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  * -----------------------
42  * the  C state is required to actually break even on this cost. CPUIDLE
44  * need is a good prediction of how long we'll be idle. Like the traditional
50  * that is based on historic behavior. For example, if in the past the actual
56  * that the ratio is dependent on the order of magnitude of the expected
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
62  * as perfect; there are no power gains for sleeping longer than this)
65  * indexed based on the magnitude of the expected duration as well as the
68  * Repeatable-interval-detector
69  * ----------------------------
79  * ---------------------------
81  * noticeable impact on workloads, which is not acceptable for most sysadmins,
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
99  * IO on this CPU.
125 	 * We keep two groups of stats; one with no  in which_bucket()
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()
234 	 * Use this result only if there is no timer to wake us up sooner.  in get_typical_interval()
255 	thresh = max - 1;  in get_typical_interval()
260  * menu_select - selects the next idle state to enter
263  * @stop_tick: indication on whether or not to stop the tick
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()
288 	data->next_timer_ns = delta;  in menu_select()
290 	nr_iowaiters = nr_iowait_cpu(dev->cpu);  in menu_select()
291 	data->bucket = which_bucket(data->next_timer_ns, nr_iowaiters);  in menu_select()
293 	if (unlikely(drv->state_count <= 1 || latency_req == 0) ||  in menu_select()
294 	    ((data->next_timer_ns < drv->states[1].target_residency_ns ||  in menu_select()
295 	      latency_req < drv->states[1].exit_latency_ns) &&  in menu_select()
296 	     !dev->states_usage[0].disable)) {  in menu_select()
298 		 * In this case state[0] will be used no matter what, so return  in menu_select()
302 		*stop_tick = !(drv->states[0].flags & CPUIDLE_FLAG_POLLING);  in menu_select()
307 	predicted_us = div_u64(data->next_timer_ns *  in menu_select()
308 			       data->correction_factor[data->bucket] +  in menu_select()
311 	/* Use the lowest expected idle interval to pick the idle state. */  in menu_select()
319 		 * idle duration misprediction is much higher, because the CPU  in menu_select()
320 		 * may be stuck in a shallow idle state for a long time as a  in menu_select()
322 		 * the known time till the closest timer event for the idle  in menu_select()
326 			predicted_ns = data->next_timer_ns;  in menu_select()
329 		 * Use the performance multiplier and the user-configurable  in menu_select()
339 	 * Find the idle state with the lowest power while satisfying  in menu_select()
342 	idx = -1;  in menu_select()
343 	for (i = 0; i < drv->state_count; i++) {  in menu_select()
344 		struct cpuidle_state *s = &drv->states[i];  in menu_select()
346 		if (dev->states_usage[i].disable)  in menu_select()
349 		if (idx == -1)  in menu_select()
352 		if (s->target_residency_ns > predicted_ns) {  in menu_select()
354 			 * Use a physical idle state, not busy polling, unless  in menu_select()
357 			if ((drv->states[idx].flags & CPUIDLE_FLAG_POLLING) &&  in menu_select()
358 			    s->exit_latency_ns <= latency_req &&  in menu_select()
359 			    s->target_residency_ns <= data->next_timer_ns) {  in menu_select()
360 				predicted_ns = s->target_residency_ns;  in menu_select()
374 				predicted_ns = drv->states[idx].target_residency_ns;  in menu_select()
384 			if (drv->states[idx].target_residency_ns < TICK_NSEC &&  in menu_select()
385 			    s->target_residency_ns <= delta_tick)  in menu_select()
390 		if (s->exit_latency_ns > latency_req)  in menu_select()
396 	if (idx == -1)  in menu_select()
397 		idx = 0; /* No states enabled. Must use 0. */  in menu_select()
401 	 * expected idle duration is shorter than the tick period length.  in menu_select()
403 	if (((drv->states[idx].flags & CPUIDLE_FLAG_POLLING) ||  in menu_select()
407 		if (idx > 0 && drv->states[idx].target_residency_ns > delta_tick) {  in menu_select()
414 			for (i = idx - 1; i >= 0; i--) {  in menu_select()
415 				if (dev->states_usage[i].disable)  in menu_select()
419 				if (drv->states[i].target_residency_ns <= delta_tick)  in menu_select()
429  * menu_reflect - records that data structures need update
440 	dev->last_state_idx = index;  in menu_reflect()
441 	data->needs_update = 1;  in menu_reflect()
442 	data->tick_wakeup = tick_nohz_idle_got_tick();  in menu_reflect()
446  * menu_update - attempts to guess what happened after entry
453 	int last_idx = dev->last_state_idx;  in menu_update()
454 	struct cpuidle_state *target = &drv->states[last_idx];  in menu_update()
462 	 * If the entered idle state didn't support residency measurements,  in menu_update()
473 	if (data->tick_wakeup && data->next_timer_ns > TICK_NSEC) {  in menu_update()
476 		 * the tick boundary (if the tick was stopped), but the idle  in menu_update()
480 		 * have been idle long (but not forever) to help the idle  in menu_update()
484 	} else if ((drv->states[last_idx].flags & CPUIDLE_FLAG_POLLING) &&  in menu_update()
485 		   dev->poll_time_limit) {  in menu_update()
488 		 * the idle duration prediction leading to the selection of that  in menu_update()
490 		 * the CPU might have been woken up from idle by the next timer.  in menu_update()
493 		measured_ns = data->next_timer_ns;  in menu_update()
496 		measured_ns = dev->last_residency_ns;  in menu_update()
499 		if (measured_ns > 2 * target->exit_latency_ns)  in menu_update()
500 			measured_ns -= target->exit_latency_ns;  in menu_update()
506 	if (measured_ns > data->next_timer_ns)  in menu_update()
507 		measured_ns = data->next_timer_ns;  in menu_update()
510 	new_factor = data->correction_factor[data->bucket];  in menu_update()
511 	new_factor -= new_factor / DECAY;  in menu_update()
513 	if (data->next_timer_ns > 0 && measured_ns < MAX_INTERESTING)  in menu_update()
515 					data->next_timer_ns);  in menu_update()
518 		 * we were idle so long that we count it as a perfect  in menu_update()
532 	data->correction_factor[data->bucket] = new_factor;  in menu_update()
534 	/* update the repeating-pattern data */  in menu_update()
535 	data->intervals[data->interval_ptr++] = ktime_to_us(measured_ns);  in menu_update()
536 	if (data->interval_ptr >= INTERVALS)  in menu_update()
537 		data->interval_ptr = 0;  in menu_update()
541  * menu_enable_device - scans a CPU's states and does setup
548 	struct menu_device *data = &per_cpu(menu_devices, dev->cpu);  in menu_enable_device()
554 	 * if the correction factor is 0 (eg first time init or cpu hotplug  in menu_enable_device()
558 		data->correction_factor[i] = RESOLUTION * DECAY;  in menu_enable_device()
572  * init_menu - initializes the governor