1 // SPDX-License-Identifier: GPL-2.0
3 * Floating proportions with flexible aging period
5 * Copyright (C) 2011, SUSE, Jan Kara <jack@suse.cz>
7 * The goal of this code is: Given different types of event, measure proportion
8 * of each type of event over time. The proportions are measured with
9 * exponentially decaying history to give smooth transitions. A formula
10 * expressing proportion of event of type 'j' is:
12 * p_{j} = (\Sum_{i>=0} x_{i,j}/2^{i+1})/(\Sum_{i>=0} x_i/2^{i+1})
14 * Where x_{i,j} is j's number of events in i-th last time period and x_i is
15 * total number of events in i-th last time period.
17 * Note that p_{j}'s are normalised, i.e.
21 * This formula can be straightforwardly computed by maintaining denominator
22 * (let's call it 'd') and for each event type its numerator (let's call it
23 * 'n_j'). When an event of type 'j' happens, we simply need to do:
26 * When a new period is declared, we could do:
31 * To avoid iteration over all event types, we instead shift numerator of event
32 * j lazily when someone asks for a proportion of event j or when event j
33 * occurs. This can bit trivially implemented by remembering last period in
34 * which something happened with proportion of type j.
36 #include <linux/flex_proportions.h>
38 int fprop_global_init(struct fprop_global *p, gfp_t gfp)
43 /* Use 1 to avoid dealing with periods with 0 events... */
44 err = percpu_counter_init(&p->events, 1, gfp);
47 seqcount_init(&p->sequence);
51 void fprop_global_destroy(struct fprop_global *p)
53 percpu_counter_destroy(&p->events);
57 * Declare @periods new periods. It is upto the caller to make sure period
58 * transitions cannot happen in parallel.
60 * The function returns true if the proportions are still defined and false
61 * if aging zeroed out all events. This can be used to detect whether declaring
62 * further periods has any effect.
64 bool fprop_new_period(struct fprop_global *p, int periods)
69 local_irq_save(flags);
70 events = percpu_counter_sum(&p->events);
72 * Don't do anything if there are no events.
75 local_irq_restore(flags);
78 write_seqcount_begin(&p->sequence);
80 events -= events >> periods;
81 /* Use addition to avoid losing events happening between sum and set */
82 percpu_counter_add(&p->events, -events);
84 write_seqcount_end(&p->sequence);
85 local_irq_restore(flags);
94 int fprop_local_init_single(struct fprop_local_single *pl)
98 raw_spin_lock_init(&pl->lock);
102 void fprop_local_destroy_single(struct fprop_local_single *pl)
106 static void fprop_reflect_period_single(struct fprop_global *p,
107 struct fprop_local_single *pl)
109 unsigned int period = p->period;
112 /* Fast path - period didn't change */
113 if (pl->period == period)
115 raw_spin_lock_irqsave(&pl->lock, flags);
116 /* Someone updated pl->period while we were spinning? */
117 if (pl->period >= period) {
118 raw_spin_unlock_irqrestore(&pl->lock, flags);
121 /* Aging zeroed our fraction? */
122 if (period - pl->period < BITS_PER_LONG)
123 pl->events >>= period - pl->period;
127 raw_spin_unlock_irqrestore(&pl->lock, flags);
130 /* Event of type pl happened */
131 void __fprop_inc_single(struct fprop_global *p, struct fprop_local_single *pl)
133 fprop_reflect_period_single(p, pl);
135 percpu_counter_add(&p->events, 1);
138 /* Return fraction of events of type pl */
139 void fprop_fraction_single(struct fprop_global *p,
140 struct fprop_local_single *pl,
141 unsigned long *numerator, unsigned long *denominator)
147 seq = read_seqcount_begin(&p->sequence);
148 fprop_reflect_period_single(p, pl);
150 den = percpu_counter_read_positive(&p->events);
151 } while (read_seqcount_retry(&p->sequence, seq));
154 * Make fraction <= 1 and denominator > 0 even in presence of percpu
170 #define PROP_BATCH (8*(1+ilog2(nr_cpu_ids)))
172 int fprop_local_init_percpu(struct fprop_local_percpu *pl, gfp_t gfp)
176 err = percpu_counter_init(&pl->events, 0, gfp);
180 raw_spin_lock_init(&pl->lock);
184 void fprop_local_destroy_percpu(struct fprop_local_percpu *pl)
186 percpu_counter_destroy(&pl->events);
189 static void fprop_reflect_period_percpu(struct fprop_global *p,
190 struct fprop_local_percpu *pl)
192 unsigned int period = p->period;
195 /* Fast path - period didn't change */
196 if (pl->period == period)
198 raw_spin_lock_irqsave(&pl->lock, flags);
199 /* Someone updated pl->period while we were spinning? */
200 if (pl->period >= period) {
201 raw_spin_unlock_irqrestore(&pl->lock, flags);
204 /* Aging zeroed our fraction? */
205 if (period - pl->period < BITS_PER_LONG) {
206 s64 val = percpu_counter_read(&pl->events);
208 if (val < (nr_cpu_ids * PROP_BATCH))
209 val = percpu_counter_sum(&pl->events);
211 percpu_counter_add_batch(&pl->events,
212 -val + (val >> (period-pl->period)), PROP_BATCH);
214 percpu_counter_set(&pl->events, 0);
216 raw_spin_unlock_irqrestore(&pl->lock, flags);
219 /* Event of type pl happened */
220 void __fprop_add_percpu(struct fprop_global *p, struct fprop_local_percpu *pl,
223 fprop_reflect_period_percpu(p, pl);
224 percpu_counter_add_batch(&pl->events, nr, PROP_BATCH);
225 percpu_counter_add(&p->events, nr);
228 void fprop_fraction_percpu(struct fprop_global *p,
229 struct fprop_local_percpu *pl,
230 unsigned long *numerator, unsigned long *denominator)
236 seq = read_seqcount_begin(&p->sequence);
237 fprop_reflect_period_percpu(p, pl);
238 num = percpu_counter_read_positive(&pl->events);
239 den = percpu_counter_read_positive(&p->events);
240 } while (read_seqcount_retry(&p->sequence, seq));
243 * Make fraction <= 1 and denominator > 0 even in presence of percpu
257 * Like __fprop_add_percpu() except that event is counted only if the given
258 * type has fraction smaller than @max_frac/FPROP_FRAC_BASE
260 void __fprop_add_percpu_max(struct fprop_global *p,
261 struct fprop_local_percpu *pl, int max_frac, long nr)
263 if (unlikely(max_frac < FPROP_FRAC_BASE)) {
264 unsigned long numerator, denominator;
267 fprop_fraction_percpu(p, pl, &numerator, &denominator);
268 /* Adding 'nr' to fraction exceeds max_frac/FPROP_FRAC_BASE? */
269 tmp = (u64)denominator * max_frac -
270 ((u64)numerator << FPROP_FRAC_SHIFT);
272 /* Maximum fraction already exceeded? */
274 } else if (tmp < nr * (FPROP_FRAC_BASE - max_frac)) {
275 /* Add just enough for the fraction to saturate */
276 nr = div_u64(tmp + FPROP_FRAC_BASE - max_frac - 1,
277 FPROP_FRAC_BASE - max_frac);
281 __fprop_add_percpu(p, pl, nr);