Merge tag 'ext4_for_linus-6.5-rc3' of git://git.kernel.org/pub/scm/linux/kernel/git...
[platform/kernel/linux-rpi.git] / block / blk-iocost.c
1 /* SPDX-License-Identifier: GPL-2.0
2  *
3  * IO cost model based controller.
4  *
5  * Copyright (C) 2019 Tejun Heo <tj@kernel.org>
6  * Copyright (C) 2019 Andy Newell <newella@fb.com>
7  * Copyright (C) 2019 Facebook
8  *
9  * One challenge of controlling IO resources is the lack of trivially
10  * observable cost metric.  This is distinguished from CPU and memory where
11  * wallclock time and the number of bytes can serve as accurate enough
12  * approximations.
13  *
14  * Bandwidth and iops are the most commonly used metrics for IO devices but
15  * depending on the type and specifics of the device, different IO patterns
16  * easily lead to multiple orders of magnitude variations rendering them
17  * useless for the purpose of IO capacity distribution.  While on-device
18  * time, with a lot of clutches, could serve as a useful approximation for
19  * non-queued rotational devices, this is no longer viable with modern
20  * devices, even the rotational ones.
21  *
22  * While there is no cost metric we can trivially observe, it isn't a
23  * complete mystery.  For example, on a rotational device, seek cost
24  * dominates while a contiguous transfer contributes a smaller amount
25  * proportional to the size.  If we can characterize at least the relative
26  * costs of these different types of IOs, it should be possible to
27  * implement a reasonable work-conserving proportional IO resource
28  * distribution.
29  *
30  * 1. IO Cost Model
31  *
32  * IO cost model estimates the cost of an IO given its basic parameters and
33  * history (e.g. the end sector of the last IO).  The cost is measured in
34  * device time.  If a given IO is estimated to cost 10ms, the device should
35  * be able to process ~100 of those IOs in a second.
36  *
37  * Currently, there's only one builtin cost model - linear.  Each IO is
38  * classified as sequential or random and given a base cost accordingly.
39  * On top of that, a size cost proportional to the length of the IO is
40  * added.  While simple, this model captures the operational
41  * characteristics of a wide varienty of devices well enough.  Default
42  * parameters for several different classes of devices are provided and the
43  * parameters can be configured from userspace via
44  * /sys/fs/cgroup/io.cost.model.
45  *
46  * If needed, tools/cgroup/iocost_coef_gen.py can be used to generate
47  * device-specific coefficients.
48  *
49  * 2. Control Strategy
50  *
51  * The device virtual time (vtime) is used as the primary control metric.
52  * The control strategy is composed of the following three parts.
53  *
54  * 2-1. Vtime Distribution
55  *
56  * When a cgroup becomes active in terms of IOs, its hierarchical share is
57  * calculated.  Please consider the following hierarchy where the numbers
58  * inside parentheses denote the configured weights.
59  *
60  *           root
61  *         /       \
62  *      A (w:100)  B (w:300)
63  *      /       \
64  *  A0 (w:100)  A1 (w:100)
65  *
66  * If B is idle and only A0 and A1 are actively issuing IOs, as the two are
67  * of equal weight, each gets 50% share.  If then B starts issuing IOs, B
68  * gets 300/(100+300) or 75% share, and A0 and A1 equally splits the rest,
69  * 12.5% each.  The distribution mechanism only cares about these flattened
70  * shares.  They're called hweights (hierarchical weights) and always add
71  * upto 1 (WEIGHT_ONE).
72  *
73  * A given cgroup's vtime runs slower in inverse proportion to its hweight.
74  * For example, with 12.5% weight, A0's time runs 8 times slower (100/12.5)
75  * against the device vtime - an IO which takes 10ms on the underlying
76  * device is considered to take 80ms on A0.
77  *
78  * This constitutes the basis of IO capacity distribution.  Each cgroup's
79  * vtime is running at a rate determined by its hweight.  A cgroup tracks
80  * the vtime consumed by past IOs and can issue a new IO if doing so
81  * wouldn't outrun the current device vtime.  Otherwise, the IO is
82  * suspended until the vtime has progressed enough to cover it.
83  *
84  * 2-2. Vrate Adjustment
85  *
86  * It's unrealistic to expect the cost model to be perfect.  There are too
87  * many devices and even on the same device the overall performance
88  * fluctuates depending on numerous factors such as IO mixture and device
89  * internal garbage collection.  The controller needs to adapt dynamically.
90  *
91  * This is achieved by adjusting the overall IO rate according to how busy
92  * the device is.  If the device becomes overloaded, we're sending down too
93  * many IOs and should generally slow down.  If there are waiting issuers
94  * but the device isn't saturated, we're issuing too few and should
95  * generally speed up.
96  *
97  * To slow down, we lower the vrate - the rate at which the device vtime
98  * passes compared to the wall clock.  For example, if the vtime is running
99  * at the vrate of 75%, all cgroups added up would only be able to issue
100  * 750ms worth of IOs per second, and vice-versa for speeding up.
101  *
102  * Device business is determined using two criteria - rq wait and
103  * completion latencies.
104  *
105  * When a device gets saturated, the on-device and then the request queues
106  * fill up and a bio which is ready to be issued has to wait for a request
107  * to become available.  When this delay becomes noticeable, it's a clear
108  * indication that the device is saturated and we lower the vrate.  This
109  * saturation signal is fairly conservative as it only triggers when both
110  * hardware and software queues are filled up, and is used as the default
111  * busy signal.
112  *
113  * As devices can have deep queues and be unfair in how the queued commands
114  * are executed, solely depending on rq wait may not result in satisfactory
115  * control quality.  For a better control quality, completion latency QoS
116  * parameters can be configured so that the device is considered saturated
117  * if N'th percentile completion latency rises above the set point.
118  *
119  * The completion latency requirements are a function of both the
120  * underlying device characteristics and the desired IO latency quality of
121  * service.  There is an inherent trade-off - the tighter the latency QoS,
122  * the higher the bandwidth lossage.  Latency QoS is disabled by default
123  * and can be set through /sys/fs/cgroup/io.cost.qos.
124  *
125  * 2-3. Work Conservation
126  *
127  * Imagine two cgroups A and B with equal weights.  A is issuing a small IO
128  * periodically while B is sending out enough parallel IOs to saturate the
129  * device on its own.  Let's say A's usage amounts to 100ms worth of IO
130  * cost per second, i.e., 10% of the device capacity.  The naive
131  * distribution of half and half would lead to 60% utilization of the
132  * device, a significant reduction in the total amount of work done
133  * compared to free-for-all competition.  This is too high a cost to pay
134  * for IO control.
135  *
136  * To conserve the total amount of work done, we keep track of how much
137  * each active cgroup is actually using and yield part of its weight if
138  * there are other cgroups which can make use of it.  In the above case,
139  * A's weight will be lowered so that it hovers above the actual usage and
140  * B would be able to use the rest.
141  *
142  * As we don't want to penalize a cgroup for donating its weight, the
143  * surplus weight adjustment factors in a margin and has an immediate
144  * snapback mechanism in case the cgroup needs more IO vtime for itself.
145  *
146  * Note that adjusting down surplus weights has the same effects as
147  * accelerating vtime for other cgroups and work conservation can also be
148  * implemented by adjusting vrate dynamically.  However, squaring who can
149  * donate and should take back how much requires hweight propagations
150  * anyway making it easier to implement and understand as a separate
151  * mechanism.
152  *
153  * 3. Monitoring
154  *
155  * Instead of debugfs or other clumsy monitoring mechanisms, this
156  * controller uses a drgn based monitoring script -
157  * tools/cgroup/iocost_monitor.py.  For details on drgn, please see
158  * https://github.com/osandov/drgn.  The output looks like the following.
159  *
160  *  sdb RUN   per=300ms cur_per=234.218:v203.695 busy= +1 vrate= 62.12%
161  *                 active      weight      hweight% inflt% dbt  delay usages%
162  *  test/a              *    50/   50  33.33/ 33.33  27.65   2  0*041 033:033:033
163  *  test/b              *   100/  100  66.67/ 66.67  17.56   0  0*000 066:079:077
164  *
165  * - per        : Timer period
166  * - cur_per    : Internal wall and device vtime clock
167  * - vrate      : Device virtual time rate against wall clock
168  * - weight     : Surplus-adjusted and configured weights
169  * - hweight    : Surplus-adjusted and configured hierarchical weights
170  * - inflt      : The percentage of in-flight IO cost at the end of last period
171  * - del_ms     : Deferred issuer delay induction level and duration
172  * - usages     : Usage history
173  */
174
175 #include <linux/kernel.h>
176 #include <linux/module.h>
177 #include <linux/timer.h>
178 #include <linux/time64.h>
179 #include <linux/parser.h>
180 #include <linux/sched/signal.h>
181 #include <asm/local.h>
182 #include <asm/local64.h>
183 #include "blk-rq-qos.h"
184 #include "blk-stat.h"
185 #include "blk-wbt.h"
186 #include "blk-cgroup.h"
187
188 #ifdef CONFIG_TRACEPOINTS
189
190 /* copied from TRACE_CGROUP_PATH, see cgroup-internal.h */
191 #define TRACE_IOCG_PATH_LEN 1024
192 static DEFINE_SPINLOCK(trace_iocg_path_lock);
193 static char trace_iocg_path[TRACE_IOCG_PATH_LEN];
194
195 #define TRACE_IOCG_PATH(type, iocg, ...)                                        \
196         do {                                                                    \
197                 unsigned long flags;                                            \
198                 if (trace_iocost_##type##_enabled()) {                          \
199                         spin_lock_irqsave(&trace_iocg_path_lock, flags);        \
200                         cgroup_path(iocg_to_blkg(iocg)->blkcg->css.cgroup,      \
201                                     trace_iocg_path, TRACE_IOCG_PATH_LEN);      \
202                         trace_iocost_##type(iocg, trace_iocg_path,              \
203                                               ##__VA_ARGS__);                   \
204                         spin_unlock_irqrestore(&trace_iocg_path_lock, flags);   \
205                 }                                                               \
206         } while (0)
207
208 #else   /* CONFIG_TRACE_POINTS */
209 #define TRACE_IOCG_PATH(type, iocg, ...)        do { } while (0)
210 #endif  /* CONFIG_TRACE_POINTS */
211
212 enum {
213         MILLION                 = 1000000,
214
215         /* timer period is calculated from latency requirements, bound it */
216         MIN_PERIOD              = USEC_PER_MSEC,
217         MAX_PERIOD              = USEC_PER_SEC,
218
219         /*
220          * iocg->vtime is targeted at 50% behind the device vtime, which
221          * serves as its IO credit buffer.  Surplus weight adjustment is
222          * immediately canceled if the vtime margin runs below 10%.
223          */
224         MARGIN_MIN_PCT          = 10,
225         MARGIN_LOW_PCT          = 20,
226         MARGIN_TARGET_PCT       = 50,
227
228         INUSE_ADJ_STEP_PCT      = 25,
229
230         /* Have some play in timer operations */
231         TIMER_SLACK_PCT         = 1,
232
233         /* 1/64k is granular enough and can easily be handled w/ u32 */
234         WEIGHT_ONE              = 1 << 16,
235 };
236
237 enum {
238         /*
239          * As vtime is used to calculate the cost of each IO, it needs to
240          * be fairly high precision.  For example, it should be able to
241          * represent the cost of a single page worth of discard with
242          * suffificient accuracy.  At the same time, it should be able to
243          * represent reasonably long enough durations to be useful and
244          * convenient during operation.
245          *
246          * 1s worth of vtime is 2^37.  This gives us both sub-nanosecond
247          * granularity and days of wrap-around time even at extreme vrates.
248          */
249         VTIME_PER_SEC_SHIFT     = 37,
250         VTIME_PER_SEC           = 1LLU << VTIME_PER_SEC_SHIFT,
251         VTIME_PER_USEC          = VTIME_PER_SEC / USEC_PER_SEC,
252         VTIME_PER_NSEC          = VTIME_PER_SEC / NSEC_PER_SEC,
253
254         /* bound vrate adjustments within two orders of magnitude */
255         VRATE_MIN_PPM           = 10000,        /* 1% */
256         VRATE_MAX_PPM           = 100000000,    /* 10000% */
257
258         VRATE_MIN               = VTIME_PER_USEC * VRATE_MIN_PPM / MILLION,
259         VRATE_CLAMP_ADJ_PCT     = 4,
260
261         /* switch iff the conditions are met for longer than this */
262         AUTOP_CYCLE_NSEC        = 10LLU * NSEC_PER_SEC,
263 };
264
265 enum {
266         /* if IOs end up waiting for requests, issue less */
267         RQ_WAIT_BUSY_PCT        = 5,
268
269         /* unbusy hysterisis */
270         UNBUSY_THR_PCT          = 75,
271
272         /*
273          * The effect of delay is indirect and non-linear and a huge amount of
274          * future debt can accumulate abruptly while unthrottled. Linearly scale
275          * up delay as debt is going up and then let it decay exponentially.
276          * This gives us quick ramp ups while delay is accumulating and long
277          * tails which can help reducing the frequency of debt explosions on
278          * unthrottle. The parameters are experimentally determined.
279          *
280          * The delay mechanism provides adequate protection and behavior in many
281          * cases. However, this is far from ideal and falls shorts on both
282          * fronts. The debtors are often throttled too harshly costing a
283          * significant level of fairness and possibly total work while the
284          * protection against their impacts on the system can be choppy and
285          * unreliable.
286          *
287          * The shortcoming primarily stems from the fact that, unlike for page
288          * cache, the kernel doesn't have well-defined back-pressure propagation
289          * mechanism and policies for anonymous memory. Fully addressing this
290          * issue will likely require substantial improvements in the area.
291          */
292         MIN_DELAY_THR_PCT       = 500,
293         MAX_DELAY_THR_PCT       = 25000,
294         MIN_DELAY               = 250,
295         MAX_DELAY               = 250 * USEC_PER_MSEC,
296
297         /* halve debts if avg usage over 100ms is under 50% */
298         DFGV_USAGE_PCT          = 50,
299         DFGV_PERIOD             = 100 * USEC_PER_MSEC,
300
301         /* don't let cmds which take a very long time pin lagging for too long */
302         MAX_LAGGING_PERIODS     = 10,
303
304         /*
305          * Count IO size in 4k pages.  The 12bit shift helps keeping
306          * size-proportional components of cost calculation in closer
307          * numbers of digits to per-IO cost components.
308          */
309         IOC_PAGE_SHIFT          = 12,
310         IOC_PAGE_SIZE           = 1 << IOC_PAGE_SHIFT,
311         IOC_SECT_TO_PAGE_SHIFT  = IOC_PAGE_SHIFT - SECTOR_SHIFT,
312
313         /* if apart further than 16M, consider randio for linear model */
314         LCOEF_RANDIO_PAGES      = 4096,
315 };
316
317 enum ioc_running {
318         IOC_IDLE,
319         IOC_RUNNING,
320         IOC_STOP,
321 };
322
323 /* io.cost.qos controls including per-dev enable of the whole controller */
324 enum {
325         QOS_ENABLE,
326         QOS_CTRL,
327         NR_QOS_CTRL_PARAMS,
328 };
329
330 /* io.cost.qos params */
331 enum {
332         QOS_RPPM,
333         QOS_RLAT,
334         QOS_WPPM,
335         QOS_WLAT,
336         QOS_MIN,
337         QOS_MAX,
338         NR_QOS_PARAMS,
339 };
340
341 /* io.cost.model controls */
342 enum {
343         COST_CTRL,
344         COST_MODEL,
345         NR_COST_CTRL_PARAMS,
346 };
347
348 /* builtin linear cost model coefficients */
349 enum {
350         I_LCOEF_RBPS,
351         I_LCOEF_RSEQIOPS,
352         I_LCOEF_RRANDIOPS,
353         I_LCOEF_WBPS,
354         I_LCOEF_WSEQIOPS,
355         I_LCOEF_WRANDIOPS,
356         NR_I_LCOEFS,
357 };
358
359 enum {
360         LCOEF_RPAGE,
361         LCOEF_RSEQIO,
362         LCOEF_RRANDIO,
363         LCOEF_WPAGE,
364         LCOEF_WSEQIO,
365         LCOEF_WRANDIO,
366         NR_LCOEFS,
367 };
368
369 enum {
370         AUTOP_INVALID,
371         AUTOP_HDD,
372         AUTOP_SSD_QD1,
373         AUTOP_SSD_DFL,
374         AUTOP_SSD_FAST,
375 };
376
377 struct ioc_params {
378         u32                             qos[NR_QOS_PARAMS];
379         u64                             i_lcoefs[NR_I_LCOEFS];
380         u64                             lcoefs[NR_LCOEFS];
381         u32                             too_fast_vrate_pct;
382         u32                             too_slow_vrate_pct;
383 };
384
385 struct ioc_margins {
386         s64                             min;
387         s64                             low;
388         s64                             target;
389 };
390
391 struct ioc_missed {
392         local_t                         nr_met;
393         local_t                         nr_missed;
394         u32                             last_met;
395         u32                             last_missed;
396 };
397
398 struct ioc_pcpu_stat {
399         struct ioc_missed               missed[2];
400
401         local64_t                       rq_wait_ns;
402         u64                             last_rq_wait_ns;
403 };
404
405 /* per device */
406 struct ioc {
407         struct rq_qos                   rqos;
408
409         bool                            enabled;
410
411         struct ioc_params               params;
412         struct ioc_margins              margins;
413         u32                             period_us;
414         u32                             timer_slack_ns;
415         u64                             vrate_min;
416         u64                             vrate_max;
417
418         spinlock_t                      lock;
419         struct timer_list               timer;
420         struct list_head                active_iocgs;   /* active cgroups */
421         struct ioc_pcpu_stat __percpu   *pcpu_stat;
422
423         enum ioc_running                running;
424         atomic64_t                      vtime_rate;
425         u64                             vtime_base_rate;
426         s64                             vtime_err;
427
428         seqcount_spinlock_t             period_seqcount;
429         u64                             period_at;      /* wallclock starttime */
430         u64                             period_at_vtime; /* vtime starttime */
431
432         atomic64_t                      cur_period;     /* inc'd each period */
433         int                             busy_level;     /* saturation history */
434
435         bool                            weights_updated;
436         atomic_t                        hweight_gen;    /* for lazy hweights */
437
438         /* debt forgivness */
439         u64                             dfgv_period_at;
440         u64                             dfgv_period_rem;
441         u64                             dfgv_usage_us_sum;
442
443         u64                             autop_too_fast_at;
444         u64                             autop_too_slow_at;
445         int                             autop_idx;
446         bool                            user_qos_params:1;
447         bool                            user_cost_model:1;
448 };
449
450 struct iocg_pcpu_stat {
451         local64_t                       abs_vusage;
452 };
453
454 struct iocg_stat {
455         u64                             usage_us;
456         u64                             wait_us;
457         u64                             indebt_us;
458         u64                             indelay_us;
459 };
460
461 /* per device-cgroup pair */
462 struct ioc_gq {
463         struct blkg_policy_data         pd;
464         struct ioc                      *ioc;
465
466         /*
467          * A iocg can get its weight from two sources - an explicit
468          * per-device-cgroup configuration or the default weight of the
469          * cgroup.  `cfg_weight` is the explicit per-device-cgroup
470          * configuration.  `weight` is the effective considering both
471          * sources.
472          *
473          * When an idle cgroup becomes active its `active` goes from 0 to
474          * `weight`.  `inuse` is the surplus adjusted active weight.
475          * `active` and `inuse` are used to calculate `hweight_active` and
476          * `hweight_inuse`.
477          *
478          * `last_inuse` remembers `inuse` while an iocg is idle to persist
479          * surplus adjustments.
480          *
481          * `inuse` may be adjusted dynamically during period. `saved_*` are used
482          * to determine and track adjustments.
483          */
484         u32                             cfg_weight;
485         u32                             weight;
486         u32                             active;
487         u32                             inuse;
488
489         u32                             last_inuse;
490         s64                             saved_margin;
491
492         sector_t                        cursor;         /* to detect randio */
493
494         /*
495          * `vtime` is this iocg's vtime cursor which progresses as IOs are
496          * issued.  If lagging behind device vtime, the delta represents
497          * the currently available IO budget.  If running ahead, the
498          * overage.
499          *
500          * `vtime_done` is the same but progressed on completion rather
501          * than issue.  The delta behind `vtime` represents the cost of
502          * currently in-flight IOs.
503          */
504         atomic64_t                      vtime;
505         atomic64_t                      done_vtime;
506         u64                             abs_vdebt;
507
508         /* current delay in effect and when it started */
509         u64                             delay;
510         u64                             delay_at;
511
512         /*
513          * The period this iocg was last active in.  Used for deactivation
514          * and invalidating `vtime`.
515          */
516         atomic64_t                      active_period;
517         struct list_head                active_list;
518
519         /* see __propagate_weights() and current_hweight() for details */
520         u64                             child_active_sum;
521         u64                             child_inuse_sum;
522         u64                             child_adjusted_sum;
523         int                             hweight_gen;
524         u32                             hweight_active;
525         u32                             hweight_inuse;
526         u32                             hweight_donating;
527         u32                             hweight_after_donation;
528
529         struct list_head                walk_list;
530         struct list_head                surplus_list;
531
532         struct wait_queue_head          waitq;
533         struct hrtimer                  waitq_timer;
534
535         /* timestamp at the latest activation */
536         u64                             activated_at;
537
538         /* statistics */
539         struct iocg_pcpu_stat __percpu  *pcpu_stat;
540         struct iocg_stat                stat;
541         struct iocg_stat                last_stat;
542         u64                             last_stat_abs_vusage;
543         u64                             usage_delta_us;
544         u64                             wait_since;
545         u64                             indebt_since;
546         u64                             indelay_since;
547
548         /* this iocg's depth in the hierarchy and ancestors including self */
549         int                             level;
550         struct ioc_gq                   *ancestors[];
551 };
552
553 /* per cgroup */
554 struct ioc_cgrp {
555         struct blkcg_policy_data        cpd;
556         unsigned int                    dfl_weight;
557 };
558
559 struct ioc_now {
560         u64                             now_ns;
561         u64                             now;
562         u64                             vnow;
563 };
564
565 struct iocg_wait {
566         struct wait_queue_entry         wait;
567         struct bio                      *bio;
568         u64                             abs_cost;
569         bool                            committed;
570 };
571
572 struct iocg_wake_ctx {
573         struct ioc_gq                   *iocg;
574         u32                             hw_inuse;
575         s64                             vbudget;
576 };
577
578 static const struct ioc_params autop[] = {
579         [AUTOP_HDD] = {
580                 .qos                            = {
581                         [QOS_RLAT]              =        250000, /* 250ms */
582                         [QOS_WLAT]              =        250000,
583                         [QOS_MIN]               = VRATE_MIN_PPM,
584                         [QOS_MAX]               = VRATE_MAX_PPM,
585                 },
586                 .i_lcoefs                       = {
587                         [I_LCOEF_RBPS]          =     174019176,
588                         [I_LCOEF_RSEQIOPS]      =         41708,
589                         [I_LCOEF_RRANDIOPS]     =           370,
590                         [I_LCOEF_WBPS]          =     178075866,
591                         [I_LCOEF_WSEQIOPS]      =         42705,
592                         [I_LCOEF_WRANDIOPS]     =           378,
593                 },
594         },
595         [AUTOP_SSD_QD1] = {
596                 .qos                            = {
597                         [QOS_RLAT]              =         25000, /* 25ms */
598                         [QOS_WLAT]              =         25000,
599                         [QOS_MIN]               = VRATE_MIN_PPM,
600                         [QOS_MAX]               = VRATE_MAX_PPM,
601                 },
602                 .i_lcoefs                       = {
603                         [I_LCOEF_RBPS]          =     245855193,
604                         [I_LCOEF_RSEQIOPS]      =         61575,
605                         [I_LCOEF_RRANDIOPS]     =          6946,
606                         [I_LCOEF_WBPS]          =     141365009,
607                         [I_LCOEF_WSEQIOPS]      =         33716,
608                         [I_LCOEF_WRANDIOPS]     =         26796,
609                 },
610         },
611         [AUTOP_SSD_DFL] = {
612                 .qos                            = {
613                         [QOS_RLAT]              =         25000, /* 25ms */
614                         [QOS_WLAT]              =         25000,
615                         [QOS_MIN]               = VRATE_MIN_PPM,
616                         [QOS_MAX]               = VRATE_MAX_PPM,
617                 },
618                 .i_lcoefs                       = {
619                         [I_LCOEF_RBPS]          =     488636629,
620                         [I_LCOEF_RSEQIOPS]      =          8932,
621                         [I_LCOEF_RRANDIOPS]     =          8518,
622                         [I_LCOEF_WBPS]          =     427891549,
623                         [I_LCOEF_WSEQIOPS]      =         28755,
624                         [I_LCOEF_WRANDIOPS]     =         21940,
625                 },
626                 .too_fast_vrate_pct             =           500,
627         },
628         [AUTOP_SSD_FAST] = {
629                 .qos                            = {
630                         [QOS_RLAT]              =          5000, /* 5ms */
631                         [QOS_WLAT]              =          5000,
632                         [QOS_MIN]               = VRATE_MIN_PPM,
633                         [QOS_MAX]               = VRATE_MAX_PPM,
634                 },
635                 .i_lcoefs                       = {
636                         [I_LCOEF_RBPS]          =    3102524156LLU,
637                         [I_LCOEF_RSEQIOPS]      =        724816,
638                         [I_LCOEF_RRANDIOPS]     =        778122,
639                         [I_LCOEF_WBPS]          =    1742780862LLU,
640                         [I_LCOEF_WSEQIOPS]      =        425702,
641                         [I_LCOEF_WRANDIOPS]     =        443193,
642                 },
643                 .too_slow_vrate_pct             =            10,
644         },
645 };
646
647 /*
648  * vrate adjust percentages indexed by ioc->busy_level.  We adjust up on
649  * vtime credit shortage and down on device saturation.
650  */
651 static u32 vrate_adj_pct[] =
652         { 0, 0, 0, 0,
653           1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
654           2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
655           4, 4, 4, 4, 4, 4, 4, 4, 8, 8, 8, 8, 8, 8, 8, 8, 16 };
656
657 static struct blkcg_policy blkcg_policy_iocost;
658
659 /* accessors and helpers */
660 static struct ioc *rqos_to_ioc(struct rq_qos *rqos)
661 {
662         return container_of(rqos, struct ioc, rqos);
663 }
664
665 static struct ioc *q_to_ioc(struct request_queue *q)
666 {
667         return rqos_to_ioc(rq_qos_id(q, RQ_QOS_COST));
668 }
669
670 static const char __maybe_unused *ioc_name(struct ioc *ioc)
671 {
672         struct gendisk *disk = ioc->rqos.disk;
673
674         if (!disk)
675                 return "<unknown>";
676         return disk->disk_name;
677 }
678
679 static struct ioc_gq *pd_to_iocg(struct blkg_policy_data *pd)
680 {
681         return pd ? container_of(pd, struct ioc_gq, pd) : NULL;
682 }
683
684 static struct ioc_gq *blkg_to_iocg(struct blkcg_gq *blkg)
685 {
686         return pd_to_iocg(blkg_to_pd(blkg, &blkcg_policy_iocost));
687 }
688
689 static struct blkcg_gq *iocg_to_blkg(struct ioc_gq *iocg)
690 {
691         return pd_to_blkg(&iocg->pd);
692 }
693
694 static struct ioc_cgrp *blkcg_to_iocc(struct blkcg *blkcg)
695 {
696         return container_of(blkcg_to_cpd(blkcg, &blkcg_policy_iocost),
697                             struct ioc_cgrp, cpd);
698 }
699
700 /*
701  * Scale @abs_cost to the inverse of @hw_inuse.  The lower the hierarchical
702  * weight, the more expensive each IO.  Must round up.
703  */
704 static u64 abs_cost_to_cost(u64 abs_cost, u32 hw_inuse)
705 {
706         return DIV64_U64_ROUND_UP(abs_cost * WEIGHT_ONE, hw_inuse);
707 }
708
709 /*
710  * The inverse of abs_cost_to_cost().  Must round up.
711  */
712 static u64 cost_to_abs_cost(u64 cost, u32 hw_inuse)
713 {
714         return DIV64_U64_ROUND_UP(cost * hw_inuse, WEIGHT_ONE);
715 }
716
717 static void iocg_commit_bio(struct ioc_gq *iocg, struct bio *bio,
718                             u64 abs_cost, u64 cost)
719 {
720         struct iocg_pcpu_stat *gcs;
721
722         bio->bi_iocost_cost = cost;
723         atomic64_add(cost, &iocg->vtime);
724
725         gcs = get_cpu_ptr(iocg->pcpu_stat);
726         local64_add(abs_cost, &gcs->abs_vusage);
727         put_cpu_ptr(gcs);
728 }
729
730 static void iocg_lock(struct ioc_gq *iocg, bool lock_ioc, unsigned long *flags)
731 {
732         if (lock_ioc) {
733                 spin_lock_irqsave(&iocg->ioc->lock, *flags);
734                 spin_lock(&iocg->waitq.lock);
735         } else {
736                 spin_lock_irqsave(&iocg->waitq.lock, *flags);
737         }
738 }
739
740 static void iocg_unlock(struct ioc_gq *iocg, bool unlock_ioc, unsigned long *flags)
741 {
742         if (unlock_ioc) {
743                 spin_unlock(&iocg->waitq.lock);
744                 spin_unlock_irqrestore(&iocg->ioc->lock, *flags);
745         } else {
746                 spin_unlock_irqrestore(&iocg->waitq.lock, *flags);
747         }
748 }
749
750 #define CREATE_TRACE_POINTS
751 #include <trace/events/iocost.h>
752
753 static void ioc_refresh_margins(struct ioc *ioc)
754 {
755         struct ioc_margins *margins = &ioc->margins;
756         u32 period_us = ioc->period_us;
757         u64 vrate = ioc->vtime_base_rate;
758
759         margins->min = (period_us * MARGIN_MIN_PCT / 100) * vrate;
760         margins->low = (period_us * MARGIN_LOW_PCT / 100) * vrate;
761         margins->target = (period_us * MARGIN_TARGET_PCT / 100) * vrate;
762 }
763
764 /* latency Qos params changed, update period_us and all the dependent params */
765 static void ioc_refresh_period_us(struct ioc *ioc)
766 {
767         u32 ppm, lat, multi, period_us;
768
769         lockdep_assert_held(&ioc->lock);
770
771         /* pick the higher latency target */
772         if (ioc->params.qos[QOS_RLAT] >= ioc->params.qos[QOS_WLAT]) {
773                 ppm = ioc->params.qos[QOS_RPPM];
774                 lat = ioc->params.qos[QOS_RLAT];
775         } else {
776                 ppm = ioc->params.qos[QOS_WPPM];
777                 lat = ioc->params.qos[QOS_WLAT];
778         }
779
780         /*
781          * We want the period to be long enough to contain a healthy number
782          * of IOs while short enough for granular control.  Define it as a
783          * multiple of the latency target.  Ideally, the multiplier should
784          * be scaled according to the percentile so that it would nominally
785          * contain a certain number of requests.  Let's be simpler and
786          * scale it linearly so that it's 2x >= pct(90) and 10x at pct(50).
787          */
788         if (ppm)
789                 multi = max_t(u32, (MILLION - ppm) / 50000, 2);
790         else
791                 multi = 2;
792         period_us = multi * lat;
793         period_us = clamp_t(u32, period_us, MIN_PERIOD, MAX_PERIOD);
794
795         /* calculate dependent params */
796         ioc->period_us = period_us;
797         ioc->timer_slack_ns = div64_u64(
798                 (u64)period_us * NSEC_PER_USEC * TIMER_SLACK_PCT,
799                 100);
800         ioc_refresh_margins(ioc);
801 }
802
803 /*
804  *  ioc->rqos.disk isn't initialized when this function is called from
805  *  the init path.
806  */
807 static int ioc_autop_idx(struct ioc *ioc, struct gendisk *disk)
808 {
809         int idx = ioc->autop_idx;
810         const struct ioc_params *p = &autop[idx];
811         u32 vrate_pct;
812         u64 now_ns;
813
814         /* rotational? */
815         if (!blk_queue_nonrot(disk->queue))
816                 return AUTOP_HDD;
817
818         /* handle SATA SSDs w/ broken NCQ */
819         if (blk_queue_depth(disk->queue) == 1)
820                 return AUTOP_SSD_QD1;
821
822         /* use one of the normal ssd sets */
823         if (idx < AUTOP_SSD_DFL)
824                 return AUTOP_SSD_DFL;
825
826         /* if user is overriding anything, maintain what was there */
827         if (ioc->user_qos_params || ioc->user_cost_model)
828                 return idx;
829
830         /* step up/down based on the vrate */
831         vrate_pct = div64_u64(ioc->vtime_base_rate * 100, VTIME_PER_USEC);
832         now_ns = ktime_get_ns();
833
834         if (p->too_fast_vrate_pct && p->too_fast_vrate_pct <= vrate_pct) {
835                 if (!ioc->autop_too_fast_at)
836                         ioc->autop_too_fast_at = now_ns;
837                 if (now_ns - ioc->autop_too_fast_at >= AUTOP_CYCLE_NSEC)
838                         return idx + 1;
839         } else {
840                 ioc->autop_too_fast_at = 0;
841         }
842
843         if (p->too_slow_vrate_pct && p->too_slow_vrate_pct >= vrate_pct) {
844                 if (!ioc->autop_too_slow_at)
845                         ioc->autop_too_slow_at = now_ns;
846                 if (now_ns - ioc->autop_too_slow_at >= AUTOP_CYCLE_NSEC)
847                         return idx - 1;
848         } else {
849                 ioc->autop_too_slow_at = 0;
850         }
851
852         return idx;
853 }
854
855 /*
856  * Take the followings as input
857  *
858  *  @bps        maximum sequential throughput
859  *  @seqiops    maximum sequential 4k iops
860  *  @randiops   maximum random 4k iops
861  *
862  * and calculate the linear model cost coefficients.
863  *
864  *  *@page      per-page cost           1s / (@bps / 4096)
865  *  *@seqio     base cost of a seq IO   max((1s / @seqiops) - *@page, 0)
866  *  @randiops   base cost of a rand IO  max((1s / @randiops) - *@page, 0)
867  */
868 static void calc_lcoefs(u64 bps, u64 seqiops, u64 randiops,
869                         u64 *page, u64 *seqio, u64 *randio)
870 {
871         u64 v;
872
873         *page = *seqio = *randio = 0;
874
875         if (bps) {
876                 u64 bps_pages = DIV_ROUND_UP_ULL(bps, IOC_PAGE_SIZE);
877
878                 if (bps_pages)
879                         *page = DIV64_U64_ROUND_UP(VTIME_PER_SEC, bps_pages);
880                 else
881                         *page = 1;
882         }
883
884         if (seqiops) {
885                 v = DIV64_U64_ROUND_UP(VTIME_PER_SEC, seqiops);
886                 if (v > *page)
887                         *seqio = v - *page;
888         }
889
890         if (randiops) {
891                 v = DIV64_U64_ROUND_UP(VTIME_PER_SEC, randiops);
892                 if (v > *page)
893                         *randio = v - *page;
894         }
895 }
896
897 static void ioc_refresh_lcoefs(struct ioc *ioc)
898 {
899         u64 *u = ioc->params.i_lcoefs;
900         u64 *c = ioc->params.lcoefs;
901
902         calc_lcoefs(u[I_LCOEF_RBPS], u[I_LCOEF_RSEQIOPS], u[I_LCOEF_RRANDIOPS],
903                     &c[LCOEF_RPAGE], &c[LCOEF_RSEQIO], &c[LCOEF_RRANDIO]);
904         calc_lcoefs(u[I_LCOEF_WBPS], u[I_LCOEF_WSEQIOPS], u[I_LCOEF_WRANDIOPS],
905                     &c[LCOEF_WPAGE], &c[LCOEF_WSEQIO], &c[LCOEF_WRANDIO]);
906 }
907
908 /*
909  * struct gendisk is required as an argument because ioc->rqos.disk
910  * is not properly initialized when called from the init path.
911  */
912 static bool ioc_refresh_params_disk(struct ioc *ioc, bool force,
913                                     struct gendisk *disk)
914 {
915         const struct ioc_params *p;
916         int idx;
917
918         lockdep_assert_held(&ioc->lock);
919
920         idx = ioc_autop_idx(ioc, disk);
921         p = &autop[idx];
922
923         if (idx == ioc->autop_idx && !force)
924                 return false;
925
926         if (idx != ioc->autop_idx) {
927                 atomic64_set(&ioc->vtime_rate, VTIME_PER_USEC);
928                 ioc->vtime_base_rate = VTIME_PER_USEC;
929         }
930
931         ioc->autop_idx = idx;
932         ioc->autop_too_fast_at = 0;
933         ioc->autop_too_slow_at = 0;
934
935         if (!ioc->user_qos_params)
936                 memcpy(ioc->params.qos, p->qos, sizeof(p->qos));
937         if (!ioc->user_cost_model)
938                 memcpy(ioc->params.i_lcoefs, p->i_lcoefs, sizeof(p->i_lcoefs));
939
940         ioc_refresh_period_us(ioc);
941         ioc_refresh_lcoefs(ioc);
942
943         ioc->vrate_min = DIV64_U64_ROUND_UP((u64)ioc->params.qos[QOS_MIN] *
944                                             VTIME_PER_USEC, MILLION);
945         ioc->vrate_max = DIV64_U64_ROUND_UP((u64)ioc->params.qos[QOS_MAX] *
946                                             VTIME_PER_USEC, MILLION);
947
948         return true;
949 }
950
951 static bool ioc_refresh_params(struct ioc *ioc, bool force)
952 {
953         return ioc_refresh_params_disk(ioc, force, ioc->rqos.disk);
954 }
955
956 /*
957  * When an iocg accumulates too much vtime or gets deactivated, we throw away
958  * some vtime, which lowers the overall device utilization. As the exact amount
959  * which is being thrown away is known, we can compensate by accelerating the
960  * vrate accordingly so that the extra vtime generated in the current period
961  * matches what got lost.
962  */
963 static void ioc_refresh_vrate(struct ioc *ioc, struct ioc_now *now)
964 {
965         s64 pleft = ioc->period_at + ioc->period_us - now->now;
966         s64 vperiod = ioc->period_us * ioc->vtime_base_rate;
967         s64 vcomp, vcomp_min, vcomp_max;
968
969         lockdep_assert_held(&ioc->lock);
970
971         /* we need some time left in this period */
972         if (pleft <= 0)
973                 goto done;
974
975         /*
976          * Calculate how much vrate should be adjusted to offset the error.
977          * Limit the amount of adjustment and deduct the adjusted amount from
978          * the error.
979          */
980         vcomp = -div64_s64(ioc->vtime_err, pleft);
981         vcomp_min = -(ioc->vtime_base_rate >> 1);
982         vcomp_max = ioc->vtime_base_rate;
983         vcomp = clamp(vcomp, vcomp_min, vcomp_max);
984
985         ioc->vtime_err += vcomp * pleft;
986
987         atomic64_set(&ioc->vtime_rate, ioc->vtime_base_rate + vcomp);
988 done:
989         /* bound how much error can accumulate */
990         ioc->vtime_err = clamp(ioc->vtime_err, -vperiod, vperiod);
991 }
992
993 static void ioc_adjust_base_vrate(struct ioc *ioc, u32 rq_wait_pct,
994                                   int nr_lagging, int nr_shortages,
995                                   int prev_busy_level, u32 *missed_ppm)
996 {
997         u64 vrate = ioc->vtime_base_rate;
998         u64 vrate_min = ioc->vrate_min, vrate_max = ioc->vrate_max;
999
1000         if (!ioc->busy_level || (ioc->busy_level < 0 && nr_lagging)) {
1001                 if (ioc->busy_level != prev_busy_level || nr_lagging)
1002                         trace_iocost_ioc_vrate_adj(ioc, vrate,
1003                                                    missed_ppm, rq_wait_pct,
1004                                                    nr_lagging, nr_shortages);
1005
1006                 return;
1007         }
1008
1009         /*
1010          * If vrate is out of bounds, apply clamp gradually as the
1011          * bounds can change abruptly.  Otherwise, apply busy_level
1012          * based adjustment.
1013          */
1014         if (vrate < vrate_min) {
1015                 vrate = div64_u64(vrate * (100 + VRATE_CLAMP_ADJ_PCT), 100);
1016                 vrate = min(vrate, vrate_min);
1017         } else if (vrate > vrate_max) {
1018                 vrate = div64_u64(vrate * (100 - VRATE_CLAMP_ADJ_PCT), 100);
1019                 vrate = max(vrate, vrate_max);
1020         } else {
1021                 int idx = min_t(int, abs(ioc->busy_level),
1022                                 ARRAY_SIZE(vrate_adj_pct) - 1);
1023                 u32 adj_pct = vrate_adj_pct[idx];
1024
1025                 if (ioc->busy_level > 0)
1026                         adj_pct = 100 - adj_pct;
1027                 else
1028                         adj_pct = 100 + adj_pct;
1029
1030                 vrate = clamp(DIV64_U64_ROUND_UP(vrate * adj_pct, 100),
1031                               vrate_min, vrate_max);
1032         }
1033
1034         trace_iocost_ioc_vrate_adj(ioc, vrate, missed_ppm, rq_wait_pct,
1035                                    nr_lagging, nr_shortages);
1036
1037         ioc->vtime_base_rate = vrate;
1038         ioc_refresh_margins(ioc);
1039 }
1040
1041 /* take a snapshot of the current [v]time and vrate */
1042 static void ioc_now(struct ioc *ioc, struct ioc_now *now)
1043 {
1044         unsigned seq;
1045         u64 vrate;
1046
1047         now->now_ns = ktime_get();
1048         now->now = ktime_to_us(now->now_ns);
1049         vrate = atomic64_read(&ioc->vtime_rate);
1050
1051         /*
1052          * The current vtime is
1053          *
1054          *   vtime at period start + (wallclock time since the start) * vrate
1055          *
1056          * As a consistent snapshot of `period_at_vtime` and `period_at` is
1057          * needed, they're seqcount protected.
1058          */
1059         do {
1060                 seq = read_seqcount_begin(&ioc->period_seqcount);
1061                 now->vnow = ioc->period_at_vtime +
1062                         (now->now - ioc->period_at) * vrate;
1063         } while (read_seqcount_retry(&ioc->period_seqcount, seq));
1064 }
1065
1066 static void ioc_start_period(struct ioc *ioc, struct ioc_now *now)
1067 {
1068         WARN_ON_ONCE(ioc->running != IOC_RUNNING);
1069
1070         write_seqcount_begin(&ioc->period_seqcount);
1071         ioc->period_at = now->now;
1072         ioc->period_at_vtime = now->vnow;
1073         write_seqcount_end(&ioc->period_seqcount);
1074
1075         ioc->timer.expires = jiffies + usecs_to_jiffies(ioc->period_us);
1076         add_timer(&ioc->timer);
1077 }
1078
1079 /*
1080  * Update @iocg's `active` and `inuse` to @active and @inuse, update level
1081  * weight sums and propagate upwards accordingly. If @save, the current margin
1082  * is saved to be used as reference for later inuse in-period adjustments.
1083  */
1084 static void __propagate_weights(struct ioc_gq *iocg, u32 active, u32 inuse,
1085                                 bool save, struct ioc_now *now)
1086 {
1087         struct ioc *ioc = iocg->ioc;
1088         int lvl;
1089
1090         lockdep_assert_held(&ioc->lock);
1091
1092         /*
1093          * For an active leaf node, its inuse shouldn't be zero or exceed
1094          * @active. An active internal node's inuse is solely determined by the
1095          * inuse to active ratio of its children regardless of @inuse.
1096          */
1097         if (list_empty(&iocg->active_list) && iocg->child_active_sum) {
1098                 inuse = DIV64_U64_ROUND_UP(active * iocg->child_inuse_sum,
1099                                            iocg->child_active_sum);
1100         } else {
1101                 inuse = clamp_t(u32, inuse, 1, active);
1102         }
1103
1104         iocg->last_inuse = iocg->inuse;
1105         if (save)
1106                 iocg->saved_margin = now->vnow - atomic64_read(&iocg->vtime);
1107
1108         if (active == iocg->active && inuse == iocg->inuse)
1109                 return;
1110
1111         for (lvl = iocg->level - 1; lvl >= 0; lvl--) {
1112                 struct ioc_gq *parent = iocg->ancestors[lvl];
1113                 struct ioc_gq *child = iocg->ancestors[lvl + 1];
1114                 u32 parent_active = 0, parent_inuse = 0;
1115
1116                 /* update the level sums */
1117                 parent->child_active_sum += (s32)(active - child->active);
1118                 parent->child_inuse_sum += (s32)(inuse - child->inuse);
1119                 /* apply the updates */
1120                 child->active = active;
1121                 child->inuse = inuse;
1122
1123                 /*
1124                  * The delta between inuse and active sums indicates that
1125                  * much of weight is being given away.  Parent's inuse
1126                  * and active should reflect the ratio.
1127                  */
1128                 if (parent->child_active_sum) {
1129                         parent_active = parent->weight;
1130                         parent_inuse = DIV64_U64_ROUND_UP(
1131                                 parent_active * parent->child_inuse_sum,
1132                                 parent->child_active_sum);
1133                 }
1134
1135                 /* do we need to keep walking up? */
1136                 if (parent_active == parent->active &&
1137                     parent_inuse == parent->inuse)
1138                         break;
1139
1140                 active = parent_active;
1141                 inuse = parent_inuse;
1142         }
1143
1144         ioc->weights_updated = true;
1145 }
1146
1147 static void commit_weights(struct ioc *ioc)
1148 {
1149         lockdep_assert_held(&ioc->lock);
1150
1151         if (ioc->weights_updated) {
1152                 /* paired with rmb in current_hweight(), see there */
1153                 smp_wmb();
1154                 atomic_inc(&ioc->hweight_gen);
1155                 ioc->weights_updated = false;
1156         }
1157 }
1158
1159 static void propagate_weights(struct ioc_gq *iocg, u32 active, u32 inuse,
1160                               bool save, struct ioc_now *now)
1161 {
1162         __propagate_weights(iocg, active, inuse, save, now);
1163         commit_weights(iocg->ioc);
1164 }
1165
1166 static void current_hweight(struct ioc_gq *iocg, u32 *hw_activep, u32 *hw_inusep)
1167 {
1168         struct ioc *ioc = iocg->ioc;
1169         int lvl;
1170         u32 hwa, hwi;
1171         int ioc_gen;
1172
1173         /* hot path - if uptodate, use cached */
1174         ioc_gen = atomic_read(&ioc->hweight_gen);
1175         if (ioc_gen == iocg->hweight_gen)
1176                 goto out;
1177
1178         /*
1179          * Paired with wmb in commit_weights(). If we saw the updated
1180          * hweight_gen, all the weight updates from __propagate_weights() are
1181          * visible too.
1182          *
1183          * We can race with weight updates during calculation and get it
1184          * wrong.  However, hweight_gen would have changed and a future
1185          * reader will recalculate and we're guaranteed to discard the
1186          * wrong result soon.
1187          */
1188         smp_rmb();
1189
1190         hwa = hwi = WEIGHT_ONE;
1191         for (lvl = 0; lvl <= iocg->level - 1; lvl++) {
1192                 struct ioc_gq *parent = iocg->ancestors[lvl];
1193                 struct ioc_gq *child = iocg->ancestors[lvl + 1];
1194                 u64 active_sum = READ_ONCE(parent->child_active_sum);
1195                 u64 inuse_sum = READ_ONCE(parent->child_inuse_sum);
1196                 u32 active = READ_ONCE(child->active);
1197                 u32 inuse = READ_ONCE(child->inuse);
1198
1199                 /* we can race with deactivations and either may read as zero */
1200                 if (!active_sum || !inuse_sum)
1201                         continue;
1202
1203                 active_sum = max_t(u64, active, active_sum);
1204                 hwa = div64_u64((u64)hwa * active, active_sum);
1205
1206                 inuse_sum = max_t(u64, inuse, inuse_sum);
1207                 hwi = div64_u64((u64)hwi * inuse, inuse_sum);
1208         }
1209
1210         iocg->hweight_active = max_t(u32, hwa, 1);
1211         iocg->hweight_inuse = max_t(u32, hwi, 1);
1212         iocg->hweight_gen = ioc_gen;
1213 out:
1214         if (hw_activep)
1215                 *hw_activep = iocg->hweight_active;
1216         if (hw_inusep)
1217                 *hw_inusep = iocg->hweight_inuse;
1218 }
1219
1220 /*
1221  * Calculate the hweight_inuse @iocg would get with max @inuse assuming all the
1222  * other weights stay unchanged.
1223  */
1224 static u32 current_hweight_max(struct ioc_gq *iocg)
1225 {
1226         u32 hwm = WEIGHT_ONE;
1227         u32 inuse = iocg->active;
1228         u64 child_inuse_sum;
1229         int lvl;
1230
1231         lockdep_assert_held(&iocg->ioc->lock);
1232
1233         for (lvl = iocg->level - 1; lvl >= 0; lvl--) {
1234                 struct ioc_gq *parent = iocg->ancestors[lvl];
1235                 struct ioc_gq *child = iocg->ancestors[lvl + 1];
1236
1237                 child_inuse_sum = parent->child_inuse_sum + inuse - child->inuse;
1238                 hwm = div64_u64((u64)hwm * inuse, child_inuse_sum);
1239                 inuse = DIV64_U64_ROUND_UP(parent->active * child_inuse_sum,
1240                                            parent->child_active_sum);
1241         }
1242
1243         return max_t(u32, hwm, 1);
1244 }
1245
1246 static void weight_updated(struct ioc_gq *iocg, struct ioc_now *now)
1247 {
1248         struct ioc *ioc = iocg->ioc;
1249         struct blkcg_gq *blkg = iocg_to_blkg(iocg);
1250         struct ioc_cgrp *iocc = blkcg_to_iocc(blkg->blkcg);
1251         u32 weight;
1252
1253         lockdep_assert_held(&ioc->lock);
1254
1255         weight = iocg->cfg_weight ?: iocc->dfl_weight;
1256         if (weight != iocg->weight && iocg->active)
1257                 propagate_weights(iocg, weight, iocg->inuse, true, now);
1258         iocg->weight = weight;
1259 }
1260
1261 static bool iocg_activate(struct ioc_gq *iocg, struct ioc_now *now)
1262 {
1263         struct ioc *ioc = iocg->ioc;
1264         u64 last_period, cur_period;
1265         u64 vtime, vtarget;
1266         int i;
1267
1268         /*
1269          * If seem to be already active, just update the stamp to tell the
1270          * timer that we're still active.  We don't mind occassional races.
1271          */
1272         if (!list_empty(&iocg->active_list)) {
1273                 ioc_now(ioc, now);
1274                 cur_period = atomic64_read(&ioc->cur_period);
1275                 if (atomic64_read(&iocg->active_period) != cur_period)
1276                         atomic64_set(&iocg->active_period, cur_period);
1277                 return true;
1278         }
1279
1280         /* racy check on internal node IOs, treat as root level IOs */
1281         if (iocg->child_active_sum)
1282                 return false;
1283
1284         spin_lock_irq(&ioc->lock);
1285
1286         ioc_now(ioc, now);
1287
1288         /* update period */
1289         cur_period = atomic64_read(&ioc->cur_period);
1290         last_period = atomic64_read(&iocg->active_period);
1291         atomic64_set(&iocg->active_period, cur_period);
1292
1293         /* already activated or breaking leaf-only constraint? */
1294         if (!list_empty(&iocg->active_list))
1295                 goto succeed_unlock;
1296         for (i = iocg->level - 1; i > 0; i--)
1297                 if (!list_empty(&iocg->ancestors[i]->active_list))
1298                         goto fail_unlock;
1299
1300         if (iocg->child_active_sum)
1301                 goto fail_unlock;
1302
1303         /*
1304          * Always start with the target budget. On deactivation, we throw away
1305          * anything above it.
1306          */
1307         vtarget = now->vnow - ioc->margins.target;
1308         vtime = atomic64_read(&iocg->vtime);
1309
1310         atomic64_add(vtarget - vtime, &iocg->vtime);
1311         atomic64_add(vtarget - vtime, &iocg->done_vtime);
1312         vtime = vtarget;
1313
1314         /*
1315          * Activate, propagate weight and start period timer if not
1316          * running.  Reset hweight_gen to avoid accidental match from
1317          * wrapping.
1318          */
1319         iocg->hweight_gen = atomic_read(&ioc->hweight_gen) - 1;
1320         list_add(&iocg->active_list, &ioc->active_iocgs);
1321
1322         propagate_weights(iocg, iocg->weight,
1323                           iocg->last_inuse ?: iocg->weight, true, now);
1324
1325         TRACE_IOCG_PATH(iocg_activate, iocg, now,
1326                         last_period, cur_period, vtime);
1327
1328         iocg->activated_at = now->now;
1329
1330         if (ioc->running == IOC_IDLE) {
1331                 ioc->running = IOC_RUNNING;
1332                 ioc->dfgv_period_at = now->now;
1333                 ioc->dfgv_period_rem = 0;
1334                 ioc_start_period(ioc, now);
1335         }
1336
1337 succeed_unlock:
1338         spin_unlock_irq(&ioc->lock);
1339         return true;
1340
1341 fail_unlock:
1342         spin_unlock_irq(&ioc->lock);
1343         return false;
1344 }
1345
1346 static bool iocg_kick_delay(struct ioc_gq *iocg, struct ioc_now *now)
1347 {
1348         struct ioc *ioc = iocg->ioc;
1349         struct blkcg_gq *blkg = iocg_to_blkg(iocg);
1350         u64 tdelta, delay, new_delay;
1351         s64 vover, vover_pct;
1352         u32 hwa;
1353
1354         lockdep_assert_held(&iocg->waitq.lock);
1355
1356         /* calculate the current delay in effect - 1/2 every second */
1357         tdelta = now->now - iocg->delay_at;
1358         if (iocg->delay)
1359                 delay = iocg->delay >> div64_u64(tdelta, USEC_PER_SEC);
1360         else
1361                 delay = 0;
1362
1363         /* calculate the new delay from the debt amount */
1364         current_hweight(iocg, &hwa, NULL);
1365         vover = atomic64_read(&iocg->vtime) +
1366                 abs_cost_to_cost(iocg->abs_vdebt, hwa) - now->vnow;
1367         vover_pct = div64_s64(100 * vover,
1368                               ioc->period_us * ioc->vtime_base_rate);
1369
1370         if (vover_pct <= MIN_DELAY_THR_PCT)
1371                 new_delay = 0;
1372         else if (vover_pct >= MAX_DELAY_THR_PCT)
1373                 new_delay = MAX_DELAY;
1374         else
1375                 new_delay = MIN_DELAY +
1376                         div_u64((MAX_DELAY - MIN_DELAY) *
1377                                 (vover_pct - MIN_DELAY_THR_PCT),
1378                                 MAX_DELAY_THR_PCT - MIN_DELAY_THR_PCT);
1379
1380         /* pick the higher one and apply */
1381         if (new_delay > delay) {
1382                 iocg->delay = new_delay;
1383                 iocg->delay_at = now->now;
1384                 delay = new_delay;
1385         }
1386
1387         if (delay >= MIN_DELAY) {
1388                 if (!iocg->indelay_since)
1389                         iocg->indelay_since = now->now;
1390                 blkcg_set_delay(blkg, delay * NSEC_PER_USEC);
1391                 return true;
1392         } else {
1393                 if (iocg->indelay_since) {
1394                         iocg->stat.indelay_us += now->now - iocg->indelay_since;
1395                         iocg->indelay_since = 0;
1396                 }
1397                 iocg->delay = 0;
1398                 blkcg_clear_delay(blkg);
1399                 return false;
1400         }
1401 }
1402
1403 static void iocg_incur_debt(struct ioc_gq *iocg, u64 abs_cost,
1404                             struct ioc_now *now)
1405 {
1406         struct iocg_pcpu_stat *gcs;
1407
1408         lockdep_assert_held(&iocg->ioc->lock);
1409         lockdep_assert_held(&iocg->waitq.lock);
1410         WARN_ON_ONCE(list_empty(&iocg->active_list));
1411
1412         /*
1413          * Once in debt, debt handling owns inuse. @iocg stays at the minimum
1414          * inuse donating all of it share to others until its debt is paid off.
1415          */
1416         if (!iocg->abs_vdebt && abs_cost) {
1417                 iocg->indebt_since = now->now;
1418                 propagate_weights(iocg, iocg->active, 0, false, now);
1419         }
1420
1421         iocg->abs_vdebt += abs_cost;
1422
1423         gcs = get_cpu_ptr(iocg->pcpu_stat);
1424         local64_add(abs_cost, &gcs->abs_vusage);
1425         put_cpu_ptr(gcs);
1426 }
1427
1428 static void iocg_pay_debt(struct ioc_gq *iocg, u64 abs_vpay,
1429                           struct ioc_now *now)
1430 {
1431         lockdep_assert_held(&iocg->ioc->lock);
1432         lockdep_assert_held(&iocg->waitq.lock);
1433
1434         /* make sure that nobody messed with @iocg */
1435         WARN_ON_ONCE(list_empty(&iocg->active_list));
1436         WARN_ON_ONCE(iocg->inuse > 1);
1437
1438         iocg->abs_vdebt -= min(abs_vpay, iocg->abs_vdebt);
1439
1440         /* if debt is paid in full, restore inuse */
1441         if (!iocg->abs_vdebt) {
1442                 iocg->stat.indebt_us += now->now - iocg->indebt_since;
1443                 iocg->indebt_since = 0;
1444
1445                 propagate_weights(iocg, iocg->active, iocg->last_inuse,
1446                                   false, now);
1447         }
1448 }
1449
1450 static int iocg_wake_fn(struct wait_queue_entry *wq_entry, unsigned mode,
1451                         int flags, void *key)
1452 {
1453         struct iocg_wait *wait = container_of(wq_entry, struct iocg_wait, wait);
1454         struct iocg_wake_ctx *ctx = key;
1455         u64 cost = abs_cost_to_cost(wait->abs_cost, ctx->hw_inuse);
1456
1457         ctx->vbudget -= cost;
1458
1459         if (ctx->vbudget < 0)
1460                 return -1;
1461
1462         iocg_commit_bio(ctx->iocg, wait->bio, wait->abs_cost, cost);
1463         wait->committed = true;
1464
1465         /*
1466          * autoremove_wake_function() removes the wait entry only when it
1467          * actually changed the task state. We want the wait always removed.
1468          * Remove explicitly and use default_wake_function(). Note that the
1469          * order of operations is important as finish_wait() tests whether
1470          * @wq_entry is removed without grabbing the lock.
1471          */
1472         default_wake_function(wq_entry, mode, flags, key);
1473         list_del_init_careful(&wq_entry->entry);
1474         return 0;
1475 }
1476
1477 /*
1478  * Calculate the accumulated budget, pay debt if @pay_debt and wake up waiters
1479  * accordingly. When @pay_debt is %true, the caller must be holding ioc->lock in
1480  * addition to iocg->waitq.lock.
1481  */
1482 static void iocg_kick_waitq(struct ioc_gq *iocg, bool pay_debt,
1483                             struct ioc_now *now)
1484 {
1485         struct ioc *ioc = iocg->ioc;
1486         struct iocg_wake_ctx ctx = { .iocg = iocg };
1487         u64 vshortage, expires, oexpires;
1488         s64 vbudget;
1489         u32 hwa;
1490
1491         lockdep_assert_held(&iocg->waitq.lock);
1492
1493         current_hweight(iocg, &hwa, NULL);
1494         vbudget = now->vnow - atomic64_read(&iocg->vtime);
1495
1496         /* pay off debt */
1497         if (pay_debt && iocg->abs_vdebt && vbudget > 0) {
1498                 u64 abs_vbudget = cost_to_abs_cost(vbudget, hwa);
1499                 u64 abs_vpay = min_t(u64, abs_vbudget, iocg->abs_vdebt);
1500                 u64 vpay = abs_cost_to_cost(abs_vpay, hwa);
1501
1502                 lockdep_assert_held(&ioc->lock);
1503
1504                 atomic64_add(vpay, &iocg->vtime);
1505                 atomic64_add(vpay, &iocg->done_vtime);
1506                 iocg_pay_debt(iocg, abs_vpay, now);
1507                 vbudget -= vpay;
1508         }
1509
1510         if (iocg->abs_vdebt || iocg->delay)
1511                 iocg_kick_delay(iocg, now);
1512
1513         /*
1514          * Debt can still be outstanding if we haven't paid all yet or the
1515          * caller raced and called without @pay_debt. Shouldn't wake up waiters
1516          * under debt. Make sure @vbudget reflects the outstanding amount and is
1517          * not positive.
1518          */
1519         if (iocg->abs_vdebt) {
1520                 s64 vdebt = abs_cost_to_cost(iocg->abs_vdebt, hwa);
1521                 vbudget = min_t(s64, 0, vbudget - vdebt);
1522         }
1523
1524         /*
1525          * Wake up the ones which are due and see how much vtime we'll need for
1526          * the next one. As paying off debt restores hw_inuse, it must be read
1527          * after the above debt payment.
1528          */
1529         ctx.vbudget = vbudget;
1530         current_hweight(iocg, NULL, &ctx.hw_inuse);
1531
1532         __wake_up_locked_key(&iocg->waitq, TASK_NORMAL, &ctx);
1533
1534         if (!waitqueue_active(&iocg->waitq)) {
1535                 if (iocg->wait_since) {
1536                         iocg->stat.wait_us += now->now - iocg->wait_since;
1537                         iocg->wait_since = 0;
1538                 }
1539                 return;
1540         }
1541
1542         if (!iocg->wait_since)
1543                 iocg->wait_since = now->now;
1544
1545         if (WARN_ON_ONCE(ctx.vbudget >= 0))
1546                 return;
1547
1548         /* determine next wakeup, add a timer margin to guarantee chunking */
1549         vshortage = -ctx.vbudget;
1550         expires = now->now_ns +
1551                 DIV64_U64_ROUND_UP(vshortage, ioc->vtime_base_rate) *
1552                 NSEC_PER_USEC;
1553         expires += ioc->timer_slack_ns;
1554
1555         /* if already active and close enough, don't bother */
1556         oexpires = ktime_to_ns(hrtimer_get_softexpires(&iocg->waitq_timer));
1557         if (hrtimer_is_queued(&iocg->waitq_timer) &&
1558             abs(oexpires - expires) <= ioc->timer_slack_ns)
1559                 return;
1560
1561         hrtimer_start_range_ns(&iocg->waitq_timer, ns_to_ktime(expires),
1562                                ioc->timer_slack_ns, HRTIMER_MODE_ABS);
1563 }
1564
1565 static enum hrtimer_restart iocg_waitq_timer_fn(struct hrtimer *timer)
1566 {
1567         struct ioc_gq *iocg = container_of(timer, struct ioc_gq, waitq_timer);
1568         bool pay_debt = READ_ONCE(iocg->abs_vdebt);
1569         struct ioc_now now;
1570         unsigned long flags;
1571
1572         ioc_now(iocg->ioc, &now);
1573
1574         iocg_lock(iocg, pay_debt, &flags);
1575         iocg_kick_waitq(iocg, pay_debt, &now);
1576         iocg_unlock(iocg, pay_debt, &flags);
1577
1578         return HRTIMER_NORESTART;
1579 }
1580
1581 static void ioc_lat_stat(struct ioc *ioc, u32 *missed_ppm_ar, u32 *rq_wait_pct_p)
1582 {
1583         u32 nr_met[2] = { };
1584         u32 nr_missed[2] = { };
1585         u64 rq_wait_ns = 0;
1586         int cpu, rw;
1587
1588         for_each_online_cpu(cpu) {
1589                 struct ioc_pcpu_stat *stat = per_cpu_ptr(ioc->pcpu_stat, cpu);
1590                 u64 this_rq_wait_ns;
1591
1592                 for (rw = READ; rw <= WRITE; rw++) {
1593                         u32 this_met = local_read(&stat->missed[rw].nr_met);
1594                         u32 this_missed = local_read(&stat->missed[rw].nr_missed);
1595
1596                         nr_met[rw] += this_met - stat->missed[rw].last_met;
1597                         nr_missed[rw] += this_missed - stat->missed[rw].last_missed;
1598                         stat->missed[rw].last_met = this_met;
1599                         stat->missed[rw].last_missed = this_missed;
1600                 }
1601
1602                 this_rq_wait_ns = local64_read(&stat->rq_wait_ns);
1603                 rq_wait_ns += this_rq_wait_ns - stat->last_rq_wait_ns;
1604                 stat->last_rq_wait_ns = this_rq_wait_ns;
1605         }
1606
1607         for (rw = READ; rw <= WRITE; rw++) {
1608                 if (nr_met[rw] + nr_missed[rw])
1609                         missed_ppm_ar[rw] =
1610                                 DIV64_U64_ROUND_UP((u64)nr_missed[rw] * MILLION,
1611                                                    nr_met[rw] + nr_missed[rw]);
1612                 else
1613                         missed_ppm_ar[rw] = 0;
1614         }
1615
1616         *rq_wait_pct_p = div64_u64(rq_wait_ns * 100,
1617                                    ioc->period_us * NSEC_PER_USEC);
1618 }
1619
1620 /* was iocg idle this period? */
1621 static bool iocg_is_idle(struct ioc_gq *iocg)
1622 {
1623         struct ioc *ioc = iocg->ioc;
1624
1625         /* did something get issued this period? */
1626         if (atomic64_read(&iocg->active_period) ==
1627             atomic64_read(&ioc->cur_period))
1628                 return false;
1629
1630         /* is something in flight? */
1631         if (atomic64_read(&iocg->done_vtime) != atomic64_read(&iocg->vtime))
1632                 return false;
1633
1634         return true;
1635 }
1636
1637 /*
1638  * Call this function on the target leaf @iocg's to build pre-order traversal
1639  * list of all the ancestors in @inner_walk. The inner nodes are linked through
1640  * ->walk_list and the caller is responsible for dissolving the list after use.
1641  */
1642 static void iocg_build_inner_walk(struct ioc_gq *iocg,
1643                                   struct list_head *inner_walk)
1644 {
1645         int lvl;
1646
1647         WARN_ON_ONCE(!list_empty(&iocg->walk_list));
1648
1649         /* find the first ancestor which hasn't been visited yet */
1650         for (lvl = iocg->level - 1; lvl >= 0; lvl--) {
1651                 if (!list_empty(&iocg->ancestors[lvl]->walk_list))
1652                         break;
1653         }
1654
1655         /* walk down and visit the inner nodes to get pre-order traversal */
1656         while (++lvl <= iocg->level - 1) {
1657                 struct ioc_gq *inner = iocg->ancestors[lvl];
1658
1659                 /* record traversal order */
1660                 list_add_tail(&inner->walk_list, inner_walk);
1661         }
1662 }
1663
1664 /* propagate the deltas to the parent */
1665 static void iocg_flush_stat_upward(struct ioc_gq *iocg)
1666 {
1667         if (iocg->level > 0) {
1668                 struct iocg_stat *parent_stat =
1669                         &iocg->ancestors[iocg->level - 1]->stat;
1670
1671                 parent_stat->usage_us +=
1672                         iocg->stat.usage_us - iocg->last_stat.usage_us;
1673                 parent_stat->wait_us +=
1674                         iocg->stat.wait_us - iocg->last_stat.wait_us;
1675                 parent_stat->indebt_us +=
1676                         iocg->stat.indebt_us - iocg->last_stat.indebt_us;
1677                 parent_stat->indelay_us +=
1678                         iocg->stat.indelay_us - iocg->last_stat.indelay_us;
1679         }
1680
1681         iocg->last_stat = iocg->stat;
1682 }
1683
1684 /* collect per-cpu counters and propagate the deltas to the parent */
1685 static void iocg_flush_stat_leaf(struct ioc_gq *iocg, struct ioc_now *now)
1686 {
1687         struct ioc *ioc = iocg->ioc;
1688         u64 abs_vusage = 0;
1689         u64 vusage_delta;
1690         int cpu;
1691
1692         lockdep_assert_held(&iocg->ioc->lock);
1693
1694         /* collect per-cpu counters */
1695         for_each_possible_cpu(cpu) {
1696                 abs_vusage += local64_read(
1697                                 per_cpu_ptr(&iocg->pcpu_stat->abs_vusage, cpu));
1698         }
1699         vusage_delta = abs_vusage - iocg->last_stat_abs_vusage;
1700         iocg->last_stat_abs_vusage = abs_vusage;
1701
1702         iocg->usage_delta_us = div64_u64(vusage_delta, ioc->vtime_base_rate);
1703         iocg->stat.usage_us += iocg->usage_delta_us;
1704
1705         iocg_flush_stat_upward(iocg);
1706 }
1707
1708 /* get stat counters ready for reading on all active iocgs */
1709 static void iocg_flush_stat(struct list_head *target_iocgs, struct ioc_now *now)
1710 {
1711         LIST_HEAD(inner_walk);
1712         struct ioc_gq *iocg, *tiocg;
1713
1714         /* flush leaves and build inner node walk list */
1715         list_for_each_entry(iocg, target_iocgs, active_list) {
1716                 iocg_flush_stat_leaf(iocg, now);
1717                 iocg_build_inner_walk(iocg, &inner_walk);
1718         }
1719
1720         /* keep flushing upwards by walking the inner list backwards */
1721         list_for_each_entry_safe_reverse(iocg, tiocg, &inner_walk, walk_list) {
1722                 iocg_flush_stat_upward(iocg);
1723                 list_del_init(&iocg->walk_list);
1724         }
1725 }
1726
1727 /*
1728  * Determine what @iocg's hweight_inuse should be after donating unused
1729  * capacity. @hwm is the upper bound and used to signal no donation. This
1730  * function also throws away @iocg's excess budget.
1731  */
1732 static u32 hweight_after_donation(struct ioc_gq *iocg, u32 old_hwi, u32 hwm,
1733                                   u32 usage, struct ioc_now *now)
1734 {
1735         struct ioc *ioc = iocg->ioc;
1736         u64 vtime = atomic64_read(&iocg->vtime);
1737         s64 excess, delta, target, new_hwi;
1738
1739         /* debt handling owns inuse for debtors */
1740         if (iocg->abs_vdebt)
1741                 return 1;
1742
1743         /* see whether minimum margin requirement is met */
1744         if (waitqueue_active(&iocg->waitq) ||
1745             time_after64(vtime, now->vnow - ioc->margins.min))
1746                 return hwm;
1747
1748         /* throw away excess above target */
1749         excess = now->vnow - vtime - ioc->margins.target;
1750         if (excess > 0) {
1751                 atomic64_add(excess, &iocg->vtime);
1752                 atomic64_add(excess, &iocg->done_vtime);
1753                 vtime += excess;
1754                 ioc->vtime_err -= div64_u64(excess * old_hwi, WEIGHT_ONE);
1755         }
1756
1757         /*
1758          * Let's say the distance between iocg's and device's vtimes as a
1759          * fraction of period duration is delta. Assuming that the iocg will
1760          * consume the usage determined above, we want to determine new_hwi so
1761          * that delta equals MARGIN_TARGET at the end of the next period.
1762          *
1763          * We need to execute usage worth of IOs while spending the sum of the
1764          * new budget (1 - MARGIN_TARGET) and the leftover from the last period
1765          * (delta):
1766          *
1767          *   usage = (1 - MARGIN_TARGET + delta) * new_hwi
1768          *
1769          * Therefore, the new_hwi is:
1770          *
1771          *   new_hwi = usage / (1 - MARGIN_TARGET + delta)
1772          */
1773         delta = div64_s64(WEIGHT_ONE * (now->vnow - vtime),
1774                           now->vnow - ioc->period_at_vtime);
1775         target = WEIGHT_ONE * MARGIN_TARGET_PCT / 100;
1776         new_hwi = div64_s64(WEIGHT_ONE * usage, WEIGHT_ONE - target + delta);
1777
1778         return clamp_t(s64, new_hwi, 1, hwm);
1779 }
1780
1781 /*
1782  * For work-conservation, an iocg which isn't using all of its share should
1783  * donate the leftover to other iocgs. There are two ways to achieve this - 1.
1784  * bumping up vrate accordingly 2. lowering the donating iocg's inuse weight.
1785  *
1786  * #1 is mathematically simpler but has the drawback of requiring synchronous
1787  * global hweight_inuse updates when idle iocg's get activated or inuse weights
1788  * change due to donation snapbacks as it has the possibility of grossly
1789  * overshooting what's allowed by the model and vrate.
1790  *
1791  * #2 is inherently safe with local operations. The donating iocg can easily
1792  * snap back to higher weights when needed without worrying about impacts on
1793  * other nodes as the impacts will be inherently correct. This also makes idle
1794  * iocg activations safe. The only effect activations have is decreasing
1795  * hweight_inuse of others, the right solution to which is for those iocgs to
1796  * snap back to higher weights.
1797  *
1798  * So, we go with #2. The challenge is calculating how each donating iocg's
1799  * inuse should be adjusted to achieve the target donation amounts. This is done
1800  * using Andy's method described in the following pdf.
1801  *
1802  *   https://drive.google.com/file/d/1PsJwxPFtjUnwOY1QJ5AeICCcsL7BM3bo
1803  *
1804  * Given the weights and target after-donation hweight_inuse values, Andy's
1805  * method determines how the proportional distribution should look like at each
1806  * sibling level to maintain the relative relationship between all non-donating
1807  * pairs. To roughly summarize, it divides the tree into donating and
1808  * non-donating parts, calculates global donation rate which is used to
1809  * determine the target hweight_inuse for each node, and then derives per-level
1810  * proportions.
1811  *
1812  * The following pdf shows that global distribution calculated this way can be
1813  * achieved by scaling inuse weights of donating leaves and propagating the
1814  * adjustments upwards proportionally.
1815  *
1816  *   https://drive.google.com/file/d/1vONz1-fzVO7oY5DXXsLjSxEtYYQbOvsE
1817  *
1818  * Combining the above two, we can determine how each leaf iocg's inuse should
1819  * be adjusted to achieve the target donation.
1820  *
1821  *   https://drive.google.com/file/d/1WcrltBOSPN0qXVdBgnKm4mdp9FhuEFQN
1822  *
1823  * The inline comments use symbols from the last pdf.
1824  *
1825  *   b is the sum of the absolute budgets in the subtree. 1 for the root node.
1826  *   f is the sum of the absolute budgets of non-donating nodes in the subtree.
1827  *   t is the sum of the absolute budgets of donating nodes in the subtree.
1828  *   w is the weight of the node. w = w_f + w_t
1829  *   w_f is the non-donating portion of w. w_f = w * f / b
1830  *   w_b is the donating portion of w. w_t = w * t / b
1831  *   s is the sum of all sibling weights. s = Sum(w) for siblings
1832  *   s_f and s_t are the non-donating and donating portions of s.
1833  *
1834  * Subscript p denotes the parent's counterpart and ' the adjusted value - e.g.
1835  * w_pt is the donating portion of the parent's weight and w'_pt the same value
1836  * after adjustments. Subscript r denotes the root node's values.
1837  */
1838 static void transfer_surpluses(struct list_head *surpluses, struct ioc_now *now)
1839 {
1840         LIST_HEAD(over_hwa);
1841         LIST_HEAD(inner_walk);
1842         struct ioc_gq *iocg, *tiocg, *root_iocg;
1843         u32 after_sum, over_sum, over_target, gamma;
1844
1845         /*
1846          * It's pretty unlikely but possible for the total sum of
1847          * hweight_after_donation's to be higher than WEIGHT_ONE, which will
1848          * confuse the following calculations. If such condition is detected,
1849          * scale down everyone over its full share equally to keep the sum below
1850          * WEIGHT_ONE.
1851          */
1852         after_sum = 0;
1853         over_sum = 0;
1854         list_for_each_entry(iocg, surpluses, surplus_list) {
1855                 u32 hwa;
1856
1857                 current_hweight(iocg, &hwa, NULL);
1858                 after_sum += iocg->hweight_after_donation;
1859
1860                 if (iocg->hweight_after_donation > hwa) {
1861                         over_sum += iocg->hweight_after_donation;
1862                         list_add(&iocg->walk_list, &over_hwa);
1863                 }
1864         }
1865
1866         if (after_sum >= WEIGHT_ONE) {
1867                 /*
1868                  * The delta should be deducted from the over_sum, calculate
1869                  * target over_sum value.
1870                  */
1871                 u32 over_delta = after_sum - (WEIGHT_ONE - 1);
1872                 WARN_ON_ONCE(over_sum <= over_delta);
1873                 over_target = over_sum - over_delta;
1874         } else {
1875                 over_target = 0;
1876         }
1877
1878         list_for_each_entry_safe(iocg, tiocg, &over_hwa, walk_list) {
1879                 if (over_target)
1880                         iocg->hweight_after_donation =
1881                                 div_u64((u64)iocg->hweight_after_donation *
1882                                         over_target, over_sum);
1883                 list_del_init(&iocg->walk_list);
1884         }
1885
1886         /*
1887          * Build pre-order inner node walk list and prepare for donation
1888          * adjustment calculations.
1889          */
1890         list_for_each_entry(iocg, surpluses, surplus_list) {
1891                 iocg_build_inner_walk(iocg, &inner_walk);
1892         }
1893
1894         root_iocg = list_first_entry(&inner_walk, struct ioc_gq, walk_list);
1895         WARN_ON_ONCE(root_iocg->level > 0);
1896
1897         list_for_each_entry(iocg, &inner_walk, walk_list) {
1898                 iocg->child_adjusted_sum = 0;
1899                 iocg->hweight_donating = 0;
1900                 iocg->hweight_after_donation = 0;
1901         }
1902
1903         /*
1904          * Propagate the donating budget (b_t) and after donation budget (b'_t)
1905          * up the hierarchy.
1906          */
1907         list_for_each_entry(iocg, surpluses, surplus_list) {
1908                 struct ioc_gq *parent = iocg->ancestors[iocg->level - 1];
1909
1910                 parent->hweight_donating += iocg->hweight_donating;
1911                 parent->hweight_after_donation += iocg->hweight_after_donation;
1912         }
1913
1914         list_for_each_entry_reverse(iocg, &inner_walk, walk_list) {
1915                 if (iocg->level > 0) {
1916                         struct ioc_gq *parent = iocg->ancestors[iocg->level - 1];
1917
1918                         parent->hweight_donating += iocg->hweight_donating;
1919                         parent->hweight_after_donation += iocg->hweight_after_donation;
1920                 }
1921         }
1922
1923         /*
1924          * Calculate inner hwa's (b) and make sure the donation values are
1925          * within the accepted ranges as we're doing low res calculations with
1926          * roundups.
1927          */
1928         list_for_each_entry(iocg, &inner_walk, walk_list) {
1929                 if (iocg->level) {
1930                         struct ioc_gq *parent = iocg->ancestors[iocg->level - 1];
1931
1932                         iocg->hweight_active = DIV64_U64_ROUND_UP(
1933                                 (u64)parent->hweight_active * iocg->active,
1934                                 parent->child_active_sum);
1935
1936                 }
1937
1938                 iocg->hweight_donating = min(iocg->hweight_donating,
1939                                              iocg->hweight_active);
1940                 iocg->hweight_after_donation = min(iocg->hweight_after_donation,
1941                                                    iocg->hweight_donating - 1);
1942                 if (WARN_ON_ONCE(iocg->hweight_active <= 1 ||
1943                                  iocg->hweight_donating <= 1 ||
1944                                  iocg->hweight_after_donation == 0)) {
1945                         pr_warn("iocg: invalid donation weights in ");
1946                         pr_cont_cgroup_path(iocg_to_blkg(iocg)->blkcg->css.cgroup);
1947                         pr_cont(": active=%u donating=%u after=%u\n",
1948                                 iocg->hweight_active, iocg->hweight_donating,
1949                                 iocg->hweight_after_donation);
1950                 }
1951         }
1952
1953         /*
1954          * Calculate the global donation rate (gamma) - the rate to adjust
1955          * non-donating budgets by.
1956          *
1957          * No need to use 64bit multiplication here as the first operand is
1958          * guaranteed to be smaller than WEIGHT_ONE (1<<16).
1959          *
1960          * We know that there are beneficiary nodes and the sum of the donating
1961          * hweights can't be whole; however, due to the round-ups during hweight
1962          * calculations, root_iocg->hweight_donating might still end up equal to
1963          * or greater than whole. Limit the range when calculating the divider.
1964          *
1965          * gamma = (1 - t_r') / (1 - t_r)
1966          */
1967         gamma = DIV_ROUND_UP(
1968                 (WEIGHT_ONE - root_iocg->hweight_after_donation) * WEIGHT_ONE,
1969                 WEIGHT_ONE - min_t(u32, root_iocg->hweight_donating, WEIGHT_ONE - 1));
1970
1971         /*
1972          * Calculate adjusted hwi, child_adjusted_sum and inuse for the inner
1973          * nodes.
1974          */
1975         list_for_each_entry(iocg, &inner_walk, walk_list) {
1976                 struct ioc_gq *parent;
1977                 u32 inuse, wpt, wptp;
1978                 u64 st, sf;
1979
1980                 if (iocg->level == 0) {
1981                         /* adjusted weight sum for 1st level: s' = s * b_pf / b'_pf */
1982                         iocg->child_adjusted_sum = DIV64_U64_ROUND_UP(
1983                                 iocg->child_active_sum * (WEIGHT_ONE - iocg->hweight_donating),
1984                                 WEIGHT_ONE - iocg->hweight_after_donation);
1985                         continue;
1986                 }
1987
1988                 parent = iocg->ancestors[iocg->level - 1];
1989
1990                 /* b' = gamma * b_f + b_t' */
1991                 iocg->hweight_inuse = DIV64_U64_ROUND_UP(
1992                         (u64)gamma * (iocg->hweight_active - iocg->hweight_donating),
1993                         WEIGHT_ONE) + iocg->hweight_after_donation;
1994
1995                 /* w' = s' * b' / b'_p */
1996                 inuse = DIV64_U64_ROUND_UP(
1997                         (u64)parent->child_adjusted_sum * iocg->hweight_inuse,
1998                         parent->hweight_inuse);
1999
2000                 /* adjusted weight sum for children: s' = s_f + s_t * w'_pt / w_pt */
2001                 st = DIV64_U64_ROUND_UP(
2002                         iocg->child_active_sum * iocg->hweight_donating,
2003                         iocg->hweight_active);
2004                 sf = iocg->child_active_sum - st;
2005                 wpt = DIV64_U64_ROUND_UP(
2006                         (u64)iocg->active * iocg->hweight_donating,
2007                         iocg->hweight_active);
2008                 wptp = DIV64_U64_ROUND_UP(
2009                         (u64)inuse * iocg->hweight_after_donation,
2010                         iocg->hweight_inuse);
2011
2012                 iocg->child_adjusted_sum = sf + DIV64_U64_ROUND_UP(st * wptp, wpt);
2013         }
2014
2015         /*
2016          * All inner nodes now have ->hweight_inuse and ->child_adjusted_sum and
2017          * we can finally determine leaf adjustments.
2018          */
2019         list_for_each_entry(iocg, surpluses, surplus_list) {
2020                 struct ioc_gq *parent = iocg->ancestors[iocg->level - 1];
2021                 u32 inuse;
2022
2023                 /*
2024                  * In-debt iocgs participated in the donation calculation with
2025                  * the minimum target hweight_inuse. Configuring inuse
2026                  * accordingly would work fine but debt handling expects
2027                  * @iocg->inuse stay at the minimum and we don't wanna
2028                  * interfere.
2029                  */
2030                 if (iocg->abs_vdebt) {
2031                         WARN_ON_ONCE(iocg->inuse > 1);
2032                         continue;
2033                 }
2034
2035                 /* w' = s' * b' / b'_p, note that b' == b'_t for donating leaves */
2036                 inuse = DIV64_U64_ROUND_UP(
2037                         parent->child_adjusted_sum * iocg->hweight_after_donation,
2038                         parent->hweight_inuse);
2039
2040                 TRACE_IOCG_PATH(inuse_transfer, iocg, now,
2041                                 iocg->inuse, inuse,
2042                                 iocg->hweight_inuse,
2043                                 iocg->hweight_after_donation);
2044
2045                 __propagate_weights(iocg, iocg->active, inuse, true, now);
2046         }
2047
2048         /* walk list should be dissolved after use */
2049         list_for_each_entry_safe(iocg, tiocg, &inner_walk, walk_list)
2050                 list_del_init(&iocg->walk_list);
2051 }
2052
2053 /*
2054  * A low weight iocg can amass a large amount of debt, for example, when
2055  * anonymous memory gets reclaimed aggressively. If the system has a lot of
2056  * memory paired with a slow IO device, the debt can span multiple seconds or
2057  * more. If there are no other subsequent IO issuers, the in-debt iocg may end
2058  * up blocked paying its debt while the IO device is idle.
2059  *
2060  * The following protects against such cases. If the device has been
2061  * sufficiently idle for a while, the debts are halved and delays are
2062  * recalculated.
2063  */
2064 static void ioc_forgive_debts(struct ioc *ioc, u64 usage_us_sum, int nr_debtors,
2065                               struct ioc_now *now)
2066 {
2067         struct ioc_gq *iocg;
2068         u64 dur, usage_pct, nr_cycles;
2069
2070         /* if no debtor, reset the cycle */
2071         if (!nr_debtors) {
2072                 ioc->dfgv_period_at = now->now;
2073                 ioc->dfgv_period_rem = 0;
2074                 ioc->dfgv_usage_us_sum = 0;
2075                 return;
2076         }
2077
2078         /*
2079          * Debtors can pass through a lot of writes choking the device and we
2080          * don't want to be forgiving debts while the device is struggling from
2081          * write bursts. If we're missing latency targets, consider the device
2082          * fully utilized.
2083          */
2084         if (ioc->busy_level > 0)
2085                 usage_us_sum = max_t(u64, usage_us_sum, ioc->period_us);
2086
2087         ioc->dfgv_usage_us_sum += usage_us_sum;
2088         if (time_before64(now->now, ioc->dfgv_period_at + DFGV_PERIOD))
2089                 return;
2090
2091         /*
2092          * At least DFGV_PERIOD has passed since the last period. Calculate the
2093          * average usage and reset the period counters.
2094          */
2095         dur = now->now - ioc->dfgv_period_at;
2096         usage_pct = div64_u64(100 * ioc->dfgv_usage_us_sum, dur);
2097
2098         ioc->dfgv_period_at = now->now;
2099         ioc->dfgv_usage_us_sum = 0;
2100
2101         /* if was too busy, reset everything */
2102         if (usage_pct > DFGV_USAGE_PCT) {
2103                 ioc->dfgv_period_rem = 0;
2104                 return;
2105         }
2106
2107         /*
2108          * Usage is lower than threshold. Let's forgive some debts. Debt
2109          * forgiveness runs off of the usual ioc timer but its period usually
2110          * doesn't match ioc's. Compensate the difference by performing the
2111          * reduction as many times as would fit in the duration since the last
2112          * run and carrying over the left-over duration in @ioc->dfgv_period_rem
2113          * - if ioc period is 75% of DFGV_PERIOD, one out of three consecutive
2114          * reductions is doubled.
2115          */
2116         nr_cycles = dur + ioc->dfgv_period_rem;
2117         ioc->dfgv_period_rem = do_div(nr_cycles, DFGV_PERIOD);
2118
2119         list_for_each_entry(iocg, &ioc->active_iocgs, active_list) {
2120                 u64 __maybe_unused old_debt, __maybe_unused old_delay;
2121
2122                 if (!iocg->abs_vdebt && !iocg->delay)
2123                         continue;
2124
2125                 spin_lock(&iocg->waitq.lock);
2126
2127                 old_debt = iocg->abs_vdebt;
2128                 old_delay = iocg->delay;
2129
2130                 if (iocg->abs_vdebt)
2131                         iocg->abs_vdebt = iocg->abs_vdebt >> nr_cycles ?: 1;
2132                 if (iocg->delay)
2133                         iocg->delay = iocg->delay >> nr_cycles ?: 1;
2134
2135                 iocg_kick_waitq(iocg, true, now);
2136
2137                 TRACE_IOCG_PATH(iocg_forgive_debt, iocg, now, usage_pct,
2138                                 old_debt, iocg->abs_vdebt,
2139                                 old_delay, iocg->delay);
2140
2141                 spin_unlock(&iocg->waitq.lock);
2142         }
2143 }
2144
2145 /*
2146  * Check the active iocgs' state to avoid oversleeping and deactive
2147  * idle iocgs.
2148  *
2149  * Since waiters determine the sleep durations based on the vrate
2150  * they saw at the time of sleep, if vrate has increased, some
2151  * waiters could be sleeping for too long. Wake up tardy waiters
2152  * which should have woken up in the last period and expire idle
2153  * iocgs.
2154  */
2155 static int ioc_check_iocgs(struct ioc *ioc, struct ioc_now *now)
2156 {
2157         int nr_debtors = 0;
2158         struct ioc_gq *iocg, *tiocg;
2159
2160         list_for_each_entry_safe(iocg, tiocg, &ioc->active_iocgs, active_list) {
2161                 if (!waitqueue_active(&iocg->waitq) && !iocg->abs_vdebt &&
2162                     !iocg->delay && !iocg_is_idle(iocg))
2163                         continue;
2164
2165                 spin_lock(&iocg->waitq.lock);
2166
2167                 /* flush wait and indebt stat deltas */
2168                 if (iocg->wait_since) {
2169                         iocg->stat.wait_us += now->now - iocg->wait_since;
2170                         iocg->wait_since = now->now;
2171                 }
2172                 if (iocg->indebt_since) {
2173                         iocg->stat.indebt_us +=
2174                                 now->now - iocg->indebt_since;
2175                         iocg->indebt_since = now->now;
2176                 }
2177                 if (iocg->indelay_since) {
2178                         iocg->stat.indelay_us +=
2179                                 now->now - iocg->indelay_since;
2180                         iocg->indelay_since = now->now;
2181                 }
2182
2183                 if (waitqueue_active(&iocg->waitq) || iocg->abs_vdebt ||
2184                     iocg->delay) {
2185                         /* might be oversleeping vtime / hweight changes, kick */
2186                         iocg_kick_waitq(iocg, true, now);
2187                         if (iocg->abs_vdebt || iocg->delay)
2188                                 nr_debtors++;
2189                 } else if (iocg_is_idle(iocg)) {
2190                         /* no waiter and idle, deactivate */
2191                         u64 vtime = atomic64_read(&iocg->vtime);
2192                         s64 excess;
2193
2194                         /*
2195                          * @iocg has been inactive for a full duration and will
2196                          * have a high budget. Account anything above target as
2197                          * error and throw away. On reactivation, it'll start
2198                          * with the target budget.
2199                          */
2200                         excess = now->vnow - vtime - ioc->margins.target;
2201                         if (excess > 0) {
2202                                 u32 old_hwi;
2203
2204                                 current_hweight(iocg, NULL, &old_hwi);
2205                                 ioc->vtime_err -= div64_u64(excess * old_hwi,
2206                                                             WEIGHT_ONE);
2207                         }
2208
2209                         TRACE_IOCG_PATH(iocg_idle, iocg, now,
2210                                         atomic64_read(&iocg->active_period),
2211                                         atomic64_read(&ioc->cur_period), vtime);
2212                         __propagate_weights(iocg, 0, 0, false, now);
2213                         list_del_init(&iocg->active_list);
2214                 }
2215
2216                 spin_unlock(&iocg->waitq.lock);
2217         }
2218
2219         commit_weights(ioc);
2220         return nr_debtors;
2221 }
2222
2223 static void ioc_timer_fn(struct timer_list *timer)
2224 {
2225         struct ioc *ioc = container_of(timer, struct ioc, timer);
2226         struct ioc_gq *iocg, *tiocg;
2227         struct ioc_now now;
2228         LIST_HEAD(surpluses);
2229         int nr_debtors, nr_shortages = 0, nr_lagging = 0;
2230         u64 usage_us_sum = 0;
2231         u32 ppm_rthr;
2232         u32 ppm_wthr;
2233         u32 missed_ppm[2], rq_wait_pct;
2234         u64 period_vtime;
2235         int prev_busy_level;
2236
2237         /* how were the latencies during the period? */
2238         ioc_lat_stat(ioc, missed_ppm, &rq_wait_pct);
2239
2240         /* take care of active iocgs */
2241         spin_lock_irq(&ioc->lock);
2242
2243         ppm_rthr = MILLION - ioc->params.qos[QOS_RPPM];
2244         ppm_wthr = MILLION - ioc->params.qos[QOS_WPPM];
2245         ioc_now(ioc, &now);
2246
2247         period_vtime = now.vnow - ioc->period_at_vtime;
2248         if (WARN_ON_ONCE(!period_vtime)) {
2249                 spin_unlock_irq(&ioc->lock);
2250                 return;
2251         }
2252
2253         nr_debtors = ioc_check_iocgs(ioc, &now);
2254
2255         /*
2256          * Wait and indebt stat are flushed above and the donation calculation
2257          * below needs updated usage stat. Let's bring stat up-to-date.
2258          */
2259         iocg_flush_stat(&ioc->active_iocgs, &now);
2260
2261         /* calc usage and see whether some weights need to be moved around */
2262         list_for_each_entry(iocg, &ioc->active_iocgs, active_list) {
2263                 u64 vdone, vtime, usage_us;
2264                 u32 hw_active, hw_inuse;
2265
2266                 /*
2267                  * Collect unused and wind vtime closer to vnow to prevent
2268                  * iocgs from accumulating a large amount of budget.
2269                  */
2270                 vdone = atomic64_read(&iocg->done_vtime);
2271                 vtime = atomic64_read(&iocg->vtime);
2272                 current_hweight(iocg, &hw_active, &hw_inuse);
2273
2274                 /*
2275                  * Latency QoS detection doesn't account for IOs which are
2276                  * in-flight for longer than a period.  Detect them by
2277                  * comparing vdone against period start.  If lagging behind
2278                  * IOs from past periods, don't increase vrate.
2279                  */
2280                 if ((ppm_rthr != MILLION || ppm_wthr != MILLION) &&
2281                     !atomic_read(&iocg_to_blkg(iocg)->use_delay) &&
2282                     time_after64(vtime, vdone) &&
2283                     time_after64(vtime, now.vnow -
2284                                  MAX_LAGGING_PERIODS * period_vtime) &&
2285                     time_before64(vdone, now.vnow - period_vtime))
2286                         nr_lagging++;
2287
2288                 /*
2289                  * Determine absolute usage factoring in in-flight IOs to avoid
2290                  * high-latency completions appearing as idle.
2291                  */
2292                 usage_us = iocg->usage_delta_us;
2293                 usage_us_sum += usage_us;
2294
2295                 /* see whether there's surplus vtime */
2296                 WARN_ON_ONCE(!list_empty(&iocg->surplus_list));
2297                 if (hw_inuse < hw_active ||
2298                     (!waitqueue_active(&iocg->waitq) &&
2299                      time_before64(vtime, now.vnow - ioc->margins.low))) {
2300                         u32 hwa, old_hwi, hwm, new_hwi, usage;
2301                         u64 usage_dur;
2302
2303                         if (vdone != vtime) {
2304                                 u64 inflight_us = DIV64_U64_ROUND_UP(
2305                                         cost_to_abs_cost(vtime - vdone, hw_inuse),
2306                                         ioc->vtime_base_rate);
2307
2308                                 usage_us = max(usage_us, inflight_us);
2309                         }
2310
2311                         /* convert to hweight based usage ratio */
2312                         if (time_after64(iocg->activated_at, ioc->period_at))
2313                                 usage_dur = max_t(u64, now.now - iocg->activated_at, 1);
2314                         else
2315                                 usage_dur = max_t(u64, now.now - ioc->period_at, 1);
2316
2317                         usage = clamp_t(u32,
2318                                 DIV64_U64_ROUND_UP(usage_us * WEIGHT_ONE,
2319                                                    usage_dur),
2320                                 1, WEIGHT_ONE);
2321
2322                         /*
2323                          * Already donating or accumulated enough to start.
2324                          * Determine the donation amount.
2325                          */
2326                         current_hweight(iocg, &hwa, &old_hwi);
2327                         hwm = current_hweight_max(iocg);
2328                         new_hwi = hweight_after_donation(iocg, old_hwi, hwm,
2329                                                          usage, &now);
2330                         /*
2331                          * Donation calculation assumes hweight_after_donation
2332                          * to be positive, a condition that a donor w/ hwa < 2
2333                          * can't meet. Don't bother with donation if hwa is
2334                          * below 2. It's not gonna make a meaningful difference
2335                          * anyway.
2336                          */
2337                         if (new_hwi < hwm && hwa >= 2) {
2338                                 iocg->hweight_donating = hwa;
2339                                 iocg->hweight_after_donation = new_hwi;
2340                                 list_add(&iocg->surplus_list, &surpluses);
2341                         } else if (!iocg->abs_vdebt) {
2342                                 /*
2343                                  * @iocg doesn't have enough to donate. Reset
2344                                  * its inuse to active.
2345                                  *
2346                                  * Don't reset debtors as their inuse's are
2347                                  * owned by debt handling. This shouldn't affect
2348                                  * donation calculuation in any meaningful way
2349                                  * as @iocg doesn't have a meaningful amount of
2350                                  * share anyway.
2351                                  */
2352                                 TRACE_IOCG_PATH(inuse_shortage, iocg, &now,
2353                                                 iocg->inuse, iocg->active,
2354                                                 iocg->hweight_inuse, new_hwi);
2355
2356                                 __propagate_weights(iocg, iocg->active,
2357                                                     iocg->active, true, &now);
2358                                 nr_shortages++;
2359                         }
2360                 } else {
2361                         /* genuinely short on vtime */
2362                         nr_shortages++;
2363                 }
2364         }
2365
2366         if (!list_empty(&surpluses) && nr_shortages)
2367                 transfer_surpluses(&surpluses, &now);
2368
2369         commit_weights(ioc);
2370
2371         /* surplus list should be dissolved after use */
2372         list_for_each_entry_safe(iocg, tiocg, &surpluses, surplus_list)
2373                 list_del_init(&iocg->surplus_list);
2374
2375         /*
2376          * If q is getting clogged or we're missing too much, we're issuing
2377          * too much IO and should lower vtime rate.  If we're not missing
2378          * and experiencing shortages but not surpluses, we're too stingy
2379          * and should increase vtime rate.
2380          */
2381         prev_busy_level = ioc->busy_level;
2382         if (rq_wait_pct > RQ_WAIT_BUSY_PCT ||
2383             missed_ppm[READ] > ppm_rthr ||
2384             missed_ppm[WRITE] > ppm_wthr) {
2385                 /* clearly missing QoS targets, slow down vrate */
2386                 ioc->busy_level = max(ioc->busy_level, 0);
2387                 ioc->busy_level++;
2388         } else if (rq_wait_pct <= RQ_WAIT_BUSY_PCT * UNBUSY_THR_PCT / 100 &&
2389                    missed_ppm[READ] <= ppm_rthr * UNBUSY_THR_PCT / 100 &&
2390                    missed_ppm[WRITE] <= ppm_wthr * UNBUSY_THR_PCT / 100) {
2391                 /* QoS targets are being met with >25% margin */
2392                 if (nr_shortages) {
2393                         /*
2394                          * We're throttling while the device has spare
2395                          * capacity.  If vrate was being slowed down, stop.
2396                          */
2397                         ioc->busy_level = min(ioc->busy_level, 0);
2398
2399                         /*
2400                          * If there are IOs spanning multiple periods, wait
2401                          * them out before pushing the device harder.
2402                          */
2403                         if (!nr_lagging)
2404                                 ioc->busy_level--;
2405                 } else {
2406                         /*
2407                          * Nobody is being throttled and the users aren't
2408                          * issuing enough IOs to saturate the device.  We
2409                          * simply don't know how close the device is to
2410                          * saturation.  Coast.
2411                          */
2412                         ioc->busy_level = 0;
2413                 }
2414         } else {
2415                 /* inside the hysterisis margin, we're good */
2416                 ioc->busy_level = 0;
2417         }
2418
2419         ioc->busy_level = clamp(ioc->busy_level, -1000, 1000);
2420
2421         ioc_adjust_base_vrate(ioc, rq_wait_pct, nr_lagging, nr_shortages,
2422                               prev_busy_level, missed_ppm);
2423
2424         ioc_refresh_params(ioc, false);
2425
2426         ioc_forgive_debts(ioc, usage_us_sum, nr_debtors, &now);
2427
2428         /*
2429          * This period is done.  Move onto the next one.  If nothing's
2430          * going on with the device, stop the timer.
2431          */
2432         atomic64_inc(&ioc->cur_period);
2433
2434         if (ioc->running != IOC_STOP) {
2435                 if (!list_empty(&ioc->active_iocgs)) {
2436                         ioc_start_period(ioc, &now);
2437                 } else {
2438                         ioc->busy_level = 0;
2439                         ioc->vtime_err = 0;
2440                         ioc->running = IOC_IDLE;
2441                 }
2442
2443                 ioc_refresh_vrate(ioc, &now);
2444         }
2445
2446         spin_unlock_irq(&ioc->lock);
2447 }
2448
2449 static u64 adjust_inuse_and_calc_cost(struct ioc_gq *iocg, u64 vtime,
2450                                       u64 abs_cost, struct ioc_now *now)
2451 {
2452         struct ioc *ioc = iocg->ioc;
2453         struct ioc_margins *margins = &ioc->margins;
2454         u32 __maybe_unused old_inuse = iocg->inuse, __maybe_unused old_hwi;
2455         u32 hwi, adj_step;
2456         s64 margin;
2457         u64 cost, new_inuse;
2458         unsigned long flags;
2459
2460         current_hweight(iocg, NULL, &hwi);
2461         old_hwi = hwi;
2462         cost = abs_cost_to_cost(abs_cost, hwi);
2463         margin = now->vnow - vtime - cost;
2464
2465         /* debt handling owns inuse for debtors */
2466         if (iocg->abs_vdebt)
2467                 return cost;
2468
2469         /*
2470          * We only increase inuse during period and do so if the margin has
2471          * deteriorated since the previous adjustment.
2472          */
2473         if (margin >= iocg->saved_margin || margin >= margins->low ||
2474             iocg->inuse == iocg->active)
2475                 return cost;
2476
2477         spin_lock_irqsave(&ioc->lock, flags);
2478
2479         /* we own inuse only when @iocg is in the normal active state */
2480         if (iocg->abs_vdebt || list_empty(&iocg->active_list)) {
2481                 spin_unlock_irqrestore(&ioc->lock, flags);
2482                 return cost;
2483         }
2484
2485         /*
2486          * Bump up inuse till @abs_cost fits in the existing budget.
2487          * adj_step must be determined after acquiring ioc->lock - we might
2488          * have raced and lost to another thread for activation and could
2489          * be reading 0 iocg->active before ioc->lock which will lead to
2490          * infinite loop.
2491          */
2492         new_inuse = iocg->inuse;
2493         adj_step = DIV_ROUND_UP(iocg->active * INUSE_ADJ_STEP_PCT, 100);
2494         do {
2495                 new_inuse = new_inuse + adj_step;
2496                 propagate_weights(iocg, iocg->active, new_inuse, true, now);
2497                 current_hweight(iocg, NULL, &hwi);
2498                 cost = abs_cost_to_cost(abs_cost, hwi);
2499         } while (time_after64(vtime + cost, now->vnow) &&
2500                  iocg->inuse != iocg->active);
2501
2502         spin_unlock_irqrestore(&ioc->lock, flags);
2503
2504         TRACE_IOCG_PATH(inuse_adjust, iocg, now,
2505                         old_inuse, iocg->inuse, old_hwi, hwi);
2506
2507         return cost;
2508 }
2509
2510 static void calc_vtime_cost_builtin(struct bio *bio, struct ioc_gq *iocg,
2511                                     bool is_merge, u64 *costp)
2512 {
2513         struct ioc *ioc = iocg->ioc;
2514         u64 coef_seqio, coef_randio, coef_page;
2515         u64 pages = max_t(u64, bio_sectors(bio) >> IOC_SECT_TO_PAGE_SHIFT, 1);
2516         u64 seek_pages = 0;
2517         u64 cost = 0;
2518
2519         /* Can't calculate cost for empty bio */
2520         if (!bio->bi_iter.bi_size)
2521                 goto out;
2522
2523         switch (bio_op(bio)) {
2524         case REQ_OP_READ:
2525                 coef_seqio      = ioc->params.lcoefs[LCOEF_RSEQIO];
2526                 coef_randio     = ioc->params.lcoefs[LCOEF_RRANDIO];
2527                 coef_page       = ioc->params.lcoefs[LCOEF_RPAGE];
2528                 break;
2529         case REQ_OP_WRITE:
2530                 coef_seqio      = ioc->params.lcoefs[LCOEF_WSEQIO];
2531                 coef_randio     = ioc->params.lcoefs[LCOEF_WRANDIO];
2532                 coef_page       = ioc->params.lcoefs[LCOEF_WPAGE];
2533                 break;
2534         default:
2535                 goto out;
2536         }
2537
2538         if (iocg->cursor) {
2539                 seek_pages = abs(bio->bi_iter.bi_sector - iocg->cursor);
2540                 seek_pages >>= IOC_SECT_TO_PAGE_SHIFT;
2541         }
2542
2543         if (!is_merge) {
2544                 if (seek_pages > LCOEF_RANDIO_PAGES) {
2545                         cost += coef_randio;
2546                 } else {
2547                         cost += coef_seqio;
2548                 }
2549         }
2550         cost += pages * coef_page;
2551 out:
2552         *costp = cost;
2553 }
2554
2555 static u64 calc_vtime_cost(struct bio *bio, struct ioc_gq *iocg, bool is_merge)
2556 {
2557         u64 cost;
2558
2559         calc_vtime_cost_builtin(bio, iocg, is_merge, &cost);
2560         return cost;
2561 }
2562
2563 static void calc_size_vtime_cost_builtin(struct request *rq, struct ioc *ioc,
2564                                          u64 *costp)
2565 {
2566         unsigned int pages = blk_rq_stats_sectors(rq) >> IOC_SECT_TO_PAGE_SHIFT;
2567
2568         switch (req_op(rq)) {
2569         case REQ_OP_READ:
2570                 *costp = pages * ioc->params.lcoefs[LCOEF_RPAGE];
2571                 break;
2572         case REQ_OP_WRITE:
2573                 *costp = pages * ioc->params.lcoefs[LCOEF_WPAGE];
2574                 break;
2575         default:
2576                 *costp = 0;
2577         }
2578 }
2579
2580 static u64 calc_size_vtime_cost(struct request *rq, struct ioc *ioc)
2581 {
2582         u64 cost;
2583
2584         calc_size_vtime_cost_builtin(rq, ioc, &cost);
2585         return cost;
2586 }
2587
2588 static void ioc_rqos_throttle(struct rq_qos *rqos, struct bio *bio)
2589 {
2590         struct blkcg_gq *blkg = bio->bi_blkg;
2591         struct ioc *ioc = rqos_to_ioc(rqos);
2592         struct ioc_gq *iocg = blkg_to_iocg(blkg);
2593         struct ioc_now now;
2594         struct iocg_wait wait;
2595         u64 abs_cost, cost, vtime;
2596         bool use_debt, ioc_locked;
2597         unsigned long flags;
2598
2599         /* bypass IOs if disabled, still initializing, or for root cgroup */
2600         if (!ioc->enabled || !iocg || !iocg->level)
2601                 return;
2602
2603         /* calculate the absolute vtime cost */
2604         abs_cost = calc_vtime_cost(bio, iocg, false);
2605         if (!abs_cost)
2606                 return;
2607
2608         if (!iocg_activate(iocg, &now))
2609                 return;
2610
2611         iocg->cursor = bio_end_sector(bio);
2612         vtime = atomic64_read(&iocg->vtime);
2613         cost = adjust_inuse_and_calc_cost(iocg, vtime, abs_cost, &now);
2614
2615         /*
2616          * If no one's waiting and within budget, issue right away.  The
2617          * tests are racy but the races aren't systemic - we only miss once
2618          * in a while which is fine.
2619          */
2620         if (!waitqueue_active(&iocg->waitq) && !iocg->abs_vdebt &&
2621             time_before_eq64(vtime + cost, now.vnow)) {
2622                 iocg_commit_bio(iocg, bio, abs_cost, cost);
2623                 return;
2624         }
2625
2626         /*
2627          * We're over budget. This can be handled in two ways. IOs which may
2628          * cause priority inversions are punted to @ioc->aux_iocg and charged as
2629          * debt. Otherwise, the issuer is blocked on @iocg->waitq. Debt handling
2630          * requires @ioc->lock, waitq handling @iocg->waitq.lock. Determine
2631          * whether debt handling is needed and acquire locks accordingly.
2632          */
2633         use_debt = bio_issue_as_root_blkg(bio) || fatal_signal_pending(current);
2634         ioc_locked = use_debt || READ_ONCE(iocg->abs_vdebt);
2635 retry_lock:
2636         iocg_lock(iocg, ioc_locked, &flags);
2637
2638         /*
2639          * @iocg must stay activated for debt and waitq handling. Deactivation
2640          * is synchronized against both ioc->lock and waitq.lock and we won't
2641          * get deactivated as long as we're waiting or has debt, so we're good
2642          * if we're activated here. In the unlikely cases that we aren't, just
2643          * issue the IO.
2644          */
2645         if (unlikely(list_empty(&iocg->active_list))) {
2646                 iocg_unlock(iocg, ioc_locked, &flags);
2647                 iocg_commit_bio(iocg, bio, abs_cost, cost);
2648                 return;
2649         }
2650
2651         /*
2652          * We're over budget. If @bio has to be issued regardless, remember
2653          * the abs_cost instead of advancing vtime. iocg_kick_waitq() will pay
2654          * off the debt before waking more IOs.
2655          *
2656          * This way, the debt is continuously paid off each period with the
2657          * actual budget available to the cgroup. If we just wound vtime, we
2658          * would incorrectly use the current hw_inuse for the entire amount
2659          * which, for example, can lead to the cgroup staying blocked for a
2660          * long time even with substantially raised hw_inuse.
2661          *
2662          * An iocg with vdebt should stay online so that the timer can keep
2663          * deducting its vdebt and [de]activate use_delay mechanism
2664          * accordingly. We don't want to race against the timer trying to
2665          * clear them and leave @iocg inactive w/ dangling use_delay heavily
2666          * penalizing the cgroup and its descendants.
2667          */
2668         if (use_debt) {
2669                 iocg_incur_debt(iocg, abs_cost, &now);
2670                 if (iocg_kick_delay(iocg, &now))
2671                         blkcg_schedule_throttle(rqos->disk,
2672                                         (bio->bi_opf & REQ_SWAP) == REQ_SWAP);
2673                 iocg_unlock(iocg, ioc_locked, &flags);
2674                 return;
2675         }
2676
2677         /* guarantee that iocgs w/ waiters have maximum inuse */
2678         if (!iocg->abs_vdebt && iocg->inuse != iocg->active) {
2679                 if (!ioc_locked) {
2680                         iocg_unlock(iocg, false, &flags);
2681                         ioc_locked = true;
2682                         goto retry_lock;
2683                 }
2684                 propagate_weights(iocg, iocg->active, iocg->active, true,
2685                                   &now);
2686         }
2687
2688         /*
2689          * Append self to the waitq and schedule the wakeup timer if we're
2690          * the first waiter.  The timer duration is calculated based on the
2691          * current vrate.  vtime and hweight changes can make it too short
2692          * or too long.  Each wait entry records the absolute cost it's
2693          * waiting for to allow re-evaluation using a custom wait entry.
2694          *
2695          * If too short, the timer simply reschedules itself.  If too long,
2696          * the period timer will notice and trigger wakeups.
2697          *
2698          * All waiters are on iocg->waitq and the wait states are
2699          * synchronized using waitq.lock.
2700          */
2701         init_waitqueue_func_entry(&wait.wait, iocg_wake_fn);
2702         wait.wait.private = current;
2703         wait.bio = bio;
2704         wait.abs_cost = abs_cost;
2705         wait.committed = false; /* will be set true by waker */
2706
2707         __add_wait_queue_entry_tail(&iocg->waitq, &wait.wait);
2708         iocg_kick_waitq(iocg, ioc_locked, &now);
2709
2710         iocg_unlock(iocg, ioc_locked, &flags);
2711
2712         while (true) {
2713                 set_current_state(TASK_UNINTERRUPTIBLE);
2714                 if (wait.committed)
2715                         break;
2716                 io_schedule();
2717         }
2718
2719         /* waker already committed us, proceed */
2720         finish_wait(&iocg->waitq, &wait.wait);
2721 }
2722
2723 static void ioc_rqos_merge(struct rq_qos *rqos, struct request *rq,
2724                            struct bio *bio)
2725 {
2726         struct ioc_gq *iocg = blkg_to_iocg(bio->bi_blkg);
2727         struct ioc *ioc = rqos_to_ioc(rqos);
2728         sector_t bio_end = bio_end_sector(bio);
2729         struct ioc_now now;
2730         u64 vtime, abs_cost, cost;
2731         unsigned long flags;
2732
2733         /* bypass if disabled, still initializing, or for root cgroup */
2734         if (!ioc->enabled || !iocg || !iocg->level)
2735                 return;
2736
2737         abs_cost = calc_vtime_cost(bio, iocg, true);
2738         if (!abs_cost)
2739                 return;
2740
2741         ioc_now(ioc, &now);
2742
2743         vtime = atomic64_read(&iocg->vtime);
2744         cost = adjust_inuse_and_calc_cost(iocg, vtime, abs_cost, &now);
2745
2746         /* update cursor if backmerging into the request at the cursor */
2747         if (blk_rq_pos(rq) < bio_end &&
2748             blk_rq_pos(rq) + blk_rq_sectors(rq) == iocg->cursor)
2749                 iocg->cursor = bio_end;
2750
2751         /*
2752          * Charge if there's enough vtime budget and the existing request has
2753          * cost assigned.
2754          */
2755         if (rq->bio && rq->bio->bi_iocost_cost &&
2756             time_before_eq64(atomic64_read(&iocg->vtime) + cost, now.vnow)) {
2757                 iocg_commit_bio(iocg, bio, abs_cost, cost);
2758                 return;
2759         }
2760
2761         /*
2762          * Otherwise, account it as debt if @iocg is online, which it should
2763          * be for the vast majority of cases. See debt handling in
2764          * ioc_rqos_throttle() for details.
2765          */
2766         spin_lock_irqsave(&ioc->lock, flags);
2767         spin_lock(&iocg->waitq.lock);
2768
2769         if (likely(!list_empty(&iocg->active_list))) {
2770                 iocg_incur_debt(iocg, abs_cost, &now);
2771                 if (iocg_kick_delay(iocg, &now))
2772                         blkcg_schedule_throttle(rqos->disk,
2773                                         (bio->bi_opf & REQ_SWAP) == REQ_SWAP);
2774         } else {
2775                 iocg_commit_bio(iocg, bio, abs_cost, cost);
2776         }
2777
2778         spin_unlock(&iocg->waitq.lock);
2779         spin_unlock_irqrestore(&ioc->lock, flags);
2780 }
2781
2782 static void ioc_rqos_done_bio(struct rq_qos *rqos, struct bio *bio)
2783 {
2784         struct ioc_gq *iocg = blkg_to_iocg(bio->bi_blkg);
2785
2786         if (iocg && bio->bi_iocost_cost)
2787                 atomic64_add(bio->bi_iocost_cost, &iocg->done_vtime);
2788 }
2789
2790 static void ioc_rqos_done(struct rq_qos *rqos, struct request *rq)
2791 {
2792         struct ioc *ioc = rqos_to_ioc(rqos);
2793         struct ioc_pcpu_stat *ccs;
2794         u64 on_q_ns, rq_wait_ns, size_nsec;
2795         int pidx, rw;
2796
2797         if (!ioc->enabled || !rq->alloc_time_ns || !rq->start_time_ns)
2798                 return;
2799
2800         switch (req_op(rq)) {
2801         case REQ_OP_READ:
2802                 pidx = QOS_RLAT;
2803                 rw = READ;
2804                 break;
2805         case REQ_OP_WRITE:
2806                 pidx = QOS_WLAT;
2807                 rw = WRITE;
2808                 break;
2809         default:
2810                 return;
2811         }
2812
2813         on_q_ns = ktime_get_ns() - rq->alloc_time_ns;
2814         rq_wait_ns = rq->start_time_ns - rq->alloc_time_ns;
2815         size_nsec = div64_u64(calc_size_vtime_cost(rq, ioc), VTIME_PER_NSEC);
2816
2817         ccs = get_cpu_ptr(ioc->pcpu_stat);
2818
2819         if (on_q_ns <= size_nsec ||
2820             on_q_ns - size_nsec <= ioc->params.qos[pidx] * NSEC_PER_USEC)
2821                 local_inc(&ccs->missed[rw].nr_met);
2822         else
2823                 local_inc(&ccs->missed[rw].nr_missed);
2824
2825         local64_add(rq_wait_ns, &ccs->rq_wait_ns);
2826
2827         put_cpu_ptr(ccs);
2828 }
2829
2830 static void ioc_rqos_queue_depth_changed(struct rq_qos *rqos)
2831 {
2832         struct ioc *ioc = rqos_to_ioc(rqos);
2833
2834         spin_lock_irq(&ioc->lock);
2835         ioc_refresh_params(ioc, false);
2836         spin_unlock_irq(&ioc->lock);
2837 }
2838
2839 static void ioc_rqos_exit(struct rq_qos *rqos)
2840 {
2841         struct ioc *ioc = rqos_to_ioc(rqos);
2842
2843         blkcg_deactivate_policy(rqos->disk, &blkcg_policy_iocost);
2844
2845         spin_lock_irq(&ioc->lock);
2846         ioc->running = IOC_STOP;
2847         spin_unlock_irq(&ioc->lock);
2848
2849         timer_shutdown_sync(&ioc->timer);
2850         free_percpu(ioc->pcpu_stat);
2851         kfree(ioc);
2852 }
2853
2854 static const struct rq_qos_ops ioc_rqos_ops = {
2855         .throttle = ioc_rqos_throttle,
2856         .merge = ioc_rqos_merge,
2857         .done_bio = ioc_rqos_done_bio,
2858         .done = ioc_rqos_done,
2859         .queue_depth_changed = ioc_rqos_queue_depth_changed,
2860         .exit = ioc_rqos_exit,
2861 };
2862
2863 static int blk_iocost_init(struct gendisk *disk)
2864 {
2865         struct ioc *ioc;
2866         int i, cpu, ret;
2867
2868         ioc = kzalloc(sizeof(*ioc), GFP_KERNEL);
2869         if (!ioc)
2870                 return -ENOMEM;
2871
2872         ioc->pcpu_stat = alloc_percpu(struct ioc_pcpu_stat);
2873         if (!ioc->pcpu_stat) {
2874                 kfree(ioc);
2875                 return -ENOMEM;
2876         }
2877
2878         for_each_possible_cpu(cpu) {
2879                 struct ioc_pcpu_stat *ccs = per_cpu_ptr(ioc->pcpu_stat, cpu);
2880
2881                 for (i = 0; i < ARRAY_SIZE(ccs->missed); i++) {
2882                         local_set(&ccs->missed[i].nr_met, 0);
2883                         local_set(&ccs->missed[i].nr_missed, 0);
2884                 }
2885                 local64_set(&ccs->rq_wait_ns, 0);
2886         }
2887
2888         spin_lock_init(&ioc->lock);
2889         timer_setup(&ioc->timer, ioc_timer_fn, 0);
2890         INIT_LIST_HEAD(&ioc->active_iocgs);
2891
2892         ioc->running = IOC_IDLE;
2893         ioc->vtime_base_rate = VTIME_PER_USEC;
2894         atomic64_set(&ioc->vtime_rate, VTIME_PER_USEC);
2895         seqcount_spinlock_init(&ioc->period_seqcount, &ioc->lock);
2896         ioc->period_at = ktime_to_us(ktime_get());
2897         atomic64_set(&ioc->cur_period, 0);
2898         atomic_set(&ioc->hweight_gen, 0);
2899
2900         spin_lock_irq(&ioc->lock);
2901         ioc->autop_idx = AUTOP_INVALID;
2902         ioc_refresh_params_disk(ioc, true, disk);
2903         spin_unlock_irq(&ioc->lock);
2904
2905         /*
2906          * rqos must be added before activation to allow ioc_pd_init() to
2907          * lookup the ioc from q. This means that the rqos methods may get
2908          * called before policy activation completion, can't assume that the
2909          * target bio has an iocg associated and need to test for NULL iocg.
2910          */
2911         ret = rq_qos_add(&ioc->rqos, disk, RQ_QOS_COST, &ioc_rqos_ops);
2912         if (ret)
2913                 goto err_free_ioc;
2914
2915         ret = blkcg_activate_policy(disk, &blkcg_policy_iocost);
2916         if (ret)
2917                 goto err_del_qos;
2918         return 0;
2919
2920 err_del_qos:
2921         rq_qos_del(&ioc->rqos);
2922 err_free_ioc:
2923         free_percpu(ioc->pcpu_stat);
2924         kfree(ioc);
2925         return ret;
2926 }
2927
2928 static struct blkcg_policy_data *ioc_cpd_alloc(gfp_t gfp)
2929 {
2930         struct ioc_cgrp *iocc;
2931
2932         iocc = kzalloc(sizeof(struct ioc_cgrp), gfp);
2933         if (!iocc)
2934                 return NULL;
2935
2936         iocc->dfl_weight = CGROUP_WEIGHT_DFL * WEIGHT_ONE;
2937         return &iocc->cpd;
2938 }
2939
2940 static void ioc_cpd_free(struct blkcg_policy_data *cpd)
2941 {
2942         kfree(container_of(cpd, struct ioc_cgrp, cpd));
2943 }
2944
2945 static struct blkg_policy_data *ioc_pd_alloc(struct gendisk *disk,
2946                 struct blkcg *blkcg, gfp_t gfp)
2947 {
2948         int levels = blkcg->css.cgroup->level + 1;
2949         struct ioc_gq *iocg;
2950
2951         iocg = kzalloc_node(struct_size(iocg, ancestors, levels), gfp,
2952                             disk->node_id);
2953         if (!iocg)
2954                 return NULL;
2955
2956         iocg->pcpu_stat = alloc_percpu_gfp(struct iocg_pcpu_stat, gfp);
2957         if (!iocg->pcpu_stat) {
2958                 kfree(iocg);
2959                 return NULL;
2960         }
2961
2962         return &iocg->pd;
2963 }
2964
2965 static void ioc_pd_init(struct blkg_policy_data *pd)
2966 {
2967         struct ioc_gq *iocg = pd_to_iocg(pd);
2968         struct blkcg_gq *blkg = pd_to_blkg(&iocg->pd);
2969         struct ioc *ioc = q_to_ioc(blkg->q);
2970         struct ioc_now now;
2971         struct blkcg_gq *tblkg;
2972         unsigned long flags;
2973
2974         ioc_now(ioc, &now);
2975
2976         iocg->ioc = ioc;
2977         atomic64_set(&iocg->vtime, now.vnow);
2978         atomic64_set(&iocg->done_vtime, now.vnow);
2979         atomic64_set(&iocg->active_period, atomic64_read(&ioc->cur_period));
2980         INIT_LIST_HEAD(&iocg->active_list);
2981         INIT_LIST_HEAD(&iocg->walk_list);
2982         INIT_LIST_HEAD(&iocg->surplus_list);
2983         iocg->hweight_active = WEIGHT_ONE;
2984         iocg->hweight_inuse = WEIGHT_ONE;
2985
2986         init_waitqueue_head(&iocg->waitq);
2987         hrtimer_init(&iocg->waitq_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
2988         iocg->waitq_timer.function = iocg_waitq_timer_fn;
2989
2990         iocg->level = blkg->blkcg->css.cgroup->level;
2991
2992         for (tblkg = blkg; tblkg; tblkg = tblkg->parent) {
2993                 struct ioc_gq *tiocg = blkg_to_iocg(tblkg);
2994                 iocg->ancestors[tiocg->level] = tiocg;
2995         }
2996
2997         spin_lock_irqsave(&ioc->lock, flags);
2998         weight_updated(iocg, &now);
2999         spin_unlock_irqrestore(&ioc->lock, flags);
3000 }
3001
3002 static void ioc_pd_free(struct blkg_policy_data *pd)
3003 {
3004         struct ioc_gq *iocg = pd_to_iocg(pd);
3005         struct ioc *ioc = iocg->ioc;
3006         unsigned long flags;
3007
3008         if (ioc) {
3009                 spin_lock_irqsave(&ioc->lock, flags);
3010
3011                 if (!list_empty(&iocg->active_list)) {
3012                         struct ioc_now now;
3013
3014                         ioc_now(ioc, &now);
3015                         propagate_weights(iocg, 0, 0, false, &now);
3016                         list_del_init(&iocg->active_list);
3017                 }
3018
3019                 WARN_ON_ONCE(!list_empty(&iocg->walk_list));
3020                 WARN_ON_ONCE(!list_empty(&iocg->surplus_list));
3021
3022                 spin_unlock_irqrestore(&ioc->lock, flags);
3023
3024                 hrtimer_cancel(&iocg->waitq_timer);
3025         }
3026         free_percpu(iocg->pcpu_stat);
3027         kfree(iocg);
3028 }
3029
3030 static void ioc_pd_stat(struct blkg_policy_data *pd, struct seq_file *s)
3031 {
3032         struct ioc_gq *iocg = pd_to_iocg(pd);
3033         struct ioc *ioc = iocg->ioc;
3034
3035         if (!ioc->enabled)
3036                 return;
3037
3038         if (iocg->level == 0) {
3039                 unsigned vp10k = DIV64_U64_ROUND_CLOSEST(
3040                         ioc->vtime_base_rate * 10000,
3041                         VTIME_PER_USEC);
3042                 seq_printf(s, " cost.vrate=%u.%02u", vp10k / 100, vp10k % 100);
3043         }
3044
3045         seq_printf(s, " cost.usage=%llu", iocg->last_stat.usage_us);
3046
3047         if (blkcg_debug_stats)
3048                 seq_printf(s, " cost.wait=%llu cost.indebt=%llu cost.indelay=%llu",
3049                         iocg->last_stat.wait_us,
3050                         iocg->last_stat.indebt_us,
3051                         iocg->last_stat.indelay_us);
3052 }
3053
3054 static u64 ioc_weight_prfill(struct seq_file *sf, struct blkg_policy_data *pd,
3055                              int off)
3056 {
3057         const char *dname = blkg_dev_name(pd->blkg);
3058         struct ioc_gq *iocg = pd_to_iocg(pd);
3059
3060         if (dname && iocg->cfg_weight)
3061                 seq_printf(sf, "%s %u\n", dname, iocg->cfg_weight / WEIGHT_ONE);
3062         return 0;
3063 }
3064
3065
3066 static int ioc_weight_show(struct seq_file *sf, void *v)
3067 {
3068         struct blkcg *blkcg = css_to_blkcg(seq_css(sf));
3069         struct ioc_cgrp *iocc = blkcg_to_iocc(blkcg);
3070
3071         seq_printf(sf, "default %u\n", iocc->dfl_weight / WEIGHT_ONE);
3072         blkcg_print_blkgs(sf, blkcg, ioc_weight_prfill,
3073                           &blkcg_policy_iocost, seq_cft(sf)->private, false);
3074         return 0;
3075 }
3076
3077 static ssize_t ioc_weight_write(struct kernfs_open_file *of, char *buf,
3078                                 size_t nbytes, loff_t off)
3079 {
3080         struct blkcg *blkcg = css_to_blkcg(of_css(of));
3081         struct ioc_cgrp *iocc = blkcg_to_iocc(blkcg);
3082         struct blkg_conf_ctx ctx;
3083         struct ioc_now now;
3084         struct ioc_gq *iocg;
3085         u32 v;
3086         int ret;
3087
3088         if (!strchr(buf, ':')) {
3089                 struct blkcg_gq *blkg;
3090
3091                 if (!sscanf(buf, "default %u", &v) && !sscanf(buf, "%u", &v))
3092                         return -EINVAL;
3093
3094                 if (v < CGROUP_WEIGHT_MIN || v > CGROUP_WEIGHT_MAX)
3095                         return -EINVAL;
3096
3097                 spin_lock_irq(&blkcg->lock);
3098                 iocc->dfl_weight = v * WEIGHT_ONE;
3099                 hlist_for_each_entry(blkg, &blkcg->blkg_list, blkcg_node) {
3100                         struct ioc_gq *iocg = blkg_to_iocg(blkg);
3101
3102                         if (iocg) {
3103                                 spin_lock(&iocg->ioc->lock);
3104                                 ioc_now(iocg->ioc, &now);
3105                                 weight_updated(iocg, &now);
3106                                 spin_unlock(&iocg->ioc->lock);
3107                         }
3108                 }
3109                 spin_unlock_irq(&blkcg->lock);
3110
3111                 return nbytes;
3112         }
3113
3114         blkg_conf_init(&ctx, buf);
3115
3116         ret = blkg_conf_prep(blkcg, &blkcg_policy_iocost, &ctx);
3117         if (ret)
3118                 goto err;
3119
3120         iocg = blkg_to_iocg(ctx.blkg);
3121
3122         if (!strncmp(ctx.body, "default", 7)) {
3123                 v = 0;
3124         } else {
3125                 if (!sscanf(ctx.body, "%u", &v))
3126                         goto einval;
3127                 if (v < CGROUP_WEIGHT_MIN || v > CGROUP_WEIGHT_MAX)
3128                         goto einval;
3129         }
3130
3131         spin_lock(&iocg->ioc->lock);
3132         iocg->cfg_weight = v * WEIGHT_ONE;
3133         ioc_now(iocg->ioc, &now);
3134         weight_updated(iocg, &now);
3135         spin_unlock(&iocg->ioc->lock);
3136
3137         blkg_conf_exit(&ctx);
3138         return nbytes;
3139
3140 einval:
3141         ret = -EINVAL;
3142 err:
3143         blkg_conf_exit(&ctx);
3144         return ret;
3145 }
3146
3147 static u64 ioc_qos_prfill(struct seq_file *sf, struct blkg_policy_data *pd,
3148                           int off)
3149 {
3150         const char *dname = blkg_dev_name(pd->blkg);
3151         struct ioc *ioc = pd_to_iocg(pd)->ioc;
3152
3153         if (!dname)
3154                 return 0;
3155
3156         spin_lock_irq(&ioc->lock);
3157         seq_printf(sf, "%s enable=%d ctrl=%s rpct=%u.%02u rlat=%u wpct=%u.%02u wlat=%u min=%u.%02u max=%u.%02u\n",
3158                    dname, ioc->enabled, ioc->user_qos_params ? "user" : "auto",
3159                    ioc->params.qos[QOS_RPPM] / 10000,
3160                    ioc->params.qos[QOS_RPPM] % 10000 / 100,
3161                    ioc->params.qos[QOS_RLAT],
3162                    ioc->params.qos[QOS_WPPM] / 10000,
3163                    ioc->params.qos[QOS_WPPM] % 10000 / 100,
3164                    ioc->params.qos[QOS_WLAT],
3165                    ioc->params.qos[QOS_MIN] / 10000,
3166                    ioc->params.qos[QOS_MIN] % 10000 / 100,
3167                    ioc->params.qos[QOS_MAX] / 10000,
3168                    ioc->params.qos[QOS_MAX] % 10000 / 100);
3169         spin_unlock_irq(&ioc->lock);
3170         return 0;
3171 }
3172
3173 static int ioc_qos_show(struct seq_file *sf, void *v)
3174 {
3175         struct blkcg *blkcg = css_to_blkcg(seq_css(sf));
3176
3177         blkcg_print_blkgs(sf, blkcg, ioc_qos_prfill,
3178                           &blkcg_policy_iocost, seq_cft(sf)->private, false);
3179         return 0;
3180 }
3181
3182 static const match_table_t qos_ctrl_tokens = {
3183         { QOS_ENABLE,           "enable=%u"     },
3184         { QOS_CTRL,             "ctrl=%s"       },
3185         { NR_QOS_CTRL_PARAMS,   NULL            },
3186 };
3187
3188 static const match_table_t qos_tokens = {
3189         { QOS_RPPM,             "rpct=%s"       },
3190         { QOS_RLAT,             "rlat=%u"       },
3191         { QOS_WPPM,             "wpct=%s"       },
3192         { QOS_WLAT,             "wlat=%u"       },
3193         { QOS_MIN,              "min=%s"        },
3194         { QOS_MAX,              "max=%s"        },
3195         { NR_QOS_PARAMS,        NULL            },
3196 };
3197
3198 static ssize_t ioc_qos_write(struct kernfs_open_file *of, char *input,
3199                              size_t nbytes, loff_t off)
3200 {
3201         struct blkg_conf_ctx ctx;
3202         struct gendisk *disk;
3203         struct ioc *ioc;
3204         u32 qos[NR_QOS_PARAMS];
3205         bool enable, user;
3206         char *body, *p;
3207         int ret;
3208
3209         blkg_conf_init(&ctx, input);
3210
3211         ret = blkg_conf_open_bdev(&ctx);
3212         if (ret)
3213                 goto err;
3214
3215         body = ctx.body;
3216         disk = ctx.bdev->bd_disk;
3217         if (!queue_is_mq(disk->queue)) {
3218                 ret = -EOPNOTSUPP;
3219                 goto err;
3220         }
3221
3222         ioc = q_to_ioc(disk->queue);
3223         if (!ioc) {
3224                 ret = blk_iocost_init(disk);
3225                 if (ret)
3226                         goto err;
3227                 ioc = q_to_ioc(disk->queue);
3228         }
3229
3230         blk_mq_freeze_queue(disk->queue);
3231         blk_mq_quiesce_queue(disk->queue);
3232
3233         spin_lock_irq(&ioc->lock);
3234         memcpy(qos, ioc->params.qos, sizeof(qos));
3235         enable = ioc->enabled;
3236         user = ioc->user_qos_params;
3237
3238         while ((p = strsep(&body, " \t\n"))) {
3239                 substring_t args[MAX_OPT_ARGS];
3240                 char buf[32];
3241                 int tok;
3242                 s64 v;
3243
3244                 if (!*p)
3245                         continue;
3246
3247                 switch (match_token(p, qos_ctrl_tokens, args)) {
3248                 case QOS_ENABLE:
3249                         if (match_u64(&args[0], &v))
3250                                 goto einval;
3251                         enable = v;
3252                         continue;
3253                 case QOS_CTRL:
3254                         match_strlcpy(buf, &args[0], sizeof(buf));
3255                         if (!strcmp(buf, "auto"))
3256                                 user = false;
3257                         else if (!strcmp(buf, "user"))
3258                                 user = true;
3259                         else
3260                                 goto einval;
3261                         continue;
3262                 }
3263
3264                 tok = match_token(p, qos_tokens, args);
3265                 switch (tok) {
3266                 case QOS_RPPM:
3267                 case QOS_WPPM:
3268                         if (match_strlcpy(buf, &args[0], sizeof(buf)) >=
3269                             sizeof(buf))
3270                                 goto einval;
3271                         if (cgroup_parse_float(buf, 2, &v))
3272                                 goto einval;
3273                         if (v < 0 || v > 10000)
3274                                 goto einval;
3275                         qos[tok] = v * 100;
3276                         break;
3277                 case QOS_RLAT:
3278                 case QOS_WLAT:
3279                         if (match_u64(&args[0], &v))
3280                                 goto einval;
3281                         qos[tok] = v;
3282                         break;
3283                 case QOS_MIN:
3284                 case QOS_MAX:
3285                         if (match_strlcpy(buf, &args[0], sizeof(buf)) >=
3286                             sizeof(buf))
3287                                 goto einval;
3288                         if (cgroup_parse_float(buf, 2, &v))
3289                                 goto einval;
3290                         if (v < 0)
3291                                 goto einval;
3292                         qos[tok] = clamp_t(s64, v * 100,
3293                                            VRATE_MIN_PPM, VRATE_MAX_PPM);
3294                         break;
3295                 default:
3296                         goto einval;
3297                 }
3298                 user = true;
3299         }
3300
3301         if (qos[QOS_MIN] > qos[QOS_MAX])
3302                 goto einval;
3303
3304         if (enable) {
3305                 blk_stat_enable_accounting(disk->queue);
3306                 blk_queue_flag_set(QUEUE_FLAG_RQ_ALLOC_TIME, disk->queue);
3307                 ioc->enabled = true;
3308         } else {
3309                 blk_queue_flag_clear(QUEUE_FLAG_RQ_ALLOC_TIME, disk->queue);
3310                 ioc->enabled = false;
3311         }
3312
3313         if (user) {
3314                 memcpy(ioc->params.qos, qos, sizeof(qos));
3315                 ioc->user_qos_params = true;
3316         } else {
3317                 ioc->user_qos_params = false;
3318         }
3319
3320         ioc_refresh_params(ioc, true);
3321         spin_unlock_irq(&ioc->lock);
3322
3323         if (enable)
3324                 wbt_disable_default(disk);
3325         else
3326                 wbt_enable_default(disk);
3327
3328         blk_mq_unquiesce_queue(disk->queue);
3329         blk_mq_unfreeze_queue(disk->queue);
3330
3331         blkg_conf_exit(&ctx);
3332         return nbytes;
3333 einval:
3334         spin_unlock_irq(&ioc->lock);
3335
3336         blk_mq_unquiesce_queue(disk->queue);
3337         blk_mq_unfreeze_queue(disk->queue);
3338
3339         ret = -EINVAL;
3340 err:
3341         blkg_conf_exit(&ctx);
3342         return ret;
3343 }
3344
3345 static u64 ioc_cost_model_prfill(struct seq_file *sf,
3346                                  struct blkg_policy_data *pd, int off)
3347 {
3348         const char *dname = blkg_dev_name(pd->blkg);
3349         struct ioc *ioc = pd_to_iocg(pd)->ioc;
3350         u64 *u = ioc->params.i_lcoefs;
3351
3352         if (!dname)
3353                 return 0;
3354
3355         spin_lock_irq(&ioc->lock);
3356         seq_printf(sf, "%s ctrl=%s model=linear "
3357                    "rbps=%llu rseqiops=%llu rrandiops=%llu "
3358                    "wbps=%llu wseqiops=%llu wrandiops=%llu\n",
3359                    dname, ioc->user_cost_model ? "user" : "auto",
3360                    u[I_LCOEF_RBPS], u[I_LCOEF_RSEQIOPS], u[I_LCOEF_RRANDIOPS],
3361                    u[I_LCOEF_WBPS], u[I_LCOEF_WSEQIOPS], u[I_LCOEF_WRANDIOPS]);
3362         spin_unlock_irq(&ioc->lock);
3363         return 0;
3364 }
3365
3366 static int ioc_cost_model_show(struct seq_file *sf, void *v)
3367 {
3368         struct blkcg *blkcg = css_to_blkcg(seq_css(sf));
3369
3370         blkcg_print_blkgs(sf, blkcg, ioc_cost_model_prfill,
3371                           &blkcg_policy_iocost, seq_cft(sf)->private, false);
3372         return 0;
3373 }
3374
3375 static const match_table_t cost_ctrl_tokens = {
3376         { COST_CTRL,            "ctrl=%s"       },
3377         { COST_MODEL,           "model=%s"      },
3378         { NR_COST_CTRL_PARAMS,  NULL            },
3379 };
3380
3381 static const match_table_t i_lcoef_tokens = {
3382         { I_LCOEF_RBPS,         "rbps=%u"       },
3383         { I_LCOEF_RSEQIOPS,     "rseqiops=%u"   },
3384         { I_LCOEF_RRANDIOPS,    "rrandiops=%u"  },
3385         { I_LCOEF_WBPS,         "wbps=%u"       },
3386         { I_LCOEF_WSEQIOPS,     "wseqiops=%u"   },
3387         { I_LCOEF_WRANDIOPS,    "wrandiops=%u"  },
3388         { NR_I_LCOEFS,          NULL            },
3389 };
3390
3391 static ssize_t ioc_cost_model_write(struct kernfs_open_file *of, char *input,
3392                                     size_t nbytes, loff_t off)
3393 {
3394         struct blkg_conf_ctx ctx;
3395         struct request_queue *q;
3396         struct ioc *ioc;
3397         u64 u[NR_I_LCOEFS];
3398         bool user;
3399         char *body, *p;
3400         int ret;
3401
3402         blkg_conf_init(&ctx, input);
3403
3404         ret = blkg_conf_open_bdev(&ctx);
3405         if (ret)
3406                 goto err;
3407
3408         body = ctx.body;
3409         q = bdev_get_queue(ctx.bdev);
3410         if (!queue_is_mq(q)) {
3411                 ret = -EOPNOTSUPP;
3412                 goto err;
3413         }
3414
3415         ioc = q_to_ioc(q);
3416         if (!ioc) {
3417                 ret = blk_iocost_init(ctx.bdev->bd_disk);
3418                 if (ret)
3419                         goto err;
3420                 ioc = q_to_ioc(q);
3421         }
3422
3423         blk_mq_freeze_queue(q);
3424         blk_mq_quiesce_queue(q);
3425
3426         spin_lock_irq(&ioc->lock);
3427         memcpy(u, ioc->params.i_lcoefs, sizeof(u));
3428         user = ioc->user_cost_model;
3429
3430         while ((p = strsep(&body, " \t\n"))) {
3431                 substring_t args[MAX_OPT_ARGS];
3432                 char buf[32];
3433                 int tok;
3434                 u64 v;
3435
3436                 if (!*p)
3437                         continue;
3438
3439                 switch (match_token(p, cost_ctrl_tokens, args)) {
3440                 case COST_CTRL:
3441                         match_strlcpy(buf, &args[0], sizeof(buf));
3442                         if (!strcmp(buf, "auto"))
3443                                 user = false;
3444                         else if (!strcmp(buf, "user"))
3445                                 user = true;
3446                         else
3447                                 goto einval;
3448                         continue;
3449                 case COST_MODEL:
3450                         match_strlcpy(buf, &args[0], sizeof(buf));
3451                         if (strcmp(buf, "linear"))
3452                                 goto einval;
3453                         continue;
3454                 }
3455
3456                 tok = match_token(p, i_lcoef_tokens, args);
3457                 if (tok == NR_I_LCOEFS)
3458                         goto einval;
3459                 if (match_u64(&args[0], &v))
3460                         goto einval;
3461                 u[tok] = v;
3462                 user = true;
3463         }
3464
3465         if (user) {
3466                 memcpy(ioc->params.i_lcoefs, u, sizeof(u));
3467                 ioc->user_cost_model = true;
3468         } else {
3469                 ioc->user_cost_model = false;
3470         }
3471         ioc_refresh_params(ioc, true);
3472         spin_unlock_irq(&ioc->lock);
3473
3474         blk_mq_unquiesce_queue(q);
3475         blk_mq_unfreeze_queue(q);
3476
3477         blkg_conf_exit(&ctx);
3478         return nbytes;
3479
3480 einval:
3481         spin_unlock_irq(&ioc->lock);
3482
3483         blk_mq_unquiesce_queue(q);
3484         blk_mq_unfreeze_queue(q);
3485
3486         ret = -EINVAL;
3487 err:
3488         blkg_conf_exit(&ctx);
3489         return ret;
3490 }
3491
3492 static struct cftype ioc_files[] = {
3493         {
3494                 .name = "weight",
3495                 .flags = CFTYPE_NOT_ON_ROOT,
3496                 .seq_show = ioc_weight_show,
3497                 .write = ioc_weight_write,
3498         },
3499         {
3500                 .name = "cost.qos",
3501                 .flags = CFTYPE_ONLY_ON_ROOT,
3502                 .seq_show = ioc_qos_show,
3503                 .write = ioc_qos_write,
3504         },
3505         {
3506                 .name = "cost.model",
3507                 .flags = CFTYPE_ONLY_ON_ROOT,
3508                 .seq_show = ioc_cost_model_show,
3509                 .write = ioc_cost_model_write,
3510         },
3511         {}
3512 };
3513
3514 static struct blkcg_policy blkcg_policy_iocost = {
3515         .dfl_cftypes    = ioc_files,
3516         .cpd_alloc_fn   = ioc_cpd_alloc,
3517         .cpd_free_fn    = ioc_cpd_free,
3518         .pd_alloc_fn    = ioc_pd_alloc,
3519         .pd_init_fn     = ioc_pd_init,
3520         .pd_free_fn     = ioc_pd_free,
3521         .pd_stat_fn     = ioc_pd_stat,
3522 };
3523
3524 static int __init ioc_init(void)
3525 {
3526         return blkcg_policy_register(&blkcg_policy_iocost);
3527 }
3528
3529 static void __exit ioc_exit(void)
3530 {
3531         blkcg_policy_unregister(&blkcg_policy_iocost);
3532 }
3533
3534 module_init(ioc_init);
3535 module_exit(ioc_exit);