid	port_gid;
	__be32		qkey;
	__be16		mlid;
	u8		mtusel_mtu;
	u8		tclass;
	__be16		pkey;
	u8		ratesel_rate;
	u8		lifetmsel_lifetm;
	__be32		sl_flowlabel_hoplimit;
	u8		scope_join_state;
	u8		proxy_join;
	u8		reserved[2];
};

struct mcast_group {
	struct ib_sa_mcmember_data rec;
	struct rb_node		node;
	struct list_head	mgid0_list;
	struct mlx4_ib_demux_ctx *demux;
	struct mcast_member	func[MAX_VFS];
	struct mutex		lock;
	struct work_struct	work;
	struct list_head	pending_list;
	int			members[3];
	enum mcast_group_state	state;
	enum mcast_group_state	prev_state;
	struct ib_sa_mad	response_sa_mad;
	__be64			last_req_tid;

	char			name[33]; /* MGID string */
	struct device_attribute	dentry;

	/* refcount is the reference count for the following:
	   1. Each queued request
	   2. Each invocation of the worker thread
	   3. Membership of the port at the SA
	*/
	atomic_t		refcount;

	/* delayed work to clean pending SM request */
	struct delayed_work	timeout_work;
	struct list_head	cleanup_list;
};

struct mcast_req {
	int			func;
	struct ib_sa_mad	sa_mad;
	struct list_head	group_list;
	struct list_head	func_list;
	struct mcast_group	*group;
	int			clean;
};


#define safe_atomic_dec(ref) \
	do {\
		if (atomic_dec_and_test(ref)) \
			mcg_warn_group(group, "did not expect to reach zero\n"); \
	} while (0)

static const char *get_state_string(enum mcast_group_state state)
{
	switch (state) {
	case MCAST_IDLE:
		return "MCAST_IDLE";
	case MCAST_JOIN_SENT:
		return "MCAST_JOIN_SENT";
	case MCAST_LEAVE_SENT:
		return "MCAST_LEAVE_SENT";
	case MCAST_RESP_READY:
		return "MCAST_RESP_READY";
	}
	return "Invalid State";
}

static struct mcast_group *mcast_find(struct mlx4_ib_demux_ctx *ctx,
				      union ib_gid *mgid)
{
	struct rb_node *node = ctx->mcg_table.rb_node;
	struct mcast_group *group;
	int ret;

	while (node) {
		group = rb_entry(node, struct mcast_group, node);
		ret = memcmp(mgid->raw, group->rec.mgid.raw, sizeof *mgid);
		if (!ret)
			return group;

		if (ret < 0)
			node = node->rb_left;
		else
			node = node->rb_right;
	}
	return NULL;
}

static struct mcast_group *mcast_insert(struct mlx4_ib_demux_ctx *ctx,
					struct mcast_group *group)
{
	struct rb_node **link = &ctx->mcg_table.rb_node;
	struct rb_node *parent = NULL;
	struct mcast_group *cur_group;
	int ret;

	while (*link) {
		parent = *link;
		cur_group = rb_entry(parent, struct mcast_group, node);

		ret = memcmp(group->rec.mgid.raw, cur_group->rec.mgid.raw,
			     sizeof group->rec.mgid);
		if (ret < 0)
			link = &(*link)->rb_left;
		else if (ret > 0)
			link = &(*link)->rb_right;
		else
			return cur_group;
	}
	rb_link_node(&group->node, parent, link);
	rb_insert_color(&group->node, &ctx->mcg_table);
	return NULL;
}

static int send_mad_to_wire(struct mlx4_ib_demux_ctx *ctx, struct ib_mad *mad)
{
	struct mlx4_ib_dev *dev = ctx->dev;
	struct ib_ah_attr	ah_attr;

	spin_lock(&dev->sm_lock);
	if (!dev->sm_ah[ctx->port - 1]) {
		/* port is not yet Active, sm_ah not ready */
		spin_unlock(&dev->sm_lock);
		return -EAGAIN;
	}
	mlx4_ib_query_ah(dev->sm_ah[ctx->port - 1], &ah_attr);
	spin_unlock(&dev->sm_lock);
	return mlx4_ib_send_to_wire(dev, mlx4_master_func_num(dev->dev), ctx->port,
				    IB_QPT_GSI, 0, 1, IB_QP1_QKEY, &ah_attr, mad);
}

static int send_mad_to_slave(int slave, struct mlx4_ib_demux_ctx *ctx,
			     struct ib_mad *mad)
{
	struct mlx4_ib_dev *dev = ctx->dev;
	struct ib_mad_agent *agent = dev->send_agent[ctx->port - 1][1];
	struct ib_wc wc;
	struct ib_ah_attr ah_attr;

	/* Our agent might not yet be registered when mads start to arrive */
	if (!agent)
		return -EAGAIN;

	ib_query_ah(dev->sm_ah[ctx->port - 1], &ah_attr);

	if (ib_find_cached_pkey(&dev->ib_dev, ctx->port, IB_DEFAULT_PKEY_FULL, &wc.pkey_index))
		return -EINVAL;
	wc.sl = 0;
	wc.dlid_path_bits = 0;
	wc.port_num = ctx->port;
	wc.slid = ah_attr.dlid;  /* opensm lid */
	wc.src_qp = 1;
	return mlx4_ib_send_to_slave(dev, slave, ctx->port, IB_QPT_GSI, &wc, NULL, mad);
}

static int send_join_to_wire(struct mcast_group *group, struct ib_sa_mad *sa_mad)
{
	struct ib_sa_mad mad;
	struct ib_sa_mcmember_data *sa_mad_data = (struct ib_sa_mcmember_data *)&mad.data;
	int ret;

	/* we rely on a mad request as arrived from a VF */
	memcpy(&mad, sa_mad, sizeof mad);

	/* fix port GID to be the real one (slave 0) */
	sa_mad_data->port_gid.global.interface_id = group->demux->guid_cache[0];

	/* assign our own TID */
	mad.mad_hdr.tid = mlx4_ib_get_new_demux_tid(group->demux);
	group->last_req_tid = mad.mad_hdr.tid; /* keep it for later validation */

	ret = send_mad_to_wire(group->demux, (struct ib_mad *)&mad);
	/* set timeout handler */
	if (!ret) {
		/* calls mlx4_ib_mcg_timeout_handler */
		queue_delayed_work(group->demux->mcg_wq, &group->timeout_work,
				msecs_to_jiffies(MAD_TIMEOUT_MS));
	}

	return ret;
}

static int send_leave_to_wire(struct mcast_group *group, u8 join_state)
{
	struct ib_sa_mad mad;
	struct ib_sa_mcmember_data *sa_data = (struct ib_sa_mcmember_data *)&mad.data;
	int ret;

	memset(&mad, 0, sizeof mad);
	mad.mad_hdr.base_version = 1;
	mad.mad_hdr.mgmt_class = IB_MGMT_CLASS_SUBN_ADM;
	mad.mad_hdr.class_version = 2;
	mad.mad_hdr.method = IB_SA_METHOD_DELETE;
	mad.mad_hdr.status = cpu_to_be16(0);
	mad.mad_hdr.class_specific = cpu_to_be16(0);
	mad.mad_hdr.tid = mlx4_ib_get_new_demux_tid(group->demux);
	group->last_req_tid = mad.mad_hdr.tid; /* keep it for later validation */
	mad.mad_hdr.attr_id = cpu_to_be16(IB_SA_ATTR_MC_MEMBER_REC);
	mad.mad_hdr.attr_mod = cpu_to_be32(0);
	mad.sa_hdr.sm_key = 0x0;
	mad.sa_hdr.attr_offset = cpu_to_be16(7);
	mad.sa_hdr.comp_mask = IB_SA_MCMEMBER_REC_MGID |
		IB_SA_MCMEMBER_REC_PORT_GID | IB_SA_MCMEMBER_REC_JOIN_STATE;

	*sa_data = group->rec;
	sa_data->scope_join_state = join_state;

	ret = send_mad_to_wire(group->demux, (struct ib_mad *)&mad);
	if (ret)
		group->state = MCAST_IDLE;

	/* set timeout handler */
	if (!ret) {
		/* calls mlx4_ib_mcg_timeout_handler */
		queue_delayed_work(group->demux->mcg_wq, &group->timeout_work,
				msecs_to_jiffies(MAD_TIMEOUT_MS));
	}

	return ret;
}

static int send_reply_to_slave(int slave, struct mcast_group *group,
		struct ib_sa_mad *req_sa_mad, u16 status)
{
	struct ib_sa_mad mad;
	struct ib_sa_mcmember_data *sa_data = (struct ib_sa_mcmember_data *)&mad.data;
	struct ib_sa_mcmember_data *req_sa_data = (struct ib_sa_mcmember_data *)&req_sa_mad->data;
	int ret;

	memset(&mad, 0, sizeof mad);
	mad.mad_hdr.base_version = 1;
	mad.mad_hdr.mgmt_class = IB_MGMT_CLASS_SUBN_ADM;
	mad.mad_hdr.class_version = 2;
	mad.mad_hdr.method = IB_MGMT_METHOD_GET_RESP;
	mad.mad_hdr.status = cpu_to_be16(status);
	mad.mad_hdr.class_specific = cpu_to_be16(0);
	mad.mad_hdr.tid = req_sa_mad->mad_hdr.tid;
	*(u8 *)&mad.mad_hdr.tid = 0; /* resetting tid to 0 */
	mad.mad_hdr.attr_id = cpu_to_be16(IB_SA_ATTR_MC_MEMBER_REC);
	mad.mad_hdr.attr_mod = cpu_to_be32(0);
	mad.sa_hdr.sm_key = req_sa_mad->sa_hdr.sm_key;
	mad.sa_hdr.attr_offset = cpu_to_be16(7);
	mad.sa_hdr.comp_mask = 0; /* ignored on responses, see IBTA spec */

	*sa_data = group->rec;

	/* reconstruct VF's requested join_state and port_gid */
	sa_data->scope_join_state &= 0xf0;
	sa_data->scope_join_state |= (group->func[slave].join_state & 0x0f);
	memcpy(&sa_data->port_gid, &req_sa_data->port_gid, sizeof req_sa_data->port_gid);

	ret = send_mad_to_slave(slave, group->demux, (struct ib_mad *)&mad);
	return ret;
}

static int check_selector(ib_sa_comp_mask comp_mask,
			  ib_sa_comp_mask selector_mask,
			  ib_sa_comp_mask value_mask,
			  u8 src_value, u8 dst_value)
{
	int err;
	u8 selector = dst_value >> 6;
	dst_value &= 0x3f;
	src_value &= 0x3f;

	if (!(comp_mask & selector_mask) || !(comp_mask & value_mask))
		return 0;

	switch (selector) {
	case IB_SA_GT:
		err = (src_value <= dst_value);
		break;
	case IB_SA_LT:
		err = (src_value >= dst_value);
		break;
	case IB_SA_EQ:
		err = (src_value != dst_value);
		break;
	default:
		err = 0;
		break;
	}

	return err;
}

static u16 cmp_rec(struct ib_sa_mcmember_data *src,
		   struct ib_sa_mcmember_data *dst, ib_sa_comp_mask comp_mask)
{
	/* src is group record, dst is request record */
	/* MGID must already match */
	/* Port_GID we always replace to our Port_GID, so it is a match */

#define MAD_STATUS_REQ_INVALID 0x0200
	if (comp_mask & IB_SA_MCMEMBER_REC_QKEY && src->qkey != dst->qkey)
		return MAD_STATUS_REQ_INVALID;
	if (comp_mask & IB_SA_MCMEMBER_REC_MLID && src->mlid != dst->mlid)
		return MAD_STATUS_REQ_INVALID;
	if (check_selector(comp_mask, IB_SA_MCMEMBER_REC_MTU_SELECTOR,
				 IB_SA_MCMEMBER_REC_MTU,
				 src->mtusel_mtu, dst->mtusel_mtu))
		return MAD_STATUS_REQ_INVALID;
	if (comp_mask & IB_SA_MCMEMBER_REC_TRAFFIC_CLASS &&
	    src->tclass != dst->tclass)
		return MAD_STATUS_REQ_INVALID;
	if (comp_mask & IB_SA_MCMEMBER_REC_PKEY && src->pkey != dst->pkey)
		return MAD_STATUS_REQ_INVALID;
	if (check_selector(comp_mask, IB_SA_MCMEMBER_REC_RATE_SELECTOR,
				 IB_SA_MCMEMBER_REC_RATE,
				 src->ratesel_rate, dst->ratesel_rate))
		return MAD_STATUS_REQ_INVALID;
	if (check_selector(comp_mask,
				 IB_SA_MCMEMBER_REC_PACKET_LIFE_TIME_SELECTOR,
				 IB_SA_MCMEMBER_REC_PACKET_LIFE_TIME,
				 src->lifetmsel_lifetm, dst->lifetmsel_lifetm))
		return MAD_STATUS_REQ_INVALID;
	if (comp_mask & IB_SA_MCMEMBER_REC_SL &&
			(be32_to_cpu(src->sl_flowlabel_hoplimit) & 0xf0000000) !=
			(be32_to_cpu(dst->sl_flowlabel_hoplimit) & 0xf0000000))
		return MAD_STATUS_REQ_INVALID;
	if (comp_mask & IB_SA_MCMEMBER_REC_FLOW_LABEL &&
			(be32_to_cpu(src->sl_flowlabel_hoplimit) & 0x0fffff00) !=
			(be32_to_cpu(dst->sl_flowlabel_hoplimit) & 0x0fffff00))
		return MAD_STATUS_REQ_INVALID;
	if (comp_mask & IB_SA_MCMEMBER_REC_HOP_LIMIT &&
			(be32_to_cpu(src->sl_flowlabel_hoplimit) & 0x000000ff) !=
			(be32_to_cpu(dst->sl_flowlabel_hoplimit) & 0x000000ff))
		return MAD_STATUS_REQ_INVALID;
	if (comp_mask & IB_SA_MCMEMBER_REC_SCOPE &&
			(src->scope_join_state & 0xf0) !=
			(dst->scope_join_state & 0xf0))
		return MAD_STATUS_REQ_INVALID;

	/* join_state checked separately, proxy_join ignored */

	return 0;
}

/* release group, return 1 if this was last release and group is destroyed
 * timout work is canceled sync */
static int release_group(struct mcast_group *group, int from_timeout_handler)
{
	struct mlx4_ib_demux_ctx *ctx = group->demux;
	int nzgroup;

	mutex_lock(&ctx->mcg_table_lock);
	mutex_lock(&group->lock);
	if (atomic_dec_and_test(&group->refcount)) {
		if (!from_timeout_handler) {
			if (group->state != MCAST_IDLE &&
			    !cancel_delayed_work(&group->timeout_work)) {
				atomic_inc(&group->refcount);
				mutex_unlock(&group->lock);
				mutex_unlock(&ctx->mcg_table_lock);
				return 0;
			}
		}

		nzgroup = memcmp(&group->rec.mgid, &mgid0, sizeof mgid0);
		if (nzgroup)
			del_sysfs_port_mcg_attr(ctx->dev, ctx->port, &group->dentry.attr);
		if (!list_empty(&group->pending_list))
			mcg_warn_group(group, "releasing a group with non empty pending list\n");
		if (nzgroup)
			rb_erase(&group->node, &ctx->mcg_table);
		list_del_init(&group->mgid0_list);
		mutex_unlock(&group->lock);
		mutex_unlock(&ctx->mcg_table_lock);
		kfree(group);
		return 1;
	} else {
		mutex_unlock(&group->lock);
		mutex_unlock(&ctx->mcg_table_lock);
	}
	return 0;
}

static void adjust_membership(struct mcast_group *group, u8 join_state, int inc)
{
	int i;

	for (i = 0; i < 3; i++, join_state >>= 1)
		if (join_state & 0x1)
			group->members[i] += inc;
}

static u8 get_leave_state(struct mcast_group *group)
{
	u8 leave_state = 0;
	int i;

	for (i = 0; i < 3; i++)
		if (!group->members[i])
			leave_state |= (1 << i);

	return leave_state & (group->rec.scope_join_state & 7);
}

static int join_group(struct mcast_group *group, int slave, u8 join_mask)
{
	int ret = 0;
	u8 join_state;

	/* remove bits that slave is already member of, and adjust */
	join_state = join_mask & (~group->func[slave].join_state);
	adjust_membership(group, join_state, 1);
	group->func[slave].join_state |= join_state;
	if (group->func[slave].state != MCAST_MEMBER && join_state) {
		group->func[slave].state = MCAST_MEMBER;
		ret = 1;
	}
	return ret;
}

static int leave_group(struct mcast_group *group, int slave, u8 leave_state)
{
	int ret = 0;

	adjust_membership(group, leave_state, -1);
	group->func[slave].join_state &= ~leave_state;
	if (!group->func[slave].join_state) {
		group->func[slave].state = MCAST_NOT_MEMBER;
		ret = 1;
	}
	return ret;
}

static int check_leave(struct mcast_group *group, int slave, u8 leave_mask)
{
	if (group->func[slave].state != MCAST_MEMBER)
		return MAD_STATUS_REQ_INVALID;

	/* make sure we're not deleting unset bits */
	if (~group->func[slave].join_state & leave_mask)
		return MAD_STATUS_REQ_INVALID;

	if (!leave_mask)
		return MAD_STATUS_REQ_INVALID;

	return 0;
}

static void mlx4_ib_mcg_timeout_handler(struct work_struct *work)
{
	struct delayed_work *delay = to_delayed_work(work);
	struct mcast_group *group;
	struct mcast_req *req = NULL;

	group = container_of(delay, typeof(*group), timeout_work);

	mutex_lock(&group->lock);
	if (group->state == MCAST_JOIN_SENT) {
		if (!list_empty(&group->pending_list)) {
			req = list_first_entry(&group->pending_list, struct mcast_req, group_list);
			list_del(&req->group_list);
			list_del(&req->func_list);
			--group->func[req->func].num_pend_reqs;
			mutex_unlock(&group->lock);
			kfree(req);
			if (memcmp(&group->rec.mgid, &mgid0, sizeof mgid0)) {
				if (release_group(group, 1))
					return;
			} else {
				kfree(group);
				return;
			}
			mutex_lock(&group->lock);
		} else
			mcg_warn_group(group, "DRIVER BUG\n");
	} else if (group->state == MCAST_LEAVE_SENT) {
		if (group->rec.scope_join_state & 7)
			group->rec.scope_join_state &= 0xf8;
		group->state = MCAST_IDLE;
		mutex_unlock(&group->lock);
		if (release_group(group, 1))
			return;
		mutex_lock(&group->lock);
	} else
		mcg_warn_group(group, "invalid state %s\n", get_state_string(group->state));
	group->state = MCAST_IDLE;
	atomic_inc(&group->refcount);
	if (!queue_work(group->demux->mcg_wq, &group->work))
		safe_atomic_dec(&group->refcount);

	mutex_unlock(&group->lock);
}

static int handle_leave_req(struct mcast_group *group, u8 leave_mask,
			    struct mcast_req *req)
{
	u16 status;

	if (req->clean)
		leave_mask = group->func[req->func].join_state;

	status = check_leave(group, req->func, leave_mask);
	if (!status)
		leave_group(group, req->func, leave_mask);

	if (!req->clean)
		send_reply_to_slave(req->func, group, &req->sa_mad, status);
	--group->func[req->func].num_pend_reqs;
	list_del(&req->group_list);
	list_del(&req->func_list);
	kfree(req);
	return 1;
}

static int handle_join_req(struct mcast_group *group, u8 join_mask,
			   struct mcast_req *req)
{
	u8 group_join_state = group->rec.scope_join_state & 7;
	int ref = 0;
	u16 status;
	struct ib_sa_mcmember_data *sa_data = (struct ib_sa_mcmember_data *)req->sa_mad.data;

	if (join_mask == (group_join_state & join_mask)) {
		/* port's membership need not change */
		status = cmp_rec(&group->rec, sa_data, req->sa_mad.sa_hdr.comp_mask);
		if (!status)
			join_group(group, req->func, join_mask);

		--group->func[req->func].num_pend_reqs;
		send_reply_to_slave(req->func, group, &req->sa_mad, status);
		list_del(&req->group_list);
		list_del(&req->func_list);
		kfree(req);
		++ref;
	} else {
		/* port's membership needs to be updated */
		group->prev_state = group->state;
		if (send_join_to_wire(group, &req->sa_mad)) {
			--group->func[req->func].num_pend_reqs;
			list_del(&req->group_list);
			list_del(&req->func_list);
			kfree(req);
			ref = 1;
			group->state = group->prev_state;
		} else
			group->state = MCAST_JOIN_SENT;
	}

	return ref;
}

static void mlx4_ib_mcg_work_handler(struct work_struct *work)
{
	struct mcast_group *group;
	struct mcast_req *req = NULL;
	struct ib_sa_mcmember_data *sa_data;
	u8 req_join_state;
	int rc = 1; /* release_count - this is for the scheduled work */
	u16 status;
	u8 method;

	group = container_of(work, typeof(*group), work);

	mutex_lock(&group->lock);

	/* First, let's see if a response from SM is waiting regarding this group.
	 * If so, we need to update the group's REC. If this is a bad response, we
	 * may need to send a bad response to a VF waiting for it. If VF is waiting
	 * and this is a good response, the VF will be answered later in this func. */
	if (group->state == MCAST_RESP_READY) {
		/* cancels mlx4_ib_mcg_timeout_handler */
		cancel_delayed_work(&group->timeout_work);
		status = be16_to_cpu(group->response_sa_mad.mad_hdr.status);
		method = group->response_sa_mad.mad_hdr.method;
		if (group->last_req_tid != group->response_sa_mad.mad_hdr.tid) {
			mcg_warn_group(group, "Got MAD response to existing MGID but wrong TID, dropping. Resp TID=%llx, group TID=%llx\n",
				be64_to_cpu(group->response_sa_mad.mad_hdr.tid),
				be64_to_cpu(group->last_req_tid));
			group->state = group->prev_state;
			goto process_requests;
		}
		if (status) {
			if (!list_empty(&group->pending_list))
				req = list_first_entry(&group->pending_list,
						struct mcast_req, group_list);
			if ((method == IB_MGMT_METHOD_GET_RESP)) {
					if (req) {
						send_reply_to_slave(req->func, group, &req->sa_mad, status);
						--group->func[req->func].num_pend_reqs;
						list_del(&req->group_list);
						list_del(&req->func_list);
						kfree(req);
						++rc;
					} else
						mcg_warn_group(group, "no request for failed join\n");
			} else if (method == IB_SA_METHOD_DELETE_RESP && group->demux->flushing)
				++rc;
		} else {
			u8 resp_join_state;
			u8 cur_join_state;

			resp_join_state = ((struct ib_sa_mcmember_data *)
						group->response_sa_mad.data)->scope_join_state & 7;
			cur_join_state = group->rec.scope_join_state & 7;

			if (method == IB_MGMT_METHOD_GET_RESP) {
				/* successfull join */
				if (!cur_join_state && resp_join_state)
					--rc;
			} else if (!resp_join_state)
					++rc;
			memcpy(&group->rec, group->response_sa_mad.data, sizeof group->rec);
		}
		group->state = MCAST_IDLE;
	}

process_requests:
	/* We should now go over pending join/leave requests, as long as we are idle. */
	while (!list_empty(&group->pending_list) && group->state == MCAST_IDLE) {
		req = list_first_entry(&group->pending_list, struct mcast_req,
				       group_list);
		sa_data = (struct ib_sa_mcmember_data *)req->sa_mad.data;
		req_join_state = sa_data->scope_join_state & 0x7;

		/* For a leave request, we will immediately answer the VF, and
		 * update our internal counters. The actual leave will be sent
		 * to SM later, if at all needed. We dequeue the request now. */
		if (req->sa_mad.mad_hdr.method == IB_SA_METHOD_DELETE)
			rc += handle_leave_req(group, req_join_state, req);
		else
			rc += handle_join_req(group, req_join_state, req);
	}

	/* Handle leaves */
	if (group->state == MCAST_IDLE) {
		req_join_state = get_leave_state(group);
		if (req_join_state) {
			group->rec.scope_join_state &= ~req_join_state;
			group->prev_state = group->state;
			if (send_leave_to_wire(group, req_join_state)) {
				group->state = group->prev_state;
				++rc;
			} else
				group->state = MCAST_LEAVE_SENT;
		}
	}

	if (!list_empty(&group->pending_list) && group->state == MCAST_IDLE)
		goto process_requests;
	mutex_unlock(&group->lock);

	while (rc--)
		release_group(group, 0);
}

static struct mcast_group *search_relocate_mgid0_group(struct mlx4_ib_demux_ctx *ctx,
						       __be64 tid,
						       union ib_gid *new_mgid)
{
	struct mcast_group *group = NULL, *cur_group;
	struct mcast_req *req;
	struct list_head *pos;
	struct list_head *n;

	mutex_lock(&ctx->mcg_table_lock);
	list_for_each_safe(pos, n, &ctx->mcg_mgid0_list) {
		group = list_entry(pos, struct mcast_group, mgid0_list);
		mutex_lock(&group->lock);
		if (group->last_req_tid == tid) {
			if (memcmp(new_mgid, &mgid0, sizeof mgid0)) {
				group->rec.mgid = *new_mgid;
				sprintf(group->name, "%016llx%016llx",
						be64_to_cpu(group->rec.mgid.global.subnet_prefix),
						be64_to_cpu(group->rec.mgid.global.interface_id));
				list_del_init(&group->mgid0_list);
				cur_group = mcast_insert(ctx, group);
				if (cur_group) {
					/* A race between our code and SM. Silently cleaning the new one */
					req = list_first_entry(&group->pending_list,
							       struct mcast_req, group_list);
					--group->func[req->func].num_pend_reqs;
					list_del(&req->group_list);
					list_del(&req->func_list);
					kfree(req);
					mutex_unlock(&group->lock);
					mutex_unlock(&ctx->mcg_table_lock);
					release_group(group, 0);
					return NULL;
				}

				atomic_inc(&group->refcount);
				add_sysfs_port_mcg_attr(ctx->dev, ctx->port, &group->dentry.attr);
				mutex_unlock(&group->lock);
				mutex_unlock(&ctx->mcg_table_lock);
				return group;
			} else {
				struct mcast_req *tmp1, *tmp2;

				list_del(&group->mgid0_list);
				if (!list_empty(&group->pending_list) && group->state != MCAST_IDLE)
					cancel_delayed_work_sync(&group->timeout_work);

				list_for_each_entry_safe(tmp1, tmp2, &group->pending_list, group_list) {
					list_del(&tmp1->group_list);
					kfree(tmp1);
				}
				mutex_unlock(&group->lock);
				mutex_unlock(&ctx->mcg_table_lock);
				kfree(group);
				return NULL;
			}
		}
		mutex_unlock(&group->lock);
	}
	mutex_unlock(&ctx->mcg_table_lock);

	return NULL;
}

static ssize_t sysfs_show_group(struct device *dev,
		struct device_attribute *attr, char *buf);

static struct mcast_group *acquire_group(struct mlx4_ib_demux_ctx *ctx,
					 union ib_gid *mgid, int create,
					 gfp_t gfp_mask)
{
	struct mcast_group *group, *cur_group;
	int is_mgid0;
	int i;

	is_mgid0 = !memcmp(&mgid0, mgid, sizeof mgid0);
	if (!is_mgid0) {
		group = mcast_find(ctx, mgid);
		if (group)
			goto found;
	}

	if (!create)
		return ERR_PTR(-ENOENT);

	group = kzalloc(sizeof *group, gfp_mask);
	if (!group)
		return ERR_PTR(-ENOMEM);

	group->demux = ctx;
	group->rec.mgid = *mgid;
	INIT_LIST_HEAD(&group->pending_list);
	INIT_LIST_HEAD(&group->mgid0_list);
	for (i = 0; i < MAX_VFS; ++i)
		INIT_LIST_HEAD(&group->func[i].pending);
	INIT_WORK(&group->work, mlx4_ib_mcg_work_handler);
	INIT_DELAYED_WORK(&group->timeout_work, mlx4_ib_mcg_timeout_handler);
	mutex_init(&group->lock);
	sprintf(group->name, "%016llx%016llx",
			be64_to_cpu(group->rec.mgid.global.subnet_prefix),
			be64_to_cpu(group->rec.mgid.global.interface_id));
	sysfs_attr_init(&group->dentry.attr);
	group->dentry.show = sysfs_show_group;
	group->dentry.store = NULL;
	group->dentry.attr.name = group->name;
	group->dentry.attr.mode = 0400;
	group->state = MCAST_IDLE;

	if (is_mgid0) {
		list_add(&group->mgid0_list, &ctx->mcg_mgid0_list);
		goto found;
	}

	cur_group = mcast_insert(ctx, group);
	if (cur_group) {
		mcg_warn("group just showed up %s - confused\n", cur_group->name);
		kfree(group);
		return ERR_PTR(-EINVAL);
	}

	add_sysfs_port_mcg_attr(ctx->dev, ctx->port, &group->dentry.attr);

found:
	atomic_inc(&group->refcount);
	return group;
}

static void queue_req(struct mcast_req *req)
{
	struct mcast_group *group = req->group;

	atomic_inc(&group->refcount); /* for the request */
	atomic_inc(&group->refcount); /* for scheduling the work */
	list_add_tail(&req->group_list, &group->pending_list);
	list_add_tail(&req->func_list, &group->func[req->func].pending);
	/* calls mlx4_ib_mcg_work_handler */
	if (!queue_work(group->demux->mcg_wq, &group->work))
		safe_atomic_dec(&group->refcount);
}

int mlx4_ib_mcg_demux_handler(struct ib_device *ibdev, int port, int slave,
			      struct ib_sa_mad *mad)
{
	struct mlx4_ib_dev *dev = to_mdev(ibdev);
	struct ib_sa_mcmember_data *rec = (struct ib_sa_mcmember_data *)mad->data;
	struct mlx4_ib_demux_ctx *ctx = &dev->sriov.demux[port - 1];
	struct mcast_group *group;

	switch (mad->mad_hdr.method) {
	case IB_MGMT_METHOD_GET_RESP:
	case IB_SA_METHOD_DELETE_RESP:
		mutex_lock(&ctx->mcg_table_lock);
		group = acquire_group(ctx, &rec->mgid, 0, GFP_KERNEL);
		mutex_unlock(&ctx->mcg_table_lock);
		if (IS_ERR(group)) {
			if (mad->mad_hdr.method == IB_MGMT_METHOD_GET_RESP) {
				__be64 tid = mad->mad_hdr.tid;
				*(u8 *)(&tid) = (u8)slave; /* in group we kept the modified TID */
				group = search_relocate_mgid0_group(ctx, tid, &rec->mgid);
			} else
				group = NULL;
		}

		if (!group)
			return 1;

		mutex_lock(&group->lock);
		group->response_sa_mad = *mad;
		group->prev_state = group->state;
		group->state = MCAST_RESP_READY;
		/* calls mlx4_ib_mcg_work_handler */
		atomic_inc(&group->refcount);
		if (!queue_work(ctx->mcg_wq, &group->work))
			safe_atomic_dec(&group->refcount);
		mutex_unlock(&group->lock);
		release_group(group, 0);
		return 1; /* consumed */
	case IB_MGMT_METHOD_SET:
	case IB_SA_METHOD_GET_TABLE:
	case IB_SA_METHOD_GET_TABLE_RESP:
	case IB_SA_METHOD_DELETE:
		return 0; /* not consumed, pass-through to guest over tunnel */
	default:
		mcg_warn("In demux, port %d: unexpected MCMember method: 0x%x, dropping\n",
			port, mad->mad_hdr.method);
		return 1; /* consumed */
	}
}

int mlx4_ib_mcg_multiplex_handler(struct ib_device *ibdev, int port,
				  int slave, struct ib_sa_mad *sa_mad)
{
	struct mlx4_ib_dev *dev = to_mdev(ibdev);
	struct ib_sa_mcmember_data *rec = (struct ib_sa_mcmember_data *)sa_mad->data;
	struct mlx4_ib_demux_ctx *ctx = &dev->sriov.demux[port - 1];
	struct mcast_group *group;
	struct mcast_req *req;
	int may_create = 0;

	if (ctx->flushing)
		return -EAGAIN;

	switch (sa_mad->mad_hdr.method) {
	case IB_MGMT_METHOD_SET:
		may_create = 1;
	case IB_SA_METHOD_DELETE:
		req = kzalloc(sizeof *req, GFP_KERNEL);
		if (!req)
			return -ENOMEM;

		req->func = slave;
		req->sa_mad = *sa_mad;

		mutex_lock(&ctx->mcg_table_lock);
		group = acquire_group(ctx, &rec->mgid, may_create, GFP_KERNEL);
		mutex_unlock(&ctx->mcg_table_lock);
		if (IS_ERR(group)) {
			kfree(req);
			return PTR_ERR(group);
		}
		mutex_lock(&group->lock);
		if (group->func[slave].num_pend_reqs > MAX_PEND_REQS_PER_FUNC) {
			mutex_unlock(&group->lock);
			mcg_warn_group(group, "Port %d, Func %d has too many pending requests (%d), dropping\n",
				       port, slave, MAX_PEND_REQS_PER_FUNC);
			release_group(group, 0);
			kfree(req);
			return -ENOMEM;
		}
		++group->func[slave].num_pend_reqs;
		req->group = group;
		queue_req(req);
		mutex_unlock(&group->lock);
		release_group(group, 0);
		return 1; /* consumed */
	case IB_SA_METHOD_GET_TABLE:
	case IB_MGMT_METHOD_GET_RESP:
	case IB_SA_METHOD_GET_TABLE_RESP:
	case IB_SA_METHOD_DELETE_RESP:
		return 0; /* not consumed, pass-through */
	default:
		mcg_warn("In multiplex, port %d, func %d: unexpected MCMember method: 0x%x, dropping\n",
			port, slave, sa_mad->mad_hdr.method);
		return 1; /* consumed */
	}
}

static ssize_t sysfs_show_group(struct device *dev,
		struct device_attribute *attr, char *buf)
{
	struct mcast_group *group =
		container_of(attr, struct mcast_group, dentry);
	struct mcast_req *req = NULL;
	char pending_str[40];
	char state_str[40];
	ssize_t len = 0;
	int f;

	if (group->state == MCAST_IDLE)
		sprintf(state_str, "%s", get_state_string(group->state));
	else
		sprintf(state_str, "%s(TID=0x%llx)",
				get_state_string(group->state),
				be64_to_cpu(group->last_req_tid));
	if (list_empty(&group->pending_list)) {
		sprintf(pending_str, "No");
	} else {
		req = list_first_entry(&group->pending_list, struct mcast_req, group_list);
		sprintf(pending_str, "Yes(TID=0x%llx)",
				be64_to_cpu(req->sa_mad.mad_hdr.tid));
	}
	len += sprintf(buf + len, "%1d [%02d,%02d,%02d] %4d %4s %5s     ",
			group->rec.scope_join_state & 0xf,
			group->members[2], group->members[1], group->members[0],
			atomic_read(&group->refcount),
			pending_str,
			state_str);
	for (f = 0; f < MAX_VFS; ++f)
		if (group->func[f].state == MCAST_MEMBER)
			len += sprintf(buf + len, "%d[%1x] ",
					f, group->func[f].join_state);

	len += sprintf(buf + len, "\t\t(%4hx %4x %2x %2x %2x %2x %2x "
		"%4x %4x %2x %2x)\n",
		be16_to_cpu(group->rec.pkey),
		be32_to_cpu(group->rec.qkey),
		(group->rec.mtusel_mtu & 0xc0) >> 6,
		group->rec.mtusel_mtu & 0x3f,
		group->rec.tclass,
		(group->rec.ratesel_rate & 0xc0) >> 6,
		group->rec.ratesel_rate & 0x3f,
		(be32_to_cpu(group->rec.sl_flowlabel_hoplimit) & 0xf0000000) >> 28,
		(be32_to_cpu(group->rec.sl_flowlabel_hoplimit) & 0x0fffff00) >> 8,
		be32_to_cpu(group->rec.sl_flowlabel_hoplimit) & 0x000000ff,
		group->rec.proxy_join);

	return len;
}

int mlx4_ib_mcg_port_init(struct mlx4_ib_demux_ctx *ctx)
{
	char name[20];

	atomic_set(&ctx->tid, 0);
	sprintf(name, "mlx4_ib_mcg%d", ctx->port);
	ctx->mcg_wq = create_singlethread_workqueue(name);
	if (!ctx->mcg_wq)
		return -ENOMEM;

	mutex_init(&ctx->mcg_table_lock);
	ctx->mcg_table = RB_ROOT;
	INIT_LIST_HEAD(&ctx->mcg_mgid0_list);
	ctx->flushing = 0;

	return 0;
}

static void force_clean_group(struct mcast_group *group)
{
	struct mcast_req *req, *tmp
		;
	list_for_each_entry_safe(req, tmp, &group->pending_list, group_list) {
		list_del(&req->group_list);
		kfree(req);
	}
	del_sysfs_port_mcg_attr(group->demux->dev, group->demux->port, &group->dentry.attr);
	rb_erase(&group->node, &group->demux->mcg_table);
	kfree(group);
}

static void _mlx4_ib_mcg_port_cleanup(struct mlx4_ib_demux_ctx *ctx, int destroy_wq)
{
	int i;
	struct rb_node *p;
	struct mcast_group *group;
	unsigned long end;
	int count;

	for (i = 0; i < MAX_VFS; ++i)
		clean_vf_mcast(ctx, i);

	end = jiffies + msecs_to_jiffies(MAD_TIMEOUT_MS + 3000);
	do {
		count = 0;
		mutex_lock(&ctx->mcg_table_lock);
		for (p = rb_first(&ctx->mcg_table); p; p = rb_next(p))
			++count;
		mutex_unlock(&ctx->mcg_table_lock);
		if (!count)
			break;

		msleep(1);
	} while (time_after(end, jiffies));

	flush_workqueue(ctx->mcg_wq);
	if (destroy_wq)
		destroy_workqueue(ctx->mcg_wq);

	mutex_lock(&ctx->mcg_table_lock);
	while ((p = rb_first(&ctx->mcg_table)) != NULL) {
		group = rb_entry(p, struct mcast_group, node);
		if (atomic_read(&group->refcount))
			mcg_warn_group(group, "group refcount %d!!! (pointer %p)\n", atomic_read(&group->refcount), group);

		force_clean_group(group);
	}
	mutex_unlock(&ctx->mcg_table_lock);
}

struct clean_work {
	struct work_struct work;
	struct mlx4_ib_demux_ctx *ctx;
	int destroy_wq;
};

static void mcg_clean_task(struct work_struct *work)
{
	struct clean_work *cw = container_of(work, struct clean_work, work);

	_mlx4_ib_mcg_port_cleanup(cw->ctx, cw->destroy_wq);
	cw->ctx->flushing = 0;
	kfree(cw);
}

void mlx4_ib_mcg_port_cleanup(struct mlx4_ib_demux_ctx *ctx, int destroy_wq)
{
	struct clean_work *work;

	if (ctx->flushing)
		return;

	ctx->flushing = 1;

	if (destroy_wq) {
		_mlx4_ib_mcg_port_cleanup(ctx, destroy_wq);
		ctx->flushing = 0;
		return;
	}

	work = kmalloc(sizeof *work, GFP_KERNEL);
	if (!work) {
		ctx->flushing = 0;
		mcg_warn("failed allocating work for cleanup\n");
		return;
	}

	work->ctx = ctx;
	work->destroy_wq = destroy_wq;
	INIT_WORK(&work->work, mcg_clean_task);
	queue_work(clean_wq, &work->work);
}

static void build_leave_mad(struct mcast_req *req)
{
	struct ib_sa_mad *mad = &req->sa_mad;

	mad->mad_hdr.method = IB_SA_METHOD_DELETE;
}


static void clear_pending_reqs(struct mcast_group *group, int vf)
{
	struct mcast_req *req, *tmp, *group_first = NULL;
	int clear;
	int pend = 0;

	if (!list_empty(&group->pending_list))
		group_first = list_first_entry(&group->pending_list, struct mcast_req, group_list);

	list_for_each_entry_safe(req, tmp, &group->func[vf].pending, func_list) {
		clear = 1;
		if (group_first == req &&
		    (group->state == MCAST_JOIN_SENT ||
		     group->state == MCAST_LEAVE_SENT)) {
			clear = cancel_delayed_work(&group->timeout_work);
			pend = !clear;
			group->state = MCAST_IDLE;
		}
		if (clear) {
			--group->func[vf].num_pend_reqs;
			list_del(&req->group_list);
			list_del(&req->func_list);
			kfree(req);
			atomic_dec(&group->refcount);
		}
	}

	if (!pend && (!list_empty(&group->func[vf].pending) || group->func[vf].num_pend_reqs)) {
		mcg_warn_group(group, "DRIVER BUG: list_empty %d, num_pend_reqs %d\n",
			       list_empty(&group->func[vf].pending), group->func[vf].num_pend_reqs);
	}
}

static int push_deleteing_req(struct mcast_group *group, int slave)
{
	struct mcast_req *req;
	struct mcast_req *pend_req;

	if (!group->func[slave].join_state)
		return 0;

	req = kzalloc(sizeof *req, GFP_KERNEL);
	if (!req) {
		mcg_warn_group(group, "failed allocation - may leave stall groups\n");
		return -ENOMEM;
	}

	if (!list_empty(&group->func[slave].pending)) {
		pend_req = list_entry(group->func[slave].pending.prev, struct mcast_req, group_list);
		if (pend_req->clean) {
			kfree(req);
			return 0;
		}
	}

	req->clean = 1;
	req->func = slave;
	req->group = group;
	++group->func[slave].num_pend_reqs;
	build_leave_mad(req);
	queue_req(req);
	return 0;
}

void clean_vf_mcast(struct mlx4_ib_demux_ctx *ctx, int slave)
{
	struct mcast_group *group;
	struct rb_node *p;

	mutex_lock(&ctx->mcg_table_lock);
	for (p = rb_first(&ctx->mcg_table); p; p = rb_next(p)) {
		group = rb_entry(p, struct mcast_group, node);
		mutex_lock(&group->lock);
		if (atomic_read(&group->refcount)) {
			/* clear pending requests of this VF */
			clear_pending_reqs(group, slave);
			push_deleteing_req(group, slave);
		}
		mutex_unlock(&group->lock);
	}
	mutex_unlock(&ctx->mcg_table_lock);
}


int mlx4_ib_mcg_init(void)
{
	clean_wq = create_singlethread_workqueue("mlx4_ib_mcg");
	if (!clean_wq)
		return -ENOMEM;

	return 0;
}

void mlx4_ib_mcg_destroy(void)
{
	destroy_workqueue(clean_wq);
}