diff options
Diffstat (limited to 'drivers/cpuidle')
| -rw-r--r-- | drivers/cpuidle/governors/menu.c | 6 | ||||
| -rw-r--r-- | drivers/cpuidle/governors/teo.c | 476 |
2 files changed, 251 insertions, 231 deletions
diff --git a/drivers/cpuidle/governors/menu.c b/drivers/cpuidle/governors/menu.c index c3aa8d6ccee3..2e5670446991 100644 --- a/drivers/cpuidle/governors/menu.c +++ b/drivers/cpuidle/governors/menu.c @@ -117,7 +117,7 @@ struct menu_device { int interval_ptr; }; -static inline int which_bucket(u64 duration_ns, unsigned long nr_iowaiters) +static inline int which_bucket(u64 duration_ns, unsigned int nr_iowaiters) { int bucket = 0; @@ -150,7 +150,7 @@ static inline int which_bucket(u64 duration_ns, unsigned long nr_iowaiters) * 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) +static inline int performance_multiplier(unsigned int nr_iowaiters) { /* for IO wait tasks (per cpu!) we add 10x each */ return 1 + 10 * nr_iowaiters; @@ -270,7 +270,7 @@ static int menu_select(struct cpuidle_driver *drv, struct cpuidle_device *dev, unsigned int predicted_us; u64 predicted_ns; u64 interactivity_req; - unsigned long nr_iowaiters; + unsigned int nr_iowaiters; ktime_t delta, delta_tick; int i, idx; diff --git a/drivers/cpuidle/governors/teo.c b/drivers/cpuidle/governors/teo.c index ac4bb27d69b0..7b91060e82f6 100644 --- a/drivers/cpuidle/governors/teo.c +++ b/drivers/cpuidle/governors/teo.c @@ -2,47 +2,103 @@ /* * Timer events oriented CPU idle governor * - * Copyright (C) 2018 Intel Corporation + * Copyright (C) 2018 - 2021 Intel Corporation * Author: Rafael J. Wysocki <rafael.j.wysocki@intel.com> + */ + +/** + * DOC: teo-description * * The idea of this governor is based on the observation that on many systems * timer events are two or more orders of magnitude more frequent than any - * other interrupts, so they are likely to be the most significant source of CPU + * other interrupts, so they are likely to be the most significant cause of CPU * wakeups from idle states. Moreover, information about what happened in the * (relatively recent) past can be used to estimate whether or not the deepest - * idle state with target residency within the time to the closest timer is - * likely to be suitable for the upcoming idle time of the CPU and, if not, then - * which of the shallower idle states to choose. + * idle state with target residency within the (known) time till the closest + * timer event, referred to as the sleep length, is likely to be suitable for + * the upcoming CPU idle period and, if not, then which of the shallower idle + * states to choose instead of it. + * + * Of course, non-timer wakeup sources are more important in some use cases + * which can be covered by taking a few most recent idle time intervals of the + * CPU into account. However, even in that context it is not necessary to + * consider idle duration values greater than the sleep length, because the + * closest timer will ultimately wake up the CPU anyway unless it is woken up + * earlier. + * + * Thus this governor estimates whether or not the prospective idle duration of + * a CPU is likely to be significantly shorter than the sleep length and selects + * an idle state for it accordingly. + * + * The computations carried out by this governor are based on using bins whose + * boundaries are aligned with the target residency parameter values of the CPU + * idle states provided by the %CPUIdle driver in the ascending order. That is, + * the first bin spans from 0 up to, but not including, the target residency of + * the second idle state (idle state 1), the second bin spans from the target + * residency of idle state 1 up to, but not including, the target residency of + * idle state 2, the third bin spans from the target residency of idle state 2 + * up to, but not including, the target residency of idle state 3 and so on. + * The last bin spans from the target residency of the deepest idle state + * supplied by the driver to infinity. + * + * Two metrics called "hits" and "intercepts" are associated with each bin. + * They are updated every time before selecting an idle state for the given CPU + * in accordance with what happened last time. + * + * The "hits" metric reflects the relative frequency of situations in which the + * sleep length and the idle duration measured after CPU wakeup fall into the + * same bin (that is, the CPU appears to wake up "on time" relative to the sleep + * length). In turn, the "intercepts" metric reflects the relative frequency of + * situations in which the measured idle duration is so much shorter than the + * sleep length that the bin it falls into corresponds to an idle state + * shallower than the one whose bin is fallen into by the sleep length (these + * situations are referred to as "intercepts" below). + * + * In addition to the metrics described above, the governor counts recent + * intercepts (that is, intercepts that have occurred during the last + * %NR_RECENT invocations of it for the given CPU) for each bin. * - * Of course, non-timer wakeup sources are more important in some use cases and - * they can be covered by taking a few most recent idle time intervals of the - * CPU into account. However, even in that case it is not necessary to consider - * idle duration values greater than the time till the closest timer, as the - * patterns that they may belong to produce average values close enough to - * the time till the closest timer (sleep length) anyway. + * In order to select an idle state for a CPU, the governor takes the following + * steps (modulo the possible latency constraint that must be taken into account + * too): * - * Thus this governor estimates whether or not the upcoming idle time of the CPU - * is likely to be significantly shorter than the sleep length and selects an - * idle state for it in accordance with that, as follows: + * 1. Find the deepest CPU idle state whose target residency does not exceed + * the current sleep length (the candidate idle state) and compute 3 sums as + * follows: * - * - Find an idle state on the basis of the sleep length and state statistics - * collected over time: + * - The sum of the "hits" and "intercepts" metrics for the candidate state + * and all of the deeper idle states (it represents the cases in which the + * CPU was idle long enough to avoid being intercepted if the sleep length + * had been equal to the current one). * - * o Find the deepest idle state whose target residency is less than or equal - * to the sleep length. + * - The sum of the "intercepts" metrics for all of the idle states shallower + * than the candidate one (it represents the cases in which the CPU was not + * idle long enough to avoid being intercepted if the sleep length had been + * equal to the current one). * - * o Select it if it matched both the sleep length and the observed idle - * duration in the past more often than it matched the sleep length alone - * (i.e. the observed idle duration was significantly shorter than the sleep - * length matched by it). + * - The sum of the numbers of recent intercepts for all of the idle states + * shallower than the candidate one. * - * o Otherwise, select the shallower state with the greatest matched "early" - * wakeups metric. + * 2. If the second sum is greater than the first one or the third sum is + * greater than %NR_RECENT / 2, the CPU is likely to wake up early, so look + * for an alternative idle state to select. * - * - If the majority of the most recent idle duration values are below the - * target residency of the idle state selected so far, use those values to - * compute the new expected idle duration and find an idle state matching it - * (which has to be shallower than the one selected so far). + * - Traverse the idle states shallower than the candidate one in the + * descending order. + * + * - For each of them compute the sum of the "intercepts" metrics and the sum + * of the numbers of recent intercepts over all of the idle states between + * it and the candidate one (including the former and excluding the + * latter). + * + * - If each of these sums that needs to be taken into account (because the + * check related to it has indicated that the CPU is likely to wake up + * early) is greater than a half of the corresponding sum computed in step + * 1 (which means that the target residency of the state in question had + * not exceeded the idle duration in over a half of the relevant cases), + * select the given idle state instead of the candidate one. + * + * 3. By default, select the candidate state. */ #include <linux/cpuidle.h> @@ -60,65 +116,51 @@ /* * Number of the most recent idle duration values to take into consideration for - * the detection of wakeup patterns. + * the detection of recent early wakeup patterns. */ -#define INTERVALS 8 +#define NR_RECENT 9 /** - * struct teo_idle_state - Idle state data used by the TEO cpuidle governor. - * @early_hits: "Early" CPU wakeups "matching" this state. - * @hits: "On time" CPU wakeups "matching" this state. - * @misses: CPU wakeups "missing" this state. - * - * A CPU wakeup is "matched" by a given idle state if the idle duration measured - * after the wakeup is between the target residency of that state and the target - * residency of the next one (or if this is the deepest available idle state, it - * "matches" a CPU wakeup when the measured idle duration is at least equal to - * its target residency). - * - * Also, from the TEO governor perspective, a CPU wakeup from idle is "early" if - * it occurs significantly earlier than the closest expected timer event (that - * is, early enough to match an idle state shallower than the one matching the - * time till the closest timer event). Otherwise, the wakeup is "on time", or - * it is a "hit". - * - * A "miss" occurs when the given state doesn't match the wakeup, but it matches - * the time till the closest timer event used for idle state selection. + * struct teo_bin - Metrics used by the TEO cpuidle governor. + * @intercepts: The "intercepts" metric. + * @hits: The "hits" metric. + * @recent: The number of recent "intercepts". */ -struct teo_idle_state { - unsigned int early_hits; +struct teo_bin { + unsigned int intercepts; unsigned int hits; - unsigned int misses; + unsigned int recent; }; /** * struct teo_cpu - CPU data used by the TEO cpuidle governor. * @time_span_ns: Time between idle state selection and post-wakeup update. * @sleep_length_ns: Time till the closest timer event (at the selection time). - * @states: Idle states data corresponding to this CPU. - * @interval_idx: Index of the most recent saved idle interval. - * @intervals: Saved idle duration values. + * @state_bins: Idle state data bins for this CPU. + * @total: Grand total of the "intercepts" and "hits" mertics for all bins. + * @next_recent_idx: Index of the next @recent_idx entry to update. + * @recent_idx: Indices of bins corresponding to recent "intercepts". */ struct teo_cpu { s64 time_span_ns; s64 sleep_length_ns; - struct teo_idle_state states[CPUIDLE_STATE_MAX]; - int interval_idx; - u64 intervals[INTERVALS]; + struct teo_bin state_bins[CPUIDLE_STATE_MAX]; + unsigned int total; + int next_recent_idx; + int recent_idx[NR_RECENT]; }; static DEFINE_PER_CPU(struct teo_cpu, teo_cpus); /** - * teo_update - Update CPU data after wakeup. + * teo_update - Update CPU metrics after wakeup. * @drv: cpuidle driver containing state data. * @dev: Target CPU. */ static void teo_update(struct cpuidle_driver *drv, struct cpuidle_device *dev) { struct teo_cpu *cpu_data = per_cpu_ptr(&teo_cpus, dev->cpu); - int i, idx_hit = 0, idx_timer = 0; - unsigned int hits, misses; + int i, idx_timer = 0, idx_duration = 0; u64 measured_ns; if (cpu_data->time_span_ns >= cpu_data->sleep_length_ns) { @@ -151,53 +193,52 @@ static void teo_update(struct cpuidle_driver *drv, struct cpuidle_device *dev) measured_ns /= 2; } + cpu_data->total = 0; + /* - * Decay the "early hits" metric for all of the states and find the - * states matching the sleep length and the measured idle duration. + * Decay the "hits" and "intercepts" metrics for all of the bins and + * find the bins that the sleep length and the measured idle duration + * fall into. */ for (i = 0; i < drv->state_count; i++) { - unsigned int early_hits = cpu_data->states[i].early_hits; + s64 target_residency_ns = drv->states[i].target_residency_ns; + struct teo_bin *bin = &cpu_data->state_bins[i]; - cpu_data->states[i].early_hits -= early_hits >> DECAY_SHIFT; + bin->hits -= bin->hits >> DECAY_SHIFT; + bin->intercepts -= bin->intercepts >> DECAY_SHIFT; - if (drv->states[i].target_residency_ns <= cpu_data->sleep_length_ns) { + cpu_data->total += bin->hits + bin->intercepts; + + if (target_residency_ns <= cpu_data->sleep_length_ns) { idx_timer = i; - if (drv->states[i].target_residency_ns <= measured_ns) - idx_hit = i; + if (target_residency_ns <= measured_ns) + idx_duration = i; } } - /* - * Update the "hits" and "misses" data for the state matching the sleep - * length. If it matches the measured idle duration too, this is a hit, - * so increase the "hits" metric for it then. Otherwise, this is a - * miss, so increase the "misses" metric for it. In the latter case - * also increase the "early hits" metric for the state that actually - * matches the measured idle duration. - */ - hits = cpu_data->states[idx_timer].hits; - hits -= hits >> DECAY_SHIFT; + i = cpu_data->next_recent_idx++; + if (cpu_data->next_recent_idx >= NR_RECENT) + cpu_data->next_recent_idx = 0; - misses = cpu_data->states[idx_timer].misses; - misses -= misses >> DECAY_SHIFT; + if (cpu_data->recent_idx[i] >= 0) + cpu_data->state_bins[cpu_data->recent_idx[i]].recent--; - if (idx_timer == idx_hit) { - hits += PULSE; + /* + * If the measured idle duration falls into the same bin as the sleep + * length, this is a "hit", so update the "hits" metric for that bin. + * Otherwise, update the "intercepts" metric for the bin fallen into by + * the measured idle duration. + */ + if (idx_timer == idx_duration) { + cpu_data->state_bins[idx_timer].hits += PULSE; + cpu_data->recent_idx[i] = -1; } else { - misses += PULSE; - cpu_data->states[idx_hit].early_hits += PULSE; + cpu_data->state_bins[idx_duration].intercepts += PULSE; + cpu_data->state_bins[idx_duration].recent++; + cpu_data->recent_idx[i] = idx_duration; } - cpu_data->states[idx_timer].misses = misses; - cpu_data->states[idx_timer].hits = hits; - - /* - * Save idle duration values corresponding to non-timer wakeups for - * pattern detection. - */ - cpu_data->intervals[cpu_data->interval_idx++] = measured_ns; - if (cpu_data->interval_idx >= INTERVALS) - cpu_data->interval_idx = 0; + cpu_data->total += PULSE; } static bool teo_time_ok(u64 interval_ns) @@ -205,6 +246,12 @@ static bool teo_time_ok(u64 interval_ns) return !tick_nohz_tick_stopped() || interval_ns >= TICK_NSEC; } +static s64 teo_middle_of_bin(int idx, struct cpuidle_driver *drv) +{ + return (drv->states[idx].target_residency_ns + + drv->states[idx+1].target_residency_ns) / 2; +} + /** * teo_find_shallower_state - Find shallower idle state matching given duration. * @drv: cpuidle driver containing state data. @@ -240,10 +287,18 @@ static int teo_select(struct cpuidle_driver *drv, struct cpuidle_device *dev, { struct teo_cpu *cpu_data = per_cpu_ptr(&teo_cpus, dev->cpu); s64 latency_req = cpuidle_governor_latency_req(dev->cpu); - int max_early_idx, prev_max_early_idx, constraint_idx, idx0, idx, i; - unsigned int hits, misses, early_hits; + unsigned int idx_intercept_sum = 0; + unsigned int intercept_sum = 0; + unsigned int idx_recent_sum = 0; + unsigned int recent_sum = 0; + unsigned int idx_hit_sum = 0; + unsigned int hit_sum = 0; + int constraint_idx = 0; + int idx0 = 0, idx = -1; + bool alt_intercepts, alt_recent; ktime_t delta_tick; s64 duration_ns; + int i; if (dev->last_state_idx >= 0) { teo_update(drv, dev); @@ -255,171 +310,136 @@ static int teo_select(struct cpuidle_driver *drv, struct cpuidle_device *dev, duration_ns = tick_nohz_get_sleep_length(&delta_tick); cpu_data->sleep_length_ns = duration_ns; - hits = 0; - misses = 0; - early_hits = 0; - max_early_idx = -1; - prev_max_early_idx = -1; - constraint_idx = drv->state_count; - idx = -1; - idx0 = idx; + /* Check if there is any choice in the first place. */ + if (drv->state_count < 2) { + idx = 0; + goto end; + } + if (!dev->states_usage[0].disable) { + idx = 0; + if (drv->states[1].target_residency_ns > duration_ns) + goto end; + } - for (i = 0; i < drv->state_count; i++) { + /* + * Find the deepest idle state whose target residency does not exceed + * the current sleep length and the deepest idle state not deeper than + * the former whose exit latency does not exceed the current latency + * constraint. Compute the sums of metrics for early wakeup pattern + * detection. + */ + for (i = 1; i < drv->state_count; i++) { + struct teo_bin *prev_bin = &cpu_data->state_bins[i-1]; struct cpuidle_state *s = &drv->states[i]; - if (dev->states_usage[i].disable) { - /* - * Ignore disabled states with target residencies beyond - * the anticipated idle duration. - */ - if (s->target_residency_ns > duration_ns) - continue; - - /* - * This state is disabled, so the range of idle duration - * values corresponding to it is covered by the current - * candidate state, but still the "hits" and "misses" - * metrics of the disabled state need to be used to - * decide whether or not the state covering the range in - * question is good enough. - */ - hits = cpu_data->states[i].hits; - misses = cpu_data->states[i].misses; - - if (early_hits >= cpu_data->states[i].early_hits || - idx < 0) - continue; - - /* - * If the current candidate state has been the one with - * the maximum "early hits" metric so far, the "early - * hits" metric of the disabled state replaces the - * current "early hits" count to avoid selecting a - * deeper state with lower "early hits" metric. - */ - if (max_early_idx == idx) { - early_hits = cpu_data->states[i].early_hits; - continue; - } - - /* - * The current candidate state is closer to the disabled - * one than the current maximum "early hits" state, so - * replace the latter with it, but in case the maximum - * "early hits" state index has not been set so far, - * check if the current candidate state is not too - * shallow for that role. - */ - if (teo_time_ok(drv->states[idx].target_residency_ns)) { - prev_max_early_idx = max_early_idx; - early_hits = cpu_data->states[i].early_hits; - max_early_idx = idx; - } + /* + * Update the sums of idle state mertics for all of the states + * shallower than the current one. + */ + intercept_sum += prev_bin->intercepts; + hit_sum += prev_bin->hits; + recent_sum += prev_bin->recent; + if (dev->states_usage[i].disable) continue; - } if (idx < 0) { idx = i; /* first enabled state */ - hits = cpu_data->states[i].hits; - misses = cpu_data->states[i].misses; idx0 = i; } if (s->target_residency_ns > duration_ns) break; - if (s->exit_latency_ns > latency_req && constraint_idx > i) + idx = i; + + if (s->exit_latency_ns <= latency_req) constraint_idx = i; - idx = i; - hits = cpu_data->states[i].hits; - misses = cpu_data->states[i].misses; - - if (early_hits < cpu_data->states[i].early_hits && - teo_time_ok(drv->states[i].target_residency_ns)) { - prev_max_early_idx = max_early_idx; - early_hits = cpu_data->states[i].early_hits; - max_early_idx = i; - } + idx_intercept_sum = intercept_sum; + idx_hit_sum = hit_sum; + idx_recent_sum = recent_sum; } - /* - * If the "hits" metric of the idle state matching the sleep length is - * greater than its "misses" metric, that is the one to use. Otherwise, - * it is more likely that one of the shallower states will match the - * idle duration observed after wakeup, so take the one with the maximum - * "early hits" metric, but if that cannot be determined, just use the - * state selected so far. - */ - if (hits <= misses) { - /* - * The current candidate state is not suitable, so take the one - * whose "early hits" metric is the maximum for the range of - * shallower states. - */ - if (idx == max_early_idx) - max_early_idx = prev_max_early_idx; - - if (max_early_idx >= 0) { - idx = max_early_idx; - duration_ns = drv->states[idx].target_residency_ns; - } + /* Avoid unnecessary overhead. */ + if (idx < 0) { + idx = 0; /* No states enabled, must use 0. */ + goto end; + } else if (idx == idx0) { + goto end; } /* - * If there is a latency constraint, it may be necessary to use a - * shallower idle state than the one selected so far. + * If the sum of the intercepts metric for all of the idle states + * shallower than the current candidate one (idx) is greater than the + * sum of the intercepts and hits metrics for the candidate state and + * all of the deeper states, or the sum of the numbers of recent + * intercepts over all of the states shallower than the candidate one + * is greater than a half of the number of recent events taken into + * account, the CPU is likely to wake up early, so find an alternative + * idle state to select. */ - if (constraint_idx < idx) - idx = constraint_idx; - - if (idx < 0) { - idx = 0; /* No states enabled. Must use 0. */ - } else if (idx > idx0) { - unsigned int count = 0; - u64 sum = 0; + alt_intercepts = 2 * idx_intercept_sum > cpu_data->total - idx_hit_sum; + alt_recent = idx_recent_sum > NR_RECENT / 2; + if (alt_recent || alt_intercepts) { + s64 last_enabled_span_ns = duration_ns; + int last_enabled_idx = idx; /* - * The target residencies of at least two different enabled idle - * states are less than or equal to the current expected idle - * duration. Try to refine the selection using the most recent - * measured idle duration values. + * Look for the deepest idle state whose target residency had + * not exceeded the idle duration in over a half of the relevant + * cases (both with respect to intercepts overall and with + * respect to the recent intercepts only) in the past. * - * Count and sum the most recent idle duration values less than - * the current expected idle duration value. + * Take the possible latency constraint and duration limitation + * present if the tick has been stopped already into account. */ - for (i = 0; i < INTERVALS; i++) { - u64 val = cpu_data->intervals[i]; + intercept_sum = 0; + recent_sum = 0; + + for (i = idx - 1; i >= idx0; i--) { + struct teo_bin *bin = &cpu_data->state_bins[i]; + s64 span_ns; - if (val >= duration_ns) + intercept_sum += bin->intercepts; + recent_sum += bin->recent; + + if (dev->states_usage[i].disable) continue; - count++; - sum += val; - } + span_ns = teo_middle_of_bin(i, drv); + if (!teo_time_ok(span_ns)) { + /* + * The current state is too shallow, so select + * the first enabled deeper state. + */ + duration_ns = last_enabled_span_ns; + idx = last_enabled_idx; + break; + } - /* - * Give up unless the majority of the most recent idle duration - * values are in the interesting range. - */ - if (count > INTERVALS / 2) { - u64 avg_ns = div64_u64(sum, count); - - /* - * Avoid spending too much time in an idle state that - * would be too shallow. - */ - if (teo_time_ok(avg_ns)) { - duration_ns = avg_ns; - if (drv->states[idx].target_residency_ns > avg_ns) - idx = teo_find_shallower_state(drv, dev, - idx, avg_ns); + if ((!alt_recent || 2 * recent_sum > idx_recent_sum) && + (!alt_intercepts || + 2 * intercept_sum > idx_intercept_sum)) { + idx = i; + duration_ns = span_ns; + break; } + + last_enabled_span_ns = span_ns; + last_enabled_idx = i; } } /* + * If there is a latency constraint, it may be necessary to select an + * idle state shallower than the current candidate one. + */ + if (idx > constraint_idx) + idx = constraint_idx; + +end: + /* * 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. */ @@ -478,8 +498,8 @@ static int teo_enable_device(struct cpuidle_driver *drv, memset(cpu_data, 0, sizeof(*cpu_data)); - for (i = 0; i < INTERVALS; i++) - cpu_data->intervals[i] = U64_MAX; + for (i = 0; i < NR_RECENT; i++) + cpu_data->recent_idx[i] = -1; return 0; } |