diff options
Diffstat (limited to 'drivers/cpuidle/governors/menu.c')
| -rw-r--r-- | drivers/cpuidle/governors/menu.c | 174 |
1 files changed, 146 insertions, 28 deletions
diff --git a/drivers/cpuidle/governors/menu.c b/drivers/cpuidle/governors/menu.c index 5b1f2c372c1..fe343a06b7d 100644 --- a/drivers/cpuidle/governors/menu.c +++ b/drivers/cpuidle/governors/menu.c @@ -28,6 +28,13 @@ #define MAX_INTERESTING 50000 #define STDDEV_THRESH 400 +/* 60 * 60 > STDDEV_THRESH * INTERVALS = 400 * 8 */ +#define MAX_DEVIATION 60 + +static DEFINE_PER_CPU(struct hrtimer, menu_hrtimer); +static DEFINE_PER_CPU(int, hrtimer_status); +/* menu hrtimer mode */ +enum {MENU_HRTIMER_STOP, MENU_HRTIMER_REPEAT, MENU_HRTIMER_GENERAL}; /* * Concepts and ideas behind the menu governor @@ -109,6 +116,13 @@ * */ +/* + * The C-state residency is so long that is is worthwhile to exit + * from the shallow C-state and re-enter into a deeper C-state. + */ +static unsigned int perfect_cstate_ms __read_mostly = 30; +module_param(perfect_cstate_ms, uint, 0000); + struct menu_device { int last_state_idx; int needs_update; @@ -191,40 +205,102 @@ static u64 div_round64(u64 dividend, u32 divisor) return div_u64(dividend + (divisor / 2), divisor); } +/* Cancel the hrtimer if it is not triggered yet */ +void menu_hrtimer_cancel(void) +{ + int cpu = smp_processor_id(); + struct hrtimer *hrtmr = &per_cpu(menu_hrtimer, cpu); + + /* The timer is still not time out*/ + if (per_cpu(hrtimer_status, cpu)) { + hrtimer_cancel(hrtmr); + per_cpu(hrtimer_status, cpu) = MENU_HRTIMER_STOP; + } +} +EXPORT_SYMBOL_GPL(menu_hrtimer_cancel); + +/* Call back for hrtimer is triggered */ +static enum hrtimer_restart menu_hrtimer_notify(struct hrtimer *hrtimer) +{ + int cpu = smp_processor_id(); + struct menu_device *data = &per_cpu(menu_devices, cpu); + + /* In general case, the expected residency is much larger than + * deepest C-state target residency, but prediction logic still + * predicts a small predicted residency, so the prediction + * history is totally broken if the timer is triggered. + * So reset the correction factor. + */ + if (per_cpu(hrtimer_status, cpu) == MENU_HRTIMER_GENERAL) + data->correction_factor[data->bucket] = RESOLUTION * DECAY; + + per_cpu(hrtimer_status, cpu) = MENU_HRTIMER_STOP; + + return HRTIMER_NORESTART; +} + /* * 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 void detect_repeating_patterns(struct menu_device *data) +static u32 get_typical_interval(struct menu_device *data) { - int i; - uint64_t avg = 0; - uint64_t stddev = 0; /* contains the square of the std deviation */ - - /* first calculate average and standard deviation of the past */ - for (i = 0; i < INTERVALS; i++) - avg += data->intervals[i]; - avg = avg / INTERVALS; - - /* if the avg is beyond the known next tick, it's worthless */ - if (avg > data->expected_us) - return; + int i = 0, divisor = 0; + uint64_t max = 0, avg = 0, stddev = 0; + int64_t thresh = LLONG_MAX; /* Discard outliers above this value. */ + unsigned int ret = 0; - for (i = 0; i < INTERVALS; i++) - stddev += (data->intervals[i] - avg) * - (data->intervals[i] - avg); +again: - stddev = stddev / INTERVALS; + /* first calculate average and standard deviation of the past */ + max = avg = divisor = stddev = 0; + for (i = 0; i < INTERVALS; i++) { + int64_t value = data->intervals[i]; + if (value <= thresh) { + avg += value; + divisor++; + if (value > max) + max = value; + } + } + do_div(avg, divisor); + for (i = 0; i < INTERVALS; i++) { + int64_t value = data->intervals[i]; + if (value <= thresh) { + int64_t diff = value - avg; + stddev += diff * diff; + } + } + do_div(stddev, divisor); + stddev = int_sqrt(stddev); /* - * now.. if stddev is small.. then assume we have a - * repeating pattern and predict we keep doing this. + * 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. + * + * The typical interval is obtained when standard deviation is small + * or standard deviation is small compared to the average interval. */ - - if (avg && stddev < STDDEV_THRESH) + if (((avg > stddev * 6) && (divisor * 4 >= INTERVALS * 3)) + || stddev <= 20) { data->predicted_us = avg; + ret = 1; + return ret; + + } else if ((divisor * 4) > INTERVALS * 3) { + /* Exclude the max interval */ + thresh = max - 1; + goto again; + } + + return ret; } /** @@ -236,10 +312,12 @@ static int menu_select(struct cpuidle_driver *drv, struct cpuidle_device *dev) { struct menu_device *data = &__get_cpu_var(menu_devices); int latency_req = pm_qos_request(PM_QOS_CPU_DMA_LATENCY); - int power_usage = -1; int i; int multiplier; struct timespec t; + int repeat = 0, low_predicted = 0; + int cpu = smp_processor_id(); + struct hrtimer *hrtmr = &per_cpu(menu_hrtimer, cpu); if (data->needs_update) { menu_update(drv, dev); @@ -274,7 +352,7 @@ static int menu_select(struct cpuidle_driver *drv, struct cpuidle_device *dev) data->predicted_us = div_round64(data->expected_us * data->correction_factor[data->bucket], RESOLUTION * DECAY); - detect_repeating_patterns(data); + repeat = get_typical_interval(data); /* * We want to default to C1 (hlt), not to busy polling @@ -295,18 +373,55 @@ static int menu_select(struct cpuidle_driver *drv, struct cpuidle_device *dev) if (s->disabled || su->disable) continue; - if (s->target_residency > data->predicted_us) + if (s->target_residency > data->predicted_us) { + low_predicted = 1; continue; + } if (s->exit_latency > latency_req) continue; if (s->exit_latency * multiplier > data->predicted_us) continue; - if (s->power_usage < power_usage) { - power_usage = s->power_usage; - data->last_state_idx = i; - data->exit_us = s->exit_latency; + data->last_state_idx = i; + data->exit_us = s->exit_latency; + } + + /* not deepest C-state chosen for low predicted residency */ + if (low_predicted) { + unsigned int timer_us = 0; + unsigned int perfect_us = 0; + + /* + * Set a timer to detect whether this sleep is much + * longer than repeat mode predicted. If the timer + * triggers, the code will evaluate whether to put + * the CPU into a deeper C-state. + * The timer is cancelled on CPU wakeup. + */ + timer_us = 2 * (data->predicted_us + MAX_DEVIATION); + + perfect_us = perfect_cstate_ms * 1000; + + if (repeat && (4 * timer_us < data->expected_us)) { + RCU_NONIDLE(hrtimer_start(hrtmr, + ns_to_ktime(1000 * timer_us), + HRTIMER_MODE_REL_PINNED)); + /* In repeat case, menu hrtimer is started */ + per_cpu(hrtimer_status, cpu) = MENU_HRTIMER_REPEAT; + } else if (perfect_us < data->expected_us) { + /* + * The next timer is long. This could be because + * we did not make a useful prediction. + * In that case, it makes sense to re-enter + * into a deeper C-state after some time. + */ + RCU_NONIDLE(hrtimer_start(hrtmr, + ns_to_ktime(1000 * timer_us), + HRTIMER_MODE_REL_PINNED)); + /* In general case, menu hrtimer is started */ + per_cpu(hrtimer_status, cpu) = MENU_HRTIMER_GENERAL; } + } return data->last_state_idx; @@ -399,6 +514,9 @@ static int menu_enable_device(struct cpuidle_driver *drv, struct cpuidle_device *dev) { struct menu_device *data = &per_cpu(menu_devices, dev->cpu); + struct hrtimer *t = &per_cpu(menu_hrtimer, dev->cpu); + hrtimer_init(t, CLOCK_MONOTONIC, HRTIMER_MODE_REL); + t->function = menu_hrtimer_notify; memset(data, 0, sizeof(struct menu_device)); |