xref: /linux/kernel/sched/pelt.c (revision f86fd32d)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Per Entity Load Tracking
4  *
5  *  Copyright (C) 2007 Red Hat, Inc., Ingo Molnar <mingo@redhat.com>
6  *
7  *  Interactivity improvements by Mike Galbraith
8  *  (C) 2007 Mike Galbraith <efault@gmx.de>
9  *
10  *  Various enhancements by Dmitry Adamushko.
11  *  (C) 2007 Dmitry Adamushko <dmitry.adamushko@gmail.com>
12  *
13  *  Group scheduling enhancements by Srivatsa Vaddagiri
14  *  Copyright IBM Corporation, 2007
15  *  Author: Srivatsa Vaddagiri <vatsa@linux.vnet.ibm.com>
16  *
17  *  Scaled math optimizations by Thomas Gleixner
18  *  Copyright (C) 2007, Thomas Gleixner <tglx@linutronix.de>
19  *
20  *  Adaptive scheduling granularity, math enhancements by Peter Zijlstra
21  *  Copyright (C) 2007 Red Hat, Inc., Peter Zijlstra
22  *
23  *  Move PELT related code from fair.c into this pelt.c file
24  *  Author: Vincent Guittot <vincent.guittot@linaro.org>
25  */
26 
27 #include <linux/sched.h>
28 #include "sched.h"
29 #include "pelt.h"
30 
31 #include <trace/events/sched.h>
32 
33 /*
34  * Approximate:
35  *   val * y^n,    where y^32 ~= 0.5 (~1 scheduling period)
36  */
37 static u64 decay_load(u64 val, u64 n)
38 {
39 	unsigned int local_n;
40 
41 	if (unlikely(n > LOAD_AVG_PERIOD * 63))
42 		return 0;
43 
44 	/* after bounds checking we can collapse to 32-bit */
45 	local_n = n;
46 
47 	/*
48 	 * As y^PERIOD = 1/2, we can combine
49 	 *    y^n = 1/2^(n/PERIOD) * y^(n%PERIOD)
50 	 * With a look-up table which covers y^n (n<PERIOD)
51 	 *
52 	 * To achieve constant time decay_load.
53 	 */
54 	if (unlikely(local_n >= LOAD_AVG_PERIOD)) {
55 		val >>= local_n / LOAD_AVG_PERIOD;
56 		local_n %= LOAD_AVG_PERIOD;
57 	}
58 
59 	val = mul_u64_u32_shr(val, runnable_avg_yN_inv[local_n], 32);
60 	return val;
61 }
62 
63 static u32 __accumulate_pelt_segments(u64 periods, u32 d1, u32 d3)
64 {
65 	u32 c1, c2, c3 = d3; /* y^0 == 1 */
66 
67 	/*
68 	 * c1 = d1 y^p
69 	 */
70 	c1 = decay_load((u64)d1, periods);
71 
72 	/*
73 	 *            p-1
74 	 * c2 = 1024 \Sum y^n
75 	 *            n=1
76 	 *
77 	 *              inf        inf
78 	 *    = 1024 ( \Sum y^n - \Sum y^n - y^0 )
79 	 *              n=0        n=p
80 	 */
81 	c2 = LOAD_AVG_MAX - decay_load(LOAD_AVG_MAX, periods) - 1024;
82 
83 	return c1 + c2 + c3;
84 }
85 
86 #define cap_scale(v, s) ((v)*(s) >> SCHED_CAPACITY_SHIFT)
87 
88 /*
89  * Accumulate the three separate parts of the sum; d1 the remainder
90  * of the last (incomplete) period, d2 the span of full periods and d3
91  * the remainder of the (incomplete) current period.
92  *
93  *           d1          d2           d3
94  *           ^           ^            ^
95  *           |           |            |
96  *         |<->|<----------------->|<--->|
97  * ... |---x---|------| ... |------|-----x (now)
98  *
99  *                           p-1
100  * u' = (u + d1) y^p + 1024 \Sum y^n + d3 y^0
101  *                           n=1
102  *
103  *    = u y^p +					(Step 1)
104  *
105  *                     p-1
106  *      d1 y^p + 1024 \Sum y^n + d3 y^0		(Step 2)
107  *                     n=1
108  */
109 static __always_inline u32
110 accumulate_sum(u64 delta, struct sched_avg *sa,
111 	       unsigned long load, unsigned long runnable, int running)
112 {
113 	u32 contrib = (u32)delta; /* p == 0 -> delta < 1024 */
114 	u64 periods;
115 
116 	delta += sa->period_contrib;
117 	periods = delta / 1024; /* A period is 1024us (~1ms) */
118 
119 	/*
120 	 * Step 1: decay old *_sum if we crossed period boundaries.
121 	 */
122 	if (periods) {
123 		sa->load_sum = decay_load(sa->load_sum, periods);
124 		sa->runnable_load_sum =
125 			decay_load(sa->runnable_load_sum, periods);
126 		sa->util_sum = decay_load((u64)(sa->util_sum), periods);
127 
128 		/*
129 		 * Step 2
130 		 */
131 		delta %= 1024;
132 		if (load) {
133 			/*
134 			 * This relies on the:
135 			 *
136 			 * if (!load)
137 			 *	runnable = running = 0;
138 			 *
139 			 * clause from ___update_load_sum(); this results in
140 			 * the below usage of @contrib to dissapear entirely,
141 			 * so no point in calculating it.
142 			 */
143 			contrib = __accumulate_pelt_segments(periods,
144 					1024 - sa->period_contrib, delta);
145 		}
146 	}
147 	sa->period_contrib = delta;
148 
149 	if (load)
150 		sa->load_sum += load * contrib;
151 	if (runnable)
152 		sa->runnable_load_sum += runnable * contrib;
153 	if (running)
154 		sa->util_sum += contrib << SCHED_CAPACITY_SHIFT;
155 
156 	return periods;
157 }
158 
159 /*
160  * We can represent the historical contribution to runnable average as the
161  * coefficients of a geometric series.  To do this we sub-divide our runnable
162  * history into segments of approximately 1ms (1024us); label the segment that
163  * occurred N-ms ago p_N, with p_0 corresponding to the current period, e.g.
164  *
165  * [<- 1024us ->|<- 1024us ->|<- 1024us ->| ...
166  *      p0            p1           p2
167  *     (now)       (~1ms ago)  (~2ms ago)
168  *
169  * Let u_i denote the fraction of p_i that the entity was runnable.
170  *
171  * We then designate the fractions u_i as our co-efficients, yielding the
172  * following representation of historical load:
173  *   u_0 + u_1*y + u_2*y^2 + u_3*y^3 + ...
174  *
175  * We choose y based on the with of a reasonably scheduling period, fixing:
176  *   y^32 = 0.5
177  *
178  * This means that the contribution to load ~32ms ago (u_32) will be weighted
179  * approximately half as much as the contribution to load within the last ms
180  * (u_0).
181  *
182  * When a period "rolls over" and we have new u_0`, multiplying the previous
183  * sum again by y is sufficient to update:
184  *   load_avg = u_0` + y*(u_0 + u_1*y + u_2*y^2 + ... )
185  *            = u_0 + u_1*y + u_2*y^2 + ... [re-labeling u_i --> u_{i+1}]
186  */
187 static __always_inline int
188 ___update_load_sum(u64 now, struct sched_avg *sa,
189 		  unsigned long load, unsigned long runnable, int running)
190 {
191 	u64 delta;
192 
193 	delta = now - sa->last_update_time;
194 	/*
195 	 * This should only happen when time goes backwards, which it
196 	 * unfortunately does during sched clock init when we swap over to TSC.
197 	 */
198 	if ((s64)delta < 0) {
199 		sa->last_update_time = now;
200 		return 0;
201 	}
202 
203 	/*
204 	 * Use 1024ns as the unit of measurement since it's a reasonable
205 	 * approximation of 1us and fast to compute.
206 	 */
207 	delta >>= 10;
208 	if (!delta)
209 		return 0;
210 
211 	sa->last_update_time += delta << 10;
212 
213 	/*
214 	 * running is a subset of runnable (weight) so running can't be set if
215 	 * runnable is clear. But there are some corner cases where the current
216 	 * se has been already dequeued but cfs_rq->curr still points to it.
217 	 * This means that weight will be 0 but not running for a sched_entity
218 	 * but also for a cfs_rq if the latter becomes idle. As an example,
219 	 * this happens during idle_balance() which calls
220 	 * update_blocked_averages().
221 	 *
222 	 * Also see the comment in accumulate_sum().
223 	 */
224 	if (!load)
225 		runnable = running = 0;
226 
227 	/*
228 	 * Now we know we crossed measurement unit boundaries. The *_avg
229 	 * accrues by two steps:
230 	 *
231 	 * Step 1: accumulate *_sum since last_update_time. If we haven't
232 	 * crossed period boundaries, finish.
233 	 */
234 	if (!accumulate_sum(delta, sa, load, runnable, running))
235 		return 0;
236 
237 	return 1;
238 }
239 
240 static __always_inline void
241 ___update_load_avg(struct sched_avg *sa, unsigned long load, unsigned long runnable)
242 {
243 	u32 divider = LOAD_AVG_MAX - 1024 + sa->period_contrib;
244 
245 	/*
246 	 * Step 2: update *_avg.
247 	 */
248 	sa->load_avg = div_u64(load * sa->load_sum, divider);
249 	sa->runnable_load_avg =	div_u64(runnable * sa->runnable_load_sum, divider);
250 	WRITE_ONCE(sa->util_avg, sa->util_sum / divider);
251 }
252 
253 /*
254  * sched_entity:
255  *
256  *   task:
257  *     se_runnable() == se_weight()
258  *
259  *   group: [ see update_cfs_group() ]
260  *     se_weight()   = tg->weight * grq->load_avg / tg->load_avg
261  *     se_runnable() = se_weight(se) * grq->runnable_load_avg / grq->load_avg
262  *
263  *   load_sum := runnable_sum
264  *   load_avg = se_weight(se) * runnable_avg
265  *
266  *   runnable_load_sum := runnable_sum
267  *   runnable_load_avg = se_runnable(se) * runnable_avg
268  *
269  * XXX collapse load_sum and runnable_load_sum
270  *
271  * cfq_rq:
272  *
273  *   load_sum = \Sum se_weight(se) * se->avg.load_sum
274  *   load_avg = \Sum se->avg.load_avg
275  *
276  *   runnable_load_sum = \Sum se_runnable(se) * se->avg.runnable_load_sum
277  *   runnable_load_avg = \Sum se->avg.runable_load_avg
278  */
279 
280 int __update_load_avg_blocked_se(u64 now, struct sched_entity *se)
281 {
282 	if (___update_load_sum(now, &se->avg, 0, 0, 0)) {
283 		___update_load_avg(&se->avg, se_weight(se), se_runnable(se));
284 		trace_pelt_se_tp(se);
285 		return 1;
286 	}
287 
288 	return 0;
289 }
290 
291 int __update_load_avg_se(u64 now, struct cfs_rq *cfs_rq, struct sched_entity *se)
292 {
293 	if (___update_load_sum(now, &se->avg, !!se->on_rq, !!se->on_rq,
294 				cfs_rq->curr == se)) {
295 
296 		___update_load_avg(&se->avg, se_weight(se), se_runnable(se));
297 		cfs_se_util_change(&se->avg);
298 		trace_pelt_se_tp(se);
299 		return 1;
300 	}
301 
302 	return 0;
303 }
304 
305 int __update_load_avg_cfs_rq(u64 now, struct cfs_rq *cfs_rq)
306 {
307 	if (___update_load_sum(now, &cfs_rq->avg,
308 				scale_load_down(cfs_rq->load.weight),
309 				scale_load_down(cfs_rq->runnable_weight),
310 				cfs_rq->curr != NULL)) {
311 
312 		___update_load_avg(&cfs_rq->avg, 1, 1);
313 		trace_pelt_cfs_tp(cfs_rq);
314 		return 1;
315 	}
316 
317 	return 0;
318 }
319 
320 /*
321  * rt_rq:
322  *
323  *   util_sum = \Sum se->avg.util_sum but se->avg.util_sum is not tracked
324  *   util_sum = cpu_scale * load_sum
325  *   runnable_load_sum = load_sum
326  *
327  *   load_avg and runnable_load_avg are not supported and meaningless.
328  *
329  */
330 
331 int update_rt_rq_load_avg(u64 now, struct rq *rq, int running)
332 {
333 	if (___update_load_sum(now, &rq->avg_rt,
334 				running,
335 				running,
336 				running)) {
337 
338 		___update_load_avg(&rq->avg_rt, 1, 1);
339 		trace_pelt_rt_tp(rq);
340 		return 1;
341 	}
342 
343 	return 0;
344 }
345 
346 /*
347  * dl_rq:
348  *
349  *   util_sum = \Sum se->avg.util_sum but se->avg.util_sum is not tracked
350  *   util_sum = cpu_scale * load_sum
351  *   runnable_load_sum = load_sum
352  *
353  */
354 
355 int update_dl_rq_load_avg(u64 now, struct rq *rq, int running)
356 {
357 	if (___update_load_sum(now, &rq->avg_dl,
358 				running,
359 				running,
360 				running)) {
361 
362 		___update_load_avg(&rq->avg_dl, 1, 1);
363 		trace_pelt_dl_tp(rq);
364 		return 1;
365 	}
366 
367 	return 0;
368 }
369 
370 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
371 /*
372  * irq:
373  *
374  *   util_sum = \Sum se->avg.util_sum but se->avg.util_sum is not tracked
375  *   util_sum = cpu_scale * load_sum
376  *   runnable_load_sum = load_sum
377  *
378  */
379 
380 int update_irq_load_avg(struct rq *rq, u64 running)
381 {
382 	int ret = 0;
383 
384 	/*
385 	 * We can't use clock_pelt because irq time is not accounted in
386 	 * clock_task. Instead we directly scale the running time to
387 	 * reflect the real amount of computation
388 	 */
389 	running = cap_scale(running, arch_scale_freq_capacity(cpu_of(rq)));
390 	running = cap_scale(running, arch_scale_cpu_capacity(cpu_of(rq)));
391 
392 	/*
393 	 * We know the time that has been used by interrupt since last update
394 	 * but we don't when. Let be pessimistic and assume that interrupt has
395 	 * happened just before the update. This is not so far from reality
396 	 * because interrupt will most probably wake up task and trig an update
397 	 * of rq clock during which the metric is updated.
398 	 * We start to decay with normal context time and then we add the
399 	 * interrupt context time.
400 	 * We can safely remove running from rq->clock because
401 	 * rq->clock += delta with delta >= running
402 	 */
403 	ret = ___update_load_sum(rq->clock - running, &rq->avg_irq,
404 				0,
405 				0,
406 				0);
407 	ret += ___update_load_sum(rq->clock, &rq->avg_irq,
408 				1,
409 				1,
410 				1);
411 
412 	if (ret) {
413 		___update_load_avg(&rq->avg_irq, 1, 1);
414 		trace_pelt_irq_tp(rq);
415 	}
416 
417 	return ret;
418 }
419 #endif
420