blob: e02a3883ae01444352775f6823d7a61a5b77dd6c [file] [log] [blame]
Jan Karaf3109a52012-05-24 18:59:10 +02001/*
2 * Floating proportions with flexible aging period
3 *
4 * Copyright (C) 2011, SUSE, Jan Kara <jack@suse.cz>
5 *
6 * The goal of this code is: Given different types of event, measure proportion
7 * of each type of event over time. The proportions are measured with
8 * exponentially decaying history to give smooth transitions. A formula
9 * expressing proportion of event of type 'j' is:
10 *
11 * p_{j} = (\Sum_{i>=0} x_{i,j}/2^{i+1})/(\Sum_{i>=0} x_i/2^{i+1})
12 *
13 * Where x_{i,j} is j's number of events in i-th last time period and x_i is
14 * total number of events in i-th last time period.
15 *
16 * Note that p_{j}'s are normalised, i.e.
17 *
18 * \Sum_{j} p_{j} = 1,
19 *
20 * This formula can be straightforwardly computed by maintaing denominator
21 * (let's call it 'd') and for each event type its numerator (let's call it
22 * 'n_j'). When an event of type 'j' happens, we simply need to do:
23 * n_j++; d++;
24 *
25 * When a new period is declared, we could do:
26 * d /= 2
27 * for each j
28 * n_j /= 2
29 *
30 * To avoid iteration over all event types, we instead shift numerator of event
31 * j lazily when someone asks for a proportion of event j or when event j
32 * occurs. This can bit trivially implemented by remembering last period in
33 * which something happened with proportion of type j.
34 */
35#include <linux/flex_proportions.h>
36
37int fprop_global_init(struct fprop_global *p)
38{
39 int err;
40
41 p->period = 0;
42 /* Use 1 to avoid dealing with periods with 0 events... */
43 err = percpu_counter_init(&p->events, 1);
44 if (err)
45 return err;
46 seqcount_init(&p->sequence);
47 return 0;
48}
49
50void fprop_global_destroy(struct fprop_global *p)
51{
52 percpu_counter_destroy(&p->events);
53}
54
55/*
56 * Declare @periods new periods. It is upto the caller to make sure period
57 * transitions cannot happen in parallel.
58 *
59 * The function returns true if the proportions are still defined and false
60 * if aging zeroed out all events. This can be used to detect whether declaring
61 * further periods has any effect.
62 */
63bool fprop_new_period(struct fprop_global *p, int periods)
64{
65 u64 events = percpu_counter_sum(&p->events);
66
67 /*
68 * Don't do anything if there are no events.
69 */
70 if (events <= 1)
71 return false;
72 write_seqcount_begin(&p->sequence);
73 if (periods < 64)
74 events -= events >> periods;
75 /* Use addition to avoid losing events happening between sum and set */
76 percpu_counter_add(&p->events, -events);
77 p->period += periods;
78 write_seqcount_end(&p->sequence);
79
80 return true;
81}
82
83/*
84 * ---- SINGLE ----
85 */
86
87int fprop_local_init_single(struct fprop_local_single *pl)
88{
89 pl->events = 0;
90 pl->period = 0;
91 raw_spin_lock_init(&pl->lock);
92 return 0;
93}
94
95void fprop_local_destroy_single(struct fprop_local_single *pl)
96{
97}
98
99static void fprop_reflect_period_single(struct fprop_global *p,
100 struct fprop_local_single *pl)
101{
102 unsigned int period = p->period;
103 unsigned long flags;
104
105 /* Fast path - period didn't change */
106 if (pl->period == period)
107 return;
108 raw_spin_lock_irqsave(&pl->lock, flags);
109 /* Someone updated pl->period while we were spinning? */
110 if (pl->period >= period) {
111 raw_spin_unlock_irqrestore(&pl->lock, flags);
112 return;
113 }
114 /* Aging zeroed our fraction? */
115 if (period - pl->period < BITS_PER_LONG)
116 pl->events >>= period - pl->period;
117 else
118 pl->events = 0;
119 pl->period = period;
120 raw_spin_unlock_irqrestore(&pl->lock, flags);
121}
122
123/* Event of type pl happened */
124void __fprop_inc_single(struct fprop_global *p, struct fprop_local_single *pl)
125{
126 fprop_reflect_period_single(p, pl);
127 pl->events++;
128 percpu_counter_add(&p->events, 1);
129}
130
131/* Return fraction of events of type pl */
132void fprop_fraction_single(struct fprop_global *p,
133 struct fprop_local_single *pl,
134 unsigned long *numerator, unsigned long *denominator)
135{
136 unsigned int seq;
137 s64 num, den;
138
139 do {
140 seq = read_seqcount_begin(&p->sequence);
141 fprop_reflect_period_single(p, pl);
142 num = pl->events;
143 den = percpu_counter_read_positive(&p->events);
144 } while (read_seqcount_retry(&p->sequence, seq));
145
146 /*
147 * Make fraction <= 1 and denominator > 0 even in presence of percpu
148 * counter errors
149 */
150 if (den <= num) {
151 if (num)
152 den = num;
153 else
154 den = 1;
155 }
156 *denominator = den;
157 *numerator = num;
158}
159
160/*
161 * ---- PERCPU ----
162 */
163#define PROP_BATCH (8*(1+ilog2(nr_cpu_ids)))
164
165int fprop_local_init_percpu(struct fprop_local_percpu *pl)
166{
167 int err;
168
169 err = percpu_counter_init(&pl->events, 0);
170 if (err)
171 return err;
172 pl->period = 0;
173 raw_spin_lock_init(&pl->lock);
174 return 0;
175}
176
177void fprop_local_destroy_percpu(struct fprop_local_percpu *pl)
178{
179 percpu_counter_destroy(&pl->events);
180}
181
182static void fprop_reflect_period_percpu(struct fprop_global *p,
183 struct fprop_local_percpu *pl)
184{
185 unsigned int period = p->period;
186 unsigned long flags;
187
188 /* Fast path - period didn't change */
189 if (pl->period == period)
190 return;
191 raw_spin_lock_irqsave(&pl->lock, flags);
192 /* Someone updated pl->period while we were spinning? */
193 if (pl->period >= period) {
194 raw_spin_unlock_irqrestore(&pl->lock, flags);
195 return;
196 }
197 /* Aging zeroed our fraction? */
198 if (period - pl->period < BITS_PER_LONG) {
199 s64 val = percpu_counter_read(&pl->events);
200
201 if (val < (nr_cpu_ids * PROP_BATCH))
202 val = percpu_counter_sum(&pl->events);
203
204 __percpu_counter_add(&pl->events,
205 -val + (val >> (period-pl->period)), PROP_BATCH);
206 } else
207 percpu_counter_set(&pl->events, 0);
208 pl->period = period;
209 raw_spin_unlock_irqrestore(&pl->lock, flags);
210}
211
212/* Event of type pl happened */
213void __fprop_inc_percpu(struct fprop_global *p, struct fprop_local_percpu *pl)
214{
215 fprop_reflect_period_percpu(p, pl);
216 __percpu_counter_add(&pl->events, 1, PROP_BATCH);
217 percpu_counter_add(&p->events, 1);
218}
219
220void fprop_fraction_percpu(struct fprop_global *p,
221 struct fprop_local_percpu *pl,
222 unsigned long *numerator, unsigned long *denominator)
223{
224 unsigned int seq;
225 s64 num, den;
226
227 do {
228 seq = read_seqcount_begin(&p->sequence);
229 fprop_reflect_period_percpu(p, pl);
230 num = percpu_counter_read_positive(&pl->events);
231 den = percpu_counter_read_positive(&p->events);
232 } while (read_seqcount_retry(&p->sequence, seq));
233
234 /*
235 * Make fraction <= 1 and denominator > 0 even in presence of percpu
236 * counter errors
237 */
238 if (den <= num) {
239 if (num)
240 den = num;
241 else
242 den = 1;
243 }
244 *denominator = den;
245 *numerator = num;
246}
247
248/*
249 * Like __fprop_inc_percpu() except that event is counted only if the given
250 * type has fraction smaller than @max_frac/FPROP_FRAC_BASE
251 */
252void __fprop_inc_percpu_max(struct fprop_global *p,
253 struct fprop_local_percpu *pl, int max_frac)
254{
255 if (unlikely(max_frac < FPROP_FRAC_BASE)) {
256 unsigned long numerator, denominator;
257
258 fprop_fraction_percpu(p, pl, &numerator, &denominator);
259 if (numerator >
260 (((u64)denominator) * max_frac) >> FPROP_FRAC_SHIFT)
261 return;
262 } else
263 fprop_reflect_period_percpu(p, pl);
264 __percpu_counter_add(&pl->events, 1, PROP_BATCH);
265 percpu_counter_add(&p->events, 1);
266}