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(rqos: rq_qos_id(q, id: 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(pd: blkg_to_pd(blkg, pol: &blkcg_policy_iocost));
687	}
688
689	static struct blkcg_gq *iocg_to_blkg(struct ioc_gq *iocg)
690	{
691	return pd_to_blkg(pd: &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(i: cost, v: &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(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(lock: &iocg->waitq.lock);
744	spin_unlock_irqrestore(lock: &iocg->ioc->lock, flags: *flags);
745	} else {
746	spin_unlock_irqrestore(lock: &iocg->waitq.lock, flags: *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	dividend: (u64)period_us * NSEC_PER_USEC * TIMER_SLACK_PCT,
799	divisor: 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(q: 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(dividend: ioc->vtime_base_rate * 100, divisor: 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(bps: u[I_LCOEF_RBPS], seqiops: u[I_LCOEF_RSEQIOPS], randiops: u[I_LCOEF_RRANDIOPS],
903	page: &c[LCOEF_RPAGE], seqio: &c[LCOEF_RSEQIO], randio: &c[LCOEF_RRANDIO]);
904	calc_lcoefs(bps: u[I_LCOEF_WBPS], seqiops: u[I_LCOEF_WSEQIOPS], randiops: u[I_LCOEF_WRANDIOPS],
905	page: &c[LCOEF_WPAGE], seqio: &c[LCOEF_WSEQIO], randio: &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(v: &ioc->vtime_rate, i: 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, disk: 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(dividend: ioc->vtime_err, divisor: 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(v: &ioc->vtime_rate, i: 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, new_vrate: 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(dividend: vrate * (100 + VRATE_CLAMP_ADJ_PCT), divisor: 100);
1016	vrate = min(vrate, vrate_min);
1017	} else if (vrate > vrate_max) {
1018	vrate = div64_u64(dividend: vrate * (100 - VRATE_CLAMP_ADJ_PCT), divisor: 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, new_vrate: 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(kt: now->now_ns);
1049	vrate = atomic64_read(v: &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(u: ioc->period_us);
1076	add_timer(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(head: &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(v: &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(v: &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(ioc: 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(v: &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(dividend: (u64)hwa * active, divisor: active_sum);
1205
1206	inuse_sum = max_t(u64, inuse, inuse_sum);
1207	hwi = div64_u64(dividend: (u64)hwi * inuse, divisor: 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(dividend: (u64)hwm * inuse, divisor: 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(blkcg: 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, active: weight, inuse: iocg->inuse, save: 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(head: &iocg->active_list)) {
1273	ioc_now(ioc, now);
1274	cur_period = atomic64_read(v: &ioc->cur_period);
1275	if (atomic64_read(v: &iocg->active_period) != cur_period)
1276	atomic64_set(v: &iocg->active_period, i: 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(lock: &ioc->lock);
1285
1286	ioc_now(ioc, now);
1287
1288	/* update period */
1289	cur_period = atomic64_read(v: &ioc->cur_period);
1290	last_period = atomic64_read(v: &iocg->active_period);
1291	atomic64_set(v: &iocg->active_period, i: cur_period);
1292
1293	/* already activated or breaking leaf-only constraint? */
1294	if (!list_empty(head: &iocg->active_list))
1295	goto succeed_unlock;
1296	for (i = iocg->level - 1; i > 0; i--)
1297	if (!list_empty(head: &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(v: &iocg->vtime);
1309
1310	atomic64_add(i: vtarget - vtime, v: &iocg->vtime);
1311	atomic64_add(i: vtarget - vtime, v: &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(v: &ioc->hweight_gen) - 1;
1320	list_add(new: &iocg->active_list, head: &ioc->active_iocgs);
1321
1322	propagate_weights(iocg, active: iocg->weight,
1323	inuse: iocg->last_inuse ?: iocg->weight, save: 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(lock: &ioc->lock);
1339	return true;
1340
1341	fail_unlock:
1342	spin_unlock_irq(lock: &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(dividend: tdelta, USEC_PER_SEC);
1360	else
1361	delay = 0;
1362
1363	/* calculate the new delay from the debt amount */
1364	current_hweight(iocg, hw_activep: &hwa, NULL);
1365	vover = atomic64_read(v: &iocg->vtime) +
1366	abs_cost_to_cost(abs_cost: iocg->abs_vdebt, hw_inuse: hwa) - now->vnow;
1367	vover_pct = div64_s64(dividend: 100 * vover,
1368	divisor: 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(dividend: (MAX_DELAY - MIN_DELAY) *
1377	(vover_pct - MIN_DELAY_THR_PCT),
1378	divisor: 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: 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, active: iocg->active, inuse: 0, save: 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, active: iocg->active, inuse: iocg->last_inuse,
1446	save: 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(abs_cost: wait->abs_cost, hw_inuse: ctx->hw_inuse);
1456
1457	ctx->vbudget -= cost;
1458
1459	if (ctx->vbudget < 0)
1460	return -1;
1461
1462	iocg_commit_bio(iocg: ctx->iocg, bio: wait->bio, abs_cost: 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(entry: &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, hw_activep: &hwa, NULL);
1494	vbudget = now->vnow - atomic64_read(v: &iocg->vtime);
1495
1496	/* pay off debt */
1497	if (pay_debt && iocg->abs_vdebt && vbudget > 0) {
1498	u64 abs_vbudget = cost_to_abs_cost(cost: vbudget, hw_inuse: hwa);
1499	u64 abs_vpay = min_t(u64, abs_vbudget, iocg->abs_vdebt);
1500	u64 vpay = abs_cost_to_cost(abs_cost: abs_vpay, hw_inuse: hwa);
1501
1502	lockdep_assert_held(&ioc->lock);
1503
1504	atomic64_add(i: vpay, v: &iocg->vtime);
1505	atomic64_add(i: vpay, v: &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(abs_cost: iocg->abs_vdebt, hw_inuse: 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, hw_inusep: &ctx.hw_inuse);
1531
1532	__wake_up_locked_key(wq_head: &iocg->waitq, TASK_NORMAL, key: &ctx);
1533
1534	if (!waitqueue_active(wq_head: &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(kt: hrtimer_get_softexpires(timer: &iocg->waitq_timer));
1557	if (hrtimer_is_queued(timer: &iocg->waitq_timer) &&
1558	abs(oexpires - expires) <= ioc->timer_slack_ns)
1559	return;
1560
1561	hrtimer_start_range_ns(timer: &iocg->waitq_timer, tim: ns_to_ktime(ns: expires),
1562	range_ns: ioc->timer_slack_ns, mode: 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(ioc: iocg->ioc, now: &now);
1573
1574	iocg_lock(iocg, lock_ioc: pay_debt, flags: &flags);
1575	iocg_kick_waitq(iocg, pay_debt, now: &now);
1576	iocg_unlock(iocg, unlock_ioc: pay_debt, flags: &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(dividend: rq_wait_ns * 100,
1617	divisor: 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(v: &iocg->active_period) ==
1627	atomic64_read(v: &ioc->cur_period))
1628	return false;
1629
1630	/* is something in flight? */
1631	if (atomic64_read(v: &iocg->done_vtime) != atomic64_read(v: &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(head: &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(new: &inner->walk_list, head: 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(dividend: vusage_delta, divisor: 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: &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(entry: &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(v: &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(wq_head: &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(i: excess, v: &iocg->vtime);
1752	atomic64_add(i: excess, v: &iocg->done_vtime);
1753	vtime += excess;
1754	ioc->vtime_err -= div64_u64(dividend: excess * old_hwi, divisor: 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(dividend: WEIGHT_ONE * (now->vnow - vtime),
1774	divisor: now->vnow - ioc->period_at_vtime);
1775	target = WEIGHT_ONE * MARGIN_TARGET_PCT / 100;
1776	new_hwi = div64_s64(dividend: WEIGHT_ONE * usage, divisor: 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, hw_activep: &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(new: &iocg->walk_list, head: &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(dividend: (u64)iocg->hweight_after_donation *
1882	over_target, divisor: over_sum);
1883	list_del_init(entry: &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: &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(cgrp: 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, active: iocg->active, inuse, save: 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(entry: &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(dividend: 100 * ioc->dfgv_usage_us_sum, divisor: 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(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, pay_debt: 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(lock: &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(wq_head: &iocg->waitq) && !iocg->abs_vdebt &&
2162	!iocg->delay && !iocg_is_idle(iocg))
2163	continue;
2164
2165	spin_lock(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(wq_head: &iocg->waitq) || iocg->abs_vdebt ||
2184	iocg->delay) {
2185	/* might be oversleeping vtime / hweight changes, kick */
2186	iocg_kick_waitq(iocg, pay_debt: 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(v: &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, hw_inusep: &old_hwi);
2205	ioc->vtime_err -= div64_u64(dividend: excess * old_hwi,
2206	divisor: 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, active: 0, inuse: 0, save: false, now);
2213	list_del_init(entry: &iocg->active_list);
2214	}
2215
2216	spin_unlock(lock: &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_ar: missed_ppm, rq_wait_pct_p: &rq_wait_pct);
2239
2240	/* take care of active iocgs */
2241	spin_lock_irq(lock: &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: &now);
2246
2247	period_vtime = now.vnow - ioc->period_at_vtime;
2248	if (WARN_ON_ONCE(!period_vtime)) {
2249	spin_unlock_irq(lock: &ioc->lock);
2250	return;
2251	}
2252
2253	nr_debtors = ioc_check_iocgs(ioc, now: &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(target_iocgs: &ioc->active_iocgs, now: &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(v: &iocg->done_vtime);
2271	vtime = atomic64_read(v: &iocg->vtime);
2272	current_hweight(iocg, hw_activep: &hw_active, hw_inusep: &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(v: &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(wq_head: &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, hw_activep: &hwa, hw_inusep: &old_hwi);
2327	hwm = current_hweight_max(iocg);
2328	new_hwi = hweight_after_donation(iocg, old_hwi, hwm,
2329	usage, now: &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(new: &iocg->surplus_list, head: &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, active: iocg->active,
2357	inuse: iocg->active, save: true, now: &now);
2358	nr_shortages++;
2359	}
2360	} else {
2361	/* genuinely short on vtime */
2362	nr_shortages++;
2363	}
2364	}
2365
2366	if (!list_empty(head: &surpluses) && nr_shortages)
2367	transfer_surpluses(surpluses: &surpluses, now: &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(entry: &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, force: false);
2425
2426	ioc_forgive_debts(ioc, usage_us_sum, nr_debtors, now: &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(v: &ioc->cur_period);
2433
2434	if (ioc->running != IOC_STOP) {
2435	if (!list_empty(head: &ioc->active_iocgs)) {
2436	ioc_start_period(ioc, now: &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: &now);
2444	}
2445
2446	spin_unlock_irq(lock: &ioc->lock);
2447	}
2448
2449	static u64 adjust_inuse_and_calc_cost(struct ioc_gq *iocg, u64 vtime,
2450	u64 abs_cost, struct ioc_now *now)
2451	{
2452	struct ioc *ioc = iocg->ioc;
2453	struct ioc_margins *margins = &ioc->margins;
2454	u32 __maybe_unused old_inuse = iocg->inuse, __maybe_unused old_hwi;
2455	u32 hwi, adj_step;
2456	s64 margin;
2457	u64 cost, new_inuse;
2458	unsigned long flags;
2459
2460	current_hweight(iocg, NULL, hw_inusep: &hwi);
2461	old_hwi = hwi;
2462	cost = abs_cost_to_cost(abs_cost, hw_inuse: hwi);
2463	margin = now->vnow - vtime - cost;
2464
2465	/* debt handling owns inuse for debtors */
2466	if (iocg->abs_vdebt)
2467	return cost;
2468
2469	/*
2470	* We only increase inuse during period and do so if the margin has
2471	* deteriorated since the previous adjustment.
2472	*/
2473	if (margin >= iocg->saved_margin || margin >= margins->low ||
2474	iocg->inuse == iocg->active)
2475	return cost;
2476
2477	spin_lock_irqsave(&ioc->lock, flags);
2478
2479	/* we own inuse only when @iocg is in the normal active state */
2480	if (iocg->abs_vdebt || list_empty(head: &iocg->active_list)) {
2481	spin_unlock_irqrestore(lock: &ioc->lock, flags);
2482	return cost;
2483	}
2484
2485	/*
2486	* Bump up inuse till @abs_cost fits in the existing budget.
2487	* adj_step must be determined after acquiring ioc->lock - we might
2488	* have raced and lost to another thread for activation and could
2489	* be reading 0 iocg->active before ioc->lock which will lead to
2490	* infinite loop.
2491	*/
2492	new_inuse = iocg->inuse;
2493	adj_step = DIV_ROUND_UP(iocg->active * INUSE_ADJ_STEP_PCT, 100);
2494	do {
2495	new_inuse = new_inuse + adj_step;
2496	propagate_weights(iocg, active: iocg->active, inuse: new_inuse, save: true, now);
2497	current_hweight(iocg, NULL, hw_inusep: &hwi);
2498	cost = abs_cost_to_cost(abs_cost, hw_inuse: hwi);
2499	} while (time_after64(vtime + cost, now->vnow) &&
2500	iocg->inuse != iocg->active);
2501
2502	spin_unlock_irqrestore(lock: &ioc->lock, flags);
2503
2504	TRACE_IOCG_PATH(inuse_adjust, iocg, now,
2505	old_inuse, iocg->inuse, old_hwi, hwi);
2506
2507	return cost;
2508	}
2509
2510	static void calc_vtime_cost_builtin(struct bio *bio, struct ioc_gq *iocg,
2511	bool is_merge, u64 *costp)
2512	{
2513	struct ioc *ioc = iocg->ioc;
2514	u64 coef_seqio, coef_randio, coef_page;
2515	u64 pages = max_t(u64, bio_sectors(bio) >> IOC_SECT_TO_PAGE_SHIFT, 1);
2516	u64 seek_pages = 0;
2517	u64 cost = 0;
2518
2519	/* Can't calculate cost for empty bio */
2520	if (!bio->bi_iter.bi_size)
2521	goto out;
2522
2523	switch (bio_op(bio)) {
2524	case REQ_OP_READ:
2525	coef_seqio = ioc->params.lcoefs[LCOEF_RSEQIO];
2526	coef_randio = ioc->params.lcoefs[LCOEF_RRANDIO];
2527	coef_page = ioc->params.lcoefs[LCOEF_RPAGE];
2528	break;
2529	case REQ_OP_WRITE:
2530	coef_seqio = ioc->params.lcoefs[LCOEF_WSEQIO];
2531	coef_randio = ioc->params.lcoefs[LCOEF_WRANDIO];
2532	coef_page = ioc->params.lcoefs[LCOEF_WPAGE];
2533	break;
2534	default:
2535	goto out;
2536	}
2537
2538	if (iocg->cursor) {
2539	seek_pages = abs(bio->bi_iter.bi_sector - iocg->cursor);
2540	seek_pages >>= IOC_SECT_TO_PAGE_SHIFT;
2541	}
2542
2543	if (!is_merge) {
2544	if (seek_pages > LCOEF_RANDIO_PAGES) {
2545	cost += coef_randio;
2546	} else {
2547	cost += coef_seqio;
2548	}
2549	}
2550	cost += pages * coef_page;
2551	out:
2552	*costp = cost;
2553	}
2554
2555	static u64 calc_vtime_cost(struct bio *bio, struct ioc_gq *iocg, bool is_merge)
2556	{
2557	u64 cost;
2558
2559	calc_vtime_cost_builtin(bio, iocg, is_merge, costp: &cost);
2560	return cost;
2561	}
2562
2563	static void calc_size_vtime_cost_builtin(struct request *rq, struct ioc *ioc,
2564	u64 *costp)
2565	{
2566	unsigned int pages = blk_rq_stats_sectors(rq) >> IOC_SECT_TO_PAGE_SHIFT;
2567
2568	switch (req_op(req: rq)) {
2569	case REQ_OP_READ:
2570	*costp = pages * ioc->params.lcoefs[LCOEF_RPAGE];
2571	break;
2572	case REQ_OP_WRITE:
2573	*costp = pages * ioc->params.lcoefs[LCOEF_WPAGE];
2574	break;
2575	default:
2576	*costp = 0;
2577	}
2578	}
2579
2580	static u64 calc_size_vtime_cost(struct request *rq, struct ioc *ioc)
2581	{
2582	u64 cost;
2583
2584	calc_size_vtime_cost_builtin(rq, ioc, costp: &cost);
2585	return cost;
2586	}
2587
2588	static void ioc_rqos_throttle(struct rq_qos *rqos, struct bio *bio)
2589	{
2590	struct blkcg_gq *blkg = bio->bi_blkg;
2591	struct ioc *ioc = rqos_to_ioc(rqos);
2592	struct ioc_gq *iocg = blkg_to_iocg(blkg);
2593	struct ioc_now now;
2594	struct iocg_wait wait;
2595	u64 abs_cost, cost, vtime;
2596	bool use_debt, ioc_locked;
2597	unsigned long flags;
2598
2599	/* bypass IOs if disabled, still initializing, or for root cgroup */
2600	if (!ioc->enabled || !iocg || !iocg->level)
2601	return;
2602
2603	/* calculate the absolute vtime cost */
2604	abs_cost = calc_vtime_cost(bio, iocg, is_merge: false);
2605	if (!abs_cost)
2606	return;
2607
2608	if (!iocg_activate(iocg, now: &now))
2609	return;
2610
2611	iocg->cursor = bio_end_sector(bio);
2612	vtime = atomic64_read(v: &iocg->vtime);
2613	cost = adjust_inuse_and_calc_cost(iocg, vtime, abs_cost, now: &now);
2614
2615	/*
2616	* If no one's waiting and within budget, issue right away. The
2617	* tests are racy but the races aren't systemic - we only miss once
2618	* in a while which is fine.
2619	*/
2620	if (!waitqueue_active(wq_head: &iocg->waitq) && !iocg->abs_vdebt &&
2621	time_before_eq64(vtime + cost, now.vnow)) {
2622	iocg_commit_bio(iocg, bio, abs_cost, cost);
2623	return;
2624	}
2625
2626	/*
2627	* We're over budget. This can be handled in two ways. IOs which may
2628	* cause priority inversions are punted to @ioc->aux_iocg and charged as
2629	* debt. Otherwise, the issuer is blocked on @iocg->waitq. Debt handling
2630	* requires @ioc->lock, waitq handling @iocg->waitq.lock. Determine
2631	* whether debt handling is needed and acquire locks accordingly.
2632	*/
2633	use_debt = bio_issue_as_root_blkg(bio) || fatal_signal_pending(current);
2634	ioc_locked = use_debt || READ_ONCE(iocg->abs_vdebt);
2635	retry_lock:
2636	iocg_lock(iocg, lock_ioc: ioc_locked, flags: &flags);
2637
2638	/*
2639	* @iocg must stay activated for debt and waitq handling. Deactivation
2640	* is synchronized against both ioc->lock and waitq.lock and we won't
2641	* get deactivated as long as we're waiting or has debt, so we're good
2642	* if we're activated here. In the unlikely cases that we aren't, just
2643	* issue the IO.
2644	*/
2645	if (unlikely(list_empty(&iocg->active_list))) {
2646	iocg_unlock(iocg, unlock_ioc: ioc_locked, flags: &flags);
2647	iocg_commit_bio(iocg, bio, abs_cost, cost);
2648	return;
2649	}
2650
2651	/*
2652	* We're over budget. If @bio has to be issued regardless, remember
2653	* the abs_cost instead of advancing vtime. iocg_kick_waitq() will pay
2654	* off the debt before waking more IOs.
2655	*
2656	* This way, the debt is continuously paid off each period with the
2657	* actual budget available to the cgroup. If we just wound vtime, we
2658	* would incorrectly use the current hw_inuse for the entire amount
2659	* which, for example, can lead to the cgroup staying blocked for a
2660	* long time even with substantially raised hw_inuse.
2661	*
2662	* An iocg with vdebt should stay online so that the timer can keep
2663	* deducting its vdebt and [de]activate use_delay mechanism
2664	* accordingly. We don't want to race against the timer trying to
2665	* clear them and leave @iocg inactive w/ dangling use_delay heavily
2666	* penalizing the cgroup and its descendants.
2667	*/
2668	if (use_debt) {
2669	iocg_incur_debt(iocg, abs_cost, now: &now);
2670	if (iocg_kick_delay(iocg, now: &now))
2671	blkcg_schedule_throttle(disk: rqos->disk,
2672	use_memdelay: (bio->bi_opf & REQ_SWAP) == REQ_SWAP);
2673	iocg_unlock(iocg, unlock_ioc: ioc_locked, flags: &flags);
2674	return;
2675	}
2676
2677	/* guarantee that iocgs w/ waiters have maximum inuse */
2678	if (!iocg->abs_vdebt && iocg->inuse != iocg->active) {
2679	if (!ioc_locked) {
2680	iocg_unlock(iocg, unlock_ioc: false, flags: &flags);
2681	ioc_locked = true;
2682	goto retry_lock;
2683	}
2684	propagate_weights(iocg, active: iocg->active, inuse: iocg->active, save: true,
2685	now: &now);
2686	}
2687
2688	/*
2689	* Append self to the waitq and schedule the wakeup timer if we're
2690	* the first waiter. The timer duration is calculated based on the
2691	* current vrate. vtime and hweight changes can make it too short
2692	* or too long. Each wait entry records the absolute cost it's
2693	* waiting for to allow re-evaluation using a custom wait entry.
2694	*
2695	* If too short, the timer simply reschedules itself. If too long,
2696	* the period timer will notice and trigger wakeups.
2697	*
2698	* All waiters are on iocg->waitq and the wait states are
2699	* synchronized using waitq.lock.
2700	*/
2701	init_waitqueue_func_entry(wq_entry: &wait.wait, func: iocg_wake_fn);
2702	wait.wait.private = current;
2703	wait.bio = bio;
2704	wait.abs_cost = abs_cost;
2705	wait.committed = false; /* will be set true by waker */
2706
2707	__add_wait_queue_entry_tail(wq_head: &iocg->waitq, wq_entry: &wait.wait);
2708	iocg_kick_waitq(iocg, pay_debt: ioc_locked, now: &now);
2709
2710	iocg_unlock(iocg, unlock_ioc: ioc_locked, flags: &flags);
2711
2712	while (true) {
2713	set_current_state(TASK_UNINTERRUPTIBLE);
2714	if (wait.committed)
2715	break;
2716	io_schedule();
2717	}
2718
2719	/* waker already committed us, proceed */
2720	finish_wait(wq_head: &iocg->waitq, wq_entry: &wait.wait);
2721	}
2722
2723	static void ioc_rqos_merge(struct rq_qos *rqos, struct request *rq,
2724	struct bio *bio)
2725	{
2726	struct ioc_gq *iocg = blkg_to_iocg(blkg: bio->bi_blkg);
2727	struct ioc *ioc = rqos_to_ioc(rqos);
2728	sector_t bio_end = bio_end_sector(bio);
2729	struct ioc_now now;
2730	u64 vtime, abs_cost, cost;
2731	unsigned long flags;
2732
2733	/* bypass if disabled, still initializing, or for root cgroup */
2734	if (!ioc->enabled || !iocg || !iocg->level)
2735	return;
2736
2737	abs_cost = calc_vtime_cost(bio, iocg, is_merge: true);
2738	if (!abs_cost)
2739	return;
2740
2741	ioc_now(ioc, now: &now);
2742
2743	vtime = atomic64_read(v: &iocg->vtime);
2744	cost = adjust_inuse_and_calc_cost(iocg, vtime, abs_cost, now: &now);
2745
2746	/* update cursor if backmerging into the request at the cursor */
2747	if (blk_rq_pos(rq) < bio_end &&
2748	blk_rq_pos(rq) + blk_rq_sectors(rq) == iocg->cursor)
2749	iocg->cursor = bio_end;
2750
2751	/*
2752	* Charge if there's enough vtime budget and the existing request has
2753	* cost assigned.
2754	*/
2755	if (rq->bio && rq->bio->bi_iocost_cost &&
2756	time_before_eq64(atomic64_read(&iocg->vtime) + cost, now.vnow)) {
2757	iocg_commit_bio(iocg, bio, abs_cost, cost);
2758	return;
2759	}
2760
2761	/*
2762	* Otherwise, account it as debt if @iocg is online, which it should
2763	* be for the vast majority of cases. See debt handling in
2764	* ioc_rqos_throttle() for details.
2765	*/
2766	spin_lock_irqsave(&ioc->lock, flags);
2767	spin_lock(lock: &iocg->waitq.lock);
2768
2769	if (likely(!list_empty(&iocg->active_list))) {
2770	iocg_incur_debt(iocg, abs_cost, now: &now);
2771	if (iocg_kick_delay(iocg, now: &now))
2772	blkcg_schedule_throttle(disk: rqos->disk,
2773	use_memdelay: (bio->bi_opf & REQ_SWAP) == REQ_SWAP);
2774	} else {
2775	iocg_commit_bio(iocg, bio, abs_cost, cost);
2776	}
2777
2778	spin_unlock(lock: &iocg->waitq.lock);
2779	spin_unlock_irqrestore(lock: &ioc->lock, flags);
2780	}
2781
2782	static void ioc_rqos_done_bio(struct rq_qos *rqos, struct bio *bio)
2783	{
2784	struct ioc_gq *iocg = blkg_to_iocg(blkg: bio->bi_blkg);
2785
2786	if (iocg && bio->bi_iocost_cost)
2787	atomic64_add(i: bio->bi_iocost_cost, v: &iocg->done_vtime);
2788	}
2789
2790	static void ioc_rqos_done(struct rq_qos *rqos, struct request *rq)
2791	{
2792	struct ioc *ioc = rqos_to_ioc(rqos);
2793	struct ioc_pcpu_stat *ccs;
2794	u64 on_q_ns, rq_wait_ns, size_nsec;
2795	int pidx, rw;
2796
2797	if (!ioc->enabled || !rq->alloc_time_ns || !rq->start_time_ns)
2798	return;
2799
2800	switch (req_op(req: rq)) {
2801	case REQ_OP_READ:
2802	pidx = QOS_RLAT;
2803	rw = READ;
2804	break;
2805	case REQ_OP_WRITE:
2806	pidx = QOS_WLAT;
2807	rw = WRITE;
2808	break;
2809	default:
2810	return;
2811	}
2812
2813	on_q_ns = ktime_get_ns() - rq->alloc_time_ns;
2814	rq_wait_ns = rq->start_time_ns - rq->alloc_time_ns;
2815	size_nsec = div64_u64(dividend: calc_size_vtime_cost(rq, ioc), divisor: VTIME_PER_NSEC);
2816
2817	ccs = get_cpu_ptr(ioc->pcpu_stat);
2818
2819	if (on_q_ns <= size_nsec ||
2820	on_q_ns - size_nsec <= ioc->params.qos[pidx] * NSEC_PER_USEC)
2821	local_inc(l: &ccs->missed[rw].nr_met);
2822	else
2823	local_inc(l: &ccs->missed[rw].nr_missed);
2824
2825	local64_add(rq_wait_ns, &ccs->rq_wait_ns);
2826
2827	put_cpu_ptr(ccs);
2828	}
2829
2830	static void ioc_rqos_queue_depth_changed(struct rq_qos *rqos)
2831	{
2832	struct ioc *ioc = rqos_to_ioc(rqos);
2833
2834	spin_lock_irq(lock: &ioc->lock);
2835	ioc_refresh_params(ioc, force: false);
2836	spin_unlock_irq(lock: &ioc->lock);
2837	}
2838
2839	static void ioc_rqos_exit(struct rq_qos *rqos)
2840	{
2841	struct ioc *ioc = rqos_to_ioc(rqos);
2842
2843	blkcg_deactivate_policy(disk: rqos->disk, pol: &blkcg_policy_iocost);
2844
2845	spin_lock_irq(lock: &ioc->lock);
2846	ioc->running = IOC_STOP;
2847	spin_unlock_irq(lock: &ioc->lock);
2848
2849	timer_shutdown_sync(timer: &ioc->timer);
2850	free_percpu(pdata: ioc->pcpu_stat);
2851	kfree(objp: ioc);
2852	}
2853
2854	static const struct rq_qos_ops ioc_rqos_ops = {
2855	.throttle = ioc_rqos_throttle,
2856	.merge = ioc_rqos_merge,
2857	.done_bio = ioc_rqos_done_bio,
2858	.done = ioc_rqos_done,
2859	.queue_depth_changed = ioc_rqos_queue_depth_changed,
2860	.exit = ioc_rqos_exit,
2861	};
2862
2863	static int blk_iocost_init(struct gendisk *disk)
2864	{
2865	struct ioc *ioc;
2866	int i, cpu, ret;
2867
2868	ioc = kzalloc(size: sizeof(*ioc), GFP_KERNEL);
2869	if (!ioc)
2870	return -ENOMEM;
2871
2872	ioc->pcpu_stat = alloc_percpu(struct ioc_pcpu_stat);
2873	if (!ioc->pcpu_stat) {
2874	kfree(objp: ioc);
2875	return -ENOMEM;
2876	}
2877
2878	for_each_possible_cpu(cpu) {
2879	struct ioc_pcpu_stat *ccs = per_cpu_ptr(ioc->pcpu_stat, cpu);
2880
2881	for (i = 0; i < ARRAY_SIZE(ccs->missed); i++) {
2882	local_set(&ccs->missed[i].nr_met, 0);
2883	local_set(&ccs->missed[i].nr_missed, 0);
2884	}
2885	local64_set(&ccs->rq_wait_ns, 0);
2886	}
2887
2888	spin_lock_init(&ioc->lock);
2889	timer_setup(&ioc->timer, ioc_timer_fn, 0);
2890	INIT_LIST_HEAD(list: &ioc->active_iocgs);
2891
2892	ioc->running = IOC_IDLE;
2893	ioc->vtime_base_rate = VTIME_PER_USEC;
2894	atomic64_set(v: &ioc->vtime_rate, i: VTIME_PER_USEC);
2895	seqcount_spinlock_init(&ioc->period_seqcount, &ioc->lock);
2896	ioc->period_at = ktime_to_us(kt: ktime_get());
2897	atomic64_set(v: &ioc->cur_period, i: 0);
2898	atomic_set(v: &ioc->hweight_gen, i: 0);
2899
2900	spin_lock_irq(lock: &ioc->lock);
2901	ioc->autop_idx = AUTOP_INVALID;
2902	ioc_refresh_params_disk(ioc, force: true, disk);
2903	spin_unlock_irq(lock: &ioc->lock);
2904
2905	/*
2906	* rqos must be added before activation to allow ioc_pd_init() to
2907	* lookup the ioc from q. This means that the rqos methods may get
2908	* called before policy activation completion, can't assume that the
2909	* target bio has an iocg associated and need to test for NULL iocg.
2910	*/
2911	ret = rq_qos_add(rqos: &ioc->rqos, disk, id: RQ_QOS_COST, ops: &ioc_rqos_ops);
2912	if (ret)
2913	goto err_free_ioc;
2914
2915	ret = blkcg_activate_policy(disk, pol: &blkcg_policy_iocost);
2916	if (ret)
2917	goto err_del_qos;
2918	return 0;
2919
2920	err_del_qos:
2921	rq_qos_del(rqos: &ioc->rqos);
2922	err_free_ioc:
2923	free_percpu(pdata: ioc->pcpu_stat);
2924	kfree(objp: ioc);
2925	return ret;
2926	}
2927
2928	static struct blkcg_policy_data *ioc_cpd_alloc(gfp_t gfp)
2929	{
2930	struct ioc_cgrp *iocc;
2931
2932	iocc = kzalloc(size: sizeof(struct ioc_cgrp), flags: gfp);
2933	if (!iocc)
2934	return NULL;
2935
2936	iocc->dfl_weight = CGROUP_WEIGHT_DFL * WEIGHT_ONE;
2937	return &iocc->cpd;
2938	}
2939
2940	static void ioc_cpd_free(struct blkcg_policy_data *cpd)
2941	{
2942	kfree(container_of(cpd, struct ioc_cgrp, cpd));
2943	}
2944
2945	static struct blkg_policy_data *ioc_pd_alloc(struct gendisk *disk,
2946	struct blkcg *blkcg, gfp_t gfp)
2947	{
2948	int levels = blkcg->css.cgroup->level + 1;
2949	struct ioc_gq *iocg;
2950
2951	iocg = kzalloc_node(struct_size(iocg, ancestors, levels), flags: gfp,
2952	node: disk->node_id);
2953	if (!iocg)
2954	return NULL;
2955
2956	iocg->pcpu_stat = alloc_percpu_gfp(struct iocg_pcpu_stat, gfp);
2957	if (!iocg->pcpu_stat) {
2958	kfree(objp: iocg);
2959	return NULL;
2960	}
2961
2962	return &iocg->pd;
2963	}
2964
2965	static void ioc_pd_init(struct blkg_policy_data *pd)
2966	{
2967	struct ioc_gq *iocg = pd_to_iocg(pd);
2968	struct blkcg_gq *blkg = pd_to_blkg(pd: &iocg->pd);
2969	struct ioc *ioc = q_to_ioc(q: blkg->q);
2970	struct ioc_now now;
2971	struct blkcg_gq *tblkg;
2972	unsigned long flags;
2973
2974	ioc_now(ioc, now: &now);
2975
2976	iocg->ioc = ioc;
2977	atomic64_set(v: &iocg->vtime, i: now.vnow);
2978	atomic64_set(v: &iocg->done_vtime, i: now.vnow);
2979	atomic64_set(v: &iocg->active_period, i: atomic64_read(v: &ioc->cur_period));
2980	INIT_LIST_HEAD(list: &iocg->active_list);
2981	INIT_LIST_HEAD(list: &iocg->walk_list);
2982	INIT_LIST_HEAD(list: &iocg->surplus_list);
2983	iocg->hweight_active = WEIGHT_ONE;
2984	iocg->hweight_inuse = WEIGHT_ONE;
2985
2986	init_waitqueue_head(&iocg->waitq);
2987	hrtimer_init(timer: &iocg->waitq_timer, CLOCK_MONOTONIC, mode: HRTIMER_MODE_ABS);
2988	iocg->waitq_timer.function = iocg_waitq_timer_fn;
2989
2990	iocg->level = blkg->blkcg->css.cgroup->level;
2991
2992	for (tblkg = blkg; tblkg; tblkg = tblkg->parent) {
2993	struct ioc_gq *tiocg = blkg_to_iocg(blkg: tblkg);
2994	iocg->ancestors[tiocg->level] = tiocg;
2995	}
2996
2997	spin_lock_irqsave(&ioc->lock, flags);
2998	weight_updated(iocg, now: &now);
2999	spin_unlock_irqrestore(lock: &ioc->lock, flags);
3000	}
3001
3002	static void ioc_pd_free(struct blkg_policy_data *pd)
3003	{
3004	struct ioc_gq *iocg = pd_to_iocg(pd);
3005	struct ioc *ioc = iocg->ioc;
3006	unsigned long flags;
3007
3008	if (ioc) {
3009	spin_lock_irqsave(&ioc->lock, flags);
3010
3011	if (!list_empty(head: &iocg->active_list)) {
3012	struct ioc_now now;
3013
3014	ioc_now(ioc, now: &now);
3015	propagate_weights(iocg, active: 0, inuse: 0, save: false, now: &now);
3016	list_del_init(entry: &iocg->active_list);
3017	}
3018
3019	WARN_ON_ONCE(!list_empty(&iocg->walk_list));
3020	WARN_ON_ONCE(!list_empty(&iocg->surplus_list));
3021
3022	spin_unlock_irqrestore(lock: &ioc->lock, flags);
3023
3024	hrtimer_cancel(timer: &iocg->waitq_timer);
3025	}
3026	free_percpu(pdata: iocg->pcpu_stat);
3027	kfree(objp: iocg);
3028	}
3029
3030	static void ioc_pd_stat(struct blkg_policy_data *pd, struct seq_file *s)
3031	{
3032	struct ioc_gq *iocg = pd_to_iocg(pd);
3033	struct ioc *ioc = iocg->ioc;
3034
3035	if (!ioc->enabled)
3036	return;
3037
3038	if (iocg->level == 0) {
3039	unsigned vp10k = DIV64_U64_ROUND_CLOSEST(
3040	ioc->vtime_base_rate * 10000,
3041	VTIME_PER_USEC);
3042	seq_printf(m: s, fmt: " cost.vrate=%u.%02u", vp10k / 100, vp10k % 100);
3043	}
3044
3045	seq_printf(m: s, fmt: " cost.usage=%llu", iocg->last_stat.usage_us);
3046
3047	if (blkcg_debug_stats)
3048	seq_printf(m: s, fmt: " cost.wait=%llu cost.indebt=%llu cost.indelay=%llu",
3049	iocg->last_stat.wait_us,
3050	iocg->last_stat.indebt_us,
3051	iocg->last_stat.indelay_us);
3052	}
3053
3054	static u64 ioc_weight_prfill(struct seq_file *sf, struct blkg_policy_data *pd,
3055	int off)
3056	{
3057	const char *dname = blkg_dev_name(blkg: pd->blkg);
3058	struct ioc_gq *iocg = pd_to_iocg(pd);
3059
3060	if (dname && iocg->cfg_weight)
3061	seq_printf(m: sf, fmt: "%s %u\n", dname, iocg->cfg_weight / WEIGHT_ONE);
3062	return 0;
3063	}
3064
3065
3066	static int ioc_weight_show(struct seq_file *sf, void *v)
3067	{
3068	struct blkcg *blkcg = css_to_blkcg(css: seq_css(seq: sf));
3069	struct ioc_cgrp *iocc = blkcg_to_iocc(blkcg);
3070
3071	seq_printf(m: sf, fmt: "default %u\n", iocc->dfl_weight / WEIGHT_ONE);
3072	blkcg_print_blkgs(sf, blkcg, prfill: ioc_weight_prfill,
3073	pol: &blkcg_policy_iocost, data: seq_cft(seq: sf)->private, show_total: false);
3074	return 0;
3075	}
3076
3077	static ssize_t ioc_weight_write(struct kernfs_open_file *of, char *buf,
3078	size_t nbytes, loff_t off)
3079	{
3080	struct blkcg *blkcg = css_to_blkcg(css: of_css(of));
3081	struct ioc_cgrp *iocc = blkcg_to_iocc(blkcg);
3082	struct blkg_conf_ctx ctx;
3083	struct ioc_now now;
3084	struct ioc_gq *iocg;
3085	u32 v;
3086	int ret;
3087
3088	if (!strchr(buf, ':')) {
3089	struct blkcg_gq *blkg;
3090
3091	if (!sscanf(buf, "default %u", &v) && !sscanf(buf, "%u", &v))
3092	return -EINVAL;
3093
3094	if (v < CGROUP_WEIGHT_MIN || v > CGROUP_WEIGHT_MAX)
3095	return -EINVAL;
3096
3097	spin_lock_irq(lock: &blkcg->lock);
3098	iocc->dfl_weight = v * WEIGHT_ONE;
3099	hlist_for_each_entry(blkg, &blkcg->blkg_list, blkcg_node) {
3100	struct ioc_gq *iocg = blkg_to_iocg(blkg);
3101
3102	if (iocg) {
3103	spin_lock(lock: &iocg->ioc->lock);
3104	ioc_now(ioc: iocg->ioc, now: &now);
3105	weight_updated(iocg, now: &now);
3106	spin_unlock(lock: &iocg->ioc->lock);
3107	}
3108	}
3109	spin_unlock_irq(lock: &blkcg->lock);
3110
3111	return nbytes;
3112	}
3113
3114	blkg_conf_init(ctx: &ctx, input: buf);
3115
3116	ret = blkg_conf_prep(blkcg, pol: &blkcg_policy_iocost, ctx: &ctx);
3117	if (ret)
3118	goto err;
3119
3120	iocg = blkg_to_iocg(blkg: ctx.blkg);
3121
3122	if (!strncmp(ctx.body, "default", 7)) {
3123	v = 0;
3124	} else {
3125	if (!sscanf(ctx.body, "%u", &v))
3126	goto einval;
3127	if (v < CGROUP_WEIGHT_MIN || v > CGROUP_WEIGHT_MAX)
3128	goto einval;
3129	}
3130
3131	spin_lock(lock: &iocg->ioc->lock);
3132	iocg->cfg_weight = v * WEIGHT_ONE;
3133	ioc_now(ioc: iocg->ioc, now: &now);
3134	weight_updated(iocg, now: &now);
3135	spin_unlock(lock: &iocg->ioc->lock);
3136
3137	blkg_conf_exit(ctx: &ctx);
3138	return nbytes;
3139
3140	einval:
3141	ret = -EINVAL;
3142	err:
3143	blkg_conf_exit(ctx: &ctx);
3144	return ret;
3145	}
3146
3147	static u64 ioc_qos_prfill(struct seq_file *sf, struct blkg_policy_data *pd,
3148	int off)
3149	{
3150	const char *dname = blkg_dev_name(blkg: pd->blkg);
3151	struct ioc *ioc = pd_to_iocg(pd)->ioc;
3152
3153	if (!dname)
3154	return 0;
3155
3156	spin_lock_irq(lock: &ioc->lock);
3157	seq_printf(m: sf, fmt: "%s enable=%d ctrl=%s rpct=%u.%02u rlat=%u wpct=%u.%02u wlat=%u min=%u.%02u max=%u.%02u\n",
3158	dname, ioc->enabled, ioc->user_qos_params ? "user" : "auto",
3159	ioc->params.qos[QOS_RPPM] / 10000,
3160	ioc->params.qos[QOS_RPPM] % 10000 / 100,
3161	ioc->params.qos[QOS_RLAT],
3162	ioc->params.qos[QOS_WPPM] / 10000,
3163	ioc->params.qos[QOS_WPPM] % 10000 / 100,
3164	ioc->params.qos[QOS_WLAT],
3165	ioc->params.qos[QOS_MIN] / 10000,
3166	ioc->params.qos[QOS_MIN] % 10000 / 100,
3167	ioc->params.qos[QOS_MAX] / 10000,
3168	ioc->params.qos[QOS_MAX] % 10000 / 100);
3169	spin_unlock_irq(lock: &ioc->lock);
3170	return 0;
3171	}
3172
3173	static int ioc_qos_show(struct seq_file *sf, void *v)
3174	{
3175	struct blkcg *blkcg = css_to_blkcg(css: seq_css(seq: sf));
3176
3177	blkcg_print_blkgs(sf, blkcg, prfill: ioc_qos_prfill,
3178	pol: &blkcg_policy_iocost, data: seq_cft(seq: sf)->private, show_total: false);
3179	return 0;
3180	}
3181
3182	static const match_table_t qos_ctrl_tokens = {
3183	{ QOS_ENABLE, "enable=%u" },
3184	{ QOS_CTRL, "ctrl=%s" },
3185	{ NR_QOS_CTRL_PARAMS, NULL },
3186	};
3187
3188	static const match_table_t qos_tokens = {
3189	{ QOS_RPPM, "rpct=%s" },
3190	{ QOS_RLAT, "rlat=%u" },
3191	{ QOS_WPPM, "wpct=%s" },
3192	{ QOS_WLAT, "wlat=%u" },
3193	{ QOS_MIN, "min=%s" },
3194	{ QOS_MAX, "max=%s" },
3195	{ NR_QOS_PARAMS, NULL },
3196	};
3197
3198	static ssize_t ioc_qos_write(struct kernfs_open_file *of, char *input,
3199	size_t nbytes, loff_t off)
3200	{
3201	struct blkg_conf_ctx ctx;
3202	struct gendisk *disk;
3203	struct ioc *ioc;
3204	u32 qos[NR_QOS_PARAMS];
3205	bool enable, user;
3206	char *body, *p;
3207	int ret;
3208
3209	blkg_conf_init(ctx: &ctx, input);
3210
3211	ret = blkg_conf_open_bdev(ctx: &ctx);
3212	if (ret)
3213	goto err;
3214
3215	body = ctx.body;
3216	disk = ctx.bdev->bd_disk;
3217	if (!queue_is_mq(q: disk->queue)) {
3218	ret = -EOPNOTSUPP;
3219	goto err;
3220	}
3221
3222	ioc = q_to_ioc(q: disk->queue);
3223	if (!ioc) {
3224	ret = blk_iocost_init(disk);
3225	if (ret)
3226	goto err;
3227	ioc = q_to_ioc(q: disk->queue);
3228	}
3229
3230	blk_mq_freeze_queue(q: disk->queue);
3231	blk_mq_quiesce_queue(q: disk->queue);
3232
3233	spin_lock_irq(lock: &ioc->lock);
3234	memcpy(qos, ioc->params.qos, sizeof(qos));
3235	enable = ioc->enabled;
3236	user = ioc->user_qos_params;
3237
3238	while ((p = strsep(&body, " \t\n"))) {
3239	substring_t args[MAX_OPT_ARGS];
3240	char buf[32];
3241	int tok;
3242	s64 v;
3243
3244	if (!*p)
3245	continue;
3246
3247	switch (match_token(p, table: qos_ctrl_tokens, args)) {
3248	case QOS_ENABLE:
3249	if (match_u64(&args[0], result: &v))
3250	goto einval;
3251	enable = v;
3252	continue;
3253	case QOS_CTRL:
3254	match_strlcpy(buf, &args[0], sizeof(buf));
3255	if (!strcmp(buf, "auto"))
3256	user = false;
3257	else if (!strcmp(buf, "user"))
3258	user = true;
3259	else
3260	goto einval;
3261	continue;
3262	}
3263
3264	tok = match_token(p, table: qos_tokens, args);
3265	switch (tok) {
3266	case QOS_RPPM:
3267	case QOS_WPPM:
3268	if (match_strlcpy(buf, &args[0], sizeof(buf)) >=
3269	sizeof(buf))
3270	goto einval;
3271	if (cgroup_parse_float(input: buf, dec_shift: 2, v: &v))
3272	goto einval;
3273	if (v < 0 || v > 10000)
3274	goto einval;
3275	qos[tok] = v * 100;
3276	break;
3277	case QOS_RLAT:
3278	case QOS_WLAT:
3279	if (match_u64(&args[0], result: &v))
3280	goto einval;
3281	qos[tok] = v;
3282	break;
3283	case QOS_MIN:
3284	case QOS_MAX:
3285	if (match_strlcpy(buf, &args[0], sizeof(buf)) >=
3286	sizeof(buf))
3287	goto einval;
3288	if (cgroup_parse_float(input: buf, dec_shift: 2, v: &v))
3289	goto einval;
3290	if (v < 0)
3291	goto einval;
3292	qos[tok] = clamp_t(s64, v * 100,
3293	VRATE_MIN_PPM, VRATE_MAX_PPM);
3294	break;
3295	default:
3296	goto einval;
3297	}
3298	user = true;
3299	}
3300
3301	if (qos[QOS_MIN] > qos[QOS_MAX])
3302	goto einval;
3303
3304	if (enable && !ioc->enabled) {
3305	blk_stat_enable_accounting(q: disk->queue);
3306	blk_queue_flag_set(QUEUE_FLAG_RQ_ALLOC_TIME, q: disk->queue);
3307	ioc->enabled = true;
3308	} else if (!enable && ioc->enabled) {
3309	blk_stat_disable_accounting(q: disk->queue);
3310	blk_queue_flag_clear(QUEUE_FLAG_RQ_ALLOC_TIME, q: disk->queue);
3311	ioc->enabled = false;
3312	}
3313
3314	if (user) {
3315	memcpy(ioc->params.qos, qos, sizeof(qos));
3316	ioc->user_qos_params = true;
3317	} else {
3318	ioc->user_qos_params = false;
3319	}
3320
3321	ioc_refresh_params(ioc, force: true);
3322	spin_unlock_irq(lock: &ioc->lock);
3323
3324	if (enable)
3325	wbt_disable_default(disk);
3326	else
3327	wbt_enable_default(disk);
3328
3329	blk_mq_unquiesce_queue(q: disk->queue);
3330	blk_mq_unfreeze_queue(q: disk->queue);
3331
3332	blkg_conf_exit(ctx: &ctx);
3333	return nbytes;
3334	einval:
3335	spin_unlock_irq(lock: &ioc->lock);
3336
3337	blk_mq_unquiesce_queue(q: disk->queue);
3338	blk_mq_unfreeze_queue(q: disk->queue);
3339
3340	ret = -EINVAL;
3341	err:
3342	blkg_conf_exit(ctx: &ctx);
3343	return ret;
3344	}
3345
3346	static u64 ioc_cost_model_prfill(struct seq_file *sf,
3347	struct blkg_policy_data *pd, int off)
3348	{
3349	const char *dname = blkg_dev_name(blkg: pd->blkg);
3350	struct ioc *ioc = pd_to_iocg(pd)->ioc;
3351	u64 *u = ioc->params.i_lcoefs;
3352
3353	if (!dname)
3354	return 0;
3355
3356	spin_lock_irq(lock: &ioc->lock);
3357	seq_printf(m: sf, fmt: "%s ctrl=%s model=linear "
3358	"rbps=%llu rseqiops=%llu rrandiops=%llu "
3359	"wbps=%llu wseqiops=%llu wrandiops=%llu\n",
3360	dname, ioc->user_cost_model ? "user" : "auto",
3361	u[I_LCOEF_RBPS], u[I_LCOEF_RSEQIOPS], u[I_LCOEF_RRANDIOPS],
3362	u[I_LCOEF_WBPS], u[I_LCOEF_WSEQIOPS], u[I_LCOEF_WRANDIOPS]);
3363	spin_unlock_irq(lock: &ioc->lock);
3364	return 0;
3365	}
3366
3367	static int ioc_cost_model_show(struct seq_file *sf, void *v)
3368	{
3369	struct blkcg *blkcg = css_to_blkcg(css: seq_css(seq: sf));
3370
3371	blkcg_print_blkgs(sf, blkcg, prfill: ioc_cost_model_prfill,
3372	pol: &blkcg_policy_iocost, data: seq_cft(seq: sf)->private, show_total: false);
3373	return 0;
3374	}
3375
3376	static const match_table_t cost_ctrl_tokens = {
3377	{ COST_CTRL, "ctrl=%s" },
3378	{ COST_MODEL, "model=%s" },
3379	{ NR_COST_CTRL_PARAMS, NULL },
3380	};
3381
3382	static const match_table_t i_lcoef_tokens = {
3383	{ I_LCOEF_RBPS, "rbps=%u" },
3384	{ I_LCOEF_RSEQIOPS, "rseqiops=%u" },
3385	{ I_LCOEF_RRANDIOPS, "rrandiops=%u" },
3386	{ I_LCOEF_WBPS, "wbps=%u" },
3387	{ I_LCOEF_WSEQIOPS, "wseqiops=%u" },
3388	{ I_LCOEF_WRANDIOPS, "wrandiops=%u" },
3389	{ NR_I_LCOEFS, NULL },
3390	};
3391
3392	static ssize_t ioc_cost_model_write(struct kernfs_open_file *of, char *input,
3393	size_t nbytes, loff_t off)
3394	{
3395	struct blkg_conf_ctx ctx;
3396	struct request_queue *q;
3397	struct ioc *ioc;
3398	u64 u[NR_I_LCOEFS];
3399	bool user;
3400	char *body, *p;
3401	int ret;
3402
3403	blkg_conf_init(ctx: &ctx, input);
3404
3405	ret = blkg_conf_open_bdev(ctx: &ctx);
3406	if (ret)
3407	goto err;
3408
3409	body = ctx.body;
3410	q = bdev_get_queue(bdev: ctx.bdev);
3411	if (!queue_is_mq(q)) {
3412	ret = -EOPNOTSUPP;
3413	goto err;
3414	}
3415
3416	ioc = q_to_ioc(q);
3417	if (!ioc) {
3418	ret = blk_iocost_init(disk: ctx.bdev->bd_disk);
3419	if (ret)
3420	goto err;
3421	ioc = q_to_ioc(q);
3422	}
3423
3424	blk_mq_freeze_queue(q);
3425	blk_mq_quiesce_queue(q);
3426
3427	spin_lock_irq(lock: &ioc->lock);
3428	memcpy(u, ioc->params.i_lcoefs, sizeof(u));
3429	user = ioc->user_cost_model;
3430
3431	while ((p = strsep(&body, " \t\n"))) {
3432	substring_t args[MAX_OPT_ARGS];
3433	char buf[32];
3434	int tok;
3435	u64 v;
3436
3437	if (!*p)
3438	continue;
3439
3440	switch (match_token(p, table: cost_ctrl_tokens, args)) {
3441	case COST_CTRL:
3442	match_strlcpy(buf, &args[0], sizeof(buf));
3443	if (!strcmp(buf, "auto"))
3444	user = false;
3445	else if (!strcmp(buf, "user"))
3446	user = true;
3447	else
3448	goto einval;
3449	continue;
3450	case COST_MODEL:
3451	match_strlcpy(buf, &args[0], sizeof(buf));
3452	if (strcmp(buf, "linear"))
3453	goto einval;
3454	continue;
3455	}
3456
3457	tok = match_token(p, table: i_lcoef_tokens, args);
3458	if (tok == NR_I_LCOEFS)
3459	goto einval;
3460	if (match_u64(&args[0], result: &v))
3461	goto einval;
3462	u[tok] = v;
3463	user = true;
3464	}
3465
3466	if (user) {
3467	memcpy(ioc->params.i_lcoefs, u, sizeof(u));
3468	ioc->user_cost_model = true;
3469	} else {
3470	ioc->user_cost_model = false;
3471	}
3472	ioc_refresh_params(ioc, force: true);
3473	spin_unlock_irq(lock: &ioc->lock);
3474
3475	blk_mq_unquiesce_queue(q);
3476	blk_mq_unfreeze_queue(q);
3477
3478	blkg_conf_exit(ctx: &ctx);
3479	return nbytes;
3480
3481	einval:
3482	spin_unlock_irq(lock: &ioc->lock);
3483
3484	blk_mq_unquiesce_queue(q);
3485	blk_mq_unfreeze_queue(q);
3486
3487	ret = -EINVAL;
3488	err:
3489	blkg_conf_exit(ctx: &ctx);
3490	return ret;
3491	}
3492
3493	static struct cftype ioc_files[] = {
3494	{
3495	.name = "weight",
3496	.flags = CFTYPE_NOT_ON_ROOT,
3497	.seq_show = ioc_weight_show,
3498	.write = ioc_weight_write,
3499	},
3500	{
3501	.name = "cost.qos",
3502	.flags = CFTYPE_ONLY_ON_ROOT,
3503	.seq_show = ioc_qos_show,
3504	.write = ioc_qos_write,
3505	},
3506	{
3507	.name = "cost.model",
3508	.flags = CFTYPE_ONLY_ON_ROOT,
3509	.seq_show = ioc_cost_model_show,
3510	.write = ioc_cost_model_write,
3511	},
3512	{}
3513	};
3514
3515	static struct blkcg_policy blkcg_policy_iocost = {
3516	.dfl_cftypes = ioc_files,
3517	.cpd_alloc_fn = ioc_cpd_alloc,
3518	.cpd_free_fn = ioc_cpd_free,
3519	.pd_alloc_fn = ioc_pd_alloc,
3520	.pd_init_fn = ioc_pd_init,
3521	.pd_free_fn = ioc_pd_free,
3522	.pd_stat_fn = ioc_pd_stat,
3523	};
3524
3525	static int __init ioc_init(void)
3526	{
3527	return blkcg_policy_register(pol: &blkcg_policy_iocost);
3528	}
3529
3530	static void __exit ioc_exit(void)
3531	{
3532	blkcg_policy_unregister(pol: &blkcg_policy_iocost);
3533	}
3534
3535	module_init(ioc_init);
3536	module_exit(ioc_exit);
3537