1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * Floating proportions with flexible aging period 4 * 5 * Copyright (C) 2011, SUSE, Jan Kara <jack@suse.cz> 6 * 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: 11 * 12 * p_{j} = (\Sum_{i>=0} x_{i,j}/2^{i+1})/(\Sum_{i>=0} x_i/2^{i+1}) 13 * 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. 16 * 17 * Note that p_{j}'s are normalised, i.e. 18 * 19 * \Sum_{j} p_{j} = 1, 20 * 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: 24 * n_j++; d++; 25 * 26 * When a new period is declared, we could do: 27 * d /= 2 28 * for each j 29 * n_j /= 2 30 * 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. 35 */ 36 #include <linux/flex_proportions.h> 37 38 int fprop_global_init(struct fprop_global *p, gfp_t gfp) 39 { 40 int err; 41 42 p->period = 0; 43 /* Use 1 to avoid dealing with periods with 0 events... */ 44 err = percpu_counter_init(&p->events, 1, gfp); 45 if (err) 46 return err; 47 seqcount_init(&p->sequence); 48 return 0; 49 } 50 51 void fprop_global_destroy(struct fprop_global *p) 52 { 53 percpu_counter_destroy(&p->events); 54 } 55 56 /* 57 * Declare @periods new periods. It is upto the caller to make sure period 58 * transitions cannot happen in parallel. 59 * 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. 63 */ 64 bool fprop_new_period(struct fprop_global *p, int periods) 65 { 66 s64 events = percpu_counter_sum(&p->events); 67 68 /* 69 * Don't do anything if there are no events. 70 */ 71 if (events <= 1) 72 return false; 73 write_seqcount_begin(&p->sequence); 74 if (periods < 64) 75 events -= events >> periods; 76 /* Use addition to avoid losing events happening between sum and set */ 77 percpu_counter_add(&p->events, -events); 78 p->period += periods; 79 write_seqcount_end(&p->sequence); 80 81 return true; 82 } 83 84 /* 85 * ---- SINGLE ---- 86 */ 87 88 int fprop_local_init_single(struct fprop_local_single *pl) 89 { 90 pl->events = 0; 91 pl->period = 0; 92 raw_spin_lock_init(&pl->lock); 93 return 0; 94 } 95 96 void fprop_local_destroy_single(struct fprop_local_single *pl) 97 { 98 } 99 100 static void fprop_reflect_period_single(struct fprop_global *p, 101 struct fprop_local_single *pl) 102 { 103 unsigned int period = p->period; 104 unsigned long flags; 105 106 /* Fast path - period didn't change */ 107 if (pl->period == period) 108 return; 109 raw_spin_lock_irqsave(&pl->lock, flags); 110 /* Someone updated pl->period while we were spinning? */ 111 if (pl->period >= period) { 112 raw_spin_unlock_irqrestore(&pl->lock, flags); 113 return; 114 } 115 /* Aging zeroed our fraction? */ 116 if (period - pl->period < BITS_PER_LONG) 117 pl->events >>= period - pl->period; 118 else 119 pl->events = 0; 120 pl->period = period; 121 raw_spin_unlock_irqrestore(&pl->lock, flags); 122 } 123 124 /* Event of type pl happened */ 125 void __fprop_inc_single(struct fprop_global *p, struct fprop_local_single *pl) 126 { 127 fprop_reflect_period_single(p, pl); 128 pl->events++; 129 percpu_counter_add(&p->events, 1); 130 } 131 132 /* Return fraction of events of type pl */ 133 void fprop_fraction_single(struct fprop_global *p, 134 struct fprop_local_single *pl, 135 unsigned long *numerator, unsigned long *denominator) 136 { 137 unsigned int seq; 138 s64 num, den; 139 140 do { 141 seq = read_seqcount_begin(&p->sequence); 142 fprop_reflect_period_single(p, pl); 143 num = pl->events; 144 den = percpu_counter_read_positive(&p->events); 145 } while (read_seqcount_retry(&p->sequence, seq)); 146 147 /* 148 * Make fraction <= 1 and denominator > 0 even in presence of percpu 149 * counter errors 150 */ 151 if (den <= num) { 152 if (num) 153 den = num; 154 else 155 den = 1; 156 } 157 *denominator = den; 158 *numerator = num; 159 } 160 161 /* 162 * ---- PERCPU ---- 163 */ 164 #define PROP_BATCH (8*(1+ilog2(nr_cpu_ids))) 165 166 int fprop_local_init_percpu(struct fprop_local_percpu *pl, gfp_t gfp) 167 { 168 int err; 169 170 err = percpu_counter_init(&pl->events, 0, gfp); 171 if (err) 172 return err; 173 pl->period = 0; 174 raw_spin_lock_init(&pl->lock); 175 return 0; 176 } 177 178 void fprop_local_destroy_percpu(struct fprop_local_percpu *pl) 179 { 180 percpu_counter_destroy(&pl->events); 181 } 182 183 static void fprop_reflect_period_percpu(struct fprop_global *p, 184 struct fprop_local_percpu *pl) 185 { 186 unsigned int period = p->period; 187 unsigned long flags; 188 189 /* Fast path - period didn't change */ 190 if (pl->period == period) 191 return; 192 raw_spin_lock_irqsave(&pl->lock, flags); 193 /* Someone updated pl->period while we were spinning? */ 194 if (pl->period >= period) { 195 raw_spin_unlock_irqrestore(&pl->lock, flags); 196 return; 197 } 198 /* Aging zeroed our fraction? */ 199 if (period - pl->period < BITS_PER_LONG) { 200 s64 val = percpu_counter_read(&pl->events); 201 202 if (val < (nr_cpu_ids * PROP_BATCH)) 203 val = percpu_counter_sum(&pl->events); 204 205 percpu_counter_add_batch(&pl->events, 206 -val + (val >> (period-pl->period)), PROP_BATCH); 207 } else 208 percpu_counter_set(&pl->events, 0); 209 pl->period = period; 210 raw_spin_unlock_irqrestore(&pl->lock, flags); 211 } 212 213 /* Event of type pl happened */ 214 void __fprop_add_percpu(struct fprop_global *p, struct fprop_local_percpu *pl, 215 long nr) 216 { 217 fprop_reflect_period_percpu(p, pl); 218 percpu_counter_add_batch(&pl->events, nr, PROP_BATCH); 219 percpu_counter_add(&p->events, nr); 220 } 221 222 void fprop_fraction_percpu(struct fprop_global *p, 223 struct fprop_local_percpu *pl, 224 unsigned long *numerator, unsigned long *denominator) 225 { 226 unsigned int seq; 227 s64 num, den; 228 229 do { 230 seq = read_seqcount_begin(&p->sequence); 231 fprop_reflect_period_percpu(p, pl); 232 num = percpu_counter_read_positive(&pl->events); 233 den = percpu_counter_read_positive(&p->events); 234 } while (read_seqcount_retry(&p->sequence, seq)); 235 236 /* 237 * Make fraction <= 1 and denominator > 0 even in presence of percpu 238 * counter errors 239 */ 240 if (den <= num) { 241 if (num) 242 den = num; 243 else 244 den = 1; 245 } 246 *denominator = den; 247 *numerator = num; 248 } 249 250 /* 251 * Like __fprop_add_percpu() except that event is counted only if the given 252 * type has fraction smaller than @max_frac/FPROP_FRAC_BASE 253 */ 254 void __fprop_add_percpu_max(struct fprop_global *p, 255 struct fprop_local_percpu *pl, int max_frac, long nr) 256 { 257 if (unlikely(max_frac < FPROP_FRAC_BASE)) { 258 unsigned long numerator, denominator; 259 s64 tmp; 260 261 fprop_fraction_percpu(p, pl, &numerator, &denominator); 262 /* Adding 'nr' to fraction exceeds max_frac/FPROP_FRAC_BASE? */ 263 tmp = (u64)denominator * max_frac - 264 ((u64)numerator << FPROP_FRAC_SHIFT); 265 if (tmp < 0) { 266 /* Maximum fraction already exceeded? */ 267 return; 268 } else if (tmp < nr * (FPROP_FRAC_BASE - max_frac)) { 269 /* Add just enough for the fraction to saturate */ 270 nr = div_u64(tmp + FPROP_FRAC_BASE - max_frac - 1, 271 FPROP_FRAC_BASE - max_frac); 272 } 273 } 274 275 __fprop_add_percpu(p, pl, nr); 276 } 277