xref: /openbmc/linux/block/blk-iocost.c (revision 9ad685db)
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 
2459 	current_hweight(iocg, NULL, &hwi);
2460 	old_hwi = hwi;
2461 	cost = abs_cost_to_cost(abs_cost, hwi);
2462 	margin = now->vnow - vtime - cost;
2463 
2464 	/* debt handling owns inuse for debtors */
2465 	if (iocg->abs_vdebt)
2466 		return cost;
2467 
2468 	/*
2469 	 * We only increase inuse during period and do so if the margin has
2470 	 * deteriorated since the previous adjustment.
2471 	 */
2472 	if (margin >= iocg->saved_margin || margin >= margins->low ||
2473 	    iocg->inuse == iocg->active)
2474 		return cost;
2475 
2476 	spin_lock_irq(&ioc->lock);
2477 
2478 	/* we own inuse only when @iocg is in the normal active state */
2479 	if (iocg->abs_vdebt || list_empty(&iocg->active_list)) {
2480 		spin_unlock_irq(&ioc->lock);
2481 		return cost;
2482 	}
2483 
2484 	/*
2485 	 * Bump up inuse till @abs_cost fits in the existing budget.
2486 	 * adj_step must be determined after acquiring ioc->lock - we might
2487 	 * have raced and lost to another thread for activation and could
2488 	 * be reading 0 iocg->active before ioc->lock which will lead to
2489 	 * infinite loop.
2490 	 */
2491 	new_inuse = iocg->inuse;
2492 	adj_step = DIV_ROUND_UP(iocg->active * INUSE_ADJ_STEP_PCT, 100);
2493 	do {
2494 		new_inuse = new_inuse + adj_step;
2495 		propagate_weights(iocg, iocg->active, new_inuse, true, now);
2496 		current_hweight(iocg, NULL, &hwi);
2497 		cost = abs_cost_to_cost(abs_cost, hwi);
2498 	} while (time_after64(vtime + cost, now->vnow) &&
2499 		 iocg->inuse != iocg->active);
2500 
2501 	spin_unlock_irq(&ioc->lock);
2502 
2503 	TRACE_IOCG_PATH(inuse_adjust, iocg, now,
2504 			old_inuse, iocg->inuse, old_hwi, hwi);
2505 
2506 	return cost;
2507 }
2508 
2509 static void calc_vtime_cost_builtin(struct bio *bio, struct ioc_gq *iocg,
2510 				    bool is_merge, u64 *costp)
2511 {
2512 	struct ioc *ioc = iocg->ioc;
2513 	u64 coef_seqio, coef_randio, coef_page;
2514 	u64 pages = max_t(u64, bio_sectors(bio) >> IOC_SECT_TO_PAGE_SHIFT, 1);
2515 	u64 seek_pages = 0;
2516 	u64 cost = 0;
2517 
2518 	switch (bio_op(bio)) {
2519 	case REQ_OP_READ:
2520 		coef_seqio	= ioc->params.lcoefs[LCOEF_RSEQIO];
2521 		coef_randio	= ioc->params.lcoefs[LCOEF_RRANDIO];
2522 		coef_page	= ioc->params.lcoefs[LCOEF_RPAGE];
2523 		break;
2524 	case REQ_OP_WRITE:
2525 		coef_seqio	= ioc->params.lcoefs[LCOEF_WSEQIO];
2526 		coef_randio	= ioc->params.lcoefs[LCOEF_WRANDIO];
2527 		coef_page	= ioc->params.lcoefs[LCOEF_WPAGE];
2528 		break;
2529 	default:
2530 		goto out;
2531 	}
2532 
2533 	if (iocg->cursor) {
2534 		seek_pages = abs(bio->bi_iter.bi_sector - iocg->cursor);
2535 		seek_pages >>= IOC_SECT_TO_PAGE_SHIFT;
2536 	}
2537 
2538 	if (!is_merge) {
2539 		if (seek_pages > LCOEF_RANDIO_PAGES) {
2540 			cost += coef_randio;
2541 		} else {
2542 			cost += coef_seqio;
2543 		}
2544 	}
2545 	cost += pages * coef_page;
2546 out:
2547 	*costp = cost;
2548 }
2549 
2550 static u64 calc_vtime_cost(struct bio *bio, struct ioc_gq *iocg, bool is_merge)
2551 {
2552 	u64 cost;
2553 
2554 	calc_vtime_cost_builtin(bio, iocg, is_merge, &cost);
2555 	return cost;
2556 }
2557 
2558 static void calc_size_vtime_cost_builtin(struct request *rq, struct ioc *ioc,
2559 					 u64 *costp)
2560 {
2561 	unsigned int pages = blk_rq_stats_sectors(rq) >> IOC_SECT_TO_PAGE_SHIFT;
2562 
2563 	switch (req_op(rq)) {
2564 	case REQ_OP_READ:
2565 		*costp = pages * ioc->params.lcoefs[LCOEF_RPAGE];
2566 		break;
2567 	case REQ_OP_WRITE:
2568 		*costp = pages * ioc->params.lcoefs[LCOEF_WPAGE];
2569 		break;
2570 	default:
2571 		*costp = 0;
2572 	}
2573 }
2574 
2575 static u64 calc_size_vtime_cost(struct request *rq, struct ioc *ioc)
2576 {
2577 	u64 cost;
2578 
2579 	calc_size_vtime_cost_builtin(rq, ioc, &cost);
2580 	return cost;
2581 }
2582 
2583 static void ioc_rqos_throttle(struct rq_qos *rqos, struct bio *bio)
2584 {
2585 	struct blkcg_gq *blkg = bio->bi_blkg;
2586 	struct ioc *ioc = rqos_to_ioc(rqos);
2587 	struct ioc_gq *iocg = blkg_to_iocg(blkg);
2588 	struct ioc_now now;
2589 	struct iocg_wait wait;
2590 	u64 abs_cost, cost, vtime;
2591 	bool use_debt, ioc_locked;
2592 	unsigned long flags;
2593 
2594 	/* bypass IOs if disabled, still initializing, or for root cgroup */
2595 	if (!ioc->enabled || !iocg || !iocg->level)
2596 		return;
2597 
2598 	/* calculate the absolute vtime cost */
2599 	abs_cost = calc_vtime_cost(bio, iocg, false);
2600 	if (!abs_cost)
2601 		return;
2602 
2603 	if (!iocg_activate(iocg, &now))
2604 		return;
2605 
2606 	iocg->cursor = bio_end_sector(bio);
2607 	vtime = atomic64_read(&iocg->vtime);
2608 	cost = adjust_inuse_and_calc_cost(iocg, vtime, abs_cost, &now);
2609 
2610 	/*
2611 	 * If no one's waiting and within budget, issue right away.  The
2612 	 * tests are racy but the races aren't systemic - we only miss once
2613 	 * in a while which is fine.
2614 	 */
2615 	if (!waitqueue_active(&iocg->waitq) && !iocg->abs_vdebt &&
2616 	    time_before_eq64(vtime + cost, now.vnow)) {
2617 		iocg_commit_bio(iocg, bio, abs_cost, cost);
2618 		return;
2619 	}
2620 
2621 	/*
2622 	 * We're over budget. This can be handled in two ways. IOs which may
2623 	 * cause priority inversions are punted to @ioc->aux_iocg and charged as
2624 	 * debt. Otherwise, the issuer is blocked on @iocg->waitq. Debt handling
2625 	 * requires @ioc->lock, waitq handling @iocg->waitq.lock. Determine
2626 	 * whether debt handling is needed and acquire locks accordingly.
2627 	 */
2628 	use_debt = bio_issue_as_root_blkg(bio) || fatal_signal_pending(current);
2629 	ioc_locked = use_debt || READ_ONCE(iocg->abs_vdebt);
2630 retry_lock:
2631 	iocg_lock(iocg, ioc_locked, &flags);
2632 
2633 	/*
2634 	 * @iocg must stay activated for debt and waitq handling. Deactivation
2635 	 * is synchronized against both ioc->lock and waitq.lock and we won't
2636 	 * get deactivated as long as we're waiting or has debt, so we're good
2637 	 * if we're activated here. In the unlikely cases that we aren't, just
2638 	 * issue the IO.
2639 	 */
2640 	if (unlikely(list_empty(&iocg->active_list))) {
2641 		iocg_unlock(iocg, ioc_locked, &flags);
2642 		iocg_commit_bio(iocg, bio, abs_cost, cost);
2643 		return;
2644 	}
2645 
2646 	/*
2647 	 * We're over budget. If @bio has to be issued regardless, remember
2648 	 * the abs_cost instead of advancing vtime. iocg_kick_waitq() will pay
2649 	 * off the debt before waking more IOs.
2650 	 *
2651 	 * This way, the debt is continuously paid off each period with the
2652 	 * actual budget available to the cgroup. If we just wound vtime, we
2653 	 * would incorrectly use the current hw_inuse for the entire amount
2654 	 * which, for example, can lead to the cgroup staying blocked for a
2655 	 * long time even with substantially raised hw_inuse.
2656 	 *
2657 	 * An iocg with vdebt should stay online so that the timer can keep
2658 	 * deducting its vdebt and [de]activate use_delay mechanism
2659 	 * accordingly. We don't want to race against the timer trying to
2660 	 * clear them and leave @iocg inactive w/ dangling use_delay heavily
2661 	 * penalizing the cgroup and its descendants.
2662 	 */
2663 	if (use_debt) {
2664 		iocg_incur_debt(iocg, abs_cost, &now);
2665 		if (iocg_kick_delay(iocg, &now))
2666 			blkcg_schedule_throttle(rqos->disk,
2667 					(bio->bi_opf & REQ_SWAP) == REQ_SWAP);
2668 		iocg_unlock(iocg, ioc_locked, &flags);
2669 		return;
2670 	}
2671 
2672 	/* guarantee that iocgs w/ waiters have maximum inuse */
2673 	if (!iocg->abs_vdebt && iocg->inuse != iocg->active) {
2674 		if (!ioc_locked) {
2675 			iocg_unlock(iocg, false, &flags);
2676 			ioc_locked = true;
2677 			goto retry_lock;
2678 		}
2679 		propagate_weights(iocg, iocg->active, iocg->active, true,
2680 				  &now);
2681 	}
2682 
2683 	/*
2684 	 * Append self to the waitq and schedule the wakeup timer if we're
2685 	 * the first waiter.  The timer duration is calculated based on the
2686 	 * current vrate.  vtime and hweight changes can make it too short
2687 	 * or too long.  Each wait entry records the absolute cost it's
2688 	 * waiting for to allow re-evaluation using a custom wait entry.
2689 	 *
2690 	 * If too short, the timer simply reschedules itself.  If too long,
2691 	 * the period timer will notice and trigger wakeups.
2692 	 *
2693 	 * All waiters are on iocg->waitq and the wait states are
2694 	 * synchronized using waitq.lock.
2695 	 */
2696 	init_waitqueue_func_entry(&wait.wait, iocg_wake_fn);
2697 	wait.wait.private = current;
2698 	wait.bio = bio;
2699 	wait.abs_cost = abs_cost;
2700 	wait.committed = false;	/* will be set true by waker */
2701 
2702 	__add_wait_queue_entry_tail(&iocg->waitq, &wait.wait);
2703 	iocg_kick_waitq(iocg, ioc_locked, &now);
2704 
2705 	iocg_unlock(iocg, ioc_locked, &flags);
2706 
2707 	while (true) {
2708 		set_current_state(TASK_UNINTERRUPTIBLE);
2709 		if (wait.committed)
2710 			break;
2711 		io_schedule();
2712 	}
2713 
2714 	/* waker already committed us, proceed */
2715 	finish_wait(&iocg->waitq, &wait.wait);
2716 }
2717 
2718 static void ioc_rqos_merge(struct rq_qos *rqos, struct request *rq,
2719 			   struct bio *bio)
2720 {
2721 	struct ioc_gq *iocg = blkg_to_iocg(bio->bi_blkg);
2722 	struct ioc *ioc = rqos_to_ioc(rqos);
2723 	sector_t bio_end = bio_end_sector(bio);
2724 	struct ioc_now now;
2725 	u64 vtime, abs_cost, cost;
2726 	unsigned long flags;
2727 
2728 	/* bypass if disabled, still initializing, or for root cgroup */
2729 	if (!ioc->enabled || !iocg || !iocg->level)
2730 		return;
2731 
2732 	abs_cost = calc_vtime_cost(bio, iocg, true);
2733 	if (!abs_cost)
2734 		return;
2735 
2736 	ioc_now(ioc, &now);
2737 
2738 	vtime = atomic64_read(&iocg->vtime);
2739 	cost = adjust_inuse_and_calc_cost(iocg, vtime, abs_cost, &now);
2740 
2741 	/* update cursor if backmerging into the request at the cursor */
2742 	if (blk_rq_pos(rq) < bio_end &&
2743 	    blk_rq_pos(rq) + blk_rq_sectors(rq) == iocg->cursor)
2744 		iocg->cursor = bio_end;
2745 
2746 	/*
2747 	 * Charge if there's enough vtime budget and the existing request has
2748 	 * cost assigned.
2749 	 */
2750 	if (rq->bio && rq->bio->bi_iocost_cost &&
2751 	    time_before_eq64(atomic64_read(&iocg->vtime) + cost, now.vnow)) {
2752 		iocg_commit_bio(iocg, bio, abs_cost, cost);
2753 		return;
2754 	}
2755 
2756 	/*
2757 	 * Otherwise, account it as debt if @iocg is online, which it should
2758 	 * be for the vast majority of cases. See debt handling in
2759 	 * ioc_rqos_throttle() for details.
2760 	 */
2761 	spin_lock_irqsave(&ioc->lock, flags);
2762 	spin_lock(&iocg->waitq.lock);
2763 
2764 	if (likely(!list_empty(&iocg->active_list))) {
2765 		iocg_incur_debt(iocg, abs_cost, &now);
2766 		if (iocg_kick_delay(iocg, &now))
2767 			blkcg_schedule_throttle(rqos->disk,
2768 					(bio->bi_opf & REQ_SWAP) == REQ_SWAP);
2769 	} else {
2770 		iocg_commit_bio(iocg, bio, abs_cost, cost);
2771 	}
2772 
2773 	spin_unlock(&iocg->waitq.lock);
2774 	spin_unlock_irqrestore(&ioc->lock, flags);
2775 }
2776 
2777 static void ioc_rqos_done_bio(struct rq_qos *rqos, struct bio *bio)
2778 {
2779 	struct ioc_gq *iocg = blkg_to_iocg(bio->bi_blkg);
2780 
2781 	if (iocg && bio->bi_iocost_cost)
2782 		atomic64_add(bio->bi_iocost_cost, &iocg->done_vtime);
2783 }
2784 
2785 static void ioc_rqos_done(struct rq_qos *rqos, struct request *rq)
2786 {
2787 	struct ioc *ioc = rqos_to_ioc(rqos);
2788 	struct ioc_pcpu_stat *ccs;
2789 	u64 on_q_ns, rq_wait_ns, size_nsec;
2790 	int pidx, rw;
2791 
2792 	if (!ioc->enabled || !rq->alloc_time_ns || !rq->start_time_ns)
2793 		return;
2794 
2795 	switch (req_op(rq)) {
2796 	case REQ_OP_READ:
2797 		pidx = QOS_RLAT;
2798 		rw = READ;
2799 		break;
2800 	case REQ_OP_WRITE:
2801 		pidx = QOS_WLAT;
2802 		rw = WRITE;
2803 		break;
2804 	default:
2805 		return;
2806 	}
2807 
2808 	on_q_ns = ktime_get_ns() - rq->alloc_time_ns;
2809 	rq_wait_ns = rq->start_time_ns - rq->alloc_time_ns;
2810 	size_nsec = div64_u64(calc_size_vtime_cost(rq, ioc), VTIME_PER_NSEC);
2811 
2812 	ccs = get_cpu_ptr(ioc->pcpu_stat);
2813 
2814 	if (on_q_ns <= size_nsec ||
2815 	    on_q_ns - size_nsec <= ioc->params.qos[pidx] * NSEC_PER_USEC)
2816 		local_inc(&ccs->missed[rw].nr_met);
2817 	else
2818 		local_inc(&ccs->missed[rw].nr_missed);
2819 
2820 	local64_add(rq_wait_ns, &ccs->rq_wait_ns);
2821 
2822 	put_cpu_ptr(ccs);
2823 }
2824 
2825 static void ioc_rqos_queue_depth_changed(struct rq_qos *rqos)
2826 {
2827 	struct ioc *ioc = rqos_to_ioc(rqos);
2828 
2829 	spin_lock_irq(&ioc->lock);
2830 	ioc_refresh_params(ioc, false);
2831 	spin_unlock_irq(&ioc->lock);
2832 }
2833 
2834 static void ioc_rqos_exit(struct rq_qos *rqos)
2835 {
2836 	struct ioc *ioc = rqos_to_ioc(rqos);
2837 
2838 	blkcg_deactivate_policy(rqos->disk, &blkcg_policy_iocost);
2839 
2840 	spin_lock_irq(&ioc->lock);
2841 	ioc->running = IOC_STOP;
2842 	spin_unlock_irq(&ioc->lock);
2843 
2844 	timer_shutdown_sync(&ioc->timer);
2845 	free_percpu(ioc->pcpu_stat);
2846 	kfree(ioc);
2847 }
2848 
2849 static const struct rq_qos_ops ioc_rqos_ops = {
2850 	.throttle = ioc_rqos_throttle,
2851 	.merge = ioc_rqos_merge,
2852 	.done_bio = ioc_rqos_done_bio,
2853 	.done = ioc_rqos_done,
2854 	.queue_depth_changed = ioc_rqos_queue_depth_changed,
2855 	.exit = ioc_rqos_exit,
2856 };
2857 
2858 static int blk_iocost_init(struct gendisk *disk)
2859 {
2860 	struct ioc *ioc;
2861 	int i, cpu, ret;
2862 
2863 	ioc = kzalloc(sizeof(*ioc), GFP_KERNEL);
2864 	if (!ioc)
2865 		return -ENOMEM;
2866 
2867 	ioc->pcpu_stat = alloc_percpu(struct ioc_pcpu_stat);
2868 	if (!ioc->pcpu_stat) {
2869 		kfree(ioc);
2870 		return -ENOMEM;
2871 	}
2872 
2873 	for_each_possible_cpu(cpu) {
2874 		struct ioc_pcpu_stat *ccs = per_cpu_ptr(ioc->pcpu_stat, cpu);
2875 
2876 		for (i = 0; i < ARRAY_SIZE(ccs->missed); i++) {
2877 			local_set(&ccs->missed[i].nr_met, 0);
2878 			local_set(&ccs->missed[i].nr_missed, 0);
2879 		}
2880 		local64_set(&ccs->rq_wait_ns, 0);
2881 	}
2882 
2883 	spin_lock_init(&ioc->lock);
2884 	timer_setup(&ioc->timer, ioc_timer_fn, 0);
2885 	INIT_LIST_HEAD(&ioc->active_iocgs);
2886 
2887 	ioc->running = IOC_IDLE;
2888 	ioc->vtime_base_rate = VTIME_PER_USEC;
2889 	atomic64_set(&ioc->vtime_rate, VTIME_PER_USEC);
2890 	seqcount_spinlock_init(&ioc->period_seqcount, &ioc->lock);
2891 	ioc->period_at = ktime_to_us(ktime_get());
2892 	atomic64_set(&ioc->cur_period, 0);
2893 	atomic_set(&ioc->hweight_gen, 0);
2894 
2895 	spin_lock_irq(&ioc->lock);
2896 	ioc->autop_idx = AUTOP_INVALID;
2897 	ioc_refresh_params_disk(ioc, true, disk);
2898 	spin_unlock_irq(&ioc->lock);
2899 
2900 	/*
2901 	 * rqos must be added before activation to allow ioc_pd_init() to
2902 	 * lookup the ioc from q. This means that the rqos methods may get
2903 	 * called before policy activation completion, can't assume that the
2904 	 * target bio has an iocg associated and need to test for NULL iocg.
2905 	 */
2906 	ret = rq_qos_add(&ioc->rqos, disk, RQ_QOS_COST, &ioc_rqos_ops);
2907 	if (ret)
2908 		goto err_free_ioc;
2909 
2910 	ret = blkcg_activate_policy(disk, &blkcg_policy_iocost);
2911 	if (ret)
2912 		goto err_del_qos;
2913 	return 0;
2914 
2915 err_del_qos:
2916 	rq_qos_del(&ioc->rqos);
2917 err_free_ioc:
2918 	free_percpu(ioc->pcpu_stat);
2919 	kfree(ioc);
2920 	return ret;
2921 }
2922 
2923 static struct blkcg_policy_data *ioc_cpd_alloc(gfp_t gfp)
2924 {
2925 	struct ioc_cgrp *iocc;
2926 
2927 	iocc = kzalloc(sizeof(struct ioc_cgrp), gfp);
2928 	if (!iocc)
2929 		return NULL;
2930 
2931 	iocc->dfl_weight = CGROUP_WEIGHT_DFL * WEIGHT_ONE;
2932 	return &iocc->cpd;
2933 }
2934 
2935 static void ioc_cpd_free(struct blkcg_policy_data *cpd)
2936 {
2937 	kfree(container_of(cpd, struct ioc_cgrp, cpd));
2938 }
2939 
2940 static struct blkg_policy_data *ioc_pd_alloc(struct gendisk *disk,
2941 		struct blkcg *blkcg, gfp_t gfp)
2942 {
2943 	int levels = blkcg->css.cgroup->level + 1;
2944 	struct ioc_gq *iocg;
2945 
2946 	iocg = kzalloc_node(struct_size(iocg, ancestors, levels), gfp,
2947 			    disk->node_id);
2948 	if (!iocg)
2949 		return NULL;
2950 
2951 	iocg->pcpu_stat = alloc_percpu_gfp(struct iocg_pcpu_stat, gfp);
2952 	if (!iocg->pcpu_stat) {
2953 		kfree(iocg);
2954 		return NULL;
2955 	}
2956 
2957 	return &iocg->pd;
2958 }
2959 
2960 static void ioc_pd_init(struct blkg_policy_data *pd)
2961 {
2962 	struct ioc_gq *iocg = pd_to_iocg(pd);
2963 	struct blkcg_gq *blkg = pd_to_blkg(&iocg->pd);
2964 	struct ioc *ioc = q_to_ioc(blkg->q);
2965 	struct ioc_now now;
2966 	struct blkcg_gq *tblkg;
2967 	unsigned long flags;
2968 
2969 	ioc_now(ioc, &now);
2970 
2971 	iocg->ioc = ioc;
2972 	atomic64_set(&iocg->vtime, now.vnow);
2973 	atomic64_set(&iocg->done_vtime, now.vnow);
2974 	atomic64_set(&iocg->active_period, atomic64_read(&ioc->cur_period));
2975 	INIT_LIST_HEAD(&iocg->active_list);
2976 	INIT_LIST_HEAD(&iocg->walk_list);
2977 	INIT_LIST_HEAD(&iocg->surplus_list);
2978 	iocg->hweight_active = WEIGHT_ONE;
2979 	iocg->hweight_inuse = WEIGHT_ONE;
2980 
2981 	init_waitqueue_head(&iocg->waitq);
2982 	hrtimer_init(&iocg->waitq_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
2983 	iocg->waitq_timer.function = iocg_waitq_timer_fn;
2984 
2985 	iocg->level = blkg->blkcg->css.cgroup->level;
2986 
2987 	for (tblkg = blkg; tblkg; tblkg = tblkg->parent) {
2988 		struct ioc_gq *tiocg = blkg_to_iocg(tblkg);
2989 		iocg->ancestors[tiocg->level] = tiocg;
2990 	}
2991 
2992 	spin_lock_irqsave(&ioc->lock, flags);
2993 	weight_updated(iocg, &now);
2994 	spin_unlock_irqrestore(&ioc->lock, flags);
2995 }
2996 
2997 static void ioc_pd_free(struct blkg_policy_data *pd)
2998 {
2999 	struct ioc_gq *iocg = pd_to_iocg(pd);
3000 	struct ioc *ioc = iocg->ioc;
3001 	unsigned long flags;
3002 
3003 	if (ioc) {
3004 		spin_lock_irqsave(&ioc->lock, flags);
3005 
3006 		if (!list_empty(&iocg->active_list)) {
3007 			struct ioc_now now;
3008 
3009 			ioc_now(ioc, &now);
3010 			propagate_weights(iocg, 0, 0, false, &now);
3011 			list_del_init(&iocg->active_list);
3012 		}
3013 
3014 		WARN_ON_ONCE(!list_empty(&iocg->walk_list));
3015 		WARN_ON_ONCE(!list_empty(&iocg->surplus_list));
3016 
3017 		spin_unlock_irqrestore(&ioc->lock, flags);
3018 
3019 		hrtimer_cancel(&iocg->waitq_timer);
3020 	}
3021 	free_percpu(iocg->pcpu_stat);
3022 	kfree(iocg);
3023 }
3024 
3025 static void ioc_pd_stat(struct blkg_policy_data *pd, struct seq_file *s)
3026 {
3027 	struct ioc_gq *iocg = pd_to_iocg(pd);
3028 	struct ioc *ioc = iocg->ioc;
3029 
3030 	if (!ioc->enabled)
3031 		return;
3032 
3033 	if (iocg->level == 0) {
3034 		unsigned vp10k = DIV64_U64_ROUND_CLOSEST(
3035 			ioc->vtime_base_rate * 10000,
3036 			VTIME_PER_USEC);
3037 		seq_printf(s, " cost.vrate=%u.%02u", vp10k / 100, vp10k % 100);
3038 	}
3039 
3040 	seq_printf(s, " cost.usage=%llu", iocg->last_stat.usage_us);
3041 
3042 	if (blkcg_debug_stats)
3043 		seq_printf(s, " cost.wait=%llu cost.indebt=%llu cost.indelay=%llu",
3044 			iocg->last_stat.wait_us,
3045 			iocg->last_stat.indebt_us,
3046 			iocg->last_stat.indelay_us);
3047 }
3048 
3049 static u64 ioc_weight_prfill(struct seq_file *sf, struct blkg_policy_data *pd,
3050 			     int off)
3051 {
3052 	const char *dname = blkg_dev_name(pd->blkg);
3053 	struct ioc_gq *iocg = pd_to_iocg(pd);
3054 
3055 	if (dname && iocg->cfg_weight)
3056 		seq_printf(sf, "%s %u\n", dname, iocg->cfg_weight / WEIGHT_ONE);
3057 	return 0;
3058 }
3059 
3060 
3061 static int ioc_weight_show(struct seq_file *sf, void *v)
3062 {
3063 	struct blkcg *blkcg = css_to_blkcg(seq_css(sf));
3064 	struct ioc_cgrp *iocc = blkcg_to_iocc(blkcg);
3065 
3066 	seq_printf(sf, "default %u\n", iocc->dfl_weight / WEIGHT_ONE);
3067 	blkcg_print_blkgs(sf, blkcg, ioc_weight_prfill,
3068 			  &blkcg_policy_iocost, seq_cft(sf)->private, false);
3069 	return 0;
3070 }
3071 
3072 static ssize_t ioc_weight_write(struct kernfs_open_file *of, char *buf,
3073 				size_t nbytes, loff_t off)
3074 {
3075 	struct blkcg *blkcg = css_to_blkcg(of_css(of));
3076 	struct ioc_cgrp *iocc = blkcg_to_iocc(blkcg);
3077 	struct blkg_conf_ctx ctx;
3078 	struct ioc_now now;
3079 	struct ioc_gq *iocg;
3080 	u32 v;
3081 	int ret;
3082 
3083 	if (!strchr(buf, ':')) {
3084 		struct blkcg_gq *blkg;
3085 
3086 		if (!sscanf(buf, "default %u", &v) && !sscanf(buf, "%u", &v))
3087 			return -EINVAL;
3088 
3089 		if (v < CGROUP_WEIGHT_MIN || v > CGROUP_WEIGHT_MAX)
3090 			return -EINVAL;
3091 
3092 		spin_lock_irq(&blkcg->lock);
3093 		iocc->dfl_weight = v * WEIGHT_ONE;
3094 		hlist_for_each_entry(blkg, &blkcg->blkg_list, blkcg_node) {
3095 			struct ioc_gq *iocg = blkg_to_iocg(blkg);
3096 
3097 			if (iocg) {
3098 				spin_lock(&iocg->ioc->lock);
3099 				ioc_now(iocg->ioc, &now);
3100 				weight_updated(iocg, &now);
3101 				spin_unlock(&iocg->ioc->lock);
3102 			}
3103 		}
3104 		spin_unlock_irq(&blkcg->lock);
3105 
3106 		return nbytes;
3107 	}
3108 
3109 	blkg_conf_init(&ctx, buf);
3110 
3111 	ret = blkg_conf_prep(blkcg, &blkcg_policy_iocost, &ctx);
3112 	if (ret)
3113 		goto err;
3114 
3115 	iocg = blkg_to_iocg(ctx.blkg);
3116 
3117 	if (!strncmp(ctx.body, "default", 7)) {
3118 		v = 0;
3119 	} else {
3120 		if (!sscanf(ctx.body, "%u", &v))
3121 			goto einval;
3122 		if (v < CGROUP_WEIGHT_MIN || v > CGROUP_WEIGHT_MAX)
3123 			goto einval;
3124 	}
3125 
3126 	spin_lock(&iocg->ioc->lock);
3127 	iocg->cfg_weight = v * WEIGHT_ONE;
3128 	ioc_now(iocg->ioc, &now);
3129 	weight_updated(iocg, &now);
3130 	spin_unlock(&iocg->ioc->lock);
3131 
3132 	blkg_conf_exit(&ctx);
3133 	return nbytes;
3134 
3135 einval:
3136 	ret = -EINVAL;
3137 err:
3138 	blkg_conf_exit(&ctx);
3139 	return ret;
3140 }
3141 
3142 static u64 ioc_qos_prfill(struct seq_file *sf, struct blkg_policy_data *pd,
3143 			  int off)
3144 {
3145 	const char *dname = blkg_dev_name(pd->blkg);
3146 	struct ioc *ioc = pd_to_iocg(pd)->ioc;
3147 
3148 	if (!dname)
3149 		return 0;
3150 
3151 	spin_lock_irq(&ioc->lock);
3152 	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",
3153 		   dname, ioc->enabled, ioc->user_qos_params ? "user" : "auto",
3154 		   ioc->params.qos[QOS_RPPM] / 10000,
3155 		   ioc->params.qos[QOS_RPPM] % 10000 / 100,
3156 		   ioc->params.qos[QOS_RLAT],
3157 		   ioc->params.qos[QOS_WPPM] / 10000,
3158 		   ioc->params.qos[QOS_WPPM] % 10000 / 100,
3159 		   ioc->params.qos[QOS_WLAT],
3160 		   ioc->params.qos[QOS_MIN] / 10000,
3161 		   ioc->params.qos[QOS_MIN] % 10000 / 100,
3162 		   ioc->params.qos[QOS_MAX] / 10000,
3163 		   ioc->params.qos[QOS_MAX] % 10000 / 100);
3164 	spin_unlock_irq(&ioc->lock);
3165 	return 0;
3166 }
3167 
3168 static int ioc_qos_show(struct seq_file *sf, void *v)
3169 {
3170 	struct blkcg *blkcg = css_to_blkcg(seq_css(sf));
3171 
3172 	blkcg_print_blkgs(sf, blkcg, ioc_qos_prfill,
3173 			  &blkcg_policy_iocost, seq_cft(sf)->private, false);
3174 	return 0;
3175 }
3176 
3177 static const match_table_t qos_ctrl_tokens = {
3178 	{ QOS_ENABLE,		"enable=%u"	},
3179 	{ QOS_CTRL,		"ctrl=%s"	},
3180 	{ NR_QOS_CTRL_PARAMS,	NULL		},
3181 };
3182 
3183 static const match_table_t qos_tokens = {
3184 	{ QOS_RPPM,		"rpct=%s"	},
3185 	{ QOS_RLAT,		"rlat=%u"	},
3186 	{ QOS_WPPM,		"wpct=%s"	},
3187 	{ QOS_WLAT,		"wlat=%u"	},
3188 	{ QOS_MIN,		"min=%s"	},
3189 	{ QOS_MAX,		"max=%s"	},
3190 	{ NR_QOS_PARAMS,	NULL		},
3191 };
3192 
3193 static ssize_t ioc_qos_write(struct kernfs_open_file *of, char *input,
3194 			     size_t nbytes, loff_t off)
3195 {
3196 	struct blkg_conf_ctx ctx;
3197 	struct gendisk *disk;
3198 	struct ioc *ioc;
3199 	u32 qos[NR_QOS_PARAMS];
3200 	bool enable, user;
3201 	char *body, *p;
3202 	int ret;
3203 
3204 	blkg_conf_init(&ctx, input);
3205 
3206 	ret = blkg_conf_open_bdev(&ctx);
3207 	if (ret)
3208 		goto err;
3209 
3210 	body = ctx.body;
3211 	disk = ctx.bdev->bd_disk;
3212 	if (!queue_is_mq(disk->queue)) {
3213 		ret = -EOPNOTSUPP;
3214 		goto err;
3215 	}
3216 
3217 	ioc = q_to_ioc(disk->queue);
3218 	if (!ioc) {
3219 		ret = blk_iocost_init(disk);
3220 		if (ret)
3221 			goto err;
3222 		ioc = q_to_ioc(disk->queue);
3223 	}
3224 
3225 	blk_mq_freeze_queue(disk->queue);
3226 	blk_mq_quiesce_queue(disk->queue);
3227 
3228 	spin_lock_irq(&ioc->lock);
3229 	memcpy(qos, ioc->params.qos, sizeof(qos));
3230 	enable = ioc->enabled;
3231 	user = ioc->user_qos_params;
3232 
3233 	while ((p = strsep(&body, " \t\n"))) {
3234 		substring_t args[MAX_OPT_ARGS];
3235 		char buf[32];
3236 		int tok;
3237 		s64 v;
3238 
3239 		if (!*p)
3240 			continue;
3241 
3242 		switch (match_token(p, qos_ctrl_tokens, args)) {
3243 		case QOS_ENABLE:
3244 			if (match_u64(&args[0], &v))
3245 				goto einval;
3246 			enable = v;
3247 			continue;
3248 		case QOS_CTRL:
3249 			match_strlcpy(buf, &args[0], sizeof(buf));
3250 			if (!strcmp(buf, "auto"))
3251 				user = false;
3252 			else if (!strcmp(buf, "user"))
3253 				user = true;
3254 			else
3255 				goto einval;
3256 			continue;
3257 		}
3258 
3259 		tok = match_token(p, qos_tokens, args);
3260 		switch (tok) {
3261 		case QOS_RPPM:
3262 		case QOS_WPPM:
3263 			if (match_strlcpy(buf, &args[0], sizeof(buf)) >=
3264 			    sizeof(buf))
3265 				goto einval;
3266 			if (cgroup_parse_float(buf, 2, &v))
3267 				goto einval;
3268 			if (v < 0 || v > 10000)
3269 				goto einval;
3270 			qos[tok] = v * 100;
3271 			break;
3272 		case QOS_RLAT:
3273 		case QOS_WLAT:
3274 			if (match_u64(&args[0], &v))
3275 				goto einval;
3276 			qos[tok] = v;
3277 			break;
3278 		case QOS_MIN:
3279 		case QOS_MAX:
3280 			if (match_strlcpy(buf, &args[0], sizeof(buf)) >=
3281 			    sizeof(buf))
3282 				goto einval;
3283 			if (cgroup_parse_float(buf, 2, &v))
3284 				goto einval;
3285 			if (v < 0)
3286 				goto einval;
3287 			qos[tok] = clamp_t(s64, v * 100,
3288 					   VRATE_MIN_PPM, VRATE_MAX_PPM);
3289 			break;
3290 		default:
3291 			goto einval;
3292 		}
3293 		user = true;
3294 	}
3295 
3296 	if (qos[QOS_MIN] > qos[QOS_MAX])
3297 		goto einval;
3298 
3299 	if (enable) {
3300 		blk_stat_enable_accounting(disk->queue);
3301 		blk_queue_flag_set(QUEUE_FLAG_RQ_ALLOC_TIME, disk->queue);
3302 		ioc->enabled = true;
3303 		wbt_disable_default(disk);
3304 	} else {
3305 		blk_queue_flag_clear(QUEUE_FLAG_RQ_ALLOC_TIME, disk->queue);
3306 		ioc->enabled = false;
3307 		wbt_enable_default(disk);
3308 	}
3309 
3310 	if (user) {
3311 		memcpy(ioc->params.qos, qos, sizeof(qos));
3312 		ioc->user_qos_params = true;
3313 	} else {
3314 		ioc->user_qos_params = false;
3315 	}
3316 
3317 	ioc_refresh_params(ioc, true);
3318 	spin_unlock_irq(&ioc->lock);
3319 
3320 	blk_mq_unquiesce_queue(disk->queue);
3321 	blk_mq_unfreeze_queue(disk->queue);
3322 
3323 	blkg_conf_exit(&ctx);
3324 	return nbytes;
3325 einval:
3326 	spin_unlock_irq(&ioc->lock);
3327 
3328 	blk_mq_unquiesce_queue(disk->queue);
3329 	blk_mq_unfreeze_queue(disk->queue);
3330 
3331 	ret = -EINVAL;
3332 err:
3333 	blkg_conf_exit(&ctx);
3334 	return ret;
3335 }
3336 
3337 static u64 ioc_cost_model_prfill(struct seq_file *sf,
3338 				 struct blkg_policy_data *pd, int off)
3339 {
3340 	const char *dname = blkg_dev_name(pd->blkg);
3341 	struct ioc *ioc = pd_to_iocg(pd)->ioc;
3342 	u64 *u = ioc->params.i_lcoefs;
3343 
3344 	if (!dname)
3345 		return 0;
3346 
3347 	spin_lock_irq(&ioc->lock);
3348 	seq_printf(sf, "%s ctrl=%s model=linear "
3349 		   "rbps=%llu rseqiops=%llu rrandiops=%llu "
3350 		   "wbps=%llu wseqiops=%llu wrandiops=%llu\n",
3351 		   dname, ioc->user_cost_model ? "user" : "auto",
3352 		   u[I_LCOEF_RBPS], u[I_LCOEF_RSEQIOPS], u[I_LCOEF_RRANDIOPS],
3353 		   u[I_LCOEF_WBPS], u[I_LCOEF_WSEQIOPS], u[I_LCOEF_WRANDIOPS]);
3354 	spin_unlock_irq(&ioc->lock);
3355 	return 0;
3356 }
3357 
3358 static int ioc_cost_model_show(struct seq_file *sf, void *v)
3359 {
3360 	struct blkcg *blkcg = css_to_blkcg(seq_css(sf));
3361 
3362 	blkcg_print_blkgs(sf, blkcg, ioc_cost_model_prfill,
3363 			  &blkcg_policy_iocost, seq_cft(sf)->private, false);
3364 	return 0;
3365 }
3366 
3367 static const match_table_t cost_ctrl_tokens = {
3368 	{ COST_CTRL,		"ctrl=%s"	},
3369 	{ COST_MODEL,		"model=%s"	},
3370 	{ NR_COST_CTRL_PARAMS,	NULL		},
3371 };
3372 
3373 static const match_table_t i_lcoef_tokens = {
3374 	{ I_LCOEF_RBPS,		"rbps=%u"	},
3375 	{ I_LCOEF_RSEQIOPS,	"rseqiops=%u"	},
3376 	{ I_LCOEF_RRANDIOPS,	"rrandiops=%u"	},
3377 	{ I_LCOEF_WBPS,		"wbps=%u"	},
3378 	{ I_LCOEF_WSEQIOPS,	"wseqiops=%u"	},
3379 	{ I_LCOEF_WRANDIOPS,	"wrandiops=%u"	},
3380 	{ NR_I_LCOEFS,		NULL		},
3381 };
3382 
3383 static ssize_t ioc_cost_model_write(struct kernfs_open_file *of, char *input,
3384 				    size_t nbytes, loff_t off)
3385 {
3386 	struct blkg_conf_ctx ctx;
3387 	struct request_queue *q;
3388 	struct ioc *ioc;
3389 	u64 u[NR_I_LCOEFS];
3390 	bool user;
3391 	char *body, *p;
3392 	int ret;
3393 
3394 	blkg_conf_init(&ctx, input);
3395 
3396 	ret = blkg_conf_open_bdev(&ctx);
3397 	if (ret)
3398 		goto err;
3399 
3400 	body = ctx.body;
3401 	q = bdev_get_queue(ctx.bdev);
3402 	if (!queue_is_mq(q)) {
3403 		ret = -EOPNOTSUPP;
3404 		goto err;
3405 	}
3406 
3407 	ioc = q_to_ioc(q);
3408 	if (!ioc) {
3409 		ret = blk_iocost_init(ctx.bdev->bd_disk);
3410 		if (ret)
3411 			goto err;
3412 		ioc = q_to_ioc(q);
3413 	}
3414 
3415 	blk_mq_freeze_queue(q);
3416 	blk_mq_quiesce_queue(q);
3417 
3418 	spin_lock_irq(&ioc->lock);
3419 	memcpy(u, ioc->params.i_lcoefs, sizeof(u));
3420 	user = ioc->user_cost_model;
3421 
3422 	while ((p = strsep(&body, " \t\n"))) {
3423 		substring_t args[MAX_OPT_ARGS];
3424 		char buf[32];
3425 		int tok;
3426 		u64 v;
3427 
3428 		if (!*p)
3429 			continue;
3430 
3431 		switch (match_token(p, cost_ctrl_tokens, args)) {
3432 		case COST_CTRL:
3433 			match_strlcpy(buf, &args[0], sizeof(buf));
3434 			if (!strcmp(buf, "auto"))
3435 				user = false;
3436 			else if (!strcmp(buf, "user"))
3437 				user = true;
3438 			else
3439 				goto einval;
3440 			continue;
3441 		case COST_MODEL:
3442 			match_strlcpy(buf, &args[0], sizeof(buf));
3443 			if (strcmp(buf, "linear"))
3444 				goto einval;
3445 			continue;
3446 		}
3447 
3448 		tok = match_token(p, i_lcoef_tokens, args);
3449 		if (tok == NR_I_LCOEFS)
3450 			goto einval;
3451 		if (match_u64(&args[0], &v))
3452 			goto einval;
3453 		u[tok] = v;
3454 		user = true;
3455 	}
3456 
3457 	if (user) {
3458 		memcpy(ioc->params.i_lcoefs, u, sizeof(u));
3459 		ioc->user_cost_model = true;
3460 	} else {
3461 		ioc->user_cost_model = false;
3462 	}
3463 	ioc_refresh_params(ioc, true);
3464 	spin_unlock_irq(&ioc->lock);
3465 
3466 	blk_mq_unquiesce_queue(q);
3467 	blk_mq_unfreeze_queue(q);
3468 
3469 	blkg_conf_exit(&ctx);
3470 	return nbytes;
3471 
3472 einval:
3473 	spin_unlock_irq(&ioc->lock);
3474 
3475 	blk_mq_unquiesce_queue(q);
3476 	blk_mq_unfreeze_queue(q);
3477 
3478 	ret = -EINVAL;
3479 err:
3480 	blkg_conf_exit(&ctx);
3481 	return ret;
3482 }
3483 
3484 static struct cftype ioc_files[] = {
3485 	{
3486 		.name = "weight",
3487 		.flags = CFTYPE_NOT_ON_ROOT,
3488 		.seq_show = ioc_weight_show,
3489 		.write = ioc_weight_write,
3490 	},
3491 	{
3492 		.name = "cost.qos",
3493 		.flags = CFTYPE_ONLY_ON_ROOT,
3494 		.seq_show = ioc_qos_show,
3495 		.write = ioc_qos_write,
3496 	},
3497 	{
3498 		.name = "cost.model",
3499 		.flags = CFTYPE_ONLY_ON_ROOT,
3500 		.seq_show = ioc_cost_model_show,
3501 		.write = ioc_cost_model_write,
3502 	},
3503 	{}
3504 };
3505 
3506 static struct blkcg_policy blkcg_policy_iocost = {
3507 	.dfl_cftypes	= ioc_files,
3508 	.cpd_alloc_fn	= ioc_cpd_alloc,
3509 	.cpd_free_fn	= ioc_cpd_free,
3510 	.pd_alloc_fn	= ioc_pd_alloc,
3511 	.pd_init_fn	= ioc_pd_init,
3512 	.pd_free_fn	= ioc_pd_free,
3513 	.pd_stat_fn	= ioc_pd_stat,
3514 };
3515 
3516 static int __init ioc_init(void)
3517 {
3518 	return blkcg_policy_register(&blkcg_policy_iocost);
3519 }
3520 
3521 static void __exit ioc_exit(void)
3522 {
3523 	blkcg_policy_unregister(&blkcg_policy_iocost);
3524 }
3525 
3526 module_init(ioc_init);
3527 module_exit(ioc_exit);
3528