Root/block/blk-throttle.c

1/*
2 * Interface for controlling IO bandwidth on a request queue
3 *
4 * Copyright (C) 2010 Vivek Goyal <vgoyal@redhat.com>
5 */
6
7#include <linux/module.h>
8#include <linux/slab.h>
9#include <linux/blkdev.h>
10#include <linux/bio.h>
11#include <linux/blktrace_api.h>
12#include "blk-cgroup.h"
13#include "blk.h"
14
15/* Max dispatch from a group in 1 round */
16static int throtl_grp_quantum = 8;
17
18/* Total max dispatch from all groups in one round */
19static int throtl_quantum = 32;
20
21/* Throttling is performed over 100ms slice and after that slice is renewed */
22static unsigned long throtl_slice = HZ/10; /* 100 ms */
23
24static struct blkcg_policy blkcg_policy_throtl;
25
26/* A workqueue to queue throttle related work */
27static struct workqueue_struct *kthrotld_workqueue;
28static void throtl_schedule_delayed_work(struct throtl_data *td,
29                unsigned long delay);
30
31struct throtl_rb_root {
32    struct rb_root rb;
33    struct rb_node *left;
34    unsigned int count;
35    unsigned long min_disptime;
36};
37
38#define THROTL_RB_ROOT (struct throtl_rb_root) { .rb = RB_ROOT, .left = NULL, \
39            .count = 0, .min_disptime = 0}
40
41#define rb_entry_tg(node) rb_entry((node), struct throtl_grp, rb_node)
42
43/* Per-cpu group stats */
44struct tg_stats_cpu {
45    /* total bytes transferred */
46    struct blkg_rwstat service_bytes;
47    /* total IOs serviced, post merge */
48    struct blkg_rwstat serviced;
49};
50
51struct throtl_grp {
52    /* must be the first member */
53    struct blkg_policy_data pd;
54
55    /* active throtl group service_tree member */
56    struct rb_node rb_node;
57
58    /*
59     * Dispatch time in jiffies. This is the estimated time when group
60     * will unthrottle and is ready to dispatch more bio. It is used as
61     * key to sort active groups in service tree.
62     */
63    unsigned long disptime;
64
65    unsigned int flags;
66
67    /* Two lists for READ and WRITE */
68    struct bio_list bio_lists[2];
69
70    /* Number of queued bios on READ and WRITE lists */
71    unsigned int nr_queued[2];
72
73    /* bytes per second rate limits */
74    uint64_t bps[2];
75
76    /* IOPS limits */
77    unsigned int iops[2];
78
79    /* Number of bytes disptached in current slice */
80    uint64_t bytes_disp[2];
81    /* Number of bio's dispatched in current slice */
82    unsigned int io_disp[2];
83
84    /* When did we start a new slice */
85    unsigned long slice_start[2];
86    unsigned long slice_end[2];
87
88    /* Some throttle limits got updated for the group */
89    int limits_changed;
90
91    /* Per cpu stats pointer */
92    struct tg_stats_cpu __percpu *stats_cpu;
93
94    /* List of tgs waiting for per cpu stats memory to be allocated */
95    struct list_head stats_alloc_node;
96};
97
98struct throtl_data
99{
100    /* service tree for active throtl groups */
101    struct throtl_rb_root tg_service_tree;
102
103    struct request_queue *queue;
104
105    /* Total Number of queued bios on READ and WRITE lists */
106    unsigned int nr_queued[2];
107
108    /*
109     * number of total undestroyed groups
110     */
111    unsigned int nr_undestroyed_grps;
112
113    /* Work for dispatching throttled bios */
114    struct delayed_work throtl_work;
115
116    int limits_changed;
117};
118
119/* list and work item to allocate percpu group stats */
120static DEFINE_SPINLOCK(tg_stats_alloc_lock);
121static LIST_HEAD(tg_stats_alloc_list);
122
123static void tg_stats_alloc_fn(struct work_struct *);
124static DECLARE_DELAYED_WORK(tg_stats_alloc_work, tg_stats_alloc_fn);
125
126static inline struct throtl_grp *pd_to_tg(struct blkg_policy_data *pd)
127{
128    return pd ? container_of(pd, struct throtl_grp, pd) : NULL;
129}
130
131static inline struct throtl_grp *blkg_to_tg(struct blkcg_gq *blkg)
132{
133    return pd_to_tg(blkg_to_pd(blkg, &blkcg_policy_throtl));
134}
135
136static inline struct blkcg_gq *tg_to_blkg(struct throtl_grp *tg)
137{
138    return pd_to_blkg(&tg->pd);
139}
140
141static inline struct throtl_grp *td_root_tg(struct throtl_data *td)
142{
143    return blkg_to_tg(td->queue->root_blkg);
144}
145
146enum tg_state_flags {
147    THROTL_TG_FLAG_on_rr = 0, /* on round-robin busy list */
148};
149
150#define THROTL_TG_FNS(name) \
151static inline void throtl_mark_tg_##name(struct throtl_grp *tg) \
152{ \
153    (tg)->flags |= (1 << THROTL_TG_FLAG_##name); \
154} \
155static inline void throtl_clear_tg_##name(struct throtl_grp *tg) \
156{ \
157    (tg)->flags &= ~(1 << THROTL_TG_FLAG_##name); \
158} \
159static inline int throtl_tg_##name(const struct throtl_grp *tg) \
160{ \
161    return ((tg)->flags & (1 << THROTL_TG_FLAG_##name)) != 0; \
162}
163
164THROTL_TG_FNS(on_rr);
165
166#define throtl_log_tg(td, tg, fmt, args...) do { \
167    char __pbuf[128]; \
168                                    \
169    blkg_path(tg_to_blkg(tg), __pbuf, sizeof(__pbuf)); \
170    blk_add_trace_msg((td)->queue, "throtl %s " fmt, __pbuf, ##args); \
171} while (0)
172
173#define throtl_log(td, fmt, args...) \
174    blk_add_trace_msg((td)->queue, "throtl " fmt, ##args)
175
176static inline unsigned int total_nr_queued(struct throtl_data *td)
177{
178    return td->nr_queued[0] + td->nr_queued[1];
179}
180
181/*
182 * Worker for allocating per cpu stat for tgs. This is scheduled on the
183 * system_wq once there are some groups on the alloc_list waiting for
184 * allocation.
185 */
186static void tg_stats_alloc_fn(struct work_struct *work)
187{
188    static struct tg_stats_cpu *stats_cpu; /* this fn is non-reentrant */
189    struct delayed_work *dwork = to_delayed_work(work);
190    bool empty = false;
191
192alloc_stats:
193    if (!stats_cpu) {
194        stats_cpu = alloc_percpu(struct tg_stats_cpu);
195        if (!stats_cpu) {
196            /* allocation failed, try again after some time */
197            schedule_delayed_work(dwork, msecs_to_jiffies(10));
198            return;
199        }
200    }
201
202    spin_lock_irq(&tg_stats_alloc_lock);
203
204    if (!list_empty(&tg_stats_alloc_list)) {
205        struct throtl_grp *tg = list_first_entry(&tg_stats_alloc_list,
206                             struct throtl_grp,
207                             stats_alloc_node);
208        swap(tg->stats_cpu, stats_cpu);
209        list_del_init(&tg->stats_alloc_node);
210    }
211
212    empty = list_empty(&tg_stats_alloc_list);
213    spin_unlock_irq(&tg_stats_alloc_lock);
214    if (!empty)
215        goto alloc_stats;
216}
217
218static void throtl_pd_init(struct blkcg_gq *blkg)
219{
220    struct throtl_grp *tg = blkg_to_tg(blkg);
221    unsigned long flags;
222
223    RB_CLEAR_NODE(&tg->rb_node);
224    bio_list_init(&tg->bio_lists[0]);
225    bio_list_init(&tg->bio_lists[1]);
226    tg->limits_changed = false;
227
228    tg->bps[READ] = -1;
229    tg->bps[WRITE] = -1;
230    tg->iops[READ] = -1;
231    tg->iops[WRITE] = -1;
232
233    /*
234     * Ugh... We need to perform per-cpu allocation for tg->stats_cpu
235     * but percpu allocator can't be called from IO path. Queue tg on
236     * tg_stats_alloc_list and allocate from work item.
237     */
238    spin_lock_irqsave(&tg_stats_alloc_lock, flags);
239    list_add(&tg->stats_alloc_node, &tg_stats_alloc_list);
240    schedule_delayed_work(&tg_stats_alloc_work, 0);
241    spin_unlock_irqrestore(&tg_stats_alloc_lock, flags);
242}
243
244static void throtl_pd_exit(struct blkcg_gq *blkg)
245{
246    struct throtl_grp *tg = blkg_to_tg(blkg);
247    unsigned long flags;
248
249    spin_lock_irqsave(&tg_stats_alloc_lock, flags);
250    list_del_init(&tg->stats_alloc_node);
251    spin_unlock_irqrestore(&tg_stats_alloc_lock, flags);
252
253    free_percpu(tg->stats_cpu);
254}
255
256static void throtl_pd_reset_stats(struct blkcg_gq *blkg)
257{
258    struct throtl_grp *tg = blkg_to_tg(blkg);
259    int cpu;
260
261    if (tg->stats_cpu == NULL)
262        return;
263
264    for_each_possible_cpu(cpu) {
265        struct tg_stats_cpu *sc = per_cpu_ptr(tg->stats_cpu, cpu);
266
267        blkg_rwstat_reset(&sc->service_bytes);
268        blkg_rwstat_reset(&sc->serviced);
269    }
270}
271
272static struct throtl_grp *throtl_lookup_tg(struct throtl_data *td,
273                       struct blkcg *blkcg)
274{
275    /*
276     * This is the common case when there are no blkcgs. Avoid lookup
277     * in this case
278     */
279    if (blkcg == &blkcg_root)
280        return td_root_tg(td);
281
282    return blkg_to_tg(blkg_lookup(blkcg, td->queue));
283}
284
285static struct throtl_grp *throtl_lookup_create_tg(struct throtl_data *td,
286                          struct blkcg *blkcg)
287{
288    struct request_queue *q = td->queue;
289    struct throtl_grp *tg = NULL;
290
291    /*
292     * This is the common case when there are no blkcgs. Avoid lookup
293     * in this case
294     */
295    if (blkcg == &blkcg_root) {
296        tg = td_root_tg(td);
297    } else {
298        struct blkcg_gq *blkg;
299
300        blkg = blkg_lookup_create(blkcg, q);
301
302        /* if %NULL and @q is alive, fall back to root_tg */
303        if (!IS_ERR(blkg))
304            tg = blkg_to_tg(blkg);
305        else if (!blk_queue_dying(q))
306            tg = td_root_tg(td);
307    }
308
309    return tg;
310}
311
312static struct throtl_grp *throtl_rb_first(struct throtl_rb_root *root)
313{
314    /* Service tree is empty */
315    if (!root->count)
316        return NULL;
317
318    if (!root->left)
319        root->left = rb_first(&root->rb);
320
321    if (root->left)
322        return rb_entry_tg(root->left);
323
324    return NULL;
325}
326
327static void rb_erase_init(struct rb_node *n, struct rb_root *root)
328{
329    rb_erase(n, root);
330    RB_CLEAR_NODE(n);
331}
332
333static void throtl_rb_erase(struct rb_node *n, struct throtl_rb_root *root)
334{
335    if (root->left == n)
336        root->left = NULL;
337    rb_erase_init(n, &root->rb);
338    --root->count;
339}
340
341static void update_min_dispatch_time(struct throtl_rb_root *st)
342{
343    struct throtl_grp *tg;
344
345    tg = throtl_rb_first(st);
346    if (!tg)
347        return;
348
349    st->min_disptime = tg->disptime;
350}
351
352static void
353tg_service_tree_add(struct throtl_rb_root *st, struct throtl_grp *tg)
354{
355    struct rb_node **node = &st->rb.rb_node;
356    struct rb_node *parent = NULL;
357    struct throtl_grp *__tg;
358    unsigned long key = tg->disptime;
359    int left = 1;
360
361    while (*node != NULL) {
362        parent = *node;
363        __tg = rb_entry_tg(parent);
364
365        if (time_before(key, __tg->disptime))
366            node = &parent->rb_left;
367        else {
368            node = &parent->rb_right;
369            left = 0;
370        }
371    }
372
373    if (left)
374        st->left = &tg->rb_node;
375
376    rb_link_node(&tg->rb_node, parent, node);
377    rb_insert_color(&tg->rb_node, &st->rb);
378}
379
380static void __throtl_enqueue_tg(struct throtl_data *td, struct throtl_grp *tg)
381{
382    struct throtl_rb_root *st = &td->tg_service_tree;
383
384    tg_service_tree_add(st, tg);
385    throtl_mark_tg_on_rr(tg);
386    st->count++;
387}
388
389static void throtl_enqueue_tg(struct throtl_data *td, struct throtl_grp *tg)
390{
391    if (!throtl_tg_on_rr(tg))
392        __throtl_enqueue_tg(td, tg);
393}
394
395static void __throtl_dequeue_tg(struct throtl_data *td, struct throtl_grp *tg)
396{
397    throtl_rb_erase(&tg->rb_node, &td->tg_service_tree);
398    throtl_clear_tg_on_rr(tg);
399}
400
401static void throtl_dequeue_tg(struct throtl_data *td, struct throtl_grp *tg)
402{
403    if (throtl_tg_on_rr(tg))
404        __throtl_dequeue_tg(td, tg);
405}
406
407static void throtl_schedule_next_dispatch(struct throtl_data *td)
408{
409    struct throtl_rb_root *st = &td->tg_service_tree;
410
411    /*
412     * If there are more bios pending, schedule more work.
413     */
414    if (!total_nr_queued(td))
415        return;
416
417    BUG_ON(!st->count);
418
419    update_min_dispatch_time(st);
420
421    if (time_before_eq(st->min_disptime, jiffies))
422        throtl_schedule_delayed_work(td, 0);
423    else
424        throtl_schedule_delayed_work(td, (st->min_disptime - jiffies));
425}
426
427static inline void
428throtl_start_new_slice(struct throtl_data *td, struct throtl_grp *tg, bool rw)
429{
430    tg->bytes_disp[rw] = 0;
431    tg->io_disp[rw] = 0;
432    tg->slice_start[rw] = jiffies;
433    tg->slice_end[rw] = jiffies + throtl_slice;
434    throtl_log_tg(td, tg, "[%c] new slice start=%lu end=%lu jiffies=%lu",
435            rw == READ ? 'R' : 'W', tg->slice_start[rw],
436            tg->slice_end[rw], jiffies);
437}
438
439static inline void throtl_set_slice_end(struct throtl_data *td,
440        struct throtl_grp *tg, bool rw, unsigned long jiffy_end)
441{
442    tg->slice_end[rw] = roundup(jiffy_end, throtl_slice);
443}
444
445static inline void throtl_extend_slice(struct throtl_data *td,
446        struct throtl_grp *tg, bool rw, unsigned long jiffy_end)
447{
448    tg->slice_end[rw] = roundup(jiffy_end, throtl_slice);
449    throtl_log_tg(td, tg, "[%c] extend slice start=%lu end=%lu jiffies=%lu",
450            rw == READ ? 'R' : 'W', tg->slice_start[rw],
451            tg->slice_end[rw], jiffies);
452}
453
454/* Determine if previously allocated or extended slice is complete or not */
455static bool
456throtl_slice_used(struct throtl_data *td, struct throtl_grp *tg, bool rw)
457{
458    if (time_in_range(jiffies, tg->slice_start[rw], tg->slice_end[rw]))
459        return 0;
460
461    return 1;
462}
463
464/* Trim the used slices and adjust slice start accordingly */
465static inline void
466throtl_trim_slice(struct throtl_data *td, struct throtl_grp *tg, bool rw)
467{
468    unsigned long nr_slices, time_elapsed, io_trim;
469    u64 bytes_trim, tmp;
470
471    BUG_ON(time_before(tg->slice_end[rw], tg->slice_start[rw]));
472
473    /*
474     * If bps are unlimited (-1), then time slice don't get
475     * renewed. Don't try to trim the slice if slice is used. A new
476     * slice will start when appropriate.
477     */
478    if (throtl_slice_used(td, tg, rw))
479        return;
480
481    /*
482     * A bio has been dispatched. Also adjust slice_end. It might happen
483     * that initially cgroup limit was very low resulting in high
484     * slice_end, but later limit was bumped up and bio was dispached
485     * sooner, then we need to reduce slice_end. A high bogus slice_end
486     * is bad because it does not allow new slice to start.
487     */
488
489    throtl_set_slice_end(td, tg, rw, jiffies + throtl_slice);
490
491    time_elapsed = jiffies - tg->slice_start[rw];
492
493    nr_slices = time_elapsed / throtl_slice;
494
495    if (!nr_slices)
496        return;
497    tmp = tg->bps[rw] * throtl_slice * nr_slices;
498    do_div(tmp, HZ);
499    bytes_trim = tmp;
500
501    io_trim = (tg->iops[rw] * throtl_slice * nr_slices)/HZ;
502
503    if (!bytes_trim && !io_trim)
504        return;
505
506    if (tg->bytes_disp[rw] >= bytes_trim)
507        tg->bytes_disp[rw] -= bytes_trim;
508    else
509        tg->bytes_disp[rw] = 0;
510
511    if (tg->io_disp[rw] >= io_trim)
512        tg->io_disp[rw] -= io_trim;
513    else
514        tg->io_disp[rw] = 0;
515
516    tg->slice_start[rw] += nr_slices * throtl_slice;
517
518    throtl_log_tg(td, tg, "[%c] trim slice nr=%lu bytes=%llu io=%lu"
519            " start=%lu end=%lu jiffies=%lu",
520            rw == READ ? 'R' : 'W', nr_slices, bytes_trim, io_trim,
521            tg->slice_start[rw], tg->slice_end[rw], jiffies);
522}
523
524static bool tg_with_in_iops_limit(struct throtl_data *td, struct throtl_grp *tg,
525        struct bio *bio, unsigned long *wait)
526{
527    bool rw = bio_data_dir(bio);
528    unsigned int io_allowed;
529    unsigned long jiffy_elapsed, jiffy_wait, jiffy_elapsed_rnd;
530    u64 tmp;
531
532    jiffy_elapsed = jiffy_elapsed_rnd = jiffies - tg->slice_start[rw];
533
534    /* Slice has just started. Consider one slice interval */
535    if (!jiffy_elapsed)
536        jiffy_elapsed_rnd = throtl_slice;
537
538    jiffy_elapsed_rnd = roundup(jiffy_elapsed_rnd, throtl_slice);
539
540    /*
541     * jiffy_elapsed_rnd should not be a big value as minimum iops can be
542     * 1 then at max jiffy elapsed should be equivalent of 1 second as we
543     * will allow dispatch after 1 second and after that slice should
544     * have been trimmed.
545     */
546
547    tmp = (u64)tg->iops[rw] * jiffy_elapsed_rnd;
548    do_div(tmp, HZ);
549
550    if (tmp > UINT_MAX)
551        io_allowed = UINT_MAX;
552    else
553        io_allowed = tmp;
554
555    if (tg->io_disp[rw] + 1 <= io_allowed) {
556        if (wait)
557            *wait = 0;
558        return 1;
559    }
560
561    /* Calc approx time to dispatch */
562    jiffy_wait = ((tg->io_disp[rw] + 1) * HZ)/tg->iops[rw] + 1;
563
564    if (jiffy_wait > jiffy_elapsed)
565        jiffy_wait = jiffy_wait - jiffy_elapsed;
566    else
567        jiffy_wait = 1;
568
569    if (wait)
570        *wait = jiffy_wait;
571    return 0;
572}
573
574static bool tg_with_in_bps_limit(struct throtl_data *td, struct throtl_grp *tg,
575        struct bio *bio, unsigned long *wait)
576{
577    bool rw = bio_data_dir(bio);
578    u64 bytes_allowed, extra_bytes, tmp;
579    unsigned long jiffy_elapsed, jiffy_wait, jiffy_elapsed_rnd;
580
581    jiffy_elapsed = jiffy_elapsed_rnd = jiffies - tg->slice_start[rw];
582
583    /* Slice has just started. Consider one slice interval */
584    if (!jiffy_elapsed)
585        jiffy_elapsed_rnd = throtl_slice;
586
587    jiffy_elapsed_rnd = roundup(jiffy_elapsed_rnd, throtl_slice);
588
589    tmp = tg->bps[rw] * jiffy_elapsed_rnd;
590    do_div(tmp, HZ);
591    bytes_allowed = tmp;
592
593    if (tg->bytes_disp[rw] + bio->bi_size <= bytes_allowed) {
594        if (wait)
595            *wait = 0;
596        return 1;
597    }
598
599    /* Calc approx time to dispatch */
600    extra_bytes = tg->bytes_disp[rw] + bio->bi_size - bytes_allowed;
601    jiffy_wait = div64_u64(extra_bytes * HZ, tg->bps[rw]);
602
603    if (!jiffy_wait)
604        jiffy_wait = 1;
605
606    /*
607     * This wait time is without taking into consideration the rounding
608     * up we did. Add that time also.
609     */
610    jiffy_wait = jiffy_wait + (jiffy_elapsed_rnd - jiffy_elapsed);
611    if (wait)
612        *wait = jiffy_wait;
613    return 0;
614}
615
616static bool tg_no_rule_group(struct throtl_grp *tg, bool rw) {
617    if (tg->bps[rw] == -1 && tg->iops[rw] == -1)
618        return 1;
619    return 0;
620}
621
622/*
623 * Returns whether one can dispatch a bio or not. Also returns approx number
624 * of jiffies to wait before this bio is with-in IO rate and can be dispatched
625 */
626static bool tg_may_dispatch(struct throtl_data *td, struct throtl_grp *tg,
627                struct bio *bio, unsigned long *wait)
628{
629    bool rw = bio_data_dir(bio);
630    unsigned long bps_wait = 0, iops_wait = 0, max_wait = 0;
631
632    /*
633      * Currently whole state machine of group depends on first bio
634     * queued in the group bio list. So one should not be calling
635     * this function with a different bio if there are other bios
636     * queued.
637     */
638    BUG_ON(tg->nr_queued[rw] && bio != bio_list_peek(&tg->bio_lists[rw]));
639
640    /* If tg->bps = -1, then BW is unlimited */
641    if (tg->bps[rw] == -1 && tg->iops[rw] == -1) {
642        if (wait)
643            *wait = 0;
644        return 1;
645    }
646
647    /*
648     * If previous slice expired, start a new one otherwise renew/extend
649     * existing slice to make sure it is at least throtl_slice interval
650     * long since now.
651     */
652    if (throtl_slice_used(td, tg, rw))
653        throtl_start_new_slice(td, tg, rw);
654    else {
655        if (time_before(tg->slice_end[rw], jiffies + throtl_slice))
656            throtl_extend_slice(td, tg, rw, jiffies + throtl_slice);
657    }
658
659    if (tg_with_in_bps_limit(td, tg, bio, &bps_wait)
660        && tg_with_in_iops_limit(td, tg, bio, &iops_wait)) {
661        if (wait)
662            *wait = 0;
663        return 1;
664    }
665
666    max_wait = max(bps_wait, iops_wait);
667
668    if (wait)
669        *wait = max_wait;
670
671    if (time_before(tg->slice_end[rw], jiffies + max_wait))
672        throtl_extend_slice(td, tg, rw, jiffies + max_wait);
673
674    return 0;
675}
676
677static void throtl_update_dispatch_stats(struct blkcg_gq *blkg, u64 bytes,
678                     int rw)
679{
680    struct throtl_grp *tg = blkg_to_tg(blkg);
681    struct tg_stats_cpu *stats_cpu;
682    unsigned long flags;
683
684    /* If per cpu stats are not allocated yet, don't do any accounting. */
685    if (tg->stats_cpu == NULL)
686        return;
687
688    /*
689     * Disabling interrupts to provide mutual exclusion between two
690     * writes on same cpu. It probably is not needed for 64bit. Not
691     * optimizing that case yet.
692     */
693    local_irq_save(flags);
694
695    stats_cpu = this_cpu_ptr(tg->stats_cpu);
696
697    blkg_rwstat_add(&stats_cpu->serviced, rw, 1);
698    blkg_rwstat_add(&stats_cpu->service_bytes, rw, bytes);
699
700    local_irq_restore(flags);
701}
702
703static void throtl_charge_bio(struct throtl_grp *tg, struct bio *bio)
704{
705    bool rw = bio_data_dir(bio);
706
707    /* Charge the bio to the group */
708    tg->bytes_disp[rw] += bio->bi_size;
709    tg->io_disp[rw]++;
710
711    throtl_update_dispatch_stats(tg_to_blkg(tg), bio->bi_size, bio->bi_rw);
712}
713
714static void throtl_add_bio_tg(struct throtl_data *td, struct throtl_grp *tg,
715            struct bio *bio)
716{
717    bool rw = bio_data_dir(bio);
718
719    bio_list_add(&tg->bio_lists[rw], bio);
720    /* Take a bio reference on tg */
721    blkg_get(tg_to_blkg(tg));
722    tg->nr_queued[rw]++;
723    td->nr_queued[rw]++;
724    throtl_enqueue_tg(td, tg);
725}
726
727static void tg_update_disptime(struct throtl_data *td, struct throtl_grp *tg)
728{
729    unsigned long read_wait = -1, write_wait = -1, min_wait = -1, disptime;
730    struct bio *bio;
731
732    if ((bio = bio_list_peek(&tg->bio_lists[READ])))
733        tg_may_dispatch(td, tg, bio, &read_wait);
734
735    if ((bio = bio_list_peek(&tg->bio_lists[WRITE])))
736        tg_may_dispatch(td, tg, bio, &write_wait);
737
738    min_wait = min(read_wait, write_wait);
739    disptime = jiffies + min_wait;
740
741    /* Update dispatch time */
742    throtl_dequeue_tg(td, tg);
743    tg->disptime = disptime;
744    throtl_enqueue_tg(td, tg);
745}
746
747static void tg_dispatch_one_bio(struct throtl_data *td, struct throtl_grp *tg,
748                bool rw, struct bio_list *bl)
749{
750    struct bio *bio;
751
752    bio = bio_list_pop(&tg->bio_lists[rw]);
753    tg->nr_queued[rw]--;
754    /* Drop bio reference on blkg */
755    blkg_put(tg_to_blkg(tg));
756
757    BUG_ON(td->nr_queued[rw] <= 0);
758    td->nr_queued[rw]--;
759
760    throtl_charge_bio(tg, bio);
761    bio_list_add(bl, bio);
762    bio->bi_rw |= REQ_THROTTLED;
763
764    throtl_trim_slice(td, tg, rw);
765}
766
767static int throtl_dispatch_tg(struct throtl_data *td, struct throtl_grp *tg,
768                struct bio_list *bl)
769{
770    unsigned int nr_reads = 0, nr_writes = 0;
771    unsigned int max_nr_reads = throtl_grp_quantum*3/4;
772    unsigned int max_nr_writes = throtl_grp_quantum - max_nr_reads;
773    struct bio *bio;
774
775    /* Try to dispatch 75% READS and 25% WRITES */
776
777    while ((bio = bio_list_peek(&tg->bio_lists[READ]))
778        && tg_may_dispatch(td, tg, bio, NULL)) {
779
780        tg_dispatch_one_bio(td, tg, bio_data_dir(bio), bl);
781        nr_reads++;
782
783        if (nr_reads >= max_nr_reads)
784            break;
785    }
786
787    while ((bio = bio_list_peek(&tg->bio_lists[WRITE]))
788        && tg_may_dispatch(td, tg, bio, NULL)) {
789
790        tg_dispatch_one_bio(td, tg, bio_data_dir(bio), bl);
791        nr_writes++;
792
793        if (nr_writes >= max_nr_writes)
794            break;
795    }
796
797    return nr_reads + nr_writes;
798}
799
800static int throtl_select_dispatch(struct throtl_data *td, struct bio_list *bl)
801{
802    unsigned int nr_disp = 0;
803    struct throtl_grp *tg;
804    struct throtl_rb_root *st = &td->tg_service_tree;
805
806    while (1) {
807        tg = throtl_rb_first(st);
808
809        if (!tg)
810            break;
811
812        if (time_before(jiffies, tg->disptime))
813            break;
814
815        throtl_dequeue_tg(td, tg);
816
817        nr_disp += throtl_dispatch_tg(td, tg, bl);
818
819        if (tg->nr_queued[0] || tg->nr_queued[1]) {
820            tg_update_disptime(td, tg);
821            throtl_enqueue_tg(td, tg);
822        }
823
824        if (nr_disp >= throtl_quantum)
825            break;
826    }
827
828    return nr_disp;
829}
830
831static void throtl_process_limit_change(struct throtl_data *td)
832{
833    struct request_queue *q = td->queue;
834    struct blkcg_gq *blkg, *n;
835
836    if (!td->limits_changed)
837        return;
838
839    xchg(&td->limits_changed, false);
840
841    throtl_log(td, "limits changed");
842
843    list_for_each_entry_safe(blkg, n, &q->blkg_list, q_node) {
844        struct throtl_grp *tg = blkg_to_tg(blkg);
845
846        if (!tg->limits_changed)
847            continue;
848
849        if (!xchg(&tg->limits_changed, false))
850            continue;
851
852        throtl_log_tg(td, tg, "limit change rbps=%llu wbps=%llu"
853            " riops=%u wiops=%u", tg->bps[READ], tg->bps[WRITE],
854            tg->iops[READ], tg->iops[WRITE]);
855
856        /*
857         * Restart the slices for both READ and WRITES. It
858         * might happen that a group's limit are dropped
859         * suddenly and we don't want to account recently
860         * dispatched IO with new low rate
861         */
862        throtl_start_new_slice(td, tg, 0);
863        throtl_start_new_slice(td, tg, 1);
864
865        if (throtl_tg_on_rr(tg))
866            tg_update_disptime(td, tg);
867    }
868}
869
870/* Dispatch throttled bios. Should be called without queue lock held. */
871static int throtl_dispatch(struct request_queue *q)
872{
873    struct throtl_data *td = q->td;
874    unsigned int nr_disp = 0;
875    struct bio_list bio_list_on_stack;
876    struct bio *bio;
877    struct blk_plug plug;
878
879    spin_lock_irq(q->queue_lock);
880
881    throtl_process_limit_change(td);
882
883    if (!total_nr_queued(td))
884        goto out;
885
886    bio_list_init(&bio_list_on_stack);
887
888    throtl_log(td, "dispatch nr_queued=%u read=%u write=%u",
889            total_nr_queued(td), td->nr_queued[READ],
890            td->nr_queued[WRITE]);
891
892    nr_disp = throtl_select_dispatch(td, &bio_list_on_stack);
893
894    if (nr_disp)
895        throtl_log(td, "bios disp=%u", nr_disp);
896
897    throtl_schedule_next_dispatch(td);
898out:
899    spin_unlock_irq(q->queue_lock);
900
901    /*
902     * If we dispatched some requests, unplug the queue to make sure
903     * immediate dispatch
904     */
905    if (nr_disp) {
906        blk_start_plug(&plug);
907        while((bio = bio_list_pop(&bio_list_on_stack)))
908            generic_make_request(bio);
909        blk_finish_plug(&plug);
910    }
911    return nr_disp;
912}
913
914void blk_throtl_work(struct work_struct *work)
915{
916    struct throtl_data *td = container_of(work, struct throtl_data,
917                    throtl_work.work);
918    struct request_queue *q = td->queue;
919
920    throtl_dispatch(q);
921}
922
923/* Call with queue lock held */
924static void
925throtl_schedule_delayed_work(struct throtl_data *td, unsigned long delay)
926{
927
928    struct delayed_work *dwork = &td->throtl_work;
929
930    /* schedule work if limits changed even if no bio is queued */
931    if (total_nr_queued(td) || td->limits_changed) {
932        mod_delayed_work(kthrotld_workqueue, dwork, delay);
933        throtl_log(td, "schedule work. delay=%lu jiffies=%lu",
934                delay, jiffies);
935    }
936}
937
938static u64 tg_prfill_cpu_rwstat(struct seq_file *sf,
939                struct blkg_policy_data *pd, int off)
940{
941    struct throtl_grp *tg = pd_to_tg(pd);
942    struct blkg_rwstat rwstat = { }, tmp;
943    int i, cpu;
944
945    for_each_possible_cpu(cpu) {
946        struct tg_stats_cpu *sc = per_cpu_ptr(tg->stats_cpu, cpu);
947
948        tmp = blkg_rwstat_read((void *)sc + off);
949        for (i = 0; i < BLKG_RWSTAT_NR; i++)
950            rwstat.cnt[i] += tmp.cnt[i];
951    }
952
953    return __blkg_prfill_rwstat(sf, pd, &rwstat);
954}
955
956static int tg_print_cpu_rwstat(struct cgroup *cgrp, struct cftype *cft,
957                   struct seq_file *sf)
958{
959    struct blkcg *blkcg = cgroup_to_blkcg(cgrp);
960
961    blkcg_print_blkgs(sf, blkcg, tg_prfill_cpu_rwstat, &blkcg_policy_throtl,
962              cft->private, true);
963    return 0;
964}
965
966static u64 tg_prfill_conf_u64(struct seq_file *sf, struct blkg_policy_data *pd,
967                  int off)
968{
969    struct throtl_grp *tg = pd_to_tg(pd);
970    u64 v = *(u64 *)((void *)tg + off);
971
972    if (v == -1)
973        return 0;
974    return __blkg_prfill_u64(sf, pd, v);
975}
976
977static u64 tg_prfill_conf_uint(struct seq_file *sf, struct blkg_policy_data *pd,
978                   int off)
979{
980    struct throtl_grp *tg = pd_to_tg(pd);
981    unsigned int v = *(unsigned int *)((void *)tg + off);
982
983    if (v == -1)
984        return 0;
985    return __blkg_prfill_u64(sf, pd, v);
986}
987
988static int tg_print_conf_u64(struct cgroup *cgrp, struct cftype *cft,
989                 struct seq_file *sf)
990{
991    blkcg_print_blkgs(sf, cgroup_to_blkcg(cgrp), tg_prfill_conf_u64,
992              &blkcg_policy_throtl, cft->private, false);
993    return 0;
994}
995
996static int tg_print_conf_uint(struct cgroup *cgrp, struct cftype *cft,
997                  struct seq_file *sf)
998{
999    blkcg_print_blkgs(sf, cgroup_to_blkcg(cgrp), tg_prfill_conf_uint,
1000              &blkcg_policy_throtl, cft->private, false);
1001    return 0;
1002}
1003
1004static int tg_set_conf(struct cgroup *cgrp, struct cftype *cft, const char *buf,
1005               bool is_u64)
1006{
1007    struct blkcg *blkcg = cgroup_to_blkcg(cgrp);
1008    struct blkg_conf_ctx ctx;
1009    struct throtl_grp *tg;
1010    struct throtl_data *td;
1011    int ret;
1012
1013    ret = blkg_conf_prep(blkcg, &blkcg_policy_throtl, buf, &ctx);
1014    if (ret)
1015        return ret;
1016
1017    tg = blkg_to_tg(ctx.blkg);
1018    td = ctx.blkg->q->td;
1019
1020    if (!ctx.v)
1021        ctx.v = -1;
1022
1023    if (is_u64)
1024        *(u64 *)((void *)tg + cft->private) = ctx.v;
1025    else
1026        *(unsigned int *)((void *)tg + cft->private) = ctx.v;
1027
1028    /* XXX: we don't need the following deferred processing */
1029    xchg(&tg->limits_changed, true);
1030    xchg(&td->limits_changed, true);
1031    throtl_schedule_delayed_work(td, 0);
1032
1033    blkg_conf_finish(&ctx);
1034    return 0;
1035}
1036
1037static int tg_set_conf_u64(struct cgroup *cgrp, struct cftype *cft,
1038               const char *buf)
1039{
1040    return tg_set_conf(cgrp, cft, buf, true);
1041}
1042
1043static int tg_set_conf_uint(struct cgroup *cgrp, struct cftype *cft,
1044                const char *buf)
1045{
1046    return tg_set_conf(cgrp, cft, buf, false);
1047}
1048
1049static struct cftype throtl_files[] = {
1050    {
1051        .name = "throttle.read_bps_device",
1052        .private = offsetof(struct throtl_grp, bps[READ]),
1053        .read_seq_string = tg_print_conf_u64,
1054        .write_string = tg_set_conf_u64,
1055        .max_write_len = 256,
1056    },
1057    {
1058        .name = "throttle.write_bps_device",
1059        .private = offsetof(struct throtl_grp, bps[WRITE]),
1060        .read_seq_string = tg_print_conf_u64,
1061        .write_string = tg_set_conf_u64,
1062        .max_write_len = 256,
1063    },
1064    {
1065        .name = "throttle.read_iops_device",
1066        .private = offsetof(struct throtl_grp, iops[READ]),
1067        .read_seq_string = tg_print_conf_uint,
1068        .write_string = tg_set_conf_uint,
1069        .max_write_len = 256,
1070    },
1071    {
1072        .name = "throttle.write_iops_device",
1073        .private = offsetof(struct throtl_grp, iops[WRITE]),
1074        .read_seq_string = tg_print_conf_uint,
1075        .write_string = tg_set_conf_uint,
1076        .max_write_len = 256,
1077    },
1078    {
1079        .name = "throttle.io_service_bytes",
1080        .private = offsetof(struct tg_stats_cpu, service_bytes),
1081        .read_seq_string = tg_print_cpu_rwstat,
1082    },
1083    {
1084        .name = "throttle.io_serviced",
1085        .private = offsetof(struct tg_stats_cpu, serviced),
1086        .read_seq_string = tg_print_cpu_rwstat,
1087    },
1088    { } /* terminate */
1089};
1090
1091static void throtl_shutdown_wq(struct request_queue *q)
1092{
1093    struct throtl_data *td = q->td;
1094
1095    cancel_delayed_work_sync(&td->throtl_work);
1096}
1097
1098static struct blkcg_policy blkcg_policy_throtl = {
1099    .pd_size = sizeof(struct throtl_grp),
1100    .cftypes = throtl_files,
1101
1102    .pd_init_fn = throtl_pd_init,
1103    .pd_exit_fn = throtl_pd_exit,
1104    .pd_reset_stats_fn = throtl_pd_reset_stats,
1105};
1106
1107bool blk_throtl_bio(struct request_queue *q, struct bio *bio)
1108{
1109    struct throtl_data *td = q->td;
1110    struct throtl_grp *tg;
1111    bool rw = bio_data_dir(bio), update_disptime = true;
1112    struct blkcg *blkcg;
1113    bool throttled = false;
1114
1115    if (bio->bi_rw & REQ_THROTTLED) {
1116        bio->bi_rw &= ~REQ_THROTTLED;
1117        goto out;
1118    }
1119
1120    /*
1121     * A throtl_grp pointer retrieved under rcu can be used to access
1122     * basic fields like stats and io rates. If a group has no rules,
1123     * just update the dispatch stats in lockless manner and return.
1124     */
1125    rcu_read_lock();
1126    blkcg = bio_blkcg(bio);
1127    tg = throtl_lookup_tg(td, blkcg);
1128    if (tg) {
1129        if (tg_no_rule_group(tg, rw)) {
1130            throtl_update_dispatch_stats(tg_to_blkg(tg),
1131                             bio->bi_size, bio->bi_rw);
1132            goto out_unlock_rcu;
1133        }
1134    }
1135
1136    /*
1137     * Either group has not been allocated yet or it is not an unlimited
1138     * IO group
1139     */
1140    spin_lock_irq(q->queue_lock);
1141    tg = throtl_lookup_create_tg(td, blkcg);
1142    if (unlikely(!tg))
1143        goto out_unlock;
1144
1145    if (tg->nr_queued[rw]) {
1146        /*
1147         * There is already another bio queued in same dir. No
1148         * need to update dispatch time.
1149         */
1150        update_disptime = false;
1151        goto queue_bio;
1152
1153    }
1154
1155    /* Bio is with-in rate limit of group */
1156    if (tg_may_dispatch(td, tg, bio, NULL)) {
1157        throtl_charge_bio(tg, bio);
1158
1159        /*
1160         * We need to trim slice even when bios are not being queued
1161         * otherwise it might happen that a bio is not queued for
1162         * a long time and slice keeps on extending and trim is not
1163         * called for a long time. Now if limits are reduced suddenly
1164         * we take into account all the IO dispatched so far at new
1165         * low rate and * newly queued IO gets a really long dispatch
1166         * time.
1167         *
1168         * So keep on trimming slice even if bio is not queued.
1169         */
1170        throtl_trim_slice(td, tg, rw);
1171        goto out_unlock;
1172    }
1173
1174queue_bio:
1175    throtl_log_tg(td, tg, "[%c] bio. bdisp=%llu sz=%u bps=%llu"
1176            " iodisp=%u iops=%u queued=%d/%d",
1177            rw == READ ? 'R' : 'W',
1178            tg->bytes_disp[rw], bio->bi_size, tg->bps[rw],
1179            tg->io_disp[rw], tg->iops[rw],
1180            tg->nr_queued[READ], tg->nr_queued[WRITE]);
1181
1182    bio_associate_current(bio);
1183    throtl_add_bio_tg(q->td, tg, bio);
1184    throttled = true;
1185
1186    if (update_disptime) {
1187        tg_update_disptime(td, tg);
1188        throtl_schedule_next_dispatch(td);
1189    }
1190
1191out_unlock:
1192    spin_unlock_irq(q->queue_lock);
1193out_unlock_rcu:
1194    rcu_read_unlock();
1195out:
1196    return throttled;
1197}
1198
1199/**
1200 * blk_throtl_drain - drain throttled bios
1201 * @q: request_queue to drain throttled bios for
1202 *
1203 * Dispatch all currently throttled bios on @q through ->make_request_fn().
1204 */
1205void blk_throtl_drain(struct request_queue *q)
1206    __releases(q->queue_lock) __acquires(q->queue_lock)
1207{
1208    struct throtl_data *td = q->td;
1209    struct throtl_rb_root *st = &td->tg_service_tree;
1210    struct throtl_grp *tg;
1211    struct bio_list bl;
1212    struct bio *bio;
1213
1214    queue_lockdep_assert_held(q);
1215
1216    bio_list_init(&bl);
1217
1218    while ((tg = throtl_rb_first(st))) {
1219        throtl_dequeue_tg(td, tg);
1220
1221        while ((bio = bio_list_peek(&tg->bio_lists[READ])))
1222            tg_dispatch_one_bio(td, tg, bio_data_dir(bio), &bl);
1223        while ((bio = bio_list_peek(&tg->bio_lists[WRITE])))
1224            tg_dispatch_one_bio(td, tg, bio_data_dir(bio), &bl);
1225    }
1226    spin_unlock_irq(q->queue_lock);
1227
1228    while ((bio = bio_list_pop(&bl)))
1229        generic_make_request(bio);
1230
1231    spin_lock_irq(q->queue_lock);
1232}
1233
1234int blk_throtl_init(struct request_queue *q)
1235{
1236    struct throtl_data *td;
1237    int ret;
1238
1239    td = kzalloc_node(sizeof(*td), GFP_KERNEL, q->node);
1240    if (!td)
1241        return -ENOMEM;
1242
1243    td->tg_service_tree = THROTL_RB_ROOT;
1244    td->limits_changed = false;
1245    INIT_DELAYED_WORK(&td->throtl_work, blk_throtl_work);
1246
1247    q->td = td;
1248    td->queue = q;
1249
1250    /* activate policy */
1251    ret = blkcg_activate_policy(q, &blkcg_policy_throtl);
1252    if (ret)
1253        kfree(td);
1254    return ret;
1255}
1256
1257void blk_throtl_exit(struct request_queue *q)
1258{
1259    BUG_ON(!q->td);
1260    throtl_shutdown_wq(q);
1261    blkcg_deactivate_policy(q, &blkcg_policy_throtl);
1262    kfree(q->td);
1263}
1264
1265static int __init throtl_init(void)
1266{
1267    kthrotld_workqueue = alloc_workqueue("kthrotld", WQ_MEM_RECLAIM, 0);
1268    if (!kthrotld_workqueue)
1269        panic("Failed to create kthrotld\n");
1270
1271    return blkcg_policy_register(&blkcg_policy_throtl);
1272}
1273
1274module_init(throtl_init);
1275

Archive Download this file

Branches:
ben-wpan
ben-wpan-stefan
javiroman/ks7010
jz-2.6.34
jz-2.6.34-rc5
jz-2.6.34-rc6
jz-2.6.34-rc7
jz-2.6.35
jz-2.6.36
jz-2.6.37
jz-2.6.38
jz-2.6.39
jz-3.0
jz-3.1
jz-3.11
jz-3.12
jz-3.13
jz-3.15
jz-3.16
jz-3.18-dt
jz-3.2
jz-3.3
jz-3.4
jz-3.5
jz-3.6
jz-3.6-rc2-pwm
jz-3.9
jz-3.9-clk
jz-3.9-rc8
jz47xx
jz47xx-2.6.38
master

Tags:
od-2011-09-04
od-2011-09-18
v2.6.34-rc5
v2.6.34-rc6
v2.6.34-rc7
v3.9



interactive