xref: /freebsd/sys/kern/kern_switch.c (revision 6829dae12bb055451fa467da4589c43bd03b1e64)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 2001 Jake Burkholder <jake@FreeBSD.org>
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  */
28 
29 
30 #include <sys/cdefs.h>
31 __FBSDID("$FreeBSD$");
32 
33 #include "opt_sched.h"
34 
35 #include <sys/param.h>
36 #include <sys/systm.h>
37 #include <sys/kdb.h>
38 #include <sys/kernel.h>
39 #include <sys/ktr.h>
40 #include <sys/lock.h>
41 #include <sys/mutex.h>
42 #include <sys/proc.h>
43 #include <sys/queue.h>
44 #include <sys/sched.h>
45 #include <sys/smp.h>
46 #include <sys/sysctl.h>
47 
48 #include <machine/cpu.h>
49 
50 /* Uncomment this to enable logging of critical_enter/exit. */
51 #if 0
52 #define	KTR_CRITICAL	KTR_SCHED
53 #else
54 #define	KTR_CRITICAL	0
55 #endif
56 
57 #ifdef FULL_PREEMPTION
58 #ifndef PREEMPTION
59 #error "The FULL_PREEMPTION option requires the PREEMPTION option"
60 #endif
61 #endif
62 
63 CTASSERT((RQB_BPW * RQB_LEN) == RQ_NQS);
64 
65 /*
66  * kern.sched.preemption allows user space to determine if preemption support
67  * is compiled in or not.  It is not currently a boot or runtime flag that
68  * can be changed.
69  */
70 #ifdef PREEMPTION
71 static int kern_sched_preemption = 1;
72 #else
73 static int kern_sched_preemption = 0;
74 #endif
75 SYSCTL_INT(_kern_sched, OID_AUTO, preemption, CTLFLAG_RD,
76     &kern_sched_preemption, 0, "Kernel preemption enabled");
77 
78 /*
79  * Support for scheduler stats exported via kern.sched.stats.  All stats may
80  * be reset with kern.sched.stats.reset = 1.  Stats may be defined elsewhere
81  * with SCHED_STAT_DEFINE().
82  */
83 #ifdef SCHED_STATS
84 SYSCTL_NODE(_kern_sched, OID_AUTO, stats, CTLFLAG_RW, 0, "switch stats");
85 
86 /* Switch reasons from mi_switch(). */
87 DPCPU_DEFINE(long, sched_switch_stats[SWT_COUNT]);
88 SCHED_STAT_DEFINE_VAR(uncategorized,
89     &DPCPU_NAME(sched_switch_stats[SWT_NONE]), "");
90 SCHED_STAT_DEFINE_VAR(preempt,
91     &DPCPU_NAME(sched_switch_stats[SWT_PREEMPT]), "");
92 SCHED_STAT_DEFINE_VAR(owepreempt,
93     &DPCPU_NAME(sched_switch_stats[SWT_OWEPREEMPT]), "");
94 SCHED_STAT_DEFINE_VAR(turnstile,
95     &DPCPU_NAME(sched_switch_stats[SWT_TURNSTILE]), "");
96 SCHED_STAT_DEFINE_VAR(sleepq,
97     &DPCPU_NAME(sched_switch_stats[SWT_SLEEPQ]), "");
98 SCHED_STAT_DEFINE_VAR(sleepqtimo,
99     &DPCPU_NAME(sched_switch_stats[SWT_SLEEPQTIMO]), "");
100 SCHED_STAT_DEFINE_VAR(relinquish,
101     &DPCPU_NAME(sched_switch_stats[SWT_RELINQUISH]), "");
102 SCHED_STAT_DEFINE_VAR(needresched,
103     &DPCPU_NAME(sched_switch_stats[SWT_NEEDRESCHED]), "");
104 SCHED_STAT_DEFINE_VAR(idle,
105     &DPCPU_NAME(sched_switch_stats[SWT_IDLE]), "");
106 SCHED_STAT_DEFINE_VAR(iwait,
107     &DPCPU_NAME(sched_switch_stats[SWT_IWAIT]), "");
108 SCHED_STAT_DEFINE_VAR(suspend,
109     &DPCPU_NAME(sched_switch_stats[SWT_SUSPEND]), "");
110 SCHED_STAT_DEFINE_VAR(remotepreempt,
111     &DPCPU_NAME(sched_switch_stats[SWT_REMOTEPREEMPT]), "");
112 SCHED_STAT_DEFINE_VAR(remotewakeidle,
113     &DPCPU_NAME(sched_switch_stats[SWT_REMOTEWAKEIDLE]), "");
114 
115 static int
116 sysctl_stats_reset(SYSCTL_HANDLER_ARGS)
117 {
118 	struct sysctl_oid *p;
119 	uintptr_t counter;
120         int error;
121 	int val;
122 	int i;
123 
124         val = 0;
125         error = sysctl_handle_int(oidp, &val, 0, req);
126         if (error != 0 || req->newptr == NULL)
127                 return (error);
128         if (val == 0)
129                 return (0);
130 	/*
131 	 * Traverse the list of children of _kern_sched_stats and reset each
132 	 * to 0.  Skip the reset entry.
133 	 */
134 	SLIST_FOREACH(p, oidp->oid_parent, oid_link) {
135 		if (p == oidp || p->oid_arg1 == NULL)
136 			continue;
137 		counter = (uintptr_t)p->oid_arg1;
138 		CPU_FOREACH(i) {
139 			*(long *)(dpcpu_off[i] + counter) = 0;
140 		}
141 	}
142 	return (0);
143 }
144 
145 SYSCTL_PROC(_kern_sched_stats, OID_AUTO, reset, CTLTYPE_INT | CTLFLAG_WR, NULL,
146     0, sysctl_stats_reset, "I", "Reset scheduler statistics");
147 #endif
148 
149 /************************************************************************
150  * Functions that manipulate runnability from a thread perspective.	*
151  ************************************************************************/
152 /*
153  * Select the thread that will be run next.
154  */
155 
156 static __noinline struct thread *
157 choosethread_panic(struct thread *td)
158 {
159 
160 	/*
161 	 * If we are in panic, only allow system threads,
162 	 * plus the one we are running in, to be run.
163 	 */
164 retry:
165 	if (((td->td_proc->p_flag & P_SYSTEM) == 0 &&
166 	    (td->td_flags & TDF_INPANIC) == 0)) {
167 		/* note that it is no longer on the run queue */
168 		TD_SET_CAN_RUN(td);
169 		td = sched_choose();
170 		goto retry;
171 	}
172 
173 	TD_SET_RUNNING(td);
174 	return (td);
175 }
176 
177 struct thread *
178 choosethread(void)
179 {
180 	struct thread *td;
181 
182 	td = sched_choose();
183 
184 	if (__predict_false(panicstr != NULL))
185 		return (choosethread_panic(td));
186 
187 	TD_SET_RUNNING(td);
188 	return (td);
189 }
190 
191 /*
192  * Kernel thread preemption implementation.  Critical sections mark
193  * regions of code in which preemptions are not allowed.
194  *
195  * It might seem a good idea to inline critical_enter() but, in order
196  * to prevent instructions reordering by the compiler, a __compiler_membar()
197  * would have to be used here (the same as sched_pin()).  The performance
198  * penalty imposed by the membar could, then, produce slower code than
199  * the function call itself, for most cases.
200  */
201 void
202 critical_enter_KBI(void)
203 {
204 #ifdef KTR
205 	struct thread *td = curthread;
206 #endif
207 	critical_enter();
208 	CTR4(KTR_CRITICAL, "critical_enter by thread %p (%ld, %s) to %d", td,
209 	    (long)td->td_proc->p_pid, td->td_name, td->td_critnest);
210 }
211 
212 void __noinline
213 critical_exit_preempt(void)
214 {
215 	struct thread *td;
216 	int flags;
217 
218 	/*
219 	 * If td_critnest is 0, it is possible that we are going to get
220 	 * preempted again before reaching the code below. This happens
221 	 * rarely and is harmless. However, this means td_owepreempt may
222 	 * now be unset.
223 	 */
224 	td = curthread;
225 	if (td->td_critnest != 0)
226 		return;
227 	if (kdb_active)
228 		return;
229 
230 	/*
231 	 * Microoptimization: we committed to switch,
232 	 * disable preemption in interrupt handlers
233 	 * while spinning for the thread lock.
234 	 */
235 	td->td_critnest = 1;
236 	thread_lock(td);
237 	td->td_critnest--;
238 	flags = SW_INVOL | SW_PREEMPT;
239 	if (TD_IS_IDLETHREAD(td))
240 		flags |= SWT_IDLE;
241 	else
242 		flags |= SWT_OWEPREEMPT;
243 	mi_switch(flags, NULL);
244 	thread_unlock(td);
245 }
246 
247 void
248 critical_exit_KBI(void)
249 {
250 #ifdef KTR
251 	struct thread *td = curthread;
252 #endif
253 	critical_exit();
254 	CTR4(KTR_CRITICAL, "critical_exit by thread %p (%ld, %s) to %d", td,
255 	    (long)td->td_proc->p_pid, td->td_name, td->td_critnest);
256 }
257 
258 /************************************************************************
259  * SYSTEM RUN QUEUE manipulations and tests				*
260  ************************************************************************/
261 /*
262  * Initialize a run structure.
263  */
264 void
265 runq_init(struct runq *rq)
266 {
267 	int i;
268 
269 	bzero(rq, sizeof *rq);
270 	for (i = 0; i < RQ_NQS; i++)
271 		TAILQ_INIT(&rq->rq_queues[i]);
272 }
273 
274 /*
275  * Clear the status bit of the queue corresponding to priority level pri,
276  * indicating that it is empty.
277  */
278 static __inline void
279 runq_clrbit(struct runq *rq, int pri)
280 {
281 	struct rqbits *rqb;
282 
283 	rqb = &rq->rq_status;
284 	CTR4(KTR_RUNQ, "runq_clrbit: bits=%#x %#x bit=%#x word=%d",
285 	    rqb->rqb_bits[RQB_WORD(pri)],
286 	    rqb->rqb_bits[RQB_WORD(pri)] & ~RQB_BIT(pri),
287 	    RQB_BIT(pri), RQB_WORD(pri));
288 	rqb->rqb_bits[RQB_WORD(pri)] &= ~RQB_BIT(pri);
289 }
290 
291 /*
292  * Find the index of the first non-empty run queue.  This is done by
293  * scanning the status bits, a set bit indicates a non-empty queue.
294  */
295 static __inline int
296 runq_findbit(struct runq *rq)
297 {
298 	struct rqbits *rqb;
299 	int pri;
300 	int i;
301 
302 	rqb = &rq->rq_status;
303 	for (i = 0; i < RQB_LEN; i++)
304 		if (rqb->rqb_bits[i]) {
305 			pri = RQB_FFS(rqb->rqb_bits[i]) + (i << RQB_L2BPW);
306 			CTR3(KTR_RUNQ, "runq_findbit: bits=%#x i=%d pri=%d",
307 			    rqb->rqb_bits[i], i, pri);
308 			return (pri);
309 		}
310 
311 	return (-1);
312 }
313 
314 static __inline int
315 runq_findbit_from(struct runq *rq, u_char pri)
316 {
317 	struct rqbits *rqb;
318 	rqb_word_t mask;
319 	int i;
320 
321 	/*
322 	 * Set the mask for the first word so we ignore priorities before 'pri'.
323 	 */
324 	mask = (rqb_word_t)-1 << (pri & (RQB_BPW - 1));
325 	rqb = &rq->rq_status;
326 again:
327 	for (i = RQB_WORD(pri); i < RQB_LEN; mask = -1, i++) {
328 		mask = rqb->rqb_bits[i] & mask;
329 		if (mask == 0)
330 			continue;
331 		pri = RQB_FFS(mask) + (i << RQB_L2BPW);
332 		CTR3(KTR_RUNQ, "runq_findbit_from: bits=%#x i=%d pri=%d",
333 		    mask, i, pri);
334 		return (pri);
335 	}
336 	if (pri == 0)
337 		return (-1);
338 	/*
339 	 * Wrap back around to the beginning of the list just once so we
340 	 * scan the whole thing.
341 	 */
342 	pri = 0;
343 	goto again;
344 }
345 
346 /*
347  * Set the status bit of the queue corresponding to priority level pri,
348  * indicating that it is non-empty.
349  */
350 static __inline void
351 runq_setbit(struct runq *rq, int pri)
352 {
353 	struct rqbits *rqb;
354 
355 	rqb = &rq->rq_status;
356 	CTR4(KTR_RUNQ, "runq_setbit: bits=%#x %#x bit=%#x word=%d",
357 	    rqb->rqb_bits[RQB_WORD(pri)],
358 	    rqb->rqb_bits[RQB_WORD(pri)] | RQB_BIT(pri),
359 	    RQB_BIT(pri), RQB_WORD(pri));
360 	rqb->rqb_bits[RQB_WORD(pri)] |= RQB_BIT(pri);
361 }
362 
363 /*
364  * Add the thread to the queue specified by its priority, and set the
365  * corresponding status bit.
366  */
367 void
368 runq_add(struct runq *rq, struct thread *td, int flags)
369 {
370 	struct rqhead *rqh;
371 	int pri;
372 
373 	pri = td->td_priority / RQ_PPQ;
374 	td->td_rqindex = pri;
375 	runq_setbit(rq, pri);
376 	rqh = &rq->rq_queues[pri];
377 	CTR4(KTR_RUNQ, "runq_add: td=%p pri=%d %d rqh=%p",
378 	    td, td->td_priority, pri, rqh);
379 	if (flags & SRQ_PREEMPTED) {
380 		TAILQ_INSERT_HEAD(rqh, td, td_runq);
381 	} else {
382 		TAILQ_INSERT_TAIL(rqh, td, td_runq);
383 	}
384 }
385 
386 void
387 runq_add_pri(struct runq *rq, struct thread *td, u_char pri, int flags)
388 {
389 	struct rqhead *rqh;
390 
391 	KASSERT(pri < RQ_NQS, ("runq_add_pri: %d out of range", pri));
392 	td->td_rqindex = pri;
393 	runq_setbit(rq, pri);
394 	rqh = &rq->rq_queues[pri];
395 	CTR4(KTR_RUNQ, "runq_add_pri: td=%p pri=%d idx=%d rqh=%p",
396 	    td, td->td_priority, pri, rqh);
397 	if (flags & SRQ_PREEMPTED) {
398 		TAILQ_INSERT_HEAD(rqh, td, td_runq);
399 	} else {
400 		TAILQ_INSERT_TAIL(rqh, td, td_runq);
401 	}
402 }
403 /*
404  * Return true if there are runnable processes of any priority on the run
405  * queue, false otherwise.  Has no side effects, does not modify the run
406  * queue structure.
407  */
408 int
409 runq_check(struct runq *rq)
410 {
411 	struct rqbits *rqb;
412 	int i;
413 
414 	rqb = &rq->rq_status;
415 	for (i = 0; i < RQB_LEN; i++)
416 		if (rqb->rqb_bits[i]) {
417 			CTR2(KTR_RUNQ, "runq_check: bits=%#x i=%d",
418 			    rqb->rqb_bits[i], i);
419 			return (1);
420 		}
421 	CTR0(KTR_RUNQ, "runq_check: empty");
422 
423 	return (0);
424 }
425 
426 /*
427  * Find the highest priority process on the run queue.
428  */
429 struct thread *
430 runq_choose_fuzz(struct runq *rq, int fuzz)
431 {
432 	struct rqhead *rqh;
433 	struct thread *td;
434 	int pri;
435 
436 	while ((pri = runq_findbit(rq)) != -1) {
437 		rqh = &rq->rq_queues[pri];
438 		/* fuzz == 1 is normal.. 0 or less are ignored */
439 		if (fuzz > 1) {
440 			/*
441 			 * In the first couple of entries, check if
442 			 * there is one for our CPU as a preference.
443 			 */
444 			int count = fuzz;
445 			int cpu = PCPU_GET(cpuid);
446 			struct thread *td2;
447 			td2 = td = TAILQ_FIRST(rqh);
448 
449 			while (count-- && td2) {
450 				if (td2->td_lastcpu == cpu) {
451 					td = td2;
452 					break;
453 				}
454 				td2 = TAILQ_NEXT(td2, td_runq);
455 			}
456 		} else
457 			td = TAILQ_FIRST(rqh);
458 		KASSERT(td != NULL, ("runq_choose_fuzz: no proc on busy queue"));
459 		CTR3(KTR_RUNQ,
460 		    "runq_choose_fuzz: pri=%d thread=%p rqh=%p", pri, td, rqh);
461 		return (td);
462 	}
463 	CTR1(KTR_RUNQ, "runq_choose_fuzz: idleproc pri=%d", pri);
464 
465 	return (NULL);
466 }
467 
468 /*
469  * Find the highest priority process on the run queue.
470  */
471 struct thread *
472 runq_choose(struct runq *rq)
473 {
474 	struct rqhead *rqh;
475 	struct thread *td;
476 	int pri;
477 
478 	while ((pri = runq_findbit(rq)) != -1) {
479 		rqh = &rq->rq_queues[pri];
480 		td = TAILQ_FIRST(rqh);
481 		KASSERT(td != NULL, ("runq_choose: no thread on busy queue"));
482 		CTR3(KTR_RUNQ,
483 		    "runq_choose: pri=%d thread=%p rqh=%p", pri, td, rqh);
484 		return (td);
485 	}
486 	CTR1(KTR_RUNQ, "runq_choose: idlethread pri=%d", pri);
487 
488 	return (NULL);
489 }
490 
491 struct thread *
492 runq_choose_from(struct runq *rq, u_char idx)
493 {
494 	struct rqhead *rqh;
495 	struct thread *td;
496 	int pri;
497 
498 	if ((pri = runq_findbit_from(rq, idx)) != -1) {
499 		rqh = &rq->rq_queues[pri];
500 		td = TAILQ_FIRST(rqh);
501 		KASSERT(td != NULL, ("runq_choose: no thread on busy queue"));
502 		CTR4(KTR_RUNQ,
503 		    "runq_choose_from: pri=%d thread=%p idx=%d rqh=%p",
504 		    pri, td, td->td_rqindex, rqh);
505 		return (td);
506 	}
507 	CTR1(KTR_RUNQ, "runq_choose_from: idlethread pri=%d", pri);
508 
509 	return (NULL);
510 }
511 /*
512  * Remove the thread from the queue specified by its priority, and clear the
513  * corresponding status bit if the queue becomes empty.
514  * Caller must set state afterwards.
515  */
516 void
517 runq_remove(struct runq *rq, struct thread *td)
518 {
519 
520 	runq_remove_idx(rq, td, NULL);
521 }
522 
523 void
524 runq_remove_idx(struct runq *rq, struct thread *td, u_char *idx)
525 {
526 	struct rqhead *rqh;
527 	u_char pri;
528 
529 	KASSERT(td->td_flags & TDF_INMEM,
530 		("runq_remove_idx: thread swapped out"));
531 	pri = td->td_rqindex;
532 	KASSERT(pri < RQ_NQS, ("runq_remove_idx: Invalid index %d\n", pri));
533 	rqh = &rq->rq_queues[pri];
534 	CTR4(KTR_RUNQ, "runq_remove_idx: td=%p, pri=%d %d rqh=%p",
535 	    td, td->td_priority, pri, rqh);
536 	TAILQ_REMOVE(rqh, td, td_runq);
537 	if (TAILQ_EMPTY(rqh)) {
538 		CTR0(KTR_RUNQ, "runq_remove_idx: empty");
539 		runq_clrbit(rq, pri);
540 		if (idx != NULL && *idx == pri)
541 			*idx = (pri + 1) % RQ_NQS;
542 	}
543 }
544