summaryrefslogtreecommitdiff
path: root/block/blk-throttle.c
blob: 8331aba9426f2c75ffb29ff65a3b89effa6b5262 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
1637
1638
1639
1640
1641
1642
1643
1644
1645
1646
1647
1648
1649
1650
1651
1652
1653
1654
1655
1656
1657
1658
1659
1660
1661
1662
1663
1664
1665
1666
1667
1668
1669
1670
1671
1672
1673
1674
1675
1676
1677
1678
1679
1680
1681
1682
1683
1684
1685
1686
1687
1688
1689
1690
1691
1692
1693
1694
1695
/*
 * Interface for controlling IO bandwidth on a request queue
 *
 * Copyright (C) 2010 Vivek Goyal <vgoyal@redhat.com>
 */

#include <linux/module.h>
#include <linux/slab.h>
#include <linux/blkdev.h>
#include <linux/bio.h>
#include <linux/blktrace_api.h>
#include "blk-cgroup.h"
#include "blk.h"

/* Max dispatch from a group in 1 round */
static int throtl_grp_quantum = 8;

/* Total max dispatch from all groups in one round */
static int throtl_quantum = 32;

/* Throttling is performed over 100ms slice and after that slice is renewed */
static unsigned long throtl_slice = HZ/10;	/* 100 ms */

static struct blkcg_policy blkcg_policy_throtl;

/* A workqueue to queue throttle related work */
static struct workqueue_struct *kthrotld_workqueue;

/*
 * To implement hierarchical throttling, throtl_grps form a tree and bios
 * are dispatched upwards level by level until they reach the top and get
 * issued.  When dispatching bios from the children and local group at each
 * level, if the bios are dispatched into a single bio_list, there's a risk
 * of a local or child group which can queue many bios at once filling up
 * the list starving others.
 *
 * To avoid such starvation, dispatched bios are queued separately
 * according to where they came from.  When they are again dispatched to
 * the parent, they're popped in round-robin order so that no single source
 * hogs the dispatch window.
 *
 * throtl_qnode is used to keep the queued bios separated by their sources.
 * Bios are queued to throtl_qnode which in turn is queued to
 * throtl_service_queue and then dispatched in round-robin order.
 *
 * It's also used to track the reference counts on blkg's.  A qnode always
 * belongs to a throtl_grp and gets queued on itself or the parent, so
 * incrementing the reference of the associated throtl_grp when a qnode is
 * queued and decrementing when dequeued is enough to keep the whole blkg
 * tree pinned while bios are in flight.
 */
struct throtl_qnode {
	struct list_head	node;		/* service_queue->queued[] */
	struct bio_list		bios;		/* queued bios */
	struct throtl_grp	*tg;		/* tg this qnode belongs to */
};

struct throtl_service_queue {
	struct throtl_service_queue *parent_sq;	/* the parent service_queue */

	/*
	 * Bios queued directly to this service_queue or dispatched from
	 * children throtl_grp's.
	 */
	struct list_head	queued[2];	/* throtl_qnode [READ/WRITE] */
	unsigned int		nr_queued[2];	/* number of queued bios */

	/*
	 * RB tree of active children throtl_grp's, which are sorted by
	 * their ->disptime.
	 */
	struct rb_root		pending_tree;	/* RB tree of active tgs */
	struct rb_node		*first_pending;	/* first node in the tree */
	unsigned int		nr_pending;	/* # queued in the tree */
	unsigned long		first_pending_disptime;	/* disptime of the first tg */
	struct timer_list	pending_timer;	/* fires on first_pending_disptime */
};

enum tg_state_flags {
	THROTL_TG_PENDING	= 1 << 0,	/* on parent's pending tree */
	THROTL_TG_WAS_EMPTY	= 1 << 1,	/* bio_lists[] became non-empty */
};

#define rb_entry_tg(node)	rb_entry((node), struct throtl_grp, rb_node)

/* Per-cpu group stats */
struct tg_stats_cpu {
	/* total bytes transferred */
	struct blkg_rwstat		service_bytes;
	/* total IOs serviced, post merge */
	struct blkg_rwstat		serviced;
};

struct throtl_grp {
	/* must be the first member */
	struct blkg_policy_data pd;

	/* active throtl group service_queue member */
	struct rb_node rb_node;

	/* throtl_data this group belongs to */
	struct throtl_data *td;

	/* this group's service queue */
	struct throtl_service_queue service_queue;

	/*
	 * qnode_on_self is used when bios are directly queued to this
	 * throtl_grp so that local bios compete fairly with bios
	 * dispatched from children.  qnode_on_parent is used when bios are
	 * dispatched from this throtl_grp into its parent and will compete
	 * with the sibling qnode_on_parents and the parent's
	 * qnode_on_self.
	 */
	struct throtl_qnode qnode_on_self[2];
	struct throtl_qnode qnode_on_parent[2];

	/*
	 * Dispatch time in jiffies. This is the estimated time when group
	 * will unthrottle and is ready to dispatch more bio. It is used as
	 * key to sort active groups in service tree.
	 */
	unsigned long disptime;

	unsigned int flags;

	/* are there any throtl rules between this group and td? */
	bool has_rules[2];

	/* bytes per second rate limits */
	uint64_t bps[2];

	/* IOPS limits */
	unsigned int iops[2];

	/* Number of bytes disptached in current slice */
	uint64_t bytes_disp[2];
	/* Number of bio's dispatched in current slice */
	unsigned int io_disp[2];

	/* When did we start a new slice */
	unsigned long slice_start[2];
	unsigned long slice_end[2];

	/* Per cpu stats pointer */
	struct tg_stats_cpu __percpu *stats_cpu;

	/* List of tgs waiting for per cpu stats memory to be allocated */
	struct list_head stats_alloc_node;
};

struct throtl_data
{
	/* service tree for active throtl groups */
	struct throtl_service_queue service_queue;

	struct request_queue *queue;

	/* Total Number of queued bios on READ and WRITE lists */
	unsigned int nr_queued[2];

	/*
	 * number of total undestroyed groups
	 */
	unsigned int nr_undestroyed_grps;

	/* Work for dispatching throttled bios */
	struct work_struct dispatch_work;
};

/* list and work item to allocate percpu group stats */
static DEFINE_SPINLOCK(tg_stats_alloc_lock);
static LIST_HEAD(tg_stats_alloc_list);

static void tg_stats_alloc_fn(struct work_struct *);
static DECLARE_DELAYED_WORK(tg_stats_alloc_work, tg_stats_alloc_fn);

static void throtl_pending_timer_fn(unsigned long arg);

static inline struct throtl_grp *pd_to_tg(struct blkg_policy_data *pd)
{
	return pd ? container_of(pd, struct throtl_grp, pd) : NULL;
}

static inline struct throtl_grp *blkg_to_tg(struct blkcg_gq *blkg)
{
	return pd_to_tg(blkg_to_pd(blkg, &blkcg_policy_throtl));
}

static inline struct blkcg_gq *tg_to_blkg(struct throtl_grp *tg)
{
	return pd_to_blkg(&tg->pd);
}

static inline struct throtl_grp *td_root_tg(struct throtl_data *td)
{
	return blkg_to_tg(td->queue->root_blkg);
}

/**
 * sq_to_tg - return the throl_grp the specified service queue belongs to
 * @sq: the throtl_service_queue of interest
 *
 * Return the throtl_grp @sq belongs to.  If @sq is the top-level one
 * embedded in throtl_data, %NULL is returned.
 */
static struct throtl_grp *sq_to_tg(struct throtl_service_queue *sq)
{
	if (sq && sq->parent_sq)
		return container_of(sq, struct throtl_grp, service_queue);
	else
		return NULL;
}

/**
 * sq_to_td - return throtl_data the specified service queue belongs to
 * @sq: the throtl_service_queue of interest
 *
 * A service_queue can be embeded in either a throtl_grp or throtl_data.
 * Determine the associated throtl_data accordingly and return it.
 */
static struct throtl_data *sq_to_td(struct throtl_service_queue *sq)
{
	struct throtl_grp *tg = sq_to_tg(sq);

	if (tg)
		return tg->td;
	else
		return container_of(sq, struct throtl_data, service_queue);
}

/**
 * throtl_log - log debug message via blktrace
 * @sq: the service_queue being reported
 * @fmt: printf format string
 * @args: printf args
 *
 * The messages are prefixed with "throtl BLKG_NAME" if @sq belongs to a
 * throtl_grp; otherwise, just "throtl".
 *
 * TODO: this should be made a function and name formatting should happen
 * after testing whether blktrace is enabled.
 */
#define throtl_log(sq, fmt, args...)	do {				\
	struct throtl_grp *__tg = sq_to_tg((sq));			\
	struct throtl_data *__td = sq_to_td((sq));			\
									\
	(void)__td;							\
	if ((__tg)) {							\
		char __pbuf[128];					\
									\
		blkg_path(tg_to_blkg(__tg), __pbuf, sizeof(__pbuf));	\
		blk_add_trace_msg(__td->queue, "throtl %s " fmt, __pbuf, ##args); \
	} else {							\
		blk_add_trace_msg(__td->queue, "throtl " fmt, ##args);	\
	}								\
} while (0)

/*
 * Worker for allocating per cpu stat for tgs. This is scheduled on the
 * system_wq once there are some groups on the alloc_list waiting for
 * allocation.
 */
static void tg_stats_alloc_fn(struct work_struct *work)
{
	static struct tg_stats_cpu *stats_cpu;	/* this fn is non-reentrant */
	struct delayed_work *dwork = to_delayed_work(work);
	bool empty = false;

alloc_stats:
	if (!stats_cpu) {
		stats_cpu = alloc_percpu(struct tg_stats_cpu);
		if (!stats_cpu) {
			/* allocation failed, try again after some time */
			schedule_delayed_work(dwork, msecs_to_jiffies(10));
			return;
		}
	}

	spin_lock_irq(&tg_stats_alloc_lock);

	if (!list_empty(&tg_stats_alloc_list)) {
		struct throtl_grp *tg = list_first_entry(&tg_stats_alloc_list,
							 struct throtl_grp,
							 stats_alloc_node);
		swap(tg->stats_cpu, stats_cpu);
		list_del_init(&tg->stats_alloc_node);
	}

	empty = list_empty(&tg_stats_alloc_list);
	spin_unlock_irq(&tg_stats_alloc_lock);
	if (!empty)
		goto alloc_stats;
}

static void throtl_qnode_init(struct throtl_qnode *qn, struct throtl_grp *tg)
{
	INIT_LIST_HEAD(&qn->node);
	bio_list_init(&qn->bios);
	qn->tg = tg;
}

/**
 * throtl_qnode_add_bio - add a bio to a throtl_qnode and activate it
 * @bio: bio being added
 * @qn: qnode to add bio to
 * @queued: the service_queue->queued[] list @qn belongs to
 *
 * Add @bio to @qn and put @qn on @queued if it's not already on.
 * @qn->tg's reference count is bumped when @qn is activated.  See the
 * comment on top of throtl_qnode definition for details.
 */
static void throtl_qnode_add_bio(struct bio *bio, struct throtl_qnode *qn,
				 struct list_head *queued)
{
	bio_list_add(&qn->bios, bio);
	if (list_empty(&qn->node)) {
		list_add_tail(&qn->node, queued);
		blkg_get(tg_to_blkg(qn->tg));
	}
}

/**
 * throtl_peek_queued - peek the first bio on a qnode list
 * @queued: the qnode list to peek
 */
static struct bio *throtl_peek_queued(struct list_head *queued)
{
	struct throtl_qnode *qn = list_first_entry(queued, struct throtl_qnode, node);
	struct bio *bio;

	if (list_empty(queued))
		return NULL;

	bio = bio_list_peek(&qn->bios);
	WARN_ON_ONCE(!bio);
	return bio;
}

/**
 * throtl_pop_queued - pop the first bio form a qnode list
 * @queued: the qnode list to pop a bio from
 * @tg_to_put: optional out argument for throtl_grp to put
 *
 * Pop the first bio from the qnode list @queued.  After popping, the first
 * qnode is removed from @queued if empty or moved to the end of @queued so
 * that the popping order is round-robin.
 *
 * When the first qnode is removed, its associated throtl_grp should be put
 * too.  If @tg_to_put is NULL, this function automatically puts it;
 * otherwise, *@tg_to_put is set to the throtl_grp to put and the caller is
 * responsible for putting it.
 */
static struct bio *throtl_pop_queued(struct list_head *queued,
				     struct throtl_grp **tg_to_put)
{
	struct throtl_qnode *qn = list_first_entry(queued, struct throtl_qnode, node);
	struct bio *bio;

	if (list_empty(queued))
		return NULL;

	bio = bio_list_pop(&qn->bios);
	WARN_ON_ONCE(!bio);

	if (bio_list_empty(&qn->bios)) {
		list_del_init(&qn->node);
		if (tg_to_put)
			*tg_to_put = qn->tg;
		else
			blkg_put(tg_to_blkg(qn->tg));
	} else {
		list_move_tail(&qn->node, queued);
	}

	return bio;
}

/* init a service_queue, assumes the caller zeroed it */
static void throtl_service_queue_init(struct throtl_service_queue *sq,
				      struct throtl_service_queue *parent_sq)
{
	INIT_LIST_HEAD(&sq->queued[0]);
	INIT_LIST_HEAD(&sq->queued[1]);
	sq->pending_tree = RB_ROOT;
	sq->parent_sq = parent_sq;
	setup_timer(&sq->pending_timer, throtl_pending_timer_fn,
		    (unsigned long)sq);
}

static void throtl_service_queue_exit(struct throtl_service_queue *sq)
{
	del_timer_sync(&sq->pending_timer);
}

static void throtl_pd_init(struct blkcg_gq *blkg)
{
	struct throtl_grp *tg = blkg_to_tg(blkg);
	struct throtl_data *td = blkg->q->td;
	struct throtl_service_queue *parent_sq;
	unsigned long flags;
	int rw;

	/*
	 * If sane_hierarchy is enabled, we switch to properly hierarchical
	 * behavior where limits on a given throtl_grp are applied to the
	 * whole subtree rather than just the group itself.  e.g. If 16M
	 * read_bps limit is set on the root group, the whole system can't
	 * exceed 16M for the device.
	 *
	 * If sane_hierarchy is not enabled, the broken flat hierarchy
	 * behavior is retained where all throtl_grps are treated as if
	 * they're all separate root groups right below throtl_data.
	 * Limits of a group don't interact with limits of other groups
	 * regardless of the position of the group in the hierarchy.
	 */
	parent_sq = &td->service_queue;

	if (cgroup_sane_behavior(blkg->blkcg->css.cgroup) && blkg->parent)
		parent_sq = &blkg_to_tg(blkg->parent)->service_queue;

	throtl_service_queue_init(&tg->service_queue, parent_sq);

	for (rw = READ; rw <= WRITE; rw++) {
		throtl_qnode_init(&tg->qnode_on_self[rw], tg);
		throtl_qnode_init(&tg->qnode_on_parent[rw], tg);
	}

	RB_CLEAR_NODE(&tg->rb_node);
	tg->td = td;

	tg->bps[READ] = -1;
	tg->bps[WRITE] = -1;
	tg->iops[READ] = -1;
	tg->iops[WRITE] = -1;

	/*
	 * Ugh... We need to perform per-cpu allocation for tg->stats_cpu
	 * but percpu allocator can't be called from IO path.  Queue tg on
	 * tg_stats_alloc_list and allocate from work item.
	 */
	spin_lock_irqsave(&tg_stats_alloc_lock, flags);
	list_add(&tg->stats_alloc_node, &tg_stats_alloc_list);
	schedule_delayed_work(&tg_stats_alloc_work, 0);
	spin_unlock_irqrestore(&tg_stats_alloc_lock, flags);
}

/*
 * Set has_rules[] if @tg or any of its parents have limits configured.
 * This doesn't require walking up to the top of the hierarchy as the
 * parent's has_rules[] is guaranteed to be correct.
 */
static void tg_update_has_rules(struct throtl_grp *tg)
{
	struct throtl_grp *parent_tg = sq_to_tg(tg->service_queue.parent_sq);
	int rw;

	for (rw = READ; rw <= WRITE; rw++)
		tg->has_rules[rw] = (parent_tg && parent_tg->has_rules[rw]) ||
				    (tg->bps[rw] != -1 || tg->iops[rw] != -1);
}

static void throtl_pd_online(struct blkcg_gq *blkg)
{
	/*
	 * We don't want new groups to escape the limits of its ancestors.
	 * Update has_rules[] after a new group is brought online.
	 */
	tg_update_has_rules(blkg_to_tg(blkg));
}

static void throtl_pd_exit(struct blkcg_gq *blkg)
{
	struct throtl_grp *tg = blkg_to_tg(blkg);
	unsigned long flags;

	spin_lock_irqsave(&tg_stats_alloc_lock, flags);
	list_del_init(&tg->stats_alloc_node);
	spin_unlock_irqrestore(&tg_stats_alloc_lock, flags);

	free_percpu(tg->stats_cpu);

	throtl_service_queue_exit(&tg->service_queue);
}

static void throtl_pd_reset_stats(struct blkcg_gq *blkg)
{
	struct throtl_grp *tg = blkg_to_tg(blkg);
	int cpu;

	if (tg->stats_cpu == NULL)
		return;

	for_each_possible_cpu(cpu) {
		struct tg_stats_cpu *sc = per_cpu_ptr(tg->stats_cpu, cpu);

		blkg_rwstat_reset(&sc->service_bytes);
		blkg_rwstat_reset(&sc->serviced);
	}
}

static struct throtl_grp *throtl_lookup_tg(struct throtl_data *td,
					   struct blkcg *blkcg)
{
	/*
	 * This is the common case when there are no blkcgs.  Avoid lookup
	 * in this case
	 */
	if (blkcg == &blkcg_root)
		return td_root_tg(td);

	return blkg_to_tg(blkg_lookup(blkcg, td->queue));
}

static struct throtl_grp *throtl_lookup_create_tg(struct throtl_data *td,
						  struct blkcg *blkcg)
{
	struct request_queue *q = td->queue;
	struct throtl_grp *tg = NULL;

	/*
	 * This is the common case when there are no blkcgs.  Avoid lookup
	 * in this case
	 */
	if (blkcg == &blkcg_root) {
		tg = td_root_tg(td);
	} else {
		struct blkcg_gq *blkg;

		blkg = blkg_lookup_create(blkcg, q);

		/* if %NULL and @q is alive, fall back to root_tg */
		if (!IS_ERR(blkg))
			tg = blkg_to_tg(blkg);
		else if (!blk_queue_dying(q))
			tg = td_root_tg(td);
	}

	return tg;
}

static struct throtl_grp *
throtl_rb_first(struct throtl_service_queue *parent_sq)
{
	/* Service tree is empty */
	if (!parent_sq->nr_pending)
		return NULL;

	if (!parent_sq->first_pending)
		parent_sq->first_pending = rb_first(&parent_sq->pending_tree);

	if (parent_sq->first_pending)
		return rb_entry_tg(parent_sq->first_pending);

	return NULL;
}

static void rb_erase_init(struct rb_node *n, struct rb_root *root)
{
	rb_erase(n, root);
	RB_CLEAR_NODE(n);
}

static void throtl_rb_erase(struct rb_node *n,
			    struct throtl_service_queue *parent_sq)
{
	if (parent_sq->first_pending == n)
		parent_sq->first_pending = NULL;
	rb_erase_init(n, &parent_sq->pending_tree);
	--parent_sq->nr_pending;
}

static void update_min_dispatch_time(struct throtl_service_queue *parent_sq)
{
	struct throtl_grp *tg;

	tg = throtl_rb_first(parent_sq);
	if (!tg)
		return;

	parent_sq->first_pending_disptime = tg->disptime;
}

static void tg_service_queue_add(struct throtl_grp *tg)
{
	struct throtl_service_queue *parent_sq = tg->service_queue.parent_sq;
	struct rb_node **node = &parent_sq->pending_tree.rb_node;
	struct rb_node *parent = NULL;
	struct throtl_grp *__tg;
	unsigned long key = tg->disptime;
	int left = 1;

	while (*node != NULL) {
		parent = *node;
		__tg = rb_entry_tg(parent);

		if (time_before(key, __tg->disptime))
			node = &parent->rb_left;
		else {
			node = &parent->rb_right;
			left = 0;
		}
	}

	if (left)
		parent_sq->first_pending = &tg->rb_node;

	rb_link_node(&tg->rb_node, parent, node);
	rb_insert_color(&tg->rb_node, &parent_sq->pending_tree);
}

static void __throtl_enqueue_tg(struct throtl_grp *tg)
{
	tg_service_queue_add(tg);
	tg->flags |= THROTL_TG_PENDING;
	tg->service_queue.parent_sq->nr_pending++;
}

static void throtl_enqueue_tg(struct throtl_grp *tg)
{
	if (!(tg->flags & THROTL_TG_PENDING))
		__throtl_enqueue_tg(tg);
}

static void __throtl_dequeue_tg(struct throtl_grp *tg)
{
	throtl_rb_erase(&tg->rb_node, tg->service_queue.parent_sq);
	tg->flags &= ~THROTL_TG_PENDING;
}

static void throtl_dequeue_tg(struct throtl_grp *tg)
{
	if (tg->flags & THROTL_TG_PENDING)
		__throtl_dequeue_tg(tg);
}

/* Call with queue lock held */
static void throtl_schedule_pending_timer(struct throtl_service_queue *sq,
					  unsigned long expires)
{
	mod_timer(&sq->pending_timer, expires);
	throtl_log(sq, "schedule timer. delay=%lu jiffies=%lu",
		   expires - jiffies, jiffies);
}

/**
 * throtl_schedule_next_dispatch - schedule the next dispatch cycle
 * @sq: the service_queue to schedule dispatch for
 * @force: force scheduling
 *
 * Arm @sq->pending_timer so that the next dispatch cycle starts on the
 * dispatch time of the first pending child.  Returns %true if either timer
 * is armed or there's no pending child left.  %false if the current
 * dispatch window is still open and the caller should continue
 * dispatching.
 *
 * If @force is %true, the dispatch timer is always scheduled and this
 * function is guaranteed to return %true.  This is to be used when the
 * caller can't dispatch itself and needs to invoke pending_timer
 * unconditionally.  Note that forced scheduling is likely to induce short
 * delay before dispatch starts even if @sq->first_pending_disptime is not
 * in the future and thus shouldn't be used in hot paths.
 */
static bool throtl_schedule_next_dispatch(struct throtl_service_queue *sq,
					  bool force)
{
	/* any pending children left? */
	if (!sq->nr_pending)
		return true;

	update_min_dispatch_time(sq);

	/* is the next dispatch time in the future? */
	if (force || time_after(sq->first_pending_disptime, jiffies)) {
		throtl_schedule_pending_timer(sq, sq->first_pending_disptime);
		return true;
	}

	/* tell the caller to continue dispatching */
	return false;
}

static inline void throtl_start_new_slice_with_credit(struct throtl_grp *tg,
		bool rw, unsigned long start)
{
	tg->bytes_disp[rw] = 0;
	tg->io_disp[rw] = 0;

	/*
	 * Previous slice has expired. We must have trimmed it after last
	 * bio dispatch. That means since start of last slice, we never used
	 * that bandwidth. Do try to make use of that bandwidth while giving
	 * credit.
	 */
	if (time_after_eq(start, tg->slice_start[rw]))
		tg->slice_start[rw] = start;

	tg->slice_end[rw] = jiffies + throtl_slice;
	throtl_log(&tg->service_queue,
		   "[%c] new slice with credit start=%lu end=%lu jiffies=%lu",
		   rw == READ ? 'R' : 'W', tg->slice_start[rw],
		   tg->slice_end[rw], jiffies);
}

static inline void throtl_start_new_slice(struct throtl_grp *tg, bool rw)
{
	tg->bytes_disp[rw] = 0;
	tg->io_disp[rw] = 0;
	tg->slice_start[rw] = jiffies;
	tg->slice_end[rw] = jiffies + throtl_slice;
	throtl_log(&tg->service_queue,
		   "[%c] new slice start=%lu end=%lu jiffies=%lu",
		   rw == READ ? 'R' : 'W', tg->slice_start[rw],
		   tg->slice_end[rw], jiffies);
}

static inline void throtl_set_slice_end(struct throtl_grp *tg, bool rw,
					unsigned long jiffy_end)
{
	tg->slice_end[rw] = roundup(jiffy_end, throtl_slice);
}

static inline void throtl_extend_slice(struct throtl_grp *tg, bool rw,
				       unsigned long jiffy_end)
{
	tg->slice_end[rw] = roundup(jiffy_end, throtl_slice);
	throtl_log(&tg->service_queue,
		   "[%c] extend slice start=%lu end=%lu jiffies=%lu",
		   rw == READ ? 'R' : 'W', tg->slice_start[rw],
		   tg->slice_end[rw], jiffies);
}

/* Determine if previously allocated or extended slice is complete or not */
static bool throtl_slice_used(struct throtl_grp *tg, bool rw)
{
	if (time_in_range(jiffies, tg->slice_start[rw], tg->slice_end[rw]))
		return 0;

	return 1;
}

/* Trim the used slices and adjust slice start accordingly */
static inline void throtl_trim_slice(struct throtl_grp *tg, bool rw)
{
	unsigned long nr_slices, time_elapsed, io_trim;
	u64 bytes_trim, tmp;

	BUG_ON(time_before(tg->slice_end[rw], tg->slice_start[rw]));

	/*
	 * If bps are unlimited (-1), then time slice don't get
	 * renewed. Don't try to trim the slice if slice is used. A new
	 * slice will start when appropriate.
	 */
	if (throtl_slice_used(tg, rw))
		return;

	/*
	 * A bio has been dispatched. Also adjust slice_end. It might happen
	 * that initially cgroup limit was very low resulting in high
	 * slice_end, but later limit was bumped up and bio was dispached
	 * sooner, then we need to reduce slice_end. A high bogus slice_end
	 * is bad because it does not allow new slice to start.
	 */

	throtl_set_slice_end(tg, rw, jiffies + throtl_slice);

	time_elapsed = jiffies - tg->slice_start[rw];

	nr_slices = time_elapsed / throtl_slice;

	if (!nr_slices)
		return;
	tmp = tg->bps[rw] * throtl_slice * nr_slices;
	do_div(tmp, HZ);
	bytes_trim = tmp;

	io_trim = (tg->iops[rw] * throtl_slice * nr_slices)/HZ;

	if (!bytes_trim && !io_trim)
		return;

	if (tg->bytes_disp[rw] >= bytes_trim)
		tg->bytes_disp[rw] -= bytes_trim;
	else
		tg->bytes_disp[rw] = 0;

	if (tg->io_disp[rw] >= io_trim)
		tg->io_disp[rw] -= io_trim;
	else
		tg->io_disp[rw] = 0;

	tg->slice_start[rw] += nr_slices * throtl_slice;

	throtl_log(&tg->service_queue,
		   "[%c] trim slice nr=%lu bytes=%llu io=%lu start=%lu end=%lu jiffies=%lu",
		   rw == READ ? 'R' : 'W', nr_slices, bytes_trim, io_trim,
		   tg->slice_start[rw], tg->slice_end[rw], jiffies);
}

static bool tg_with_in_iops_limit(struct throtl_grp *tg, struct bio *bio,
				  unsigned long *wait)
{
	bool rw = bio_data_dir(bio);
	unsigned int io_allowed;
	unsigned long jiffy_elapsed, jiffy_wait, jiffy_elapsed_rnd;
	u64 tmp;

	jiffy_elapsed = jiffy_elapsed_rnd = jiffies - tg->slice_start[rw];

	/* Slice has just started. Consider one slice interval */
	if (!jiffy_elapsed)
		jiffy_elapsed_rnd = throtl_slice;

	jiffy_elapsed_rnd = roundup(jiffy_elapsed_rnd, throtl_slice);

	/*
	 * jiffy_elapsed_rnd should not be a big value as minimum iops can be
	 * 1 then at max jiffy elapsed should be equivalent of 1 second as we
	 * will allow dispatch after 1 second and after that slice should
	 * have been trimmed.
	 */

	tmp = (u64)tg->iops[rw] * jiffy_elapsed_rnd;
	do_div(tmp, HZ);

	if (tmp > UINT_MAX)
		io_allowed = UINT_MAX;
	else
		io_allowed = tmp;

	if (tg->io_disp[rw] + 1 <= io_allowed) {
		if (wait)
			*wait = 0;
		return 1;
	}

	/* Calc approx time to dispatch */
	jiffy_wait = ((tg->io_disp[rw] + 1) * HZ)/tg->iops[rw] + 1;

	if (jiffy_wait > jiffy_elapsed)
		jiffy_wait = jiffy_wait - jiffy_elapsed;
	else
		jiffy_wait = 1;

	if (wait)
		*wait = jiffy_wait;
	return 0;
}

static bool tg_with_in_bps_limit(struct throtl_grp *tg, struct bio *bio,
				 unsigned long *wait)
{
	bool rw = bio_data_dir(bio);
	u64 bytes_allowed, extra_bytes, tmp;
	unsigned long jiffy_elapsed, jiffy_wait, jiffy_elapsed_rnd;

	jiffy_elapsed = jiffy_elapsed_rnd = jiffies - tg->slice_start[rw];

	/* Slice has just started. Consider one slice interval */
	if (!jiffy_elapsed)
		jiffy_elapsed_rnd = throtl_slice;

	jiffy_elapsed_rnd = roundup(jiffy_elapsed_rnd, throtl_slice);

	tmp = tg->bps[rw] * jiffy_elapsed_rnd;
	do_div(tmp, HZ);
	bytes_allowed = tmp;

	if (tg->bytes_disp[rw] + bio->bi_size <= bytes_allowed) {
		if (wait)
			*wait = 0;
		return 1;
	}

	/* Calc approx time to dispatch */
	extra_bytes = tg->bytes_disp[rw] + bio->bi_size - bytes_allowed;
	jiffy_wait = div64_u64(extra_bytes * HZ, tg->bps[rw]);

	if (!jiffy_wait)
		jiffy_wait = 1;

	/*
	 * This wait time is without taking into consideration the rounding
	 * up we did. Add that time also.
	 */
	jiffy_wait = jiffy_wait + (jiffy_elapsed_rnd - jiffy_elapsed);
	if (wait)
		*wait = jiffy_wait;
	return 0;
}

/*
 * Returns whether one can dispatch a bio or not. Also returns approx number
 * of jiffies to wait before this bio is with-in IO rate and can be dispatched
 */
static bool tg_may_dispatch(struct throtl_grp *tg, struct bio *bio,
			    unsigned long *wait)
{
	bool rw = bio_data_dir(bio);
	unsigned long bps_wait = 0, iops_wait = 0, max_wait = 0;

	/*
 	 * Currently whole state machine of group depends on first bio
	 * queued in the group bio list. So one should not be calling
	 * this function with a different bio if there are other bios
	 * queued.
	 */
	BUG_ON(tg->service_queue.nr_queued[rw] &&
	       bio != throtl_peek_queued(&tg->service_queue.queued[rw]));

	/* If tg->bps = -1, then BW is unlimited */
	if (tg->bps[rw] == -1 && tg->iops[rw] == -1) {
		if (wait)
			*wait = 0;
		return 1;
	}

	/*
	 * If previous slice expired, start a new one otherwise renew/extend
	 * existing slice to make sure it is at least throtl_slice interval
	 * long since now.
	 */
	if (throtl_slice_used(tg, rw))
		throtl_start_new_slice(tg, rw);
	else {
		if (time_before(tg->slice_end[rw], jiffies + throtl_slice))
			throtl_extend_slice(tg, rw, jiffies + throtl_slice);
	}

	if (tg_with_in_bps_limit(tg, bio, &bps_wait) &&
	    tg_with_in_iops_limit(tg, bio, &iops_wait)) {
		if (wait)
			*wait = 0;
		return 1;
	}

	max_wait = max(bps_wait, iops_wait);

	if (wait)
		*wait = max_wait;

	if (time_before(tg->slice_end[rw], jiffies + max_wait))
		throtl_extend_slice(tg, rw, jiffies + max_wait);

	return 0;
}

static void throtl_update_dispatch_stats(struct blkcg_gq *blkg, u64 bytes,
					 int rw)
{
	struct throtl_grp *tg = blkg_to_tg(blkg);
	struct tg_stats_cpu *stats_cpu;
	unsigned long flags;

	/* If per cpu stats are not allocated yet, don't do any accounting. */
	if (tg->stats_cpu == NULL)
		return;

	/*
	 * Disabling interrupts to provide mutual exclusion between two
	 * writes on same cpu. It probably is not needed for 64bit. Not
	 * optimizing that case yet.
	 */
	local_irq_save(flags);

	stats_cpu = this_cpu_ptr(tg->stats_cpu);

	blkg_rwstat_add(&stats_cpu->serviced, rw, 1);
	blkg_rwstat_add(&stats_cpu->service_bytes, rw, bytes);

	local_irq_restore(flags);
}

static void throtl_charge_bio(struct throtl_grp *tg, struct bio *bio)
{
	bool rw = bio_data_dir(bio);

	/* Charge the bio to the group */
	tg->bytes_disp[rw] += bio->bi_size;
	tg->io_disp[rw]++;

	/*
	 * REQ_THROTTLED is used to prevent the same bio to be throttled
	 * more than once as a throttled bio will go through blk-throtl the
	 * second time when it eventually gets issued.  Set it when a bio
	 * is being charged to a tg.
	 *
	 * Dispatch stats aren't recursive and each @bio should only be
	 * accounted by the @tg it was originally associated with.  Let's
	 * update the stats when setting REQ_THROTTLED for the first time
	 * which is guaranteed to be for the @bio's original tg.
	 */
	if (!(bio->bi_rw & REQ_THROTTLED)) {
		bio->bi_rw |= REQ_THROTTLED;
		throtl_update_dispatch_stats(tg_to_blkg(tg), bio->bi_size,
					     bio->bi_rw);
	}
}

/**
 * throtl_add_bio_tg - add a bio to the specified throtl_grp
 * @bio: bio to add
 * @qn: qnode to use
 * @tg: the target throtl_grp
 *
 * Add @bio to @tg's service_queue using @qn.  If @qn is not specified,
 * tg->qnode_on_self[] is used.
 */
static void throtl_add_bio_tg(struct bio *bio, struct throtl_qnode *qn,
			      struct throtl_grp *tg)
{
	struct throtl_service_queue *sq = &tg->service_queue;
	bool rw = bio_data_dir(bio);

	if (!qn)
		qn = &tg->qnode_on_self[rw];

	/*
	 * If @tg doesn't currently have any bios queued in the same
	 * direction, queueing @bio can change when @tg should be
	 * dispatched.  Mark that @tg was empty.  This is automatically
	 * cleaered on the next tg_update_disptime().
	 */
	if (!sq->nr_queued[rw])
		tg->flags |= THROTL_TG_WAS_EMPTY;

	throtl_qnode_add_bio(bio, qn, &sq->queued[rw]);

	sq->nr_queued[rw]++;
	throtl_enqueue_tg(tg);
}

static void tg_update_disptime(struct throtl_grp *tg)
{
	struct throtl_service_queue *sq = &tg->service_queue;
	unsigned long read_wait = -1, write_wait = -1, min_wait = -1, disptime;
	struct bio *bio;

	if ((bio = throtl_peek_queued(&sq->queued[READ])))
		tg_may_dispatch(tg, bio, &read_wait);

	if ((bio = throtl_peek_queued(&sq->queued[WRITE])))
		tg_may_dispatch(tg, bio, &write_wait);

	min_wait = min(read_wait, write_wait);
	disptime = jiffies + min_wait;

	/* Update dispatch time */
	throtl_dequeue_tg(tg);
	tg->disptime = disptime;
	throtl_enqueue_tg(tg);

	/* see throtl_add_bio_tg() */
	tg->flags &= ~THROTL_TG_WAS_EMPTY;
}

static void start_parent_slice_with_credit(struct throtl_grp *child_tg,
					struct throtl_grp *parent_tg, bool rw)
{
	if (throtl_slice_used(parent_tg, rw)) {
		throtl_start_new_slice_with_credit(parent_tg, rw,
				child_tg->slice_start[rw]);
	}

}

static void tg_dispatch_one_bio(struct throtl_grp *tg, bool rw)
{
	struct throtl_service_queue *sq = &tg->service_queue;
	struct throtl_service_queue *parent_sq = sq->parent_sq;
	struct throtl_grp *parent_tg = sq_to_tg(parent_sq);
	struct throtl_grp *tg_to_put = NULL;
	struct bio *bio;

	/*
	 * @bio is being transferred from @tg to @parent_sq.  Popping a bio
	 * from @tg may put its reference and @parent_sq might end up
	 * getting released prematurely.  Remember the tg to put and put it
	 * after @bio is transferred to @parent_sq.
	 */
	bio = throtl_pop_queued(&sq->queued[rw], &tg_to_put);
	sq->nr_queued[rw]--;

	throtl_charge_bio(tg, bio);

	/*
	 * If our parent is another tg, we just need to transfer @bio to
	 * the parent using throtl_add_bio_tg().  If our parent is
	 * @td->service_queue, @bio is ready to be issued.  Put it on its
	 * bio_lists[] and decrease total number queued.  The caller is
	 * responsible for issuing these bios.
	 */
	if (parent_tg) {
		throtl_add_bio_tg(bio, &tg->qnode_on_parent[rw], parent_tg);
		start_parent_slice_with_credit(tg, parent_tg, rw);
	} else {
		throtl_qnode_add_bio(bio, &tg->qnode_on_parent[rw],
				     &parent_sq->queued[rw]);
		BUG_ON(tg->td->nr_queued[rw] <= 0);
		tg->td->nr_queued[rw]--;
	}

	throtl_trim_slice(tg, rw);

	if (tg_to_put)
		blkg_put(tg_to_blkg(tg_to_put));
}

static int throtl_dispatch_tg(struct throtl_grp *tg)
{
	struct throtl_service_queue *sq = &tg->service_queue;
	unsigned int nr_reads = 0, nr_writes = 0;
	unsigned int max_nr_reads = throtl_grp_quantum*3/4;
	unsigned int max_nr_writes = throtl_grp_quantum - max_nr_reads;
	struct bio *bio;

	/* Try to dispatch 75% READS and 25% WRITES */

	while ((bio = throtl_peek_queued(&sq->queued[READ])) &&
	       tg_may_dispatch(tg, bio, NULL)) {

		tg_dispatch_one_bio(tg, bio_data_dir(bio));
		nr_reads++;

		if (nr_reads >= max_nr_reads)
			break;
	}

	while ((bio = throtl_peek_queued(&sq->queued[WRITE])) &&
	       tg_may_dispatch(tg, bio, NULL)) {

		tg_dispatch_one_bio(tg, bio_data_dir(bio));
		nr_writes++;

		if (nr_writes >= max_nr_writes)
			break;
	}

	return nr_reads + nr_writes;
}

static int throtl_select_dispatch(struct throtl_service_queue *parent_sq)
{
	unsigned int nr_disp = 0;

	while (1) {
		struct throtl_grp *tg = throtl_rb_first(parent_sq);
		struct throtl_service_queue *sq = &tg->service_queue;

		if (!tg)
			break;

		if (time_before(jiffies, tg->disptime))
			break;

		throtl_dequeue_tg(tg);

		nr_disp += throtl_dispatch_tg(tg);

		if (sq->nr_queued[0] || sq->nr_queued[1])
			tg_update_disptime(tg);

		if (nr_disp >= throtl_quantum)
			break;
	}

	return nr_disp;
}

/**
 * throtl_pending_timer_fn - timer function for service_queue->pending_timer
 * @arg: the throtl_service_queue being serviced
 *
 * This timer is armed when a child throtl_grp with active bio's become
 * pending and queued on the service_queue's pending_tree and expires when
 * the first child throtl_grp should be dispatched.  This function
 * dispatches bio's from the children throtl_grps to the parent
 * service_queue.
 *
 * If the parent's parent is another throtl_grp, dispatching is propagated
 * by either arming its pending_timer or repeating dispatch directly.  If
 * the top-level service_tree is reached, throtl_data->dispatch_work is
 * kicked so that the ready bio's are issued.
 */
static void throtl_pending_timer_fn(unsigned long arg)
{
	struct throtl_service_queue *sq = (void *)arg;
	struct throtl_grp *tg = sq_to_tg(sq);
	struct throtl_data *td = sq_to_td(sq);
	struct request_queue *q = td->queue;
	struct throtl_service_queue *parent_sq;
	bool dispatched;
	int ret;

	spin_lock_irq(q->queue_lock);
again:
	parent_sq = sq->parent_sq;
	dispatched = false;

	while (true) {
		throtl_log(sq, "dispatch nr_queued=%u read=%u write=%u",
			   sq->nr_queued[READ] + sq->nr_queued[WRITE],
			   sq->nr_queued[READ], sq->nr_queued[WRITE]);

		ret = throtl_select_dispatch(sq);
		if (ret) {
			throtl_log(sq, "bios disp=%u", ret);
			dispatched = true;
		}

		if (throtl_schedule_next_dispatch(sq, false))
			break;

		/* this dispatch windows is still open, relax and repeat */
		spin_unlock_irq(q->queue_lock);
		cpu_relax();
		spin_lock_irq(q->queue_lock);
	}

	if (!dispatched)
		goto out_unlock;

	if (parent_sq) {
		/* @parent_sq is another throl_grp, propagate dispatch */
		if (tg->flags & THROTL_TG_WAS_EMPTY) {
			tg_update_disptime(tg);
			if (!throtl_schedule_next_dispatch(parent_sq, false)) {
				/* window is already open, repeat dispatching */
				sq = parent_sq;
				tg = sq_to_tg(sq);
				goto again;
			}
		}
	} else {
		/* reached the top-level, queue issueing */
		queue_work(kthrotld_workqueue, &td->dispatch_work);
	}
out_unlock:
	spin_unlock_irq(q->queue_lock);
}

/**
 * blk_throtl_dispatch_work_fn - work function for throtl_data->dispatch_work
 * @work: work item being executed
 *
 * This function is queued for execution when bio's reach the bio_lists[]
 * of throtl_data->service_queue.  Those bio's are ready and issued by this
 * function.
 */
void blk_throtl_dispatch_work_fn(struct work_struct *work)
{
	struct throtl_data *td = container_of(work, struct throtl_data,
					      dispatch_work);
	struct throtl_service_queue *td_sq = &td->service_queue;
	struct request_queue *q = td->queue;
	struct bio_list bio_list_on_stack;
	struct bio *bio;
	struct blk_plug plug;
	int rw;

	bio_list_init(&bio_list_on_stack);

	spin_lock_irq(q->queue_lock);
	for (rw = READ; rw <= WRITE; rw++)
		while ((bio = throtl_pop_queued(&td_sq->queued[rw], NULL)))
			bio_list_add(&bio_list_on_stack, bio);
	spin_unlock_irq(q->queue_lock);

	if (!bio_list_empty(&bio_list_on_stack)) {
		blk_start_plug(&plug);
		while((bio = bio_list_pop(&bio_list_on_stack)))
			generic_make_request(bio);
		blk_finish_plug(&plug);
	}
}

static u64 tg_prfill_cpu_rwstat(struct seq_file *sf,
				struct blkg_policy_data *pd, int off)
{
	struct throtl_grp *tg = pd_to_tg(pd);
	struct blkg_rwstat rwstat = { }, tmp;
	int i, cpu;

	for_each_possible_cpu(cpu) {
		struct tg_stats_cpu *sc = per_cpu_ptr(tg->stats_cpu, cpu);

		tmp = blkg_rwstat_read((void *)sc + off);
		for (i = 0; i < BLKG_RWSTAT_NR; i++)
			rwstat.cnt[i] += tmp.cnt[i];
	}

	return __blkg_prfill_rwstat(sf, pd, &rwstat);
}

static int tg_print_cpu_rwstat(struct cgroup_subsys_state *css,
			       struct cftype *cft, struct seq_file *sf)
{
	struct blkcg *blkcg = css_to_blkcg(css);

	blkcg_print_blkgs(sf, blkcg, tg_prfill_cpu_rwstat, &blkcg_policy_throtl,
			  cft->private, true);
	return 0;
}

static u64 tg_prfill_conf_u64(struct seq_file *sf, struct blkg_policy_data *pd,
			      int off)
{
	struct throtl_grp *tg = pd_to_tg(pd);
	u64 v = *(u64 *)((void *)tg + off);

	if (v == -1)
		return 0;
	return __blkg_prfill_u64(sf, pd, v);
}

static u64 tg_prfill_conf_uint(struct seq_file *sf, struct blkg_policy_data *pd,
			       int off)
{
	struct throtl_grp *tg = pd_to_tg(pd);
	unsigned int v = *(unsigned int *)((void *)tg + off);

	if (v == -1)
		return 0;
	return __blkg_prfill_u64(sf, pd, v);
}

static int tg_print_conf_u64(struct cgroup_subsys_state *css,
			     struct cftype *cft, struct seq_file *sf)
{
	blkcg_print_blkgs(sf, css_to_blkcg(css), tg_prfill_conf_u64,
			  &blkcg_policy_throtl, cft->private, false);
	return 0;
}

static int tg_print_conf_uint(struct cgroup_subsys_state *css,
			      struct cftype *cft, struct seq_file *sf)
{
	blkcg_print_blkgs(sf, css_to_blkcg(css), tg_prfill_conf_uint,
			  &blkcg_policy_throtl, cft->private, false);
	return 0;
}

static int tg_set_conf(struct cgroup_subsys_state *css, struct cftype *cft,
		       const char *buf, bool is_u64)
{
	struct blkcg *blkcg = css_to_blkcg(css);
	struct blkg_conf_ctx ctx;
	struct throtl_grp *tg;
	struct throtl_service_queue *sq;
	struct blkcg_gq *blkg;
	struct cgroup_subsys_state *pos_css;
	int ret;

	ret = blkg_conf_prep(blkcg, &blkcg_policy_throtl, buf, &ctx);
	if (ret)
		return ret;

	tg = blkg_to_tg(ctx.blkg);
	sq = &tg->service_queue;

	if (!ctx.v)
		ctx.v = -1;

	if (is_u64)
		*(u64 *)((void *)tg + cft->private) = ctx.v;
	else
		*(unsigned int *)((void *)tg + cft->private) = ctx.v;

	throtl_log(&tg->service_queue,
		   "limit change rbps=%llu wbps=%llu riops=%u wiops=%u",
		   tg->bps[READ], tg->bps[WRITE],
		   tg->iops[READ], tg->iops[WRITE]);

	/*
	 * Update has_rules[] flags for the updated tg's subtree.  A tg is
	 * considered to have rules if either the tg itself or any of its
	 * ancestors has rules.  This identifies groups without any
	 * restrictions in the whole hierarchy and allows them to bypass
	 * blk-throttle.
	 */
	blkg_for_each_descendant_pre(blkg, pos_css, ctx.blkg)
		tg_update_has_rules(blkg_to_tg(blkg));

	/*
	 * We're already holding queue_lock and know @tg is valid.  Let's
	 * apply the new config directly.
	 *
	 * Restart the slices for both READ and WRITES. It might happen
	 * that a group's limit are dropped suddenly and we don't want to
	 * account recently dispatched IO with new low rate.
	 */
	throtl_start_new_slice(tg, 0);
	throtl_start_new_slice(tg, 1);

	if (tg->flags & THROTL_TG_PENDING) {
		tg_update_disptime(tg);
		throtl_schedule_next_dispatch(sq->parent_sq, true);
	}

	blkg_conf_finish(&ctx);
	return 0;
}

static int tg_set_conf_u64(struct cgroup_subsys_state *css, struct cftype *cft,
			   const char *buf)
{
	return tg_set_conf(css, cft, buf, true);
}

static int tg_set_conf_uint(struct cgroup_subsys_state *css, struct cftype *cft,
			    const char *buf)
{
	return tg_set_conf(css, cft, buf, false);
}

static struct cftype throtl_files[] = {
	{
		.name = "throttle.read_bps_device",
		.private = offsetof(struct throtl_grp, bps[READ]),
		.read_seq_string = tg_print_conf_u64,
		.write_string = tg_set_conf_u64,
		.max_write_len = 256,
	},
	{
		.name = "throttle.write_bps_device",
		.private = offsetof(struct throtl_grp, bps[WRITE]),
		.read_seq_string = tg_print_conf_u64,
		.write_string = tg_set_conf_u64,
		.max_write_len = 256,
	},
	{
		.name = "throttle.read_iops_device",
		.private = offsetof(struct throtl_grp, iops[READ]),
		.read_seq_string = tg_print_conf_uint,
		.write_string = tg_set_conf_uint,
		.max_write_len = 256,
	},
	{
		.name = "throttle.write_iops_device",
		.private = offsetof(struct throtl_grp, iops[WRITE]),
		.read_seq_string = tg_print_conf_uint,
		.write_string = tg_set_conf_uint,
		.max_write_len = 256,
	},
	{
		.name = "throttle.io_service_bytes",
		.private = offsetof(struct tg_stats_cpu, service_bytes),
		.read_seq_string = tg_print_cpu_rwstat,
	},
	{
		.name = "throttle.io_serviced",
		.private = offsetof(struct tg_stats_cpu, serviced),
		.read_seq_string = tg_print_cpu_rwstat,
	},
	{ }	/* terminate */
};

static void throtl_shutdown_wq(struct request_queue *q)
{
	struct throtl_data *td = q->td;

	cancel_work_sync(&td->dispatch_work);
}

static struct blkcg_policy blkcg_policy_throtl = {
	.pd_size		= sizeof(struct throtl_grp),
	.cftypes		= throtl_files,

	.pd_init_fn		= throtl_pd_init,
	.pd_online_fn		= throtl_pd_online,
	.pd_exit_fn		= throtl_pd_exit,
	.pd_reset_stats_fn	= throtl_pd_reset_stats,
};

bool blk_throtl_bio(struct request_queue *q, struct bio *bio)
{
	struct throtl_data *td = q->td;
	struct throtl_qnode *qn = NULL;
	struct throtl_grp *tg;
	struct throtl_service_queue *sq;
	bool rw = bio_data_dir(bio);
	struct blkcg *blkcg;
	bool throttled = false;

	/* see throtl_charge_bio() */
	if (bio->bi_rw & REQ_THROTTLED)
		goto out;

	/*
	 * A throtl_grp pointer retrieved under rcu can be used to access
	 * basic fields like stats and io rates. If a group has no rules,
	 * just update the dispatch stats in lockless manner and return.
	 */
	rcu_read_lock();
	blkcg = bio_blkcg(bio);
	tg = throtl_lookup_tg(td, blkcg);
	if (tg) {
		if (!tg->has_rules[rw]) {
			throtl_update_dispatch_stats(tg_to_blkg(tg),
						     bio->bi_size, bio->bi_rw);
			goto out_unlock_rcu;
		}
	}

	/*
	 * Either group has not been allocated yet or it is not an unlimited
	 * IO group
	 */
	spin_lock_irq(q->queue_lock);
	tg = throtl_lookup_create_tg(td, blkcg);
	if (unlikely(!tg))
		goto out_unlock;

	sq = &tg->service_queue;

	while (true) {
		/* throtl is FIFO - if bios are already queued, should queue */
		if (sq->nr_queued[rw])
			break;

		/* if above limits, break to queue */
		if (!tg_may_dispatch(tg, bio, NULL))
			break;

		/* within limits, let's charge and dispatch directly */
		throtl_charge_bio(tg, bio);

		/*
		 * We need to trim slice even when bios are not being queued
		 * otherwise it might happen that a bio is not queued for
		 * a long time and slice keeps on extending and trim is not
		 * called for a long time. Now if limits are reduced suddenly
		 * we take into account all the IO dispatched so far at new
		 * low rate and * newly queued IO gets a really long dispatch
		 * time.
		 *
		 * So keep on trimming slice even if bio is not queued.
		 */
		throtl_trim_slice(tg, rw);

		/*
		 * @bio passed through this layer without being throttled.
		 * Climb up the ladder.  If we''re already at the top, it
		 * can be executed directly.
		 */
		qn = &tg->qnode_on_parent[rw];
		sq = sq->parent_sq;
		tg = sq_to_tg(sq);
		if (!tg)
			goto out_unlock;
	}

	/* out-of-limit, queue to @tg */
	throtl_log(sq, "[%c] bio. bdisp=%llu sz=%u bps=%llu iodisp=%u iops=%u queued=%d/%d",
		   rw == READ ? 'R' : 'W',
		   tg->bytes_disp[rw], bio->bi_size, tg->bps[rw],
		   tg->io_disp[rw], tg->iops[rw],
		   sq->nr_queued[READ], sq->nr_queued[WRITE]);

	bio_associate_current(bio);
	tg->td->nr_queued[rw]++;
	throtl_add_bio_tg(bio, qn, tg);
	throttled = true;

	/*
	 * Update @tg's dispatch time and force schedule dispatch if @tg
	 * was empty before @bio.  The forced scheduling isn't likely to
	 * cause undue delay as @bio is likely to be dispatched directly if
	 * its @tg's disptime is not in the future.
	 */
	if (tg->flags & THROTL_TG_WAS_EMPTY) {
		tg_update_disptime(tg);
		throtl_schedule_next_dispatch(tg->service_queue.parent_sq, true);
	}

out_unlock:
	spin_unlock_irq(q->queue_lock);
out_unlock_rcu:
	rcu_read_unlock();
out:
	/*
	 * As multiple blk-throtls may stack in the same issue path, we
	 * don't want bios to leave with the flag set.  Clear the flag if
	 * being issued.
	 */
	if (!throttled)
		bio->bi_rw &= ~REQ_THROTTLED;
	return throttled;
}

/*
 * Dispatch all bios from all children tg's queued on @parent_sq.  On
 * return, @parent_sq is guaranteed to not have any active children tg's
 * and all bios from previously active tg's are on @parent_sq->bio_lists[].
 */
static void tg_drain_bios(struct throtl_service_queue *parent_sq)
{
	struct throtl_grp *tg;

	while ((tg = throtl_rb_first(parent_sq))) {
		struct throtl_service_queue *sq = &tg->service_queue;
		struct bio *bio;

		throtl_dequeue_tg(tg);

		while ((bio = throtl_peek_queued(&sq->queued[READ])))
			tg_dispatch_one_bio(tg, bio_data_dir(bio));
		while ((bio = throtl_peek_queued(&sq->queued[WRITE])))
			tg_dispatch_one_bio(tg, bio_data_dir(bio));
	}
}

/**
 * blk_throtl_drain - drain throttled bios
 * @q: request_queue to drain throttled bios for
 *
 * Dispatch all currently throttled bios on @q through ->make_request_fn().
 */
void blk_throtl_drain(struct request_queue *q)
	__releases(q->queue_lock) __acquires(q->queue_lock)
{
	struct throtl_data *td = q->td;
	struct blkcg_gq *blkg;
	struct cgroup_subsys_state *pos_css;
	struct bio *bio;
	int rw;

	queue_lockdep_assert_held(q);
	rcu_read_lock();

	/*
	 * Drain each tg while doing post-order walk on the blkg tree, so
	 * that all bios are propagated to td->service_queue.  It'd be
	 * better to walk service_queue tree directly but blkg walk is
	 * easier.
	 */
	blkg_for_each_descendant_post(blkg, pos_css, td->queue->root_blkg)
		tg_drain_bios(&blkg_to_tg(blkg)->service_queue);

	/* finally, transfer bios from top-level tg's into the td */
	tg_drain_bios(&td->service_queue);

	rcu_read_unlock();
	spin_unlock_irq(q->queue_lock);

	/* all bios now should be in td->service_queue, issue them */
	for (rw = READ; rw <= WRITE; rw++)
		while ((bio = throtl_pop_queued(&td->service_queue.queued[rw],
						NULL)))
			generic_make_request(bio);

	spin_lock_irq(q->queue_lock);
}

int blk_throtl_init(struct request_queue *q)
{
	struct throtl_data *td;
	int ret;

	td = kzalloc_node(sizeof(*td), GFP_KERNEL, q->node);
	if (!td)
		return -ENOMEM;

	INIT_WORK(&td->dispatch_work, blk_throtl_dispatch_work_fn);
	throtl_service_queue_init(&td->service_queue, NULL);

	q->td = td;
	td->queue = q;

	/* activate policy */
	ret = blkcg_activate_policy(q, &blkcg_policy_throtl);
	if (ret)
		kfree(td);
	return ret;
}

void blk_throtl_exit(struct request_queue *q)
{
	BUG_ON(!q->td);
	throtl_shutdown_wq(q);
	blkcg_deactivate_policy(q, &blkcg_policy_throtl);
	kfree(q->td);
}

static int __init throtl_init(void)
{
	kthrotld_workqueue = alloc_workqueue("kthrotld", WQ_MEM_RECLAIM, 0);
	if (!kthrotld_workqueue)
		panic("Failed to create kthrotld\n");

	return blkcg_policy_register(&blkcg_policy_throtl);
}

module_init(throtl_init);