1 /* 2 * menu.c - the menu idle governor 3 * 4 * Copyright (C) 2006-2007 Adam Belay <abelay@novell.com> 5 * Copyright (C) 2009 Intel Corporation 6 * Author: 7 * Arjan van de Ven <arjan@linux.intel.com> 8 * 9 * This code is licenced under the GPL version 2 as described 10 * in the COPYING file that acompanies the Linux Kernel. 11 */ 12 13 #include <linux/kernel.h> 14 #include <linux/cpuidle.h> 15 #include <linux/pm_qos.h> 16 #include <linux/time.h> 17 #include <linux/ktime.h> 18 #include <linux/hrtimer.h> 19 #include <linux/tick.h> 20 #include <linux/sched.h> 21 #include <linux/math64.h> 22 #include <linux/module.h> 23 24 /* 25 * Please note when changing the tuning values: 26 * If (MAX_INTERESTING-1) * RESOLUTION > UINT_MAX, the result of 27 * a scaling operation multiplication may overflow on 32 bit platforms. 28 * In that case, #define RESOLUTION as ULL to get 64 bit result: 29 * #define RESOLUTION 1024ULL 30 * 31 * The default values do not overflow. 32 */ 33 #define BUCKETS 12 34 #define INTERVAL_SHIFT 3 35 #define INTERVALS (1UL << INTERVAL_SHIFT) 36 #define RESOLUTION 1024 37 #define DECAY 8 38 #define MAX_INTERESTING 50000 39 40 41 /* 42 * Concepts and ideas behind the menu governor 43 * 44 * For the menu governor, there are 3 decision factors for picking a C 45 * state: 46 * 1) Energy break even point 47 * 2) Performance impact 48 * 3) Latency tolerance (from pmqos infrastructure) 49 * These these three factors are treated independently. 50 * 51 * Energy break even point 52 * ----------------------- 53 * C state entry and exit have an energy cost, and a certain amount of time in 54 * the C state is required to actually break even on this cost. CPUIDLE 55 * provides us this duration in the "target_residency" field. So all that we 56 * need is a good prediction of how long we'll be idle. Like the traditional 57 * menu governor, we start with the actual known "next timer event" time. 58 * 59 * Since there are other source of wakeups (interrupts for example) than 60 * the next timer event, this estimation is rather optimistic. To get a 61 * more realistic estimate, a correction factor is applied to the estimate, 62 * that is based on historic behavior. For example, if in the past the actual 63 * duration always was 50% of the next timer tick, the correction factor will 64 * be 0.5. 65 * 66 * menu uses a running average for this correction factor, however it uses a 67 * set of factors, not just a single factor. This stems from the realization 68 * that the ratio is dependent on the order of magnitude of the expected 69 * duration; if we expect 500 milliseconds of idle time the likelihood of 70 * getting an interrupt very early is much higher than if we expect 50 micro 71 * seconds of idle time. A second independent factor that has big impact on 72 * the actual factor is if there is (disk) IO outstanding or not. 73 * (as a special twist, we consider every sleep longer than 50 milliseconds 74 * as perfect; there are no power gains for sleeping longer than this) 75 * 76 * For these two reasons we keep an array of 12 independent factors, that gets 77 * indexed based on the magnitude of the expected duration as well as the 78 * "is IO outstanding" property. 79 * 80 * Repeatable-interval-detector 81 * ---------------------------- 82 * There are some cases where "next timer" is a completely unusable predictor: 83 * Those cases where the interval is fixed, for example due to hardware 84 * interrupt mitigation, but also due to fixed transfer rate devices such as 85 * mice. 86 * For this, we use a different predictor: We track the duration of the last 8 87 * intervals and if the stand deviation of these 8 intervals is below a 88 * threshold value, we use the average of these intervals as prediction. 89 * 90 * Limiting Performance Impact 91 * --------------------------- 92 * C states, especially those with large exit latencies, can have a real 93 * noticeable impact on workloads, which is not acceptable for most sysadmins, 94 * and in addition, less performance has a power price of its own. 95 * 96 * As a general rule of thumb, menu assumes that the following heuristic 97 * holds: 98 * The busier the system, the less impact of C states is acceptable 99 * 100 * This rule-of-thumb is implemented using a performance-multiplier: 101 * If the exit latency times the performance multiplier is longer than 102 * the predicted duration, the C state is not considered a candidate 103 * for selection due to a too high performance impact. So the higher 104 * this multiplier is, the longer we need to be idle to pick a deep C 105 * state, and thus the less likely a busy CPU will hit such a deep 106 * C state. 107 * 108 * Two factors are used in determing this multiplier: 109 * a value of 10 is added for each point of "per cpu load average" we have. 110 * a value of 5 points is added for each process that is waiting for 111 * IO on this CPU. 112 * (these values are experimentally determined) 113 * 114 * The load average factor gives a longer term (few seconds) input to the 115 * decision, while the iowait value gives a cpu local instantanious input. 116 * The iowait factor may look low, but realize that this is also already 117 * represented in the system load average. 118 * 119 */ 120 121 struct menu_device { 122 int last_state_idx; 123 int needs_update; 124 125 unsigned int next_timer_us; 126 unsigned int predicted_us; 127 unsigned int bucket; 128 unsigned int correction_factor[BUCKETS]; 129 unsigned int intervals[INTERVALS]; 130 int interval_ptr; 131 }; 132 133 134 #define LOAD_INT(x) ((x) >> FSHIFT) 135 #define LOAD_FRAC(x) LOAD_INT(((x) & (FIXED_1-1)) * 100) 136 137 static inline int get_loadavg(unsigned long load) 138 { 139 return LOAD_INT(load) * 10 + LOAD_FRAC(load) / 10; 140 } 141 142 static inline int which_bucket(unsigned int duration, unsigned long nr_iowaiters) 143 { 144 int bucket = 0; 145 146 /* 147 * We keep two groups of stats; one with no 148 * IO pending, one without. 149 * This allows us to calculate 150 * E(duration)|iowait 151 */ 152 if (nr_iowaiters) 153 bucket = BUCKETS/2; 154 155 if (duration < 10) 156 return bucket; 157 if (duration < 100) 158 return bucket + 1; 159 if (duration < 1000) 160 return bucket + 2; 161 if (duration < 10000) 162 return bucket + 3; 163 if (duration < 100000) 164 return bucket + 4; 165 return bucket + 5; 166 } 167 168 /* 169 * Return a multiplier for the exit latency that is intended 170 * to take performance requirements into account. 171 * The more performance critical we estimate the system 172 * to be, the higher this multiplier, and thus the higher 173 * the barrier to go to an expensive C state. 174 */ 175 static inline int performance_multiplier(unsigned long nr_iowaiters, unsigned long load) 176 { 177 int mult = 1; 178 179 /* for higher loadavg, we are more reluctant */ 180 181 mult += 2 * get_loadavg(load); 182 183 /* for IO wait tasks (per cpu!) we add 5x each */ 184 mult += 10 * nr_iowaiters; 185 186 return mult; 187 } 188 189 static DEFINE_PER_CPU(struct menu_device, menu_devices); 190 191 static void menu_update(struct cpuidle_driver *drv, struct cpuidle_device *dev); 192 193 /* 194 * Try detecting repeating patterns by keeping track of the last 8 195 * intervals, and checking if the standard deviation of that set 196 * of points is below a threshold. If it is... then use the 197 * average of these 8 points as the estimated value. 198 */ 199 static unsigned int get_typical_interval(struct menu_device *data) 200 { 201 int i, divisor; 202 unsigned int max, thresh, avg; 203 uint64_t sum, variance; 204 205 thresh = UINT_MAX; /* Discard outliers above this value */ 206 207 again: 208 209 /* First calculate the average of past intervals */ 210 max = 0; 211 sum = 0; 212 divisor = 0; 213 for (i = 0; i < INTERVALS; i++) { 214 unsigned int value = data->intervals[i]; 215 if (value <= thresh) { 216 sum += value; 217 divisor++; 218 if (value > max) 219 max = value; 220 } 221 } 222 if (divisor == INTERVALS) 223 avg = sum >> INTERVAL_SHIFT; 224 else 225 avg = div_u64(sum, divisor); 226 227 /* Then try to determine variance */ 228 variance = 0; 229 for (i = 0; i < INTERVALS; i++) { 230 unsigned int value = data->intervals[i]; 231 if (value <= thresh) { 232 int64_t diff = (int64_t)value - avg; 233 variance += diff * diff; 234 } 235 } 236 if (divisor == INTERVALS) 237 variance >>= INTERVAL_SHIFT; 238 else 239 do_div(variance, divisor); 240 241 /* 242 * The typical interval is obtained when standard deviation is 243 * small (stddev <= 20 us, variance <= 400 us^2) or standard 244 * deviation is small compared to the average interval (avg > 245 * 6*stddev, avg^2 > 36*variance). The average is smaller than 246 * UINT_MAX aka U32_MAX, so computing its square does not 247 * overflow a u64. We simply reject this candidate average if 248 * the standard deviation is greater than 715 s (which is 249 * rather unlikely). 250 * 251 * Use this result only if there is no timer to wake us up sooner. 252 */ 253 if (likely(variance <= U64_MAX/36)) { 254 if ((((u64)avg*avg > variance*36) && (divisor * 4 >= INTERVALS * 3)) 255 || variance <= 400) { 256 return avg; 257 } 258 } 259 260 /* 261 * If we have outliers to the upside in our distribution, discard 262 * those by setting the threshold to exclude these outliers, then 263 * calculate the average and standard deviation again. Once we get 264 * down to the bottom 3/4 of our samples, stop excluding samples. 265 * 266 * This can deal with workloads that have long pauses interspersed 267 * with sporadic activity with a bunch of short pauses. 268 */ 269 if ((divisor * 4) <= INTERVALS * 3) 270 return UINT_MAX; 271 272 thresh = max - 1; 273 goto again; 274 } 275 276 /** 277 * menu_select - selects the next idle state to enter 278 * @drv: cpuidle driver containing state data 279 * @dev: the CPU 280 */ 281 static int menu_select(struct cpuidle_driver *drv, struct cpuidle_device *dev) 282 { 283 struct menu_device *data = this_cpu_ptr(&menu_devices); 284 int latency_req = pm_qos_request(PM_QOS_CPU_DMA_LATENCY); 285 int i; 286 unsigned int interactivity_req; 287 unsigned int expected_interval; 288 unsigned long nr_iowaiters, cpu_load; 289 290 if (data->needs_update) { 291 menu_update(drv, dev); 292 data->needs_update = 0; 293 } 294 295 /* Special case when user has set very strict latency requirement */ 296 if (unlikely(latency_req == 0)) 297 return 0; 298 299 /* determine the expected residency time, round up */ 300 data->next_timer_us = ktime_to_us(tick_nohz_get_sleep_length()); 301 302 get_iowait_load(&nr_iowaiters, &cpu_load); 303 data->bucket = which_bucket(data->next_timer_us, nr_iowaiters); 304 305 /* 306 * Force the result of multiplication to be 64 bits even if both 307 * operands are 32 bits. 308 * Make sure to round up for half microseconds. 309 */ 310 data->predicted_us = DIV_ROUND_CLOSEST_ULL((uint64_t)data->next_timer_us * 311 data->correction_factor[data->bucket], 312 RESOLUTION * DECAY); 313 314 expected_interval = get_typical_interval(data); 315 expected_interval = min(expected_interval, data->next_timer_us); 316 317 if (CPUIDLE_DRIVER_STATE_START > 0) { 318 struct cpuidle_state *s = &drv->states[CPUIDLE_DRIVER_STATE_START]; 319 unsigned int polling_threshold; 320 321 /* 322 * We want to default to C1 (hlt), not to busy polling 323 * unless the timer is happening really really soon, or 324 * C1's exit latency exceeds the user configured limit. 325 */ 326 polling_threshold = max_t(unsigned int, 20, s->target_residency); 327 if (data->next_timer_us > polling_threshold && 328 latency_req > s->exit_latency && !s->disabled && 329 !dev->states_usage[CPUIDLE_DRIVER_STATE_START].disable) 330 data->last_state_idx = CPUIDLE_DRIVER_STATE_START; 331 else 332 data->last_state_idx = CPUIDLE_DRIVER_STATE_START - 1; 333 } else { 334 data->last_state_idx = CPUIDLE_DRIVER_STATE_START; 335 } 336 337 /* 338 * Use the lowest expected idle interval to pick the idle state. 339 */ 340 data->predicted_us = min(data->predicted_us, expected_interval); 341 342 /* 343 * Use the performance multiplier and the user-configurable 344 * latency_req to determine the maximum exit latency. 345 */ 346 interactivity_req = data->predicted_us / performance_multiplier(nr_iowaiters, cpu_load); 347 if (latency_req > interactivity_req) 348 latency_req = interactivity_req; 349 350 /* 351 * Find the idle state with the lowest power while satisfying 352 * our constraints. 353 */ 354 for (i = data->last_state_idx + 1; i < drv->state_count; i++) { 355 struct cpuidle_state *s = &drv->states[i]; 356 struct cpuidle_state_usage *su = &dev->states_usage[i]; 357 358 if (s->disabled || su->disable) 359 continue; 360 if (s->target_residency > data->predicted_us) 361 continue; 362 if (s->exit_latency > latency_req) 363 continue; 364 365 data->last_state_idx = i; 366 } 367 368 return data->last_state_idx; 369 } 370 371 /** 372 * menu_reflect - records that data structures need update 373 * @dev: the CPU 374 * @index: the index of actual entered state 375 * 376 * NOTE: it's important to be fast here because this operation will add to 377 * the overall exit latency. 378 */ 379 static void menu_reflect(struct cpuidle_device *dev, int index) 380 { 381 struct menu_device *data = this_cpu_ptr(&menu_devices); 382 383 data->last_state_idx = index; 384 data->needs_update = 1; 385 } 386 387 /** 388 * menu_update - attempts to guess what happened after entry 389 * @drv: cpuidle driver containing state data 390 * @dev: the CPU 391 */ 392 static void menu_update(struct cpuidle_driver *drv, struct cpuidle_device *dev) 393 { 394 struct menu_device *data = this_cpu_ptr(&menu_devices); 395 int last_idx = data->last_state_idx; 396 struct cpuidle_state *target = &drv->states[last_idx]; 397 unsigned int measured_us; 398 unsigned int new_factor; 399 400 /* 401 * Try to figure out how much time passed between entry to low 402 * power state and occurrence of the wakeup event. 403 * 404 * If the entered idle state didn't support residency measurements, 405 * we use them anyway if they are short, and if long, 406 * truncate to the whole expected time. 407 * 408 * Any measured amount of time will include the exit latency. 409 * Since we are interested in when the wakeup begun, not when it 410 * was completed, we must subtract the exit latency. However, if 411 * the measured amount of time is less than the exit latency, 412 * assume the state was never reached and the exit latency is 0. 413 */ 414 415 /* measured value */ 416 measured_us = cpuidle_get_last_residency(dev); 417 418 /* Deduct exit latency */ 419 if (measured_us > 2 * target->exit_latency) 420 measured_us -= target->exit_latency; 421 else 422 measured_us /= 2; 423 424 /* Make sure our coefficients do not exceed unity */ 425 if (measured_us > data->next_timer_us) 426 measured_us = data->next_timer_us; 427 428 /* Update our correction ratio */ 429 new_factor = data->correction_factor[data->bucket]; 430 new_factor -= new_factor / DECAY; 431 432 if (data->next_timer_us > 0 && measured_us < MAX_INTERESTING) 433 new_factor += RESOLUTION * measured_us / data->next_timer_us; 434 else 435 /* 436 * we were idle so long that we count it as a perfect 437 * prediction 438 */ 439 new_factor += RESOLUTION; 440 441 /* 442 * We don't want 0 as factor; we always want at least 443 * a tiny bit of estimated time. Fortunately, due to rounding, 444 * new_factor will stay nonzero regardless of measured_us values 445 * and the compiler can eliminate this test as long as DECAY > 1. 446 */ 447 if (DECAY == 1 && unlikely(new_factor == 0)) 448 new_factor = 1; 449 450 data->correction_factor[data->bucket] = new_factor; 451 452 /* update the repeating-pattern data */ 453 data->intervals[data->interval_ptr++] = measured_us; 454 if (data->interval_ptr >= INTERVALS) 455 data->interval_ptr = 0; 456 } 457 458 /** 459 * menu_enable_device - scans a CPU's states and does setup 460 * @drv: cpuidle driver 461 * @dev: the CPU 462 */ 463 static int menu_enable_device(struct cpuidle_driver *drv, 464 struct cpuidle_device *dev) 465 { 466 struct menu_device *data = &per_cpu(menu_devices, dev->cpu); 467 int i; 468 469 memset(data, 0, sizeof(struct menu_device)); 470 471 /* 472 * if the correction factor is 0 (eg first time init or cpu hotplug 473 * etc), we actually want to start out with a unity factor. 474 */ 475 for(i = 0; i < BUCKETS; i++) 476 data->correction_factor[i] = RESOLUTION * DECAY; 477 478 return 0; 479 } 480 481 static struct cpuidle_governor menu_governor = { 482 .name = "menu", 483 .rating = 20, 484 .enable = menu_enable_device, 485 .select = menu_select, 486 .reflect = menu_reflect, 487 .owner = THIS_MODULE, 488 }; 489 490 /** 491 * init_menu - initializes the governor 492 */ 493 static int __init init_menu(void) 494 { 495 return cpuidle_register_governor(&menu_governor); 496 } 497 498 postcore_initcall(init_menu); 499