xref: /openbmc/linux/drivers/cpuidle/governors/teo.c (revision a61127c2)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Timer events oriented CPU idle governor
4  *
5  * Copyright (C) 2018 Intel Corporation
6  * Author: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
7  *
8  * The idea of this governor is based on the observation that on many systems
9  * timer events are two or more orders of magnitude more frequent than any
10  * other interrupts, so they are likely to be the most significant source of CPU
11  * wakeups from idle states.  Moreover, information about what happened in the
12  * (relatively recent) past can be used to estimate whether or not the deepest
13  * idle state with target residency within the time to the closest timer is
14  * likely to be suitable for the upcoming idle time of the CPU and, if not, then
15  * which of the shallower idle states to choose.
16  *
17  * Of course, non-timer wakeup sources are more important in some use cases and
18  * they can be covered by taking a few most recent idle time intervals of the
19  * CPU into account.  However, even in that case it is not necessary to consider
20  * idle duration values greater than the time till the closest timer, as the
21  * patterns that they may belong to produce average values close enough to
22  * the time till the closest timer (sleep length) anyway.
23  *
24  * Thus this governor estimates whether or not the upcoming idle time of the CPU
25  * is likely to be significantly shorter than the sleep length and selects an
26  * idle state for it in accordance with that, as follows:
27  *
28  * - Find an idle state on the basis of the sleep length and state statistics
29  *   collected over time:
30  *
31  *   o Find the deepest idle state whose target residency is less than or equal
32  *     to the sleep length.
33  *
34  *   o Select it if it matched both the sleep length and the observed idle
35  *     duration in the past more often than it matched the sleep length alone
36  *     (i.e. the observed idle duration was significantly shorter than the sleep
37  *     length matched by it).
38  *
39  *   o Otherwise, select the shallower state with the greatest matched "early"
40  *     wakeups metric.
41  *
42  * - If the majority of the most recent idle duration values are below the
43  *   target residency of the idle state selected so far, use those values to
44  *   compute the new expected idle duration and find an idle state matching it
45  *   (which has to be shallower than the one selected so far).
46  */
47 
48 #include <linux/cpuidle.h>
49 #include <linux/jiffies.h>
50 #include <linux/kernel.h>
51 #include <linux/sched/clock.h>
52 #include <linux/tick.h>
53 
54 /*
55  * The PULSE value is added to metrics when they grow and the DECAY_SHIFT value
56  * is used for decreasing metrics on a regular basis.
57  */
58 #define PULSE		1024
59 #define DECAY_SHIFT	3
60 
61 /*
62  * Number of the most recent idle duration values to take into consideration for
63  * the detection of wakeup patterns.
64  */
65 #define INTERVALS	8
66 
67 /**
68  * struct teo_idle_state - Idle state data used by the TEO cpuidle governor.
69  * @early_hits: "Early" CPU wakeups "matching" this state.
70  * @hits: "On time" CPU wakeups "matching" this state.
71  * @misses: CPU wakeups "missing" this state.
72  *
73  * A CPU wakeup is "matched" by a given idle state if the idle duration measured
74  * after the wakeup is between the target residency of that state and the target
75  * residency of the next one (or if this is the deepest available idle state, it
76  * "matches" a CPU wakeup when the measured idle duration is at least equal to
77  * its target residency).
78  *
79  * Also, from the TEO governor perspective, a CPU wakeup from idle is "early" if
80  * it occurs significantly earlier than the closest expected timer event (that
81  * is, early enough to match an idle state shallower than the one matching the
82  * time till the closest timer event).  Otherwise, the wakeup is "on time", or
83  * it is a "hit".
84  *
85  * A "miss" occurs when the given state doesn't match the wakeup, but it matches
86  * the time till the closest timer event used for idle state selection.
87  */
88 struct teo_idle_state {
89 	unsigned int early_hits;
90 	unsigned int hits;
91 	unsigned int misses;
92 };
93 
94 /**
95  * struct teo_cpu - CPU data used by the TEO cpuidle governor.
96  * @time_span_ns: Time between idle state selection and post-wakeup update.
97  * @sleep_length_ns: Time till the closest timer event (at the selection time).
98  * @states: Idle states data corresponding to this CPU.
99  * @last_state: Idle state entered by the CPU last time.
100  * @interval_idx: Index of the most recent saved idle interval.
101  * @intervals: Saved idle duration values.
102  */
103 struct teo_cpu {
104 	u64 time_span_ns;
105 	u64 sleep_length_ns;
106 	struct teo_idle_state states[CPUIDLE_STATE_MAX];
107 	int last_state;
108 	int interval_idx;
109 	unsigned int intervals[INTERVALS];
110 };
111 
112 static DEFINE_PER_CPU(struct teo_cpu, teo_cpus);
113 
114 /**
115  * teo_update - Update CPU data after wakeup.
116  * @drv: cpuidle driver containing state data.
117  * @dev: Target CPU.
118  */
119 static void teo_update(struct cpuidle_driver *drv, struct cpuidle_device *dev)
120 {
121 	struct teo_cpu *cpu_data = per_cpu_ptr(&teo_cpus, dev->cpu);
122 	unsigned int sleep_length_us = ktime_to_us(cpu_data->sleep_length_ns);
123 	int i, idx_hit = -1, idx_timer = -1;
124 	unsigned int measured_us;
125 
126 	if (cpu_data->time_span_ns >= cpu_data->sleep_length_ns) {
127 		/*
128 		 * One of the safety nets has triggered or this was a timer
129 		 * wakeup (or equivalent).
130 		 */
131 		measured_us = sleep_length_us;
132 	} else {
133 		unsigned int lat = drv->states[cpu_data->last_state].exit_latency;
134 
135 		measured_us = ktime_to_us(cpu_data->time_span_ns);
136 		/*
137 		 * The delay between the wakeup and the first instruction
138 		 * executed by the CPU is not likely to be worst-case every
139 		 * time, so take 1/2 of the exit latency as a very rough
140 		 * approximation of the average of it.
141 		 */
142 		if (measured_us >= lat)
143 			measured_us -= lat / 2;
144 		else
145 			measured_us /= 2;
146 	}
147 
148 	/*
149 	 * Decay the "early hits" metric for all of the states and find the
150 	 * states matching the sleep length and the measured idle duration.
151 	 */
152 	for (i = 0; i < drv->state_count; i++) {
153 		unsigned int early_hits = cpu_data->states[i].early_hits;
154 
155 		cpu_data->states[i].early_hits -= early_hits >> DECAY_SHIFT;
156 
157 		if (drv->states[i].target_residency <= sleep_length_us) {
158 			idx_timer = i;
159 			if (drv->states[i].target_residency <= measured_us)
160 				idx_hit = i;
161 		}
162 	}
163 
164 	/*
165 	 * Update the "hits" and "misses" data for the state matching the sleep
166 	 * length.  If it matches the measured idle duration too, this is a hit,
167 	 * so increase the "hits" metric for it then.  Otherwise, this is a
168 	 * miss, so increase the "misses" metric for it.  In the latter case
169 	 * also increase the "early hits" metric for the state that actually
170 	 * matches the measured idle duration.
171 	 */
172 	if (idx_timer >= 0) {
173 		unsigned int hits = cpu_data->states[idx_timer].hits;
174 		unsigned int misses = cpu_data->states[idx_timer].misses;
175 
176 		hits -= hits >> DECAY_SHIFT;
177 		misses -= misses >> DECAY_SHIFT;
178 
179 		if (idx_timer > idx_hit) {
180 			misses += PULSE;
181 			if (idx_hit >= 0)
182 				cpu_data->states[idx_hit].early_hits += PULSE;
183 		} else {
184 			hits += PULSE;
185 		}
186 
187 		cpu_data->states[idx_timer].misses = misses;
188 		cpu_data->states[idx_timer].hits = hits;
189 	}
190 
191 	/*
192 	 * If the total time span between idle state selection and the "reflect"
193 	 * callback is greater than or equal to the sleep length determined at
194 	 * the idle state selection time, the wakeup is likely to be due to a
195 	 * timer event.
196 	 */
197 	if (cpu_data->time_span_ns >= cpu_data->sleep_length_ns)
198 		measured_us = UINT_MAX;
199 
200 	/*
201 	 * Save idle duration values corresponding to non-timer wakeups for
202 	 * pattern detection.
203 	 */
204 	cpu_data->intervals[cpu_data->interval_idx++] = measured_us;
205 	if (cpu_data->interval_idx > INTERVALS)
206 		cpu_data->interval_idx = 0;
207 }
208 
209 /**
210  * teo_find_shallower_state - Find shallower idle state matching given duration.
211  * @drv: cpuidle driver containing state data.
212  * @dev: Target CPU.
213  * @state_idx: Index of the capping idle state.
214  * @duration_us: Idle duration value to match.
215  */
216 static int teo_find_shallower_state(struct cpuidle_driver *drv,
217 				    struct cpuidle_device *dev, int state_idx,
218 				    unsigned int duration_us)
219 {
220 	int i;
221 
222 	for (i = state_idx - 1; i >= 0; i--) {
223 		if (drv->states[i].disabled || dev->states_usage[i].disable)
224 			continue;
225 
226 		state_idx = i;
227 		if (drv->states[i].target_residency <= duration_us)
228 			break;
229 	}
230 	return state_idx;
231 }
232 
233 /**
234  * teo_select - Selects the next idle state to enter.
235  * @drv: cpuidle driver containing state data.
236  * @dev: Target CPU.
237  * @stop_tick: Indication on whether or not to stop the scheduler tick.
238  */
239 static int teo_select(struct cpuidle_driver *drv, struct cpuidle_device *dev,
240 		      bool *stop_tick)
241 {
242 	struct teo_cpu *cpu_data = per_cpu_ptr(&teo_cpus, dev->cpu);
243 	int latency_req = cpuidle_governor_latency_req(dev->cpu);
244 	unsigned int duration_us, count;
245 	int max_early_idx, idx, i;
246 	ktime_t delta_tick;
247 
248 	if (cpu_data->last_state >= 0) {
249 		teo_update(drv, dev);
250 		cpu_data->last_state = -1;
251 	}
252 
253 	cpu_data->time_span_ns = local_clock();
254 
255 	cpu_data->sleep_length_ns = tick_nohz_get_sleep_length(&delta_tick);
256 	duration_us = ktime_to_us(cpu_data->sleep_length_ns);
257 
258 	count = 0;
259 	max_early_idx = -1;
260 	idx = -1;
261 
262 	for (i = 0; i < drv->state_count; i++) {
263 		struct cpuidle_state *s = &drv->states[i];
264 		struct cpuidle_state_usage *su = &dev->states_usage[i];
265 
266 		if (s->disabled || su->disable) {
267 			/*
268 			 * If the "early hits" metric of a disabled state is
269 			 * greater than the current maximum, it should be taken
270 			 * into account, because it would be a mistake to select
271 			 * a deeper state with lower "early hits" metric.  The
272 			 * index cannot be changed to point to it, however, so
273 			 * just increase the max count alone and let the index
274 			 * still point to a shallower idle state.
275 			 */
276 			if (max_early_idx >= 0 &&
277 			    count < cpu_data->states[i].early_hits)
278 				count = cpu_data->states[i].early_hits;
279 
280 			continue;
281 		}
282 
283 		if (idx < 0)
284 			idx = i; /* first enabled state */
285 
286 		if (s->target_residency > duration_us)
287 			break;
288 
289 		if (s->exit_latency > latency_req) {
290 			/*
291 			 * If we break out of the loop for latency reasons, use
292 			 * the target residency of the selected state as the
293 			 * expected idle duration to avoid stopping the tick
294 			 * as long as that target residency is low enough.
295 			 */
296 			duration_us = drv->states[idx].target_residency;
297 			goto refine;
298 		}
299 
300 		idx = i;
301 
302 		if (count < cpu_data->states[i].early_hits &&
303 		    !(tick_nohz_tick_stopped() &&
304 		      drv->states[i].target_residency < TICK_USEC)) {
305 			count = cpu_data->states[i].early_hits;
306 			max_early_idx = i;
307 		}
308 	}
309 
310 	/*
311 	 * If the "hits" metric of the idle state matching the sleep length is
312 	 * greater than its "misses" metric, that is the one to use.  Otherwise,
313 	 * it is more likely that one of the shallower states will match the
314 	 * idle duration observed after wakeup, so take the one with the maximum
315 	 * "early hits" metric, but if that cannot be determined, just use the
316 	 * state selected so far.
317 	 */
318 	if (cpu_data->states[idx].hits <= cpu_data->states[idx].misses &&
319 	    max_early_idx >= 0) {
320 		idx = max_early_idx;
321 		duration_us = drv->states[idx].target_residency;
322 	}
323 
324 refine:
325 	if (idx < 0) {
326 		idx = 0; /* No states enabled. Must use 0. */
327 	} else if (idx > 0) {
328 		u64 sum = 0;
329 
330 		count = 0;
331 
332 		/*
333 		 * Count and sum the most recent idle duration values less than
334 		 * the target residency of the state selected so far, find the
335 		 * max.
336 		 */
337 		for (i = 0; i < INTERVALS; i++) {
338 			unsigned int val = cpu_data->intervals[i];
339 
340 			if (val >= drv->states[idx].target_residency)
341 				continue;
342 
343 			count++;
344 			sum += val;
345 		}
346 
347 		/*
348 		 * Give up unless the majority of the most recent idle duration
349 		 * values are in the interesting range.
350 		 */
351 		if (count > INTERVALS / 2) {
352 			unsigned int avg_us = div64_u64(sum, count);
353 
354 			/*
355 			 * Avoid spending too much time in an idle state that
356 			 * would be too shallow.
357 			 */
358 			if (!(tick_nohz_tick_stopped() && avg_us < TICK_USEC)) {
359 				idx = teo_find_shallower_state(drv, dev, idx, avg_us);
360 				duration_us = avg_us;
361 			}
362 		}
363 	}
364 
365 	/*
366 	 * Don't stop the tick if the selected state is a polling one or if the
367 	 * expected idle duration is shorter than the tick period length.
368 	 */
369 	if (((drv->states[idx].flags & CPUIDLE_FLAG_POLLING) ||
370 	    duration_us < TICK_USEC) && !tick_nohz_tick_stopped()) {
371 		unsigned int delta_tick_us = ktime_to_us(delta_tick);
372 
373 		*stop_tick = false;
374 
375 		/*
376 		 * The tick is not going to be stopped, so if the target
377 		 * residency of the state to be returned is not within the time
378 		 * till the closest timer including the tick, try to correct
379 		 * that.
380 		 */
381 		if (idx > 0 && drv->states[idx].target_residency > delta_tick_us)
382 			idx = teo_find_shallower_state(drv, dev, idx, delta_tick_us);
383 	}
384 
385 	return idx;
386 }
387 
388 /**
389  * teo_reflect - Note that governor data for the CPU need to be updated.
390  * @dev: Target CPU.
391  * @state: Entered state.
392  */
393 static void teo_reflect(struct cpuidle_device *dev, int state)
394 {
395 	struct teo_cpu *cpu_data = per_cpu_ptr(&teo_cpus, dev->cpu);
396 
397 	cpu_data->last_state = state;
398 	/*
399 	 * If the wakeup was not "natural", but triggered by one of the safety
400 	 * nets, assume that the CPU might have been idle for the entire sleep
401 	 * length time.
402 	 */
403 	if (dev->poll_time_limit ||
404 	    (tick_nohz_idle_got_tick() && cpu_data->sleep_length_ns > TICK_NSEC)) {
405 		dev->poll_time_limit = false;
406 		cpu_data->time_span_ns = cpu_data->sleep_length_ns;
407 	} else {
408 		cpu_data->time_span_ns = local_clock() - cpu_data->time_span_ns;
409 	}
410 }
411 
412 /**
413  * teo_enable_device - Initialize the governor's data for the target CPU.
414  * @drv: cpuidle driver (not used).
415  * @dev: Target CPU.
416  */
417 static int teo_enable_device(struct cpuidle_driver *drv,
418 			     struct cpuidle_device *dev)
419 {
420 	struct teo_cpu *cpu_data = per_cpu_ptr(&teo_cpus, dev->cpu);
421 	int i;
422 
423 	memset(cpu_data, 0, sizeof(*cpu_data));
424 
425 	for (i = 0; i < INTERVALS; i++)
426 		cpu_data->intervals[i] = UINT_MAX;
427 
428 	return 0;
429 }
430 
431 static struct cpuidle_governor teo_governor = {
432 	.name =		"teo",
433 	.rating =	19,
434 	.enable =	teo_enable_device,
435 	.select =	teo_select,
436 	.reflect =	teo_reflect,
437 };
438 
439 static int __init teo_governor_init(void)
440 {
441 	return cpuidle_register_governor(&teo_governor);
442 }
443 
444 postcore_initcall(teo_governor_init);
445