1 // SPDX-License-Identifier: GPL-2.0-only
3  * menu.c - the menu idle governor
5  * Copyright (C) 2006-2007 Adam Belay <abelay@novell.com>
40  * -----------------------
44  * need is a good prediction of how long we'll be idle. Like the traditional
57  * duration; if we expect 500 milliseconds of idle time the likelihood of
59  * seconds of idle time. A second independent factor that has big impact on
68  * Repeatable-interval-detector
69  * ----------------------------
79  * ---------------------------
80  * C states, especially those with large exit latencies, can have a real
86  *     The busier the system, the less impact of C states is acceptable
88  * This rule-of-thumb is implemented using a performance-multiplier:
92  * this multiplier is, the longer we need to be idle to pick a deep C
93  * state, and thus the less likely a busy CPU will hit such a deep
97  * a value of 10 is added for each point of "per cpu load average" we have.
99  * IO on this CPU.
103  * decision, while the iowait value gives a cpu local instantanious input.
155 	/* for IO wait tasks (per cpu!) we add 10x each */  in performance_multiplier()
186 		unsigned int value = data->intervals[i];  in get_typical_interval()
213 		unsigned int value = data->intervals[i];  in get_typical_interval()
215 			int64_t diff = (int64_t)value - avg;  in get_typical_interval()
255 	thresh = max - 1;  in get_typical_interval()
260  * menu_select - selects the next idle state to enter
262  * @dev: the CPU
269 	s64 latency_req = cpuidle_governor_latency_req(dev->cpu);  in menu_select()
277 	if (data->needs_update) {  in menu_select()
279 		data->needs_update = 0;  in menu_select()
283 	data->next_timer_ns = tick_nohz_get_sleep_length(&delta_next);  in menu_select()
285 	nr_iowaiters = nr_iowait_cpu(dev->cpu);  in menu_select()
286 	data->bucket = which_bucket(data->next_timer_ns, nr_iowaiters);  in menu_select()
288 	if (unlikely(drv->state_count <= 1 || latency_req == 0) ||  in menu_select()
289 	    ((data->next_timer_ns < drv->states[1].target_residency_ns ||  in menu_select()
290 	      latency_req < drv->states[1].exit_latency_ns) &&  in menu_select()
291 	     !dev->states_usage[0].disable)) {  in menu_select()
297 		*stop_tick = !(drv->states[0].flags & CPUIDLE_FLAG_POLLING);  in menu_select()
302 	predicted_us = div_u64(data->next_timer_ns *  in menu_select()
303 			       data->correction_factor[data->bucket] +  in menu_select()
306 	/* Use the lowest expected idle interval to pick the idle state. */  in menu_select()
314 		 * idle duration misprediction is much higher, because the CPU  in menu_select()
315 		 * may be stuck in a shallow idle state for a long time as a  in menu_select()
317 		 * the known time till the closest timer event for the idle  in menu_select()
324 		 * Use the performance multiplier and the user-configurable  in menu_select()
334 	 * Find the idle state with the lowest power while satisfying  in menu_select()
337 	idx = -1;  in menu_select()
338 	for (i = 0; i < drv->state_count; i++) {  in menu_select()
339 		struct cpuidle_state *s = &drv->states[i];  in menu_select()
341 		if (dev->states_usage[i].disable)  in menu_select()
344 		if (idx == -1)  in menu_select()
347 		if (s->target_residency_ns > predicted_ns) {  in menu_select()
349 			 * Use a physical idle state, not busy polling, unless  in menu_select()
352 			if ((drv->states[idx].flags & CPUIDLE_FLAG_POLLING) &&  in menu_select()
353 			    s->exit_latency_ns <= latency_req &&  in menu_select()
354 			    s->target_residency_ns <= data->next_timer_ns) {  in menu_select()
355 				predicted_ns = s->target_residency_ns;  in menu_select()
369 				predicted_ns = drv->states[idx].target_residency_ns;  in menu_select()
379 			if (drv->states[idx].target_residency_ns < TICK_NSEC &&  in menu_select()
380 			    s->target_residency_ns <= delta_next)  in menu_select()
385 		if (s->exit_latency_ns > latency_req)  in menu_select()
391 	if (idx == -1)  in menu_select()
392 		idx = 0; /* No states enabled. Must use 0. */  in menu_select()
396 	 * expected idle duration is shorter than the tick period length.  in menu_select()
398 	if (((drv->states[idx].flags & CPUIDLE_FLAG_POLLING) ||  in menu_select()
402 		if (idx > 0 && drv->states[idx].target_residency_ns > delta_next) {  in menu_select()
409 			for (i = idx - 1; i >= 0; i--) {  in menu_select()
410 				if (dev->states_usage[i].disable)  in menu_select()
414 				if (drv->states[i].target_residency_ns <= delta_next)  in menu_select()
424  * menu_reflect - records that data structures need update
425  * @dev: the CPU
435 	dev->last_state_idx = index;  in menu_reflect()
436 	data->needs_update = 1;  in menu_reflect()
437 	data->tick_wakeup = tick_nohz_idle_got_tick();  in menu_reflect()
441  * menu_update - attempts to guess what happened after entry
443  * @dev: the CPU
448 	int last_idx = dev->last_state_idx;  in menu_update()
449 	struct cpuidle_state *target = &drv->states[last_idx];  in menu_update()
457 	 * If the entered idle state didn't support residency measurements,  in menu_update()
468 	if (data->tick_wakeup && data->next_timer_ns > TICK_NSEC) {  in menu_update()
471 		 * the tick boundary (if the tick was stopped), but the idle  in menu_update()
472 		 * duration predictor had a differing opinion.  Since the CPU  in menu_update()
474 		 * predictor was not quite right, so assume that the CPU could  in menu_update()
475 		 * have been idle long (but not forever) to help the idle  in menu_update()
479 	} else if ((drv->states[last_idx].flags & CPUIDLE_FLAG_POLLING) &&  in menu_update()
480 		   dev->poll_time_limit) {  in menu_update()
482 		 * The CPU exited the "polling" state due to a time limit, so  in menu_update()
483 		 * the idle duration prediction leading to the selection of that  in menu_update()
485 		 * the CPU might have been woken up from idle by the next timer.  in menu_update()
488 		measured_ns = data->next_timer_ns;  in menu_update()
491 		measured_ns = dev->last_residency_ns;  in menu_update()
494 		if (measured_ns > 2 * target->exit_latency_ns)  in menu_update()
495 			measured_ns -= target->exit_latency_ns;  in menu_update()
501 	if (measured_ns > data->next_timer_ns)  in menu_update()
502 		measured_ns = data->next_timer_ns;  in menu_update()
505 	new_factor = data->correction_factor[data->bucket];  in menu_update()
506 	new_factor -= new_factor / DECAY;  in menu_update()
508 	if (data->next_timer_ns > 0 && measured_ns < MAX_INTERESTING)  in menu_update()
510 					data->next_timer_ns);  in menu_update()
513 		 * we were idle so long that we count it as a perfect  in menu_update()
527 	data->correction_factor[data->bucket] = new_factor;  in menu_update()
529 	/* update the repeating-pattern data */  in menu_update()
530 	data->intervals[data->interval_ptr++] = ktime_to_us(measured_ns);  in menu_update()
531 	if (data->interval_ptr >= INTERVALS)  in menu_update()
532 		data->interval_ptr = 0;  in menu_update()
536  * menu_enable_device - scans a CPU's states and does setup
538  * @dev: the CPU
543 	struct menu_device *data = &per_cpu(menu_devices, dev->cpu);  in menu_enable_device()
549 	 * if the correction factor is 0 (eg first time init or cpu hotplug  in menu_enable_device()
553 		data->correction_factor[i] = RESOLUTION * DECAY;  in menu_enable_device()
567  * init_menu - initializes the governor