xref: /openbmc/linux/drivers/cpuidle/governors/teo.c (revision 08283d30)
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  * @interval_idx: Index of the most recent saved idle interval.
100  * @intervals: Saved idle duration values.
101  */
102 struct teo_cpu {
103 	u64 time_span_ns;
104 	u64 sleep_length_ns;
105 	struct teo_idle_state states[CPUIDLE_STATE_MAX];
106 	int interval_idx;
107 	unsigned int intervals[INTERVALS];
108 };
109 
110 static DEFINE_PER_CPU(struct teo_cpu, teo_cpus);
111 
112 /**
113  * teo_update - Update CPU data after wakeup.
114  * @drv: cpuidle driver containing state data.
115  * @dev: Target CPU.
116  */
117 static void teo_update(struct cpuidle_driver *drv, struct cpuidle_device *dev)
118 {
119 	struct teo_cpu *cpu_data = per_cpu_ptr(&teo_cpus, dev->cpu);
120 	unsigned int sleep_length_us = ktime_to_us(cpu_data->sleep_length_ns);
121 	int i, idx_hit = -1, idx_timer = -1;
122 	unsigned int measured_us;
123 
124 	if (cpu_data->time_span_ns >= cpu_data->sleep_length_ns) {
125 		/*
126 		 * One of the safety nets has triggered or the wakeup was close
127 		 * enough to the closest timer event expected at the idle state
128 		 * selection time to be discarded.
129 		 */
130 		measured_us = UINT_MAX;
131 	} else {
132 		unsigned int lat;
133 
134 		lat = drv->states[dev->last_state_idx].exit_latency;
135 
136 		measured_us = ktime_to_us(cpu_data->time_span_ns);
137 		/*
138 		 * The delay between the wakeup and the first instruction
139 		 * executed by the CPU is not likely to be worst-case every
140 		 * time, so take 1/2 of the exit latency as a very rough
141 		 * approximation of the average of it.
142 		 */
143 		if (measured_us >= lat)
144 			measured_us -= lat / 2;
145 		else
146 			measured_us /= 2;
147 	}
148 
149 	/*
150 	 * Decay the "early hits" metric for all of the states and find the
151 	 * states matching the sleep length and the measured idle duration.
152 	 */
153 	for (i = 0; i < drv->state_count; i++) {
154 		unsigned int early_hits = cpu_data->states[i].early_hits;
155 
156 		cpu_data->states[i].early_hits -= early_hits >> DECAY_SHIFT;
157 
158 		if (drv->states[i].target_residency <= sleep_length_us) {
159 			idx_timer = i;
160 			if (drv->states[i].target_residency <= measured_us)
161 				idx_hit = i;
162 		}
163 	}
164 
165 	/*
166 	 * Update the "hits" and "misses" data for the state matching the sleep
167 	 * length.  If it matches the measured idle duration too, this is a hit,
168 	 * so increase the "hits" metric for it then.  Otherwise, this is a
169 	 * miss, so increase the "misses" metric for it.  In the latter case
170 	 * also increase the "early hits" metric for the state that actually
171 	 * matches the measured idle duration.
172 	 */
173 	if (idx_timer >= 0) {
174 		unsigned int hits = cpu_data->states[idx_timer].hits;
175 		unsigned int misses = cpu_data->states[idx_timer].misses;
176 
177 		hits -= hits >> DECAY_SHIFT;
178 		misses -= misses >> DECAY_SHIFT;
179 
180 		if (idx_timer > idx_hit) {
181 			misses += PULSE;
182 			if (idx_hit >= 0)
183 				cpu_data->states[idx_hit].early_hits += PULSE;
184 		} else {
185 			hits += PULSE;
186 		}
187 
188 		cpu_data->states[idx_timer].misses = misses;
189 		cpu_data->states[idx_timer].hits = hits;
190 	}
191 
192 	/*
193 	 * Save idle duration values corresponding to non-timer wakeups for
194 	 * pattern detection.
195 	 */
196 	cpu_data->intervals[cpu_data->interval_idx++] = measured_us;
197 	if (cpu_data->interval_idx > INTERVALS)
198 		cpu_data->interval_idx = 0;
199 }
200 
201 /**
202  * teo_find_shallower_state - Find shallower idle state matching given duration.
203  * @drv: cpuidle driver containing state data.
204  * @dev: Target CPU.
205  * @state_idx: Index of the capping idle state.
206  * @duration_us: Idle duration value to match.
207  */
208 static int teo_find_shallower_state(struct cpuidle_driver *drv,
209 				    struct cpuidle_device *dev, int state_idx,
210 				    unsigned int duration_us)
211 {
212 	int i;
213 
214 	for (i = state_idx - 1; i >= 0; i--) {
215 		if (drv->states[i].disabled || dev->states_usage[i].disable)
216 			continue;
217 
218 		state_idx = i;
219 		if (drv->states[i].target_residency <= duration_us)
220 			break;
221 	}
222 	return state_idx;
223 }
224 
225 /**
226  * teo_select - Selects the next idle state to enter.
227  * @drv: cpuidle driver containing state data.
228  * @dev: Target CPU.
229  * @stop_tick: Indication on whether or not to stop the scheduler tick.
230  */
231 static int teo_select(struct cpuidle_driver *drv, struct cpuidle_device *dev,
232 		      bool *stop_tick)
233 {
234 	struct teo_cpu *cpu_data = per_cpu_ptr(&teo_cpus, dev->cpu);
235 	int latency_req = cpuidle_governor_latency_req(dev->cpu);
236 	unsigned int duration_us, count;
237 	int max_early_idx, constraint_idx, idx, i;
238 	ktime_t delta_tick;
239 
240 	if (dev->last_state_idx >= 0) {
241 		teo_update(drv, dev);
242 		dev->last_state_idx = -1;
243 	}
244 
245 	cpu_data->time_span_ns = local_clock();
246 
247 	cpu_data->sleep_length_ns = tick_nohz_get_sleep_length(&delta_tick);
248 	duration_us = ktime_to_us(cpu_data->sleep_length_ns);
249 
250 	count = 0;
251 	max_early_idx = -1;
252 	constraint_idx = drv->state_count;
253 	idx = -1;
254 
255 	for (i = 0; i < drv->state_count; i++) {
256 		struct cpuidle_state *s = &drv->states[i];
257 		struct cpuidle_state_usage *su = &dev->states_usage[i];
258 
259 		if (s->disabled || su->disable) {
260 			/*
261 			 * If the "early hits" metric of a disabled state is
262 			 * greater than the current maximum, it should be taken
263 			 * into account, because it would be a mistake to select
264 			 * a deeper state with lower "early hits" metric.  The
265 			 * index cannot be changed to point to it, however, so
266 			 * just increase the max count alone and let the index
267 			 * still point to a shallower idle state.
268 			 */
269 			if (max_early_idx >= 0 &&
270 			    count < cpu_data->states[i].early_hits)
271 				count = cpu_data->states[i].early_hits;
272 
273 			continue;
274 		}
275 
276 		if (idx < 0)
277 			idx = i; /* first enabled state */
278 
279 		if (s->target_residency > duration_us)
280 			break;
281 
282 		if (s->exit_latency > latency_req && constraint_idx > i)
283 			constraint_idx = i;
284 
285 		idx = i;
286 
287 		if (count < cpu_data->states[i].early_hits &&
288 		    !(tick_nohz_tick_stopped() &&
289 		      drv->states[i].target_residency < TICK_USEC)) {
290 			count = cpu_data->states[i].early_hits;
291 			max_early_idx = i;
292 		}
293 	}
294 
295 	/*
296 	 * If the "hits" metric of the idle state matching the sleep length is
297 	 * greater than its "misses" metric, that is the one to use.  Otherwise,
298 	 * it is more likely that one of the shallower states will match the
299 	 * idle duration observed after wakeup, so take the one with the maximum
300 	 * "early hits" metric, but if that cannot be determined, just use the
301 	 * state selected so far.
302 	 */
303 	if (cpu_data->states[idx].hits <= cpu_data->states[idx].misses &&
304 	    max_early_idx >= 0) {
305 		idx = max_early_idx;
306 		duration_us = drv->states[idx].target_residency;
307 	}
308 
309 	/*
310 	 * If there is a latency constraint, it may be necessary to use a
311 	 * shallower idle state than the one selected so far.
312 	 */
313 	if (constraint_idx < idx)
314 		idx = constraint_idx;
315 
316 	if (idx < 0) {
317 		idx = 0; /* No states enabled. Must use 0. */
318 	} else if (idx > 0) {
319 		u64 sum = 0;
320 
321 		count = 0;
322 
323 		/*
324 		 * Count and sum the most recent idle duration values less than
325 		 * the current expected idle duration value.
326 		 */
327 		for (i = 0; i < INTERVALS; i++) {
328 			unsigned int val = cpu_data->intervals[i];
329 
330 			if (val >= duration_us)
331 				continue;
332 
333 			count++;
334 			sum += val;
335 		}
336 
337 		/*
338 		 * Give up unless the majority of the most recent idle duration
339 		 * values are in the interesting range.
340 		 */
341 		if (count > INTERVALS / 2) {
342 			unsigned int avg_us = div64_u64(sum, count);
343 
344 			/*
345 			 * Avoid spending too much time in an idle state that
346 			 * would be too shallow.
347 			 */
348 			if (!(tick_nohz_tick_stopped() && avg_us < TICK_USEC)) {
349 				duration_us = avg_us;
350 				if (drv->states[idx].target_residency > avg_us)
351 					idx = teo_find_shallower_state(drv, dev,
352 								       idx, avg_us);
353 			}
354 		}
355 	}
356 
357 	/*
358 	 * Don't stop the tick if the selected state is a polling one or if the
359 	 * expected idle duration is shorter than the tick period length.
360 	 */
361 	if (((drv->states[idx].flags & CPUIDLE_FLAG_POLLING) ||
362 	    duration_us < TICK_USEC) && !tick_nohz_tick_stopped()) {
363 		unsigned int delta_tick_us = ktime_to_us(delta_tick);
364 
365 		*stop_tick = false;
366 
367 		/*
368 		 * The tick is not going to be stopped, so if the target
369 		 * residency of the state to be returned is not within the time
370 		 * till the closest timer including the tick, try to correct
371 		 * that.
372 		 */
373 		if (idx > 0 && drv->states[idx].target_residency > delta_tick_us)
374 			idx = teo_find_shallower_state(drv, dev, idx, delta_tick_us);
375 	}
376 
377 	return idx;
378 }
379 
380 /**
381  * teo_reflect - Note that governor data for the CPU need to be updated.
382  * @dev: Target CPU.
383  * @state: Entered state.
384  */
385 static void teo_reflect(struct cpuidle_device *dev, int state)
386 {
387 	struct teo_cpu *cpu_data = per_cpu_ptr(&teo_cpus, dev->cpu);
388 
389 	dev->last_state_idx = state;
390 	/*
391 	 * If the wakeup was not "natural", but triggered by one of the safety
392 	 * nets, assume that the CPU might have been idle for the entire sleep
393 	 * length time.
394 	 */
395 	if (dev->poll_time_limit ||
396 	    (tick_nohz_idle_got_tick() && cpu_data->sleep_length_ns > TICK_NSEC)) {
397 		dev->poll_time_limit = false;
398 		cpu_data->time_span_ns = cpu_data->sleep_length_ns;
399 	} else {
400 		cpu_data->time_span_ns = local_clock() - cpu_data->time_span_ns;
401 	}
402 }
403 
404 /**
405  * teo_enable_device - Initialize the governor's data for the target CPU.
406  * @drv: cpuidle driver (not used).
407  * @dev: Target CPU.
408  */
409 static int teo_enable_device(struct cpuidle_driver *drv,
410 			     struct cpuidle_device *dev)
411 {
412 	struct teo_cpu *cpu_data = per_cpu_ptr(&teo_cpus, dev->cpu);
413 	int i;
414 
415 	memset(cpu_data, 0, sizeof(*cpu_data));
416 
417 	for (i = 0; i < INTERVALS; i++)
418 		cpu_data->intervals[i] = UINT_MAX;
419 
420 	return 0;
421 }
422 
423 static struct cpuidle_governor teo_governor = {
424 	.name =		"teo",
425 	.rating =	19,
426 	.enable =	teo_enable_device,
427 	.select =	teo_select,
428 	.reflect =	teo_reflect,
429 };
430 
431 static int __init teo_governor_init(void)
432 {
433 	return cpuidle_register_governor(&teo_governor);
434 }
435 
436 postcore_initcall(teo_governor_init);
437