aboutsummaryrefslogtreecommitdiffstatshomepage
path: root/drivers/platform/surface/aggregator/ssh_request_layer.c
blob: 69132976d297e24948de68f3ccb0f8067e54f1e4 (plain) (blame)
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
// SPDX-License-Identifier: GPL-2.0+
/*
 * SSH request transport layer.
 *
 * Copyright (C) 2019-2022 Maximilian Luz <luzmaximilian@gmail.com>
 */

#include <asm/unaligned.h>
#include <linux/atomic.h>
#include <linux/completion.h>
#include <linux/error-injection.h>
#include <linux/ktime.h>
#include <linux/limits.h>
#include <linux/list.h>
#include <linux/slab.h>
#include <linux/spinlock.h>
#include <linux/types.h>
#include <linux/workqueue.h>

#include <linux/surface_aggregator/serial_hub.h>
#include <linux/surface_aggregator/controller.h>

#include "ssh_packet_layer.h"
#include "ssh_request_layer.h"

#include "trace.h"

/*
 * SSH_RTL_REQUEST_TIMEOUT - Request timeout.
 *
 * Timeout as ktime_t delta for request responses. If we have not received a
 * response in this time-frame after finishing the underlying packet
 * transmission, the request will be completed with %-ETIMEDOUT as status
 * code.
 */
#define SSH_RTL_REQUEST_TIMEOUT			ms_to_ktime(3000)

/*
 * SSH_RTL_REQUEST_TIMEOUT_RESOLUTION - Request timeout granularity.
 *
 * Time-resolution for timeouts. Should be larger than one jiffy to avoid
 * direct re-scheduling of reaper work_struct.
 */
#define SSH_RTL_REQUEST_TIMEOUT_RESOLUTION	ms_to_ktime(max(2000 / HZ, 50))

/*
 * SSH_RTL_MAX_PENDING - Maximum number of pending requests.
 *
 * Maximum number of requests concurrently waiting to be completed (i.e.
 * waiting for the corresponding packet transmission to finish if they don't
 * have a response or waiting for a response if they have one).
 */
#define SSH_RTL_MAX_PENDING		3

/*
 * SSH_RTL_TX_BATCH - Maximum number of requests processed per work execution.
 * Used to prevent livelocking of the workqueue. Value chosen via educated
 * guess, may be adjusted.
 */
#define SSH_RTL_TX_BATCH		10

#ifdef CONFIG_SURFACE_AGGREGATOR_ERROR_INJECTION

/**
 * ssh_rtl_should_drop_response() - Error injection hook to drop request
 * responses.
 *
 * Useful to cause request transmission timeouts in the driver by dropping the
 * response to a request.
 */
static noinline bool ssh_rtl_should_drop_response(void)
{
	return false;
}
ALLOW_ERROR_INJECTION(ssh_rtl_should_drop_response, TRUE);

#else

static inline bool ssh_rtl_should_drop_response(void)
{
	return false;
}

#endif

static u16 ssh_request_get_rqid(struct ssh_request *rqst)
{
	return get_unaligned_le16(rqst->packet.data.ptr
				  + SSH_MSGOFFSET_COMMAND(rqid));
}

static u32 ssh_request_get_rqid_safe(struct ssh_request *rqst)
{
	if (!rqst->packet.data.ptr)
		return U32_MAX;

	return ssh_request_get_rqid(rqst);
}

static void ssh_rtl_queue_remove(struct ssh_request *rqst)
{
	struct ssh_rtl *rtl = ssh_request_rtl(rqst);

	spin_lock(&rtl->queue.lock);

	if (!test_and_clear_bit(SSH_REQUEST_SF_QUEUED_BIT, &rqst->state)) {
		spin_unlock(&rtl->queue.lock);
		return;
	}

	list_del(&rqst->node);

	spin_unlock(&rtl->queue.lock);
	ssh_request_put(rqst);
}

static bool ssh_rtl_queue_empty(struct ssh_rtl *rtl)
{
	bool empty;

	spin_lock(&rtl->queue.lock);
	empty = list_empty(&rtl->queue.head);
	spin_unlock(&rtl->queue.lock);

	return empty;
}

static void ssh_rtl_pending_remove(struct ssh_request *rqst)
{
	struct ssh_rtl *rtl = ssh_request_rtl(rqst);

	spin_lock(&rtl->pending.lock);

	if (!test_and_clear_bit(SSH_REQUEST_SF_PENDING_BIT, &rqst->state)) {
		spin_unlock(&rtl->pending.lock);
		return;
	}

	atomic_dec(&rtl->pending.count);
	list_del(&rqst->node);

	spin_unlock(&rtl->pending.lock);

	ssh_request_put(rqst);
}

static int ssh_rtl_tx_pending_push(struct ssh_request *rqst)
{
	struct ssh_rtl *rtl = ssh_request_rtl(rqst);

	spin_lock(&rtl->pending.lock);

	if (test_bit(SSH_REQUEST_SF_LOCKED_BIT, &rqst->state)) {
		spin_unlock(&rtl->pending.lock);
		return -EINVAL;
	}

	if (test_and_set_bit(SSH_REQUEST_SF_PENDING_BIT, &rqst->state)) {
		spin_unlock(&rtl->pending.lock);
		return -EALREADY;
	}

	atomic_inc(&rtl->pending.count);
	list_add_tail(&ssh_request_get(rqst)->node, &rtl->pending.head);

	spin_unlock(&rtl->pending.lock);
	return 0;
}

static void ssh_rtl_complete_with_status(struct ssh_request *rqst, int status)
{
	struct ssh_rtl *rtl = ssh_request_rtl(rqst);

	trace_ssam_request_complete(rqst, status);

	/* rtl/ptl may not be set if we're canceling before submitting. */
	rtl_dbg_cond(rtl, "rtl: completing request (rqid: %#06x, status: %d)\n",
		     ssh_request_get_rqid_safe(rqst), status);

	rqst->ops->complete(rqst, NULL, NULL, status);
}

static void ssh_rtl_complete_with_rsp(struct ssh_request *rqst,
				      const struct ssh_command *cmd,
				      const struct ssam_span *data)
{
	struct ssh_rtl *rtl = ssh_request_rtl(rqst);

	trace_ssam_request_complete(rqst, 0);

	rtl_dbg(rtl, "rtl: completing request with response (rqid: %#06x)\n",
		ssh_request_get_rqid(rqst));

	rqst->ops->complete(rqst, cmd, data, 0);
}

static bool ssh_rtl_tx_can_process(struct ssh_request *rqst)
{
	struct ssh_rtl *rtl = ssh_request_rtl(rqst);

	if (test_bit(SSH_REQUEST_TY_FLUSH_BIT, &rqst->state))
		return !atomic_read(&rtl->pending.count);

	return atomic_read(&rtl->pending.count) < SSH_RTL_MAX_PENDING;
}

static struct ssh_request *ssh_rtl_tx_next(struct ssh_rtl *rtl)
{
	struct ssh_request *rqst = ERR_PTR(-ENOENT);
	struct ssh_request *p, *n;

	spin_lock(&rtl->queue.lock);

	/* Find first non-locked request and remove it. */
	list_for_each_entry_safe(p, n, &rtl->queue.head, node) {
		if (unlikely(test_bit(SSH_REQUEST_SF_LOCKED_BIT, &p->state)))
			continue;

		if (!ssh_rtl_tx_can_process(p)) {
			rqst = ERR_PTR(-EBUSY);
			break;
		}

		/* Remove from queue and mark as transmitting. */
		set_bit(SSH_REQUEST_SF_TRANSMITTING_BIT, &p->state);
		/* Ensure state never gets zero. */
		smp_mb__before_atomic();
		clear_bit(SSH_REQUEST_SF_QUEUED_BIT, &p->state);

		list_del(&p->node);

		rqst = p;
		break;
	}

	spin_unlock(&rtl->queue.lock);
	return rqst;
}

static int ssh_rtl_tx_try_process_one(struct ssh_rtl *rtl)
{
	struct ssh_request *rqst;
	int status;

	/* Get and prepare next request for transmit. */
	rqst = ssh_rtl_tx_next(rtl);
	if (IS_ERR(rqst))
		return PTR_ERR(rqst);

	/* Add it to/mark it as pending. */
	status = ssh_rtl_tx_pending_push(rqst);
	if (status) {
		ssh_request_put(rqst);
		return -EAGAIN;
	}

	/* Submit packet. */
	status = ssh_ptl_submit(&rtl->ptl, &rqst->packet);
	if (status == -ESHUTDOWN) {
		/*
		 * Packet has been refused due to the packet layer shutting
		 * down. Complete it here.
		 */
		set_bit(SSH_REQUEST_SF_LOCKED_BIT, &rqst->state);
		/*
		 * Note: A barrier is not required here, as there are only two
		 * references in the system at this point: The one that we have,
		 * and the other one that belongs to the pending set. Due to the
		 * request being marked as "transmitting", our process is the
		 * only one allowed to remove the pending node and change the
		 * state. Normally, the task would fall to the packet callback,
		 * but as this is a path where submission failed, this callback
		 * will never be executed.
		 */

		ssh_rtl_pending_remove(rqst);
		ssh_rtl_complete_with_status(rqst, -ESHUTDOWN);

		ssh_request_put(rqst);
		return -ESHUTDOWN;

	} else if (status) {
		/*
		 * If submitting the packet failed and the packet layer isn't
		 * shutting down, the packet has either been submitted/queued
		 * before (-EALREADY, which cannot happen as we have
		 * guaranteed that requests cannot be re-submitted), or the
		 * packet was marked as locked (-EINVAL). To mark the packet
		 * locked at this stage, the request, and thus the packets
		 * itself, had to have been canceled. Simply drop the
		 * reference. Cancellation itself will remove it from the set
		 * of pending requests.
		 */

		WARN_ON(status != -EINVAL);

		ssh_request_put(rqst);
		return -EAGAIN;
	}

	ssh_request_put(rqst);
	return 0;
}

static bool ssh_rtl_tx_schedule(struct ssh_rtl *rtl)
{
	if (atomic_read(&rtl->pending.count) >= SSH_RTL_MAX_PENDING)
		return false;

	if (ssh_rtl_queue_empty(rtl))
		return false;

	return schedule_work(&rtl->tx.work);
}

static void ssh_rtl_tx_work_fn(struct work_struct *work)
{
	struct ssh_rtl *rtl = to_ssh_rtl(work, tx.work);
	unsigned int iterations = SSH_RTL_TX_BATCH;
	int status;

	/*
	 * Try to be nice and not block/live-lock the workqueue: Run a maximum
	 * of 10 tries, then re-submit if necessary. This should not be
	 * necessary for normal execution, but guarantee it anyway.
	 */
	do {
		status = ssh_rtl_tx_try_process_one(rtl);
		if (status == -ENOENT || status == -EBUSY)
			return;		/* No more requests to process. */

		if (status == -ESHUTDOWN) {
			/*
			 * Packet system shutting down. No new packets can be
			 * transmitted. Return silently, the party initiating
			 * the shutdown should handle the rest.
			 */
			return;
		}

		WARN_ON(status != 0 && status != -EAGAIN);
	} while (--iterations);

	/* Out of tries, reschedule. */
	ssh_rtl_tx_schedule(rtl);
}

/**
 * ssh_rtl_submit() - Submit a request to the transport layer.
 * @rtl:  The request transport layer.
 * @rqst: The request to submit.
 *
 * Submits a request to the transport layer. A single request may not be
 * submitted multiple times without reinitializing it.
 *
 * Return: Returns zero on success, %-EINVAL if the request type is invalid or
 * the request has been canceled prior to submission, %-EALREADY if the
 * request has already been submitted, or %-ESHUTDOWN in case the request
 * transport layer has been shut down.
 */
int ssh_rtl_submit(struct ssh_rtl *rtl, struct ssh_request *rqst)
{
	trace_ssam_request_submit(rqst);

	/*
	 * Ensure that requests expecting a response are sequenced. If this
	 * invariant ever changes, see the comment in ssh_rtl_complete() on what
	 * is required to be changed in the code.
	 */
	if (test_bit(SSH_REQUEST_TY_HAS_RESPONSE_BIT, &rqst->state))
		if (!test_bit(SSH_PACKET_TY_SEQUENCED_BIT, &rqst->packet.state))
			return -EINVAL;

	spin_lock(&rtl->queue.lock);

	/*
	 * Try to set ptl and check if this request has already been submitted.
	 *
	 * Must be inside lock as we might run into a lost update problem
	 * otherwise: If this were outside of the lock, cancellation in
	 * ssh_rtl_cancel_nonpending() may run after we've set the ptl
	 * reference but before we enter the lock. In that case, we'd detect
	 * that the request is being added to the queue and would try to remove
	 * it from that, but removal might fail because it hasn't actually been
	 * added yet. By putting this cmpxchg in the critical section, we
	 * ensure that the queuing detection only triggers when we are already
	 * in the critical section and the remove process will wait until the
	 * push operation has been completed (via lock) due to that. Only then,
	 * we can safely try to remove it.
	 */
	if (cmpxchg(&rqst->packet.ptl, NULL, &rtl->ptl)) {
		spin_unlock(&rtl->queue.lock);
		return -EALREADY;
	}

	/*
	 * Ensure that we set ptl reference before we continue modifying state.
	 * This is required for non-pending cancellation. This barrier is paired
	 * with the one in ssh_rtl_cancel_nonpending().
	 *
	 * By setting the ptl reference before we test for "locked", we can
	 * check if the "locked" test may have already run. See comments in
	 * ssh_rtl_cancel_nonpending() for more detail.
	 */
	smp_mb__after_atomic();

	if (test_bit(SSH_RTL_SF_SHUTDOWN_BIT, &rtl->state)) {
		spin_unlock(&rtl->queue.lock);
		return -ESHUTDOWN;
	}

	if (test_bit(SSH_REQUEST_SF_LOCKED_BIT, &rqst->state)) {
		spin_unlock(&rtl->queue.lock);
		return -EINVAL;
	}

	set_bit(SSH_REQUEST_SF_QUEUED_BIT, &rqst->state);
	list_add_tail(&ssh_request_get(rqst)->node, &rtl->queue.head);

	spin_unlock(&rtl->queue.lock);

	ssh_rtl_tx_schedule(rtl);
	return 0;
}

static void ssh_rtl_timeout_reaper_mod(struct ssh_rtl *rtl, ktime_t now,
				       ktime_t expires)
{
	unsigned long delta = msecs_to_jiffies(ktime_ms_delta(expires, now));
	ktime_t aexp = ktime_add(expires, SSH_RTL_REQUEST_TIMEOUT_RESOLUTION);

	spin_lock(&rtl->rtx_timeout.lock);

	/* Re-adjust / schedule reaper only if it is above resolution delta. */
	if (ktime_before(aexp, rtl->rtx_timeout.expires)) {
		rtl->rtx_timeout.expires = expires;
		mod_delayed_work(system_wq, &rtl->rtx_timeout.reaper, delta);
	}

	spin_unlock(&rtl->rtx_timeout.lock);
}

static void ssh_rtl_timeout_start(struct ssh_request *rqst)
{
	struct ssh_rtl *rtl = ssh_request_rtl(rqst);
	ktime_t timestamp = ktime_get_coarse_boottime();
	ktime_t timeout = rtl->rtx_timeout.timeout;

	if (test_bit(SSH_REQUEST_SF_LOCKED_BIT, &rqst->state))
		return;

	/*
	 * Note: The timestamp gets set only once. This happens on the packet
	 * callback. All other access to it is read-only.
	 */
	WRITE_ONCE(rqst->timestamp, timestamp);
	/*
	 * Ensure timestamp is set before starting the reaper. Paired with
	 * implicit barrier following check on ssh_request_get_expiration() in
	 * ssh_rtl_timeout_reap.
	 */
	smp_mb__after_atomic();

	ssh_rtl_timeout_reaper_mod(rtl, timestamp, timestamp + timeout);
}

static void ssh_rtl_complete(struct ssh_rtl *rtl,
			     const struct ssh_command *command,
			     const struct ssam_span *command_data)
{
	struct ssh_request *r = NULL;
	struct ssh_request *p, *n;
	u16 rqid = get_unaligned_le16(&command->rqid);

	trace_ssam_rx_response_received(command, command_data->len);

	/*
	 * Get request from pending based on request ID and mark it as response
	 * received and locked.
	 */
	spin_lock(&rtl->pending.lock);
	list_for_each_entry_safe(p, n, &rtl->pending.head, node) {
		/* We generally expect requests to be processed in order. */
		if (unlikely(ssh_request_get_rqid(p) != rqid))
			continue;

		/* Simulate response timeout. */
		if (ssh_rtl_should_drop_response()) {
			spin_unlock(&rtl->pending.lock);

			trace_ssam_ei_rx_drop_response(p);
			rtl_info(rtl, "request error injection: dropping response for request %p\n",
				 &p->packet);
			return;
		}

		/*
		 * Mark as "response received" and "locked" as we're going to
		 * complete it.
		 */
		set_bit(SSH_REQUEST_SF_LOCKED_BIT, &p->state);
		set_bit(SSH_REQUEST_SF_RSPRCVD_BIT, &p->state);
		/* Ensure state never gets zero. */
		smp_mb__before_atomic();
		clear_bit(SSH_REQUEST_SF_PENDING_BIT, &p->state);

		atomic_dec(&rtl->pending.count);
		list_del(&p->node);

		r = p;
		break;
	}
	spin_unlock(&rtl->pending.lock);

	if (!r) {
		rtl_warn(rtl, "rtl: dropping unexpected command message (rqid = %#06x)\n",
			 rqid);
		return;
	}

	/* If the request hasn't been completed yet, we will do this now. */
	if (test_and_set_bit(SSH_REQUEST_SF_COMPLETED_BIT, &r->state)) {
		ssh_request_put(r);
		ssh_rtl_tx_schedule(rtl);
		return;
	}

	/*
	 * Make sure the request has been transmitted. In case of a sequenced
	 * request, we are guaranteed that the completion callback will run on
	 * the receiver thread directly when the ACK for the packet has been
	 * received. Similarly, this function is guaranteed to run on the
	 * receiver thread. Thus we are guaranteed that if the packet has been
	 * successfully transmitted and received an ACK, the transmitted flag
	 * has been set and is visible here.
	 *
	 * We are currently not handling unsequenced packets here, as those
	 * should never expect a response as ensured in ssh_rtl_submit. If this
	 * ever changes, one would have to test for
	 *
	 *	(r->state & (transmitting | transmitted))
	 *
	 * on unsequenced packets to determine if they could have been
	 * transmitted. There are no synchronization guarantees as in the
	 * sequenced case, since, in this case, the callback function will not
	 * run on the same thread. Thus an exact determination is impossible.
	 */
	if (!test_bit(SSH_REQUEST_SF_TRANSMITTED_BIT, &r->state)) {
		rtl_err(rtl, "rtl: received response before ACK for request (rqid = %#06x)\n",
			rqid);

		/*
		 * NB: Timeout has already been canceled, request already been
		 * removed from pending and marked as locked and completed. As
		 * we receive a "false" response, the packet might still be
		 * queued though.
		 */
		ssh_rtl_queue_remove(r);

		ssh_rtl_complete_with_status(r, -EREMOTEIO);
		ssh_request_put(r);

		ssh_rtl_tx_schedule(rtl);
		return;
	}

	/*
	 * NB: Timeout has already been canceled, request already been
	 * removed from pending and marked as locked and completed. The request
	 * can also not be queued any more, as it has been marked as
	 * transmitting and later transmitted. Thus no need to remove it from
	 * anywhere.
	 */

	ssh_rtl_complete_with_rsp(r, command, command_data);
	ssh_request_put(r);

	ssh_rtl_tx_schedule(rtl);
}

static bool ssh_rtl_cancel_nonpending(struct ssh_request *r)
{
	struct ssh_rtl *rtl;
	unsigned long flags, fixed;
	bool remove;

	/*
	 * Handle unsubmitted request: Try to mark the packet as locked,
	 * expecting the state to be zero (i.e. unsubmitted). Note that, if
	 * setting the state worked, we might still be adding the packet to the
	 * queue in a currently executing submit call. In that case, however,
	 * ptl reference must have been set previously, as locked is checked
	 * after setting ptl. Furthermore, when the ptl reference is set, the
	 * submission process is guaranteed to have entered the critical
	 * section. Thus only if we successfully locked this request and ptl is
	 * NULL, we have successfully removed the request, i.e. we are
	 * guaranteed that, due to the "locked" check in ssh_rtl_submit(), the
	 * packet will never be added. Otherwise, we need to try and grab it
	 * from the queue, where we are now guaranteed that the packet is or has
	 * been due to the critical section.
	 *
	 * Note that if the cmpxchg() fails, we are guaranteed that ptl has
	 * been set and is non-NULL, as states can only be nonzero after this
	 * has been set. Also note that we need to fetch the static (type)
	 * flags to ensure that they don't cause the cmpxchg() to fail.
	 */
	fixed = READ_ONCE(r->state) & SSH_REQUEST_FLAGS_TY_MASK;
	flags = cmpxchg(&r->state, fixed, SSH_REQUEST_SF_LOCKED_BIT);

	/*
	 * Force correct ordering with regards to state and ptl reference access
	 * to safe-guard cancellation to concurrent submission against a
	 * lost-update problem. First try to exchange state, then also check
	 * ptl if that worked. This barrier is paired with the
	 * one in ssh_rtl_submit().
	 */
	smp_mb__after_atomic();

	if (flags == fixed && !READ_ONCE(r->packet.ptl)) {
		if (test_and_set_bit(SSH_REQUEST_SF_COMPLETED_BIT, &r->state))
			return true;

		ssh_rtl_complete_with_status(r, -ECANCELED);
		return true;
	}

	rtl = ssh_request_rtl(r);
	spin_lock(&rtl->queue.lock);

	/*
	 * Note: 1) Requests cannot be re-submitted. 2) If a request is
	 * queued, it cannot be "transmitting"/"pending" yet. Thus, if we
	 * successfully remove the request here, we have removed all its
	 * occurrences in the system.
	 */

	remove = test_and_clear_bit(SSH_REQUEST_SF_QUEUED_BIT, &r->state);
	if (!remove) {
		spin_unlock(&rtl->queue.lock);
		return false;
	}

	set_bit(SSH_REQUEST_SF_LOCKED_BIT, &r->state);
	list_del(&r->node);

	spin_unlock(&rtl->queue.lock);

	ssh_request_put(r);	/* Drop reference obtained from queue. */

	if (test_and_set_bit(SSH_REQUEST_SF_COMPLETED_BIT, &r->state))
		return true;

	ssh_rtl_complete_with_status(r, -ECANCELED);
	return true;
}

static bool ssh_rtl_cancel_pending(struct ssh_request *r)
{
	/* If the packet is already locked, it's going to be removed shortly. */
	if (test_and_set_bit(SSH_REQUEST_SF_LOCKED_BIT, &r->state))
		return true;

	/*
	 * Now that we have locked the packet, we have guaranteed that it can't
	 * be added to the system any more. If ptl is NULL, the locked
	 * check in ssh_rtl_submit() has not been run and any submission,
	 * currently in progress or called later, won't add the packet. Thus we
	 * can directly complete it.
	 *
	 * The implicit memory barrier of test_and_set_bit() should be enough
	 * to ensure that the correct order (first lock, then check ptl) is
	 * ensured. This is paired with the barrier in ssh_rtl_submit().
	 */
	if (!READ_ONCE(r->packet.ptl)) {
		if (test_and_set_bit(SSH_REQUEST_SF_COMPLETED_BIT, &r->state))
			return true;

		ssh_rtl_complete_with_status(r, -ECANCELED);
		return true;
	}

	/*
	 * Try to cancel the packet. If the packet has not been completed yet,
	 * this will subsequently (and synchronously) call the completion
	 * callback of the packet, which will complete the request.
	 */
	ssh_ptl_cancel(&r->packet);

	/*
	 * If the packet has been completed with success, i.e. has not been
	 * canceled by the above call, the request may not have been completed
	 * yet (may be waiting for a response). Check if we need to do this
	 * here.
	 */
	if (test_and_set_bit(SSH_REQUEST_SF_COMPLETED_BIT, &r->state))
		return true;

	ssh_rtl_queue_remove(r);
	ssh_rtl_pending_remove(r);
	ssh_rtl_complete_with_status(r, -ECANCELED);

	return true;
}

/**
 * ssh_rtl_cancel() - Cancel request.
 * @rqst:    The request to cancel.
 * @pending: Whether to also cancel pending requests.
 *
 * Cancels the given request. If @pending is %false, this will not cancel
 * pending requests, i.e. requests that have already been submitted to the
 * packet layer but not been completed yet. If @pending is %true, this will
 * cancel the given request regardless of the state it is in.
 *
 * If the request has been canceled by calling this function, both completion
 * and release callbacks of the request will be executed in a reasonable
 * time-frame. This may happen during execution of this function, however,
 * there is no guarantee for this. For example, a request currently
 * transmitting will be canceled/completed only after transmission has
 * completed, and the respective callbacks will be executed on the transmitter
 * thread, which may happen during, but also some time after execution of the
 * cancel function.
 *
 * Return: Returns %true if the given request has been canceled or completed,
 * either by this function or prior to calling this function, %false
 * otherwise. If @pending is %true, this function will always return %true.
 */
bool ssh_rtl_cancel(struct ssh_request *rqst, bool pending)
{
	struct ssh_rtl *rtl;
	bool canceled;

	if (test_and_set_bit(SSH_REQUEST_SF_CANCELED_BIT, &rqst->state))
		return true;

	trace_ssam_request_cancel(rqst);

	if (pending)
		canceled = ssh_rtl_cancel_pending(rqst);
	else
		canceled = ssh_rtl_cancel_nonpending(rqst);

	/* Note: rtl may be NULL if request has not been submitted yet. */
	rtl = ssh_request_rtl(rqst);
	if (canceled && rtl)
		ssh_rtl_tx_schedule(rtl);

	return canceled;
}

static void ssh_rtl_packet_callback(struct ssh_packet *p, int status)
{
	struct ssh_request *r = to_ssh_request(p);

	if (unlikely(status)) {
		set_bit(SSH_REQUEST_SF_LOCKED_BIT, &r->state);

		if (test_and_set_bit(SSH_REQUEST_SF_COMPLETED_BIT, &r->state))
			return;

		/*
		 * The packet may get canceled even though it has not been
		 * submitted yet. The request may still be queued. Check the
		 * queue and remove it if necessary. As the timeout would have
		 * been started in this function on success, there's no need
		 * to cancel it here.
		 */
		ssh_rtl_queue_remove(r);
		ssh_rtl_pending_remove(r);
		ssh_rtl_complete_with_status(r, status);

		ssh_rtl_tx_schedule(ssh_request_rtl(r));
		return;
	}

	/* Update state: Mark as transmitted and clear transmitting. */
	set_bit(SSH_REQUEST_SF_TRANSMITTED_BIT, &r->state);
	/* Ensure state never gets zero. */
	smp_mb__before_atomic();
	clear_bit(SSH_REQUEST_SF_TRANSMITTING_BIT, &r->state);

	/* If we expect a response, we just need to start the timeout. */
	if (test_bit(SSH_REQUEST_TY_HAS_RESPONSE_BIT, &r->state)) {
		/*
		 * Note: This is the only place where the timestamp gets set,
		 * all other access to it is read-only.
		 */
		ssh_rtl_timeout_start(r);
		return;
	}

	/*
	 * If we don't expect a response, lock, remove, and complete the
	 * request. Note that, at this point, the request is guaranteed to have
	 * left the queue and no timeout has been started. Thus we only need to
	 * remove it from pending. If the request has already been completed (it
	 * may have been canceled) return.
	 */

	set_bit(SSH_REQUEST_SF_LOCKED_BIT, &r->state);
	if (test_and_set_bit(SSH_REQUEST_SF_COMPLETED_BIT, &r->state))
		return;

	ssh_rtl_pending_remove(r);
	ssh_rtl_complete_with_status(r, 0);

	ssh_rtl_tx_schedule(ssh_request_rtl(r));
}

static ktime_t ssh_request_get_expiration(struct ssh_request *r, ktime_t timeout)
{
	ktime_t timestamp = READ_ONCE(r->timestamp);

	if (timestamp != KTIME_MAX)
		return ktime_add(timestamp, timeout);
	else
		return KTIME_MAX;
}

static void ssh_rtl_timeout_reap(struct work_struct *work)
{
	struct ssh_rtl *rtl = to_ssh_rtl(work, rtx_timeout.reaper.work);
	struct ssh_request *r, *n;
	LIST_HEAD(claimed);
	ktime_t now = ktime_get_coarse_boottime();
	ktime_t timeout = rtl->rtx_timeout.timeout;
	ktime_t next = KTIME_MAX;

	trace_ssam_rtl_timeout_reap(atomic_read(&rtl->pending.count));

	/*
	 * Mark reaper as "not pending". This is done before checking any
	 * requests to avoid lost-update type problems.
	 */
	spin_lock(&rtl->rtx_timeout.lock);
	rtl->rtx_timeout.expires = KTIME_MAX;
	spin_unlock(&rtl->rtx_timeout.lock);

	spin_lock(&rtl->pending.lock);
	list_for_each_entry_safe(r, n, &rtl->pending.head, node) {
		ktime_t expires = ssh_request_get_expiration(r, timeout);

		/*
		 * Check if the timeout hasn't expired yet. Find out next
		 * expiration date to be handled after this run.
		 */
		if (ktime_after(expires, now)) {
			next = ktime_before(expires, next) ? expires : next;
			continue;
		}

		/* Avoid further transitions if locked. */
		if (test_and_set_bit(SSH_REQUEST_SF_LOCKED_BIT, &r->state))
			continue;

		/*
		 * We have now marked the packet as locked. Thus it cannot be
		 * added to the pending or queued lists again after we've
		 * removed it here. We can therefore re-use the node of this
		 * packet temporarily.
		 */

		clear_bit(SSH_REQUEST_SF_PENDING_BIT, &r->state);

		atomic_dec(&rtl->pending.count);
		list_move_tail(&r->node, &claimed);
	}
	spin_unlock(&rtl->pending.lock);

	/* Cancel and complete the request. */
	list_for_each_entry_safe(r, n, &claimed, node) {
		trace_ssam_request_timeout(r);

		/*
		 * At this point we've removed the packet from pending. This
		 * means that we've obtained the last (only) reference of the
		 * system to it. Thus we can just complete it.
		 */
		if (!test_and_set_bit(SSH_REQUEST_SF_COMPLETED_BIT, &r->state))
			ssh_rtl_complete_with_status(r, -ETIMEDOUT);

		/*
		 * Drop the reference we've obtained by removing it from the
		 * pending set.
		 */
		list_del(&r->node);
		ssh_request_put(r);
	}

	/* Ensure that the reaper doesn't run again immediately. */
	next = max(next, ktime_add(now, SSH_RTL_REQUEST_TIMEOUT_RESOLUTION));
	if (next != KTIME_MAX)
		ssh_rtl_timeout_reaper_mod(rtl, now, next);

	ssh_rtl_tx_schedule(rtl);
}

static void ssh_rtl_rx_event(struct ssh_rtl *rtl, const struct ssh_command *cmd,
			     const struct ssam_span *data)
{
	trace_ssam_rx_event_received(cmd, data->len);

	rtl_dbg(rtl, "rtl: handling event (rqid: %#06x)\n",
		get_unaligned_le16(&cmd->rqid));

	rtl->ops.handle_event(rtl, cmd, data);
}

static void ssh_rtl_rx_command(struct ssh_ptl *p, const struct ssam_span *data)
{
	struct ssh_rtl *rtl = to_ssh_rtl(p, ptl);
	struct device *dev = &p->serdev->dev;
	struct ssh_command *command;
	struct ssam_span command_data;

	if (sshp_parse_command(dev, data, &command, &command_data))
		return;

	/*
	 * Check if the message was intended for us. If not, drop it.
	 *
	 * Note: We will need to change this to handle debug messages. On newer
	 * generation devices, these seem to be sent to tid_out=0x03. We as
	 * host can still receive them as they can be forwarded via an override
	 * option on SAM, but doing so does not change tid_out=0x00.
	 */
	if (command->tid_out != 0x00) {
		rtl_warn(rtl, "rtl: dropping message not intended for us (tid = %#04x)\n",
			 command->tid_out);
		return;
	}

	if (ssh_rqid_is_event(get_unaligned_le16(&command->rqid)))
		ssh_rtl_rx_event(rtl, command, &command_data);
	else
		ssh_rtl_complete(rtl, command, &command_data);
}

static void ssh_rtl_rx_data(struct ssh_ptl *p, const struct ssam_span *data)
{
	if (!data->len) {
		ptl_err(p, "rtl: rx: no data frame payload\n");
		return;
	}

	switch (data->ptr[0]) {
	case SSH_PLD_TYPE_CMD:
		ssh_rtl_rx_command(p, data);
		break;

	default:
		ptl_err(p, "rtl: rx: unknown frame payload type (type: %#04x)\n",
			data->ptr[0]);
		break;
	}
}

static void ssh_rtl_packet_release(struct ssh_packet *p)
{
	struct ssh_request *rqst;

	rqst = to_ssh_request(p);
	rqst->ops->release(rqst);
}

static const struct ssh_packet_ops ssh_rtl_packet_ops = {
	.complete = ssh_rtl_packet_callback,
	.release = ssh_rtl_packet_release,
};

/**
 * ssh_request_init() - Initialize SSH request.
 * @rqst:  The request to initialize.
 * @flags: Request flags, determining the type of the request.
 * @ops:   Request operations.
 *
 * Initializes the given SSH request and underlying packet. Sets the message
 * buffer pointer to %NULL and the message buffer length to zero. This buffer
 * has to be set separately via ssh_request_set_data() before submission and
 * must contain a valid SSH request message.
 *
 * Return: Returns zero on success or %-EINVAL if the given flags are invalid.
 */
int ssh_request_init(struct ssh_request *rqst, enum ssam_request_flags flags,
		     const struct ssh_request_ops *ops)
{
	unsigned long type = BIT(SSH_PACKET_TY_BLOCKING_BIT);

	/* Unsequenced requests cannot have a response. */
	if (flags & SSAM_REQUEST_UNSEQUENCED && flags & SSAM_REQUEST_HAS_RESPONSE)
		return -EINVAL;

	if (!(flags & SSAM_REQUEST_UNSEQUENCED))
		type |= BIT(SSH_PACKET_TY_SEQUENCED_BIT);

	ssh_packet_init(&rqst->packet, type, SSH_PACKET_PRIORITY(DATA, 0),
			&ssh_rtl_packet_ops);

	INIT_LIST_HEAD(&rqst->node);

	rqst->state = 0;
	if (flags & SSAM_REQUEST_HAS_RESPONSE)
		rqst->state |= BIT(SSH_REQUEST_TY_HAS_RESPONSE_BIT);

	rqst->timestamp = KTIME_MAX;
	rqst->ops = ops;

	return 0;
}

/**
 * ssh_rtl_init() - Initialize request transport layer.
 * @rtl:    The request transport layer to initialize.
 * @serdev: The underlying serial device, i.e. the lower-level transport.
 * @ops:    Request transport layer operations.
 *
 * Initializes the given request transport layer and associated packet
 * transport layer. Transmitter and receiver threads must be started
 * separately via ssh_rtl_start(), after the request-layer has been
 * initialized and the lower-level serial device layer has been set up.
 *
 * Return: Returns zero on success and a nonzero error code on failure.
 */
int ssh_rtl_init(struct ssh_rtl *rtl, struct serdev_device *serdev,
		 const struct ssh_rtl_ops *ops)
{
	struct ssh_ptl_ops ptl_ops;
	int status;

	ptl_ops.data_received = ssh_rtl_rx_data;

	status = ssh_ptl_init(&rtl->ptl, serdev, &ptl_ops);
	if (status)
		return status;

	spin_lock_init(&rtl->queue.lock);
	INIT_LIST_HEAD(&rtl->queue.head);

	spin_lock_init(&rtl->pending.lock);
	INIT_LIST_HEAD(&rtl->pending.head);
	atomic_set_release(&rtl->pending.count, 0);

	INIT_WORK(&rtl->tx.work, ssh_rtl_tx_work_fn);

	spin_lock_init(&rtl->rtx_timeout.lock);
	rtl->rtx_timeout.timeout = SSH_RTL_REQUEST_TIMEOUT;
	rtl->rtx_timeout.expires = KTIME_MAX;
	INIT_DELAYED_WORK(&rtl->rtx_timeout.reaper, ssh_rtl_timeout_reap);

	rtl->ops = *ops;

	return 0;
}

/**
 * ssh_rtl_destroy() - Deinitialize request transport layer.
 * @rtl: The request transport layer to deinitialize.
 *
 * Deinitializes the given request transport layer and frees resources
 * associated with it. If receiver and/or transmitter threads have been
 * started, the layer must first be shut down via ssh_rtl_shutdown() before
 * this function can be called.
 */
void ssh_rtl_destroy(struct ssh_rtl *rtl)
{
	ssh_ptl_destroy(&rtl->ptl);
}

/**
 * ssh_rtl_start() - Start request transmitter and receiver.
 * @rtl: The request transport layer.
 *
 * Return: Returns zero on success, a negative error code on failure.
 */
int ssh_rtl_start(struct ssh_rtl *rtl)
{
	int status;

	status = ssh_ptl_tx_start(&rtl->ptl);
	if (status)
		return status;

	ssh_rtl_tx_schedule(rtl);

	status = ssh_ptl_rx_start(&rtl->ptl);
	if (status) {
		ssh_rtl_flush(rtl, msecs_to_jiffies(5000));
		ssh_ptl_tx_stop(&rtl->ptl);
		return status;
	}

	return 0;
}

struct ssh_flush_request {
	struct ssh_request base;
	struct completion completion;
	int status;
};

static void ssh_rtl_flush_request_complete(struct ssh_request *r,
					   const struct ssh_command *cmd,
					   const struct ssam_span *data,
					   int status)
{
	struct ssh_flush_request *rqst;

	rqst = container_of(r, struct ssh_flush_request, base);
	rqst->status = status;
}

static void ssh_rtl_flush_request_release(struct ssh_request *r)
{
	struct ssh_flush_request *rqst;

	rqst = container_of(r, struct ssh_flush_request, base);
	complete_all(&rqst->completion);
}

static const struct ssh_request_ops ssh_rtl_flush_request_ops = {
	.complete = ssh_rtl_flush_request_complete,
	.release = ssh_rtl_flush_request_release,
};

/**
 * ssh_rtl_flush() - Flush the request transport layer.
 * @rtl:     request transport layer
 * @timeout: timeout for the flush operation in jiffies
 *
 * Queue a special flush request and wait for its completion. This request
 * will be completed after all other currently queued and pending requests
 * have been completed. Instead of a normal data packet, this request submits
 * a special flush packet, meaning that upon completion, also the underlying
 * packet transport layer has been flushed.
 *
 * Flushing the request layer guarantees that all previously submitted
 * requests have been fully completed before this call returns. Additionally,
 * flushing blocks execution of all later submitted requests until the flush
 * has been completed.
 *
 * If the caller ensures that no new requests are submitted after a call to
 * this function, the request transport layer is guaranteed to have no
 * remaining requests when this call returns. The same guarantee does not hold
 * for the packet layer, on which control packets may still be queued after
 * this call.
 *
 * Return: Returns zero on success, %-ETIMEDOUT if the flush timed out and has
 * been canceled as a result of the timeout, or %-ESHUTDOWN if the packet
 * and/or request transport layer has been shut down before this call. May
 * also return %-EINTR if the underlying packet transmission has been
 * interrupted.
 */
int ssh_rtl_flush(struct ssh_rtl *rtl, unsigned long timeout)
{
	const unsigned int init_flags = SSAM_REQUEST_UNSEQUENCED;
	struct ssh_flush_request rqst;
	int status;

	ssh_request_init(&rqst.base, init_flags, &ssh_rtl_flush_request_ops);
	rqst.base.packet.state |= BIT(SSH_PACKET_TY_FLUSH_BIT);
	rqst.base.packet.priority = SSH_PACKET_PRIORITY(FLUSH, 0);
	rqst.base.state |= BIT(SSH_REQUEST_TY_FLUSH_BIT);

	init_completion(&rqst.completion);

	status = ssh_rtl_submit(rtl, &rqst.base);
	if (status)
		return status;

	ssh_request_put(&rqst.base);

	if (!wait_for_completion_timeout(&rqst.completion, timeout)) {
		ssh_rtl_cancel(&rqst.base, true);
		wait_for_completion(&rqst.completion);
	}

	WARN_ON(rqst.status != 0 && rqst.status != -ECANCELED &&
		rqst.status != -ESHUTDOWN && rqst.status != -EINTR);

	return rqst.status == -ECANCELED ? -ETIMEDOUT : rqst.status;
}

/**
 * ssh_rtl_shutdown() - Shut down request transport layer.
 * @rtl: The request transport layer.
 *
 * Shuts down the request transport layer, removing and canceling all queued
 * and pending requests. Requests canceled by this operation will be completed
 * with %-ESHUTDOWN as status. Receiver and transmitter threads will be
 * stopped, the lower-level packet layer will be shutdown.
 *
 * As a result of this function, the transport layer will be marked as shut
 * down. Submission of requests after the transport layer has been shut down
 * will fail with %-ESHUTDOWN.
 */
void ssh_rtl_shutdown(struct ssh_rtl *rtl)
{
	struct ssh_request *r, *n;
	LIST_HEAD(claimed);
	int pending;

	set_bit(SSH_RTL_SF_SHUTDOWN_BIT, &rtl->state);
	/*
	 * Ensure that the layer gets marked as shut-down before actually
	 * stopping it. In combination with the check in ssh_rtl_submit(),
	 * this guarantees that no new requests can be added and all already
	 * queued requests are properly canceled.
	 */
	smp_mb__after_atomic();

	/* Remove requests from queue. */
	spin_lock(&rtl->queue.lock);
	list_for_each_entry_safe(r, n, &rtl->queue.head, node) {
		set_bit(SSH_REQUEST_SF_LOCKED_BIT, &r->state);
		/* Ensure state never gets zero. */
		smp_mb__before_atomic();
		clear_bit(SSH_REQUEST_SF_QUEUED_BIT, &r->state);

		list_move_tail(&r->node, &claimed);
	}
	spin_unlock(&rtl->queue.lock);

	/*
	 * We have now guaranteed that the queue is empty and no more new
	 * requests can be submitted (i.e. it will stay empty). This means that
	 * calling ssh_rtl_tx_schedule() will not schedule tx.work any more. So
	 * we can simply call cancel_work_sync() on tx.work here and when that
	 * returns, we've locked it down. This also means that after this call,
	 * we don't submit any more packets to the underlying packet layer, so
	 * we can also shut that down.
	 */

	cancel_work_sync(&rtl->tx.work);
	ssh_ptl_shutdown(&rtl->ptl);
	cancel_delayed_work_sync(&rtl->rtx_timeout.reaper);

	/*
	 * Shutting down the packet layer should also have canceled all
	 * requests. Thus the pending set should be empty. Attempt to handle
	 * this gracefully anyways, even though this should be dead code.
	 */

	pending = atomic_read(&rtl->pending.count);
	if (WARN_ON(pending)) {
		spin_lock(&rtl->pending.lock);
		list_for_each_entry_safe(r, n, &rtl->pending.head, node) {
			set_bit(SSH_REQUEST_SF_LOCKED_BIT, &r->state);
			/* Ensure state never gets zero. */
			smp_mb__before_atomic();
			clear_bit(SSH_REQUEST_SF_PENDING_BIT, &r->state);

			list_move_tail(&r->node, &claimed);
		}
		spin_unlock(&rtl->pending.lock);
	}

	/* Finally, cancel and complete the requests we claimed before. */
	list_for_each_entry_safe(r, n, &claimed, node) {
		/*
		 * We need test_and_set() because we still might compete with
		 * cancellation.
		 */
		if (!test_and_set_bit(SSH_REQUEST_SF_COMPLETED_BIT, &r->state))
			ssh_rtl_complete_with_status(r, -ESHUTDOWN);

		/*
		 * Drop the reference we've obtained by removing it from the
		 * lists.
		 */
		list_del(&r->node);
		ssh_request_put(r);
	}
}