| Author | Tokens | Token Proportion | Commits | Commit Proportion |
|---|---|---|---|---|
| Rafael J. Wysocki | 487 | 30.69% | 17 | 25.76% |
| Arjan van de Ven | 297 | 18.71% | 2 | 3.03% |
| Len Brown | 271 | 17.08% | 2 | 3.03% |
| Youquan Song | 110 | 6.93% | 2 | 3.03% |
| Tuukka Tikkanen | 86 | 5.42% | 12 | 18.18% |
| Corrado Zoccolo | 58 | 3.65% | 1 | 1.52% |
| Mel Gorman | 45 | 2.84% | 4 | 6.06% |
| Deepthi Dharwar | 37 | 2.33% | 2 | 3.03% |
| Nicholas Piggin | 37 | 2.33% | 1 | 1.52% |
| Rasmus Villemoes | 35 | 2.21% | 2 | 3.03% |
| Chander Kashyap | 28 | 1.76% | 1 | 1.52% |
| Liu ShuoX | 21 | 1.32% | 2 | 3.03% |
| Venkatesh Pallipadi | 17 | 1.07% | 2 | 3.03% |
| Rik Van Riel | 15 | 0.95% | 2 | 3.03% |
| Daniel Lezcano | 10 | 0.63% | 2 | 3.03% |
| Alex Shi | 9 | 0.57% | 2 | 3.03% |
| Christoph Lameter | 6 | 0.38% | 1 | 1.52% |
| Ingo Molnar | 6 | 0.38% | 2 | 3.03% |
| Stephen Hemminger | 3 | 0.19% | 1 | 1.52% |
| Tero Kristo | 2 | 0.13% | 1 | 1.52% |
| Thomas Gleixner | 2 | 0.13% | 1 | 1.52% |
| Fieah Lim | 2 | 0.13% | 1 | 1.52% |
| Lucas De Marchi | 1 | 0.06% | 1 | 1.52% |
| Ai Li | 1 | 0.06% | 1 | 1.52% |
| Javi Merino | 1 | 0.06% | 1 | 1.52% |
| Total | 1587 | 66 |
// SPDX-License-Identifier: GPL-2.0-only /* * menu.c - the menu idle governor * * Copyright (C) 2006-2007 Adam Belay <abelay@novell.com> * Copyright (C) 2009 Intel Corporation * Author: * Arjan van de Ven <arjan@linux.intel.com> */ #include <linux/kernel.h> #include <linux/cpuidle.h> #include <linux/time.h> #include <linux/ktime.h> #include <linux/hrtimer.h> #include <linux/tick.h> #include <linux/sched.h> #include <linux/sched/loadavg.h> #include <linux/sched/stat.h> #include <linux/math64.h> /* * Please note when changing the tuning values: * If (MAX_INTERESTING-1) * RESOLUTION > UINT_MAX, the result of * a scaling operation multiplication may overflow on 32 bit platforms. * In that case, #define RESOLUTION as ULL to get 64 bit result: * #define RESOLUTION 1024ULL * * The default values do not overflow. */ #define BUCKETS 12 #define INTERVAL_SHIFT 3 #define INTERVALS (1UL << INTERVAL_SHIFT) #define RESOLUTION 1024 #define DECAY 8 #define MAX_INTERESTING 50000 /* * Concepts and ideas behind the menu governor * * For the menu governor, there are 3 decision factors for picking a C * state: * 1) Energy break even point * 2) Performance impact * 3) Latency tolerance (from pmqos infrastructure) * These these three factors are treated independently. * * Energy break even point * ----------------------- * C state entry and exit have an energy cost, and a certain amount of time in * the C state is required to actually break even on this cost. CPUIDLE * provides us this duration in the "target_residency" field. So all that we * need is a good prediction of how long we'll be idle. Like the traditional * menu governor, we start with the actual known "next timer event" time. * * Since there are other source of wakeups (interrupts for example) than * the next timer event, this estimation is rather optimistic. To get a * more realistic estimate, a correction factor is applied to the estimate, * that is based on historic behavior. For example, if in the past the actual * duration always was 50% of the next timer tick, the correction factor will * be 0.5. * * menu uses a running average for this correction factor, however it uses a * set of factors, not just a single factor. This stems from the realization * that the ratio is dependent on the order of magnitude of the expected * duration; if we expect 500 milliseconds of idle time the likelihood of * getting an interrupt very early is much higher than if we expect 50 micro * seconds of idle time. A second independent factor that has big impact on * the actual factor is if there is (disk) IO outstanding or not. * (as a special twist, we consider every sleep longer than 50 milliseconds * as perfect; there are no power gains for sleeping longer than this) * * For these two reasons we keep an array of 12 independent factors, that gets * indexed based on the magnitude of the expected duration as well as the * "is IO outstanding" property. * * Repeatable-interval-detector * ---------------------------- * There are some cases where "next timer" is a completely unusable predictor: * Those cases where the interval is fixed, for example due to hardware * interrupt mitigation, but also due to fixed transfer rate devices such as * mice. * For this, we use a different predictor: We track the duration of the last 8 * intervals and if the stand deviation of these 8 intervals is below a * threshold value, we use the average of these intervals as prediction. * * Limiting Performance Impact * --------------------------- * C states, especially those with large exit latencies, can have a real * noticeable impact on workloads, which is not acceptable for most sysadmins, * and in addition, less performance has a power price of its own. * * As a general rule of thumb, menu assumes that the following heuristic * holds: * The busier the system, the less impact of C states is acceptable * * This rule-of-thumb is implemented using a performance-multiplier: * If the exit latency times the performance multiplier is longer than * the predicted duration, the C state is not considered a candidate * for selection due to a too high performance impact. So the higher * this multiplier is, the longer we need to be idle to pick a deep C * state, and thus the less likely a busy CPU will hit such a deep * C state. * * Two factors are used in determing this multiplier: * a value of 10 is added for each point of "per cpu load average" we have. * a value of 5 points is added for each process that is waiting for * IO on this CPU. * (these values are experimentally determined) * * The load average factor gives a longer term (few seconds) input to the * decision, while the iowait value gives a cpu local instantanious input. * The iowait factor may look low, but realize that this is also already * represented in the system load average. * */ struct menu_device { int last_state_idx; int needs_update; int tick_wakeup; unsigned int next_timer_us; unsigned int bucket; unsigned int correction_factor[BUCKETS]; unsigned int intervals[INTERVALS]; int interval_ptr; }; static inline int which_bucket(unsigned int duration, unsigned long nr_iowaiters) { int bucket = 0; /* * We keep two groups of stats; one with no * IO pending, one without. * This allows us to calculate * E(duration)|iowait */ if (nr_iowaiters) bucket = BUCKETS/2; if (duration < 10) return bucket; if (duration < 100) return bucket + 1; if (duration < 1000) return bucket + 2; if (duration < 10000) return bucket + 3; if (duration < 100000) return bucket + 4; return bucket + 5; } /* * Return a multiplier for the exit latency that is intended * to take performance requirements into account. * The more performance critical we estimate the system * to be, the higher this multiplier, and thus the higher * the barrier to go to an expensive C state. */ static inline int performance_multiplier(unsigned long nr_iowaiters) { /* for IO wait tasks (per cpu!) we add 10x each */ return 1 + 10 * nr_iowaiters; } static DEFINE_PER_CPU(struct menu_device, menu_devices); static void menu_update(struct cpuidle_driver *drv, struct cpuidle_device *dev); /* * Try detecting repeating patterns by keeping track of the last 8 * intervals, and checking if the standard deviation of that set * of points is below a threshold. If it is... then use the * average of these 8 points as the estimated value. */ static unsigned int get_typical_interval(struct menu_device *data, unsigned int predicted_us) { int i, divisor; unsigned int min, max, thresh, avg; uint64_t sum, variance; thresh = INT_MAX; /* Discard outliers above this value */ again: /* First calculate the average of past intervals */ min = UINT_MAX; max = 0; sum = 0; divisor = 0; for (i = 0; i < INTERVALS; i++) { unsigned int value = data->intervals[i]; if (value <= thresh) { sum += value; divisor++; if (value > max) max = value; if (value < min) min = value; } } /* * If the result of the computation is going to be discarded anyway, * avoid the computation altogether. */ if (min >= predicted_us) return UINT_MAX; if (divisor == INTERVALS) avg = sum >> INTERVAL_SHIFT; else avg = div_u64(sum, divisor); /* Then try to determine variance */ variance = 0; for (i = 0; i < INTERVALS; i++) { unsigned int value = data->intervals[i]; if (value <= thresh) { int64_t diff = (int64_t)value - avg; variance += diff * diff; } } if (divisor == INTERVALS) variance >>= INTERVAL_SHIFT; else do_div(variance, divisor); /* * The typical interval is obtained when standard deviation is * small (stddev <= 20 us, variance <= 400 us^2) or standard * deviation is small compared to the average interval (avg > * 6*stddev, avg^2 > 36*variance). The average is smaller than * UINT_MAX aka U32_MAX, so computing its square does not * overflow a u64. We simply reject this candidate average if * the standard deviation is greater than 715 s (which is * rather unlikely). * * Use this result only if there is no timer to wake us up sooner. */ if (likely(variance <= U64_MAX/36)) { if ((((u64)avg*avg > variance*36) && (divisor * 4 >= INTERVALS * 3)) || variance <= 400) { return avg; } } /* * If we have outliers to the upside in our distribution, discard * those by setting the threshold to exclude these outliers, then * calculate the average and standard deviation again. Once we get * down to the bottom 3/4 of our samples, stop excluding samples. * * This can deal with workloads that have long pauses interspersed * with sporadic activity with a bunch of short pauses. */ if ((divisor * 4) <= INTERVALS * 3) return UINT_MAX; thresh = max - 1; goto again; } /** * menu_select - selects the next idle state to enter * @drv: cpuidle driver containing state data * @dev: the CPU * @stop_tick: indication on whether or not to stop the tick */ static int menu_select(struct cpuidle_driver *drv, struct cpuidle_device *dev, bool *stop_tick) { struct menu_device *data = this_cpu_ptr(&menu_devices); int latency_req = cpuidle_governor_latency_req(dev->cpu); int i; int idx; unsigned int interactivity_req; unsigned int predicted_us; unsigned long nr_iowaiters; ktime_t delta_next; if (data->needs_update) { menu_update(drv, dev); data->needs_update = 0; } /* determine the expected residency time, round up */ data->next_timer_us = ktime_to_us(tick_nohz_get_sleep_length(&delta_next)); nr_iowaiters = nr_iowait_cpu(dev->cpu); data->bucket = which_bucket(data->next_timer_us, nr_iowaiters); if (unlikely(drv->state_count <= 1 || latency_req == 0) || ((data->next_timer_us < drv->states[1].target_residency || latency_req < drv->states[1].exit_latency) && !drv->states[0].disabled && !dev->states_usage[0].disable)) { /* * In this case state[0] will be used no matter what, so return * it right away and keep the tick running. */ *stop_tick = false; return 0; } /* * Force the result of multiplication to be 64 bits even if both * operands are 32 bits. * Make sure to round up for half microseconds. */ predicted_us = DIV_ROUND_CLOSEST_ULL((uint64_t)data->next_timer_us * data->correction_factor[data->bucket], RESOLUTION * DECAY); /* * Use the lowest expected idle interval to pick the idle state. */ predicted_us = min(predicted_us, get_typical_interval(data, predicted_us)); if (tick_nohz_tick_stopped()) { /* * If the tick is already stopped, the cost of possible short * idle duration misprediction is much higher, because the CPU * may be stuck in a shallow idle state for a long time as a * result of it. In that case say we might mispredict and use * the known time till the closest timer event for the idle * state selection. */ if (predicted_us < TICK_USEC) predicted_us = ktime_to_us(delta_next); } else { /* * Use the performance multiplier and the user-configurable * latency_req to determine the maximum exit latency. */ interactivity_req = predicted_us / performance_multiplier(nr_iowaiters); if (latency_req > interactivity_req) latency_req = interactivity_req; } /* * Find the idle state with the lowest power while satisfying * our constraints. */ idx = -1; for (i = 0; i < drv->state_count; i++) { struct cpuidle_state *s = &drv->states[i]; struct cpuidle_state_usage *su = &dev->states_usage[i]; if (s->disabled || su->disable) continue; if (idx == -1) idx = i; /* first enabled state */ if (s->target_residency > predicted_us) { /* * Use a physical idle state, not busy polling, unless * a timer is going to trigger soon enough. */ if ((drv->states[idx].flags & CPUIDLE_FLAG_POLLING) && s->exit_latency <= latency_req && s->target_residency <= data->next_timer_us) { predicted_us = s->target_residency; idx = i; break; } if (predicted_us < TICK_USEC) break; if (!tick_nohz_tick_stopped()) { /* * If the state selected so far is shallow, * waking up early won't hurt, so retain the * tick in that case and let the governor run * again in the next iteration of the loop. */ predicted_us = drv->states[idx].target_residency; break; } /* * If the state selected so far is shallow and this * state's target residency matches the time till the * closest timer event, select this one to avoid getting * stuck in the shallow one for too long. */ if (drv->states[idx].target_residency < TICK_USEC && s->target_residency <= ktime_to_us(delta_next)) idx = i; return idx; } if (s->exit_latency > latency_req) { /* * If we break out of the loop for latency reasons, use * the target residency of the selected state as the * expected idle duration so that the tick is retained * as long as that target residency is low enough. */ predicted_us = drv->states[idx].target_residency; break; } idx = i; } if (idx == -1) idx = 0; /* No states enabled. Must use 0. */ /* * Don't stop the tick if the selected state is a polling one or if the * expected idle duration is shorter than the tick period length. */ if (((drv->states[idx].flags & CPUIDLE_FLAG_POLLING) || predicted_us < TICK_USEC) && !tick_nohz_tick_stopped()) { unsigned int delta_next_us = ktime_to_us(delta_next); *stop_tick = false; if (idx > 0 && drv->states[idx].target_residency > delta_next_us) { /* * The tick is not going to be stopped and the target * residency of the state to be returned is not within * the time until the next timer event including the * tick, so try to correct that. */ for (i = idx - 1; i >= 0; i--) { if (drv->states[i].disabled || dev->states_usage[i].disable) continue; idx = i; if (drv->states[i].target_residency <= delta_next_us) break; } } } return idx; } /** * menu_reflect - records that data structures need update * @dev: the CPU * @index: the index of actual entered state * * NOTE: it's important to be fast here because this operation will add to * the overall exit latency. */ static void menu_reflect(struct cpuidle_device *dev, int index) { struct menu_device *data = this_cpu_ptr(&menu_devices); data->last_state_idx = index; data->needs_update = 1; data->tick_wakeup = tick_nohz_idle_got_tick(); } /** * menu_update - attempts to guess what happened after entry * @drv: cpuidle driver containing state data * @dev: the CPU */ static void menu_update(struct cpuidle_driver *drv, struct cpuidle_device *dev) { struct menu_device *data = this_cpu_ptr(&menu_devices); int last_idx = data->last_state_idx; struct cpuidle_state *target = &drv->states[last_idx]; unsigned int measured_us; unsigned int new_factor; /* * Try to figure out how much time passed between entry to low * power state and occurrence of the wakeup event. * * If the entered idle state didn't support residency measurements, * we use them anyway if they are short, and if long, * truncate to the whole expected time. * * Any measured amount of time will include the exit latency. * Since we are interested in when the wakeup begun, not when it * was completed, we must subtract the exit latency. However, if * the measured amount of time is less than the exit latency, * assume the state was never reached and the exit latency is 0. */ if (data->tick_wakeup && data->next_timer_us > TICK_USEC) { /* * The nohz code said that there wouldn't be any events within * the tick boundary (if the tick was stopped), but the idle * duration predictor had a differing opinion. Since the CPU * was woken up by a tick (that wasn't stopped after all), the * predictor was not quite right, so assume that the CPU could * have been idle long (but not forever) to help the idle * duration predictor do a better job next time. */ measured_us = 9 * MAX_INTERESTING / 10; } else if ((drv->states[last_idx].flags & CPUIDLE_FLAG_POLLING) && dev->poll_time_limit) { /* * The CPU exited the "polling" state due to a time limit, so * the idle duration prediction leading to the selection of that * state was inaccurate. If a better prediction had been made, * the CPU might have been woken up from idle by the next timer. * Assume that to be the case. */ measured_us = data->next_timer_us; } else { /* measured value */ measured_us = dev->last_residency; /* Deduct exit latency */ if (measured_us > 2 * target->exit_latency) measured_us -= target->exit_latency; else measured_us /= 2; } /* Make sure our coefficients do not exceed unity */ if (measured_us > data->next_timer_us) measured_us = data->next_timer_us; /* Update our correction ratio */ new_factor = data->correction_factor[data->bucket]; new_factor -= new_factor / DECAY; if (data->next_timer_us > 0 && measured_us < MAX_INTERESTING) new_factor += RESOLUTION * measured_us / data->next_timer_us; else /* * we were idle so long that we count it as a perfect * prediction */ new_factor += RESOLUTION; /* * We don't want 0 as factor; we always want at least * a tiny bit of estimated time. Fortunately, due to rounding, * new_factor will stay nonzero regardless of measured_us values * and the compiler can eliminate this test as long as DECAY > 1. */ if (DECAY == 1 && unlikely(new_factor == 0)) new_factor = 1; data->correction_factor[data->bucket] = new_factor; /* update the repeating-pattern data */ data->intervals[data->interval_ptr++] = measured_us; if (data->interval_ptr >= INTERVALS) data->interval_ptr = 0; } /** * menu_enable_device - scans a CPU's states and does setup * @drv: cpuidle driver * @dev: the CPU */ static int menu_enable_device(struct cpuidle_driver *drv, struct cpuidle_device *dev) { struct menu_device *data = &per_cpu(menu_devices, dev->cpu); int i; memset(data, 0, sizeof(struct menu_device)); /* * if the correction factor is 0 (eg first time init or cpu hotplug * etc), we actually want to start out with a unity factor. */ for(i = 0; i < BUCKETS; i++) data->correction_factor[i] = RESOLUTION * DECAY; return 0; } static struct cpuidle_governor menu_governor = { .name = "menu", .rating = 20, .enable = menu_enable_device, .select = menu_select, .reflect = menu_reflect, }; /** * init_menu - initializes the governor */ static int __init init_menu(void) { return cpuidle_register_governor(&menu_governor); } postcore_initcall(init_menu);
Information contained on this website is for historical information purposes only and does not indicate or represent copyright ownership.
Created with Cregit http://github.com/cregit/cregit
Version 2.0-RC1