-
Notifications
You must be signed in to change notification settings - Fork 1.3k
/
Copy pathrdma.c
1416 lines (1181 loc) · 32.7 KB
/
rdma.c
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
/*
* RDMA I/O engine
*
* RDMA I/O engine based on the IB verbs and RDMA/CM user space libraries.
* Supports both RDMA memory semantics and channel semantics
* for the InfiniBand, RoCE and iWARP protocols.
*
* You will need the Linux RDMA software installed, either
* from your Linux distributor or directly from openfabrics.org:
*
* http://www.openfabrics.org/downloads/OFED/
*
* Exchanging steps of RDMA ioengine control messages:
* 1. client side sends test mode (RDMA_WRITE/RDMA_READ/SEND)
* to server side.
* 2. server side parses test mode, and sends back confirmation
* to client side. In RDMA WRITE/READ test, this confirmation
* includes memory information, such as rkey, address.
* 3. client side initiates test loop.
* 4. In RDMA WRITE/READ test, client side sends a completion
* notification to server side. Server side updates its
* td->done as true.
*
*/
#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#include <errno.h>
#include <assert.h>
#include <netinet/in.h>
#include <arpa/inet.h>
#include <netdb.h>
#include <poll.h>
#include <sys/types.h>
#include <sys/socket.h>
#include <sys/time.h>
#include <sys/resource.h>
#include <pthread.h>
#include <inttypes.h>
#include "../fio.h"
#include "../hash.h"
#include "../optgroup.h"
#include <rdma/rdma_cma.h>
#define FIO_RDMA_MAX_IO_DEPTH 512
enum rdma_io_mode {
FIO_RDMA_UNKNOWN = 0,
FIO_RDMA_MEM_WRITE,
FIO_RDMA_MEM_READ,
FIO_RDMA_CHA_SEND,
FIO_RDMA_CHA_RECV
};
struct rdmaio_options {
struct thread_data *td;
unsigned int port;
enum rdma_io_mode verb;
char *bindname;
};
static int str_hostname_cb(void *data, const char *input)
{
struct rdmaio_options *o = data;
if (o->td->o.filename)
free(o->td->o.filename);
o->td->o.filename = strdup(input);
return 0;
}
static struct fio_option options[] = {
{
.name = "hostname",
.lname = "rdma engine hostname",
.type = FIO_OPT_STR_STORE,
.cb = str_hostname_cb,
.help = "Hostname for RDMA IO engine",
.category = FIO_OPT_C_ENGINE,
.group = FIO_OPT_G_RDMA,
},
{
.name = "bindname",
.lname = "rdma engine bindname",
.type = FIO_OPT_STR_STORE,
.off1 = offsetof(struct rdmaio_options, bindname),
.help = "Bind for RDMA IO engine",
.def = "",
.category = FIO_OPT_C_ENGINE,
.group = FIO_OPT_G_RDMA,
},
{
.name = "port",
.lname = "rdma engine port",
.type = FIO_OPT_INT,
.off1 = offsetof(struct rdmaio_options, port),
.minval = 1,
.maxval = 65535,
.help = "Port to use for RDMA connections",
.category = FIO_OPT_C_ENGINE,
.group = FIO_OPT_G_RDMA,
},
{
.name = "verb",
.lname = "RDMA engine verb",
.alias = "proto",
.type = FIO_OPT_STR,
.off1 = offsetof(struct rdmaio_options, verb),
.help = "RDMA engine verb",
.def = "write",
.posval = {
{ .ival = "write",
.oval = FIO_RDMA_MEM_WRITE,
.help = "Memory Write",
},
{ .ival = "read",
.oval = FIO_RDMA_MEM_READ,
.help = "Memory Read",
},
{ .ival = "send",
.oval = FIO_RDMA_CHA_SEND,
.help = "Posted Send",
},
{ .ival = "recv",
.oval = FIO_RDMA_CHA_RECV,
.help = "Posted Receive",
},
},
.category = FIO_OPT_C_ENGINE,
.group = FIO_OPT_G_RDMA,
},
{
.name = NULL,
},
};
struct remote_u {
uint64_t buf;
uint32_t rkey;
uint32_t size;
};
struct rdma_info_blk {
uint32_t mode; /* channel semantic or memory semantic */
uint32_t nr; /* client: io depth
server: number of records for memory semantic
*/
uint32_t max_bs; /* maximum block size */
struct remote_u rmt_us[FIO_RDMA_MAX_IO_DEPTH];
};
struct rdma_io_u_data {
uint64_t wr_id;
struct ibv_send_wr sq_wr;
struct ibv_recv_wr rq_wr;
struct ibv_sge rdma_sgl;
};
struct rdmaio_data {
int is_client;
enum rdma_io_mode rdma_protocol;
char host[64];
struct sockaddr_in addr;
struct ibv_recv_wr rq_wr;
struct ibv_sge recv_sgl;
struct rdma_info_blk recv_buf;
struct ibv_mr *recv_mr;
struct ibv_send_wr sq_wr;
struct ibv_sge send_sgl;
struct rdma_info_blk send_buf;
struct ibv_mr *send_mr;
struct ibv_comp_channel *channel;
struct ibv_cq *cq;
struct ibv_pd *pd;
struct ibv_qp *qp;
pthread_t cmthread;
struct rdma_event_channel *cm_channel;
struct rdma_cm_id *cm_id;
struct rdma_cm_id *child_cm_id;
int cq_event_num;
struct remote_u *rmt_us;
int rmt_nr;
struct io_u **io_us_queued;
int io_u_queued_nr;
struct io_u **io_us_flight;
int io_u_flight_nr;
struct io_u **io_us_completed;
int io_u_completed_nr;
struct frand_state rand_state;
};
static int client_recv(struct thread_data *td, struct ibv_wc *wc)
{
struct rdmaio_data *rd = td->io_ops_data;
unsigned int max_bs;
if (wc->byte_len != sizeof(rd->recv_buf)) {
log_err("Received bogus data, size %d\n", wc->byte_len);
return 1;
}
max_bs = max(td->o.max_bs[DDIR_READ], td->o.max_bs[DDIR_WRITE]);
if (max_bs > ntohl(rd->recv_buf.max_bs)) {
log_err("fio: Server's block size (%d) must be greater than or "
"equal to the client's block size (%d)!\n",
ntohl(rd->recv_buf.max_bs), max_bs);
return 1;
}
/* store mr info for MEMORY semantic */
if ((rd->rdma_protocol == FIO_RDMA_MEM_WRITE) ||
(rd->rdma_protocol == FIO_RDMA_MEM_READ)) {
/* struct flist_head *entry; */
int i = 0;
rd->rmt_nr = ntohl(rd->recv_buf.nr);
for (i = 0; i < rd->rmt_nr; i++) {
rd->rmt_us[i].buf = __be64_to_cpu(
rd->recv_buf.rmt_us[i].buf);
rd->rmt_us[i].rkey = ntohl(rd->recv_buf.rmt_us[i].rkey);
rd->rmt_us[i].size = ntohl(rd->recv_buf.rmt_us[i].size);
dprint(FD_IO,
"fio: Received rkey %x addr %" PRIx64
" len %d from peer\n", rd->rmt_us[i].rkey,
rd->rmt_us[i].buf, rd->rmt_us[i].size);
}
}
return 0;
}
static int server_recv(struct thread_data *td, struct ibv_wc *wc)
{
struct rdmaio_data *rd = td->io_ops_data;
unsigned int max_bs;
if (wc->wr_id == FIO_RDMA_MAX_IO_DEPTH) {
rd->rdma_protocol = ntohl(rd->recv_buf.mode);
/* CHANNEL semantic, do nothing */
if (rd->rdma_protocol == FIO_RDMA_CHA_SEND)
rd->rdma_protocol = FIO_RDMA_CHA_RECV;
max_bs = max(td->o.max_bs[DDIR_READ], td->o.max_bs[DDIR_WRITE]);
if (max_bs < ntohl(rd->recv_buf.max_bs)) {
log_err("fio: Server's block size (%d) must be greater than or "
"equal to the client's block size (%d)!\n",
ntohl(rd->recv_buf.max_bs), max_bs);
return 1;
}
}
return 0;
}
static int cq_event_handler(struct thread_data *td, enum ibv_wc_opcode opcode)
{
struct rdmaio_data *rd = td->io_ops_data;
struct ibv_wc wc;
struct rdma_io_u_data *r_io_u_d;
int ret;
int compevnum = 0;
int i;
while ((ret = ibv_poll_cq(rd->cq, 1, &wc)) == 1) {
compevnum++;
if (wc.status) {
log_err("fio: cq completion status %d(%s)\n",
wc.status, ibv_wc_status_str(wc.status));
return -1;
}
switch (wc.opcode) {
case IBV_WC_RECV:
if (rd->is_client == 1)
ret = client_recv(td, &wc);
else
ret = server_recv(td, &wc);
if (ret)
return -1;
if (wc.wr_id == FIO_RDMA_MAX_IO_DEPTH)
break;
for (i = 0; i < rd->io_u_flight_nr; i++) {
r_io_u_d = rd->io_us_flight[i]->engine_data;
if (wc.wr_id == r_io_u_d->rq_wr.wr_id) {
rd->io_us_flight[i]->resid =
rd->io_us_flight[i]->buflen
- wc.byte_len;
rd->io_us_flight[i]->error = 0;
rd->io_us_completed[rd->
io_u_completed_nr]
= rd->io_us_flight[i];
rd->io_u_completed_nr++;
break;
}
}
if (i == rd->io_u_flight_nr)
log_err("fio: recv wr %" PRId64 " not found\n",
wc.wr_id);
else {
/* put the last one into middle of the list */
rd->io_us_flight[i] =
rd->io_us_flight[rd->io_u_flight_nr - 1];
rd->io_u_flight_nr--;
}
break;
case IBV_WC_SEND:
case IBV_WC_RDMA_WRITE:
case IBV_WC_RDMA_READ:
if (wc.wr_id == FIO_RDMA_MAX_IO_DEPTH)
break;
for (i = 0; i < rd->io_u_flight_nr; i++) {
r_io_u_d = rd->io_us_flight[i]->engine_data;
if (wc.wr_id == r_io_u_d->sq_wr.wr_id) {
rd->io_us_completed[rd->
io_u_completed_nr]
= rd->io_us_flight[i];
rd->io_u_completed_nr++;
break;
}
}
if (i == rd->io_u_flight_nr)
log_err("fio: send wr %" PRId64 " not found\n",
wc.wr_id);
else {
/* put the last one into middle of the list */
rd->io_us_flight[i] =
rd->io_us_flight[rd->io_u_flight_nr - 1];
rd->io_u_flight_nr--;
}
break;
default:
log_info("fio: unknown completion event %d\n",
wc.opcode);
return -1;
}
rd->cq_event_num++;
}
if (ret) {
log_err("fio: poll error %d\n", ret);
return 1;
}
return compevnum;
}
/*
* Return -1 for error and 'nr events' for a positive number
* of events
*/
static int rdma_poll_wait(struct thread_data *td, enum ibv_wc_opcode opcode)
{
struct rdmaio_data *rd = td->io_ops_data;
struct ibv_cq *ev_cq;
void *ev_ctx;
int ret;
if (rd->cq_event_num > 0) { /* previous left */
rd->cq_event_num--;
return 0;
}
again:
if (ibv_get_cq_event(rd->channel, &ev_cq, &ev_ctx) != 0) {
log_err("fio: Failed to get cq event!\n");
return -1;
}
if (ev_cq != rd->cq) {
log_err("fio: Unknown CQ!\n");
return -1;
}
if (ibv_req_notify_cq(rd->cq, 0) != 0) {
log_err("fio: Failed to set notify!\n");
return -1;
}
ret = cq_event_handler(td, opcode);
if (ret == 0)
goto again;
ibv_ack_cq_events(rd->cq, ret);
rd->cq_event_num--;
return ret;
}
static int fio_rdmaio_setup_qp(struct thread_data *td)
{
struct rdmaio_data *rd = td->io_ops_data;
struct ibv_qp_init_attr init_attr;
int qp_depth = td->o.iodepth * 2; /* 2 times of io depth */
if (rd->is_client == 0)
rd->pd = ibv_alloc_pd(rd->child_cm_id->verbs);
else
rd->pd = ibv_alloc_pd(rd->cm_id->verbs);
if (rd->pd == NULL) {
log_err("fio: ibv_alloc_pd fail: %m\n");
return 1;
}
if (rd->is_client == 0)
rd->channel = ibv_create_comp_channel(rd->child_cm_id->verbs);
else
rd->channel = ibv_create_comp_channel(rd->cm_id->verbs);
if (rd->channel == NULL) {
log_err("fio: ibv_create_comp_channel fail: %m\n");
goto err1;
}
if (qp_depth < 16)
qp_depth = 16;
if (rd->is_client == 0)
rd->cq = ibv_create_cq(rd->child_cm_id->verbs,
qp_depth, rd, rd->channel, 0);
else
rd->cq = ibv_create_cq(rd->cm_id->verbs,
qp_depth, rd, rd->channel, 0);
if (rd->cq == NULL) {
log_err("fio: ibv_create_cq failed: %m\n");
goto err2;
}
if (ibv_req_notify_cq(rd->cq, 0) != 0) {
log_err("fio: ibv_req_notify_cq failed: %m\n");
goto err3;
}
/* create queue pair */
memset(&init_attr, 0, sizeof(init_attr));
init_attr.cap.max_send_wr = qp_depth;
init_attr.cap.max_recv_wr = qp_depth;
init_attr.cap.max_recv_sge = 1;
init_attr.cap.max_send_sge = 1;
init_attr.qp_type = IBV_QPT_RC;
init_attr.send_cq = rd->cq;
init_attr.recv_cq = rd->cq;
if (rd->is_client == 0) {
if (rdma_create_qp(rd->child_cm_id, rd->pd, &init_attr) != 0) {
log_err("fio: rdma_create_qp failed: %m\n");
goto err3;
}
rd->qp = rd->child_cm_id->qp;
} else {
if (rdma_create_qp(rd->cm_id, rd->pd, &init_attr) != 0) {
log_err("fio: rdma_create_qp failed: %m\n");
goto err3;
}
rd->qp = rd->cm_id->qp;
}
return 0;
err3:
ibv_destroy_cq(rd->cq);
err2:
ibv_destroy_comp_channel(rd->channel);
err1:
ibv_dealloc_pd(rd->pd);
return 1;
}
static int fio_rdmaio_setup_control_msg_buffers(struct thread_data *td)
{
struct rdmaio_data *rd = td->io_ops_data;
rd->recv_mr = ibv_reg_mr(rd->pd, &rd->recv_buf, sizeof(rd->recv_buf),
IBV_ACCESS_LOCAL_WRITE);
if (rd->recv_mr == NULL) {
log_err("fio: recv_buf reg_mr failed: %m\n");
return 1;
}
rd->send_mr = ibv_reg_mr(rd->pd, &rd->send_buf, sizeof(rd->send_buf),
0);
if (rd->send_mr == NULL) {
log_err("fio: send_buf reg_mr failed: %m\n");
ibv_dereg_mr(rd->recv_mr);
return 1;
}
/* setup work request */
/* recv wq */
rd->recv_sgl.addr = (uint64_t) (unsigned long)&rd->recv_buf;
rd->recv_sgl.length = sizeof(rd->recv_buf);
rd->recv_sgl.lkey = rd->recv_mr->lkey;
rd->rq_wr.sg_list = &rd->recv_sgl;
rd->rq_wr.num_sge = 1;
rd->rq_wr.wr_id = FIO_RDMA_MAX_IO_DEPTH;
/* send wq */
rd->send_sgl.addr = (uint64_t) (unsigned long)&rd->send_buf;
rd->send_sgl.length = sizeof(rd->send_buf);
rd->send_sgl.lkey = rd->send_mr->lkey;
rd->sq_wr.opcode = IBV_WR_SEND;
rd->sq_wr.send_flags = IBV_SEND_SIGNALED;
rd->sq_wr.sg_list = &rd->send_sgl;
rd->sq_wr.num_sge = 1;
rd->sq_wr.wr_id = FIO_RDMA_MAX_IO_DEPTH;
return 0;
}
static int get_next_channel_event(struct thread_data *td,
struct rdma_event_channel *channel,
enum rdma_cm_event_type wait_event)
{
struct rdmaio_data *rd = td->io_ops_data;
struct rdma_cm_event *event;
int ret;
ret = rdma_get_cm_event(channel, &event);
if (ret) {
log_err("fio: rdma_get_cm_event: %d\n", ret);
return 1;
}
if (event->event != wait_event) {
log_err("fio: event is %s instead of %s\n",
rdma_event_str(event->event),
rdma_event_str(wait_event));
return 1;
}
switch (event->event) {
case RDMA_CM_EVENT_CONNECT_REQUEST:
rd->child_cm_id = event->id;
break;
default:
break;
}
rdma_ack_cm_event(event);
return 0;
}
static int fio_rdmaio_prep(struct thread_data *td, struct io_u *io_u)
{
struct rdmaio_data *rd = td->io_ops_data;
struct rdma_io_u_data *r_io_u_d;
r_io_u_d = io_u->engine_data;
switch (rd->rdma_protocol) {
case FIO_RDMA_MEM_WRITE:
case FIO_RDMA_MEM_READ:
r_io_u_d->rdma_sgl.addr = (uint64_t) (unsigned long)io_u->buf;
r_io_u_d->rdma_sgl.lkey = io_u->mr->lkey;
r_io_u_d->sq_wr.wr_id = r_io_u_d->wr_id;
r_io_u_d->sq_wr.send_flags = IBV_SEND_SIGNALED;
r_io_u_d->sq_wr.sg_list = &r_io_u_d->rdma_sgl;
r_io_u_d->sq_wr.num_sge = 1;
break;
case FIO_RDMA_CHA_SEND:
r_io_u_d->rdma_sgl.addr = (uint64_t) (unsigned long)io_u->buf;
r_io_u_d->rdma_sgl.lkey = io_u->mr->lkey;
r_io_u_d->rdma_sgl.length = io_u->buflen;
r_io_u_d->sq_wr.wr_id = r_io_u_d->wr_id;
r_io_u_d->sq_wr.opcode = IBV_WR_SEND;
r_io_u_d->sq_wr.send_flags = IBV_SEND_SIGNALED;
r_io_u_d->sq_wr.sg_list = &r_io_u_d->rdma_sgl;
r_io_u_d->sq_wr.num_sge = 1;
break;
case FIO_RDMA_CHA_RECV:
r_io_u_d->rdma_sgl.addr = (uint64_t) (unsigned long)io_u->buf;
r_io_u_d->rdma_sgl.lkey = io_u->mr->lkey;
r_io_u_d->rdma_sgl.length = io_u->buflen;
r_io_u_d->rq_wr.wr_id = r_io_u_d->wr_id;
r_io_u_d->rq_wr.sg_list = &r_io_u_d->rdma_sgl;
r_io_u_d->rq_wr.num_sge = 1;
break;
default:
log_err("fio: unknown rdma protocol - %d\n", rd->rdma_protocol);
break;
}
return 0;
}
static struct io_u *fio_rdmaio_event(struct thread_data *td, int event)
{
struct rdmaio_data *rd = td->io_ops_data;
struct io_u *io_u;
int i;
io_u = rd->io_us_completed[0];
for (i = 0; i < rd->io_u_completed_nr - 1; i++)
rd->io_us_completed[i] = rd->io_us_completed[i + 1];
rd->io_u_completed_nr--;
dprint_io_u(io_u, "fio_rdmaio_event");
return io_u;
}
static int fio_rdmaio_getevents(struct thread_data *td, unsigned int min,
unsigned int max, const struct timespec *t)
{
struct rdmaio_data *rd = td->io_ops_data;
enum ibv_wc_opcode comp_opcode;
struct ibv_cq *ev_cq;
void *ev_ctx;
int ret, r = 0;
comp_opcode = IBV_WC_RDMA_WRITE;
switch (rd->rdma_protocol) {
case FIO_RDMA_MEM_WRITE:
comp_opcode = IBV_WC_RDMA_WRITE;
break;
case FIO_RDMA_MEM_READ:
comp_opcode = IBV_WC_RDMA_READ;
break;
case FIO_RDMA_CHA_SEND:
comp_opcode = IBV_WC_SEND;
break;
case FIO_RDMA_CHA_RECV:
comp_opcode = IBV_WC_RECV;
break;
default:
log_err("fio: unknown rdma protocol - %d\n", rd->rdma_protocol);
break;
}
if (rd->cq_event_num > 0) { /* previous left */
rd->cq_event_num--;
return 0;
}
again:
if (ibv_get_cq_event(rd->channel, &ev_cq, &ev_ctx) != 0) {
log_err("fio: Failed to get cq event!\n");
return -1;
}
if (ev_cq != rd->cq) {
log_err("fio: Unknown CQ!\n");
return -1;
}
if (ibv_req_notify_cq(rd->cq, 0) != 0) {
log_err("fio: Failed to set notify!\n");
return -1;
}
ret = cq_event_handler(td, comp_opcode);
if (ret < 1)
goto again;
ibv_ack_cq_events(rd->cq, ret);
r += ret;
if (r < min)
goto again;
rd->cq_event_num -= r;
return r;
}
static int fio_rdmaio_send(struct thread_data *td, struct io_u **io_us,
unsigned int nr)
{
struct rdmaio_data *rd = td->io_ops_data;
struct ibv_send_wr *bad_wr;
#if 0
enum ibv_wc_opcode comp_opcode;
comp_opcode = IBV_WC_RDMA_WRITE;
#endif
int i;
long index;
struct rdma_io_u_data *r_io_u_d;
r_io_u_d = NULL;
for (i = 0; i < nr; i++) {
/* RDMA_WRITE or RDMA_READ */
switch (rd->rdma_protocol) {
case FIO_RDMA_MEM_WRITE:
/* compose work request */
r_io_u_d = io_us[i]->engine_data;
index = __rand(&rd->rand_state) % rd->rmt_nr;
r_io_u_d->sq_wr.opcode = IBV_WR_RDMA_WRITE;
r_io_u_d->sq_wr.wr.rdma.rkey = rd->rmt_us[index].rkey;
r_io_u_d->sq_wr.wr.rdma.remote_addr = \
rd->rmt_us[index].buf;
r_io_u_d->sq_wr.sg_list->length = io_us[i]->buflen;
break;
case FIO_RDMA_MEM_READ:
/* compose work request */
r_io_u_d = io_us[i]->engine_data;
index = __rand(&rd->rand_state) % rd->rmt_nr;
r_io_u_d->sq_wr.opcode = IBV_WR_RDMA_READ;
r_io_u_d->sq_wr.wr.rdma.rkey = rd->rmt_us[index].rkey;
r_io_u_d->sq_wr.wr.rdma.remote_addr = \
rd->rmt_us[index].buf;
r_io_u_d->sq_wr.sg_list->length = io_us[i]->buflen;
break;
case FIO_RDMA_CHA_SEND:
r_io_u_d = io_us[i]->engine_data;
r_io_u_d->sq_wr.opcode = IBV_WR_SEND;
r_io_u_d->sq_wr.send_flags = IBV_SEND_SIGNALED;
break;
default:
log_err("fio: unknown rdma protocol - %d\n",
rd->rdma_protocol);
break;
}
if (ibv_post_send(rd->qp, &r_io_u_d->sq_wr, &bad_wr) != 0) {
log_err("fio: ibv_post_send fail: %m\n");
return -1;
}
dprint_io_u(io_us[i], "fio_rdmaio_send");
}
/* wait for completion
rdma_poll_wait(td, comp_opcode); */
return i;
}
static int fio_rdmaio_recv(struct thread_data *td, struct io_u **io_us,
unsigned int nr)
{
struct rdmaio_data *rd = td->io_ops_data;
struct ibv_recv_wr *bad_wr;
struct rdma_io_u_data *r_io_u_d;
int i;
i = 0;
if (rd->rdma_protocol == FIO_RDMA_CHA_RECV) {
/* post io_u into recv queue */
for (i = 0; i < nr; i++) {
r_io_u_d = io_us[i]->engine_data;
if (ibv_post_recv(rd->qp, &r_io_u_d->rq_wr, &bad_wr) !=
0) {
log_err("fio: ibv_post_recv fail: %m\n");
return 1;
}
}
} else if ((rd->rdma_protocol == FIO_RDMA_MEM_READ)
|| (rd->rdma_protocol == FIO_RDMA_MEM_WRITE)) {
/* re-post the rq_wr */
if (ibv_post_recv(rd->qp, &rd->rq_wr, &bad_wr) != 0) {
log_err("fio: ibv_post_recv fail: %m\n");
return 1;
}
rdma_poll_wait(td, IBV_WC_RECV);
dprint(FD_IO, "fio: recv FINISH message\n");
td->done = 1;
return 0;
}
return i;
}
static enum fio_q_status fio_rdmaio_queue(struct thread_data *td,
struct io_u *io_u)
{
struct rdmaio_data *rd = td->io_ops_data;
fio_ro_check(td, io_u);
if (rd->io_u_queued_nr == (int)td->o.iodepth)
return FIO_Q_BUSY;
rd->io_us_queued[rd->io_u_queued_nr] = io_u;
rd->io_u_queued_nr++;
dprint_io_u(io_u, "fio_rdmaio_queue");
return FIO_Q_QUEUED;
}
static void fio_rdmaio_queued(struct thread_data *td, struct io_u **io_us,
unsigned int nr)
{
struct rdmaio_data *rd = td->io_ops_data;
struct timespec now;
unsigned int i;
if (!fio_fill_issue_time(td))
return;
fio_gettime(&now, NULL);
for (i = 0; i < nr; i++) {
struct io_u *io_u = io_us[i];
/* queued -> flight */
rd->io_us_flight[rd->io_u_flight_nr] = io_u;
rd->io_u_flight_nr++;
memcpy(&io_u->issue_time, &now, sizeof(now));
io_u_queued(td, io_u);
}
/*
* only used for iolog
*/
if (td->o.read_iolog_file)
memcpy(&td->last_issue, &now, sizeof(now));
}
static int fio_rdmaio_commit(struct thread_data *td)
{
struct rdmaio_data *rd = td->io_ops_data;
struct io_u **io_us;
int ret;
if (!rd->io_us_queued)
return 0;
io_us = rd->io_us_queued;
do {
/* RDMA_WRITE or RDMA_READ */
if (rd->is_client)
ret = fio_rdmaio_send(td, io_us, rd->io_u_queued_nr);
else if (!rd->is_client)
ret = fio_rdmaio_recv(td, io_us, rd->io_u_queued_nr);
if (ret > 0) {
fio_rdmaio_queued(td, io_us, ret);
io_u_mark_submit(td, ret);
rd->io_u_queued_nr -= ret;
io_us += ret;
ret = 0;
} else
break;
} while (rd->io_u_queued_nr);
return ret;
}
static int fio_rdmaio_connect(struct thread_data *td, struct fio_file *f)
{
struct rdmaio_data *rd = td->io_ops_data;
struct rdma_conn_param conn_param;
struct ibv_send_wr *bad_wr;
memset(&conn_param, 0, sizeof(conn_param));
conn_param.responder_resources = 1;
conn_param.initiator_depth = 1;
conn_param.retry_count = 10;
if (rdma_connect(rd->cm_id, &conn_param) != 0) {
log_err("fio: rdma_connect fail: %m\n");
return 1;
}
if (get_next_channel_event
(td, rd->cm_channel, RDMA_CM_EVENT_ESTABLISHED) != 0) {
log_err("fio: wait for RDMA_CM_EVENT_ESTABLISHED\n");
return 1;
}
/* send task request */
rd->send_buf.mode = htonl(rd->rdma_protocol);
rd->send_buf.nr = htonl(td->o.iodepth);
if (ibv_post_send(rd->qp, &rd->sq_wr, &bad_wr) != 0) {
log_err("fio: ibv_post_send fail: %m\n");
return 1;
}
if (rdma_poll_wait(td, IBV_WC_SEND) < 0)
return 1;
/* wait for remote MR info from server side */
if (rdma_poll_wait(td, IBV_WC_RECV) < 0)
return 1;
/* In SEND/RECV test, it's a good practice to setup the iodepth of
* of the RECV side deeper than that of the SEND side to
* avoid RNR (receiver not ready) error. The
* SEND side may send so many unsolicited message before
* RECV side commits sufficient recv buffers into recv queue.
* This may lead to RNR error. Here, SEND side pauses for a while
* during which RECV side commits sufficient recv buffers.
*/
usleep(500000);
return 0;
}
static int fio_rdmaio_accept(struct thread_data *td, struct fio_file *f)
{
struct rdmaio_data *rd = td->io_ops_data;
struct rdma_conn_param conn_param;
struct ibv_send_wr *bad_wr;
int ret = 0;
/* rdma_accept() - then wait for accept success */
memset(&conn_param, 0, sizeof(conn_param));
conn_param.responder_resources = 1;
conn_param.initiator_depth = 1;
if (rdma_accept(rd->child_cm_id, &conn_param) != 0) {
log_err("fio: rdma_accept: %m\n");
return 1;
}
if (get_next_channel_event
(td, rd->cm_channel, RDMA_CM_EVENT_ESTABLISHED) != 0) {
log_err("fio: wait for RDMA_CM_EVENT_ESTABLISHED\n");
return 1;
}
/* wait for request */
ret = rdma_poll_wait(td, IBV_WC_RECV) < 0;
if (ibv_post_send(rd->qp, &rd->sq_wr, &bad_wr) != 0) {
log_err("fio: ibv_post_send fail: %m\n");
return 1;
}
if (rdma_poll_wait(td, IBV_WC_SEND) < 0)
return 1;
return ret;
}
static int fio_rdmaio_open_file(struct thread_data *td, struct fio_file *f)
{
if (td_read(td))
return fio_rdmaio_accept(td, f);
else
return fio_rdmaio_connect(td, f);
}
static int fio_rdmaio_close_file(struct thread_data *td, struct fio_file *f)
{
struct rdmaio_data *rd = td->io_ops_data;
struct ibv_send_wr *bad_wr;
/* unregister rdma buffer */
/*
* Client sends notification to the server side
*/
/* refer to: http://linux.die.net/man/7/rdma_cm */
if ((rd->is_client == 1) && ((rd->rdma_protocol == FIO_RDMA_MEM_WRITE)
|| (rd->rdma_protocol ==
FIO_RDMA_MEM_READ))) {
if (ibv_post_send(rd->qp, &rd->sq_wr, &bad_wr) != 0) {
log_err("fio: ibv_post_send fail: %m\n");
return 1;
}
dprint(FD_IO, "fio: close information sent success\n");
rdma_poll_wait(td, IBV_WC_SEND);
}
if (rd->is_client == 1)
rdma_disconnect(rd->cm_id);
else {
rdma_disconnect(rd->child_cm_id);
#if 0
rdma_disconnect(rd->cm_id);
#endif
}
#if 0