xref: /freebsd/sys/kern/kern_switch.c (revision 75dfc66c1b2b44609e5a7c3e1d6a751be4922689)
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 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
31 
32 #include "opt_sched.h"
33 
34 #include <sys/param.h>
35 #include <sys/systm.h>
36 #include <sys/kdb.h>
37 #include <sys/kernel.h>
38 #include <sys/ktr.h>
39 #include <sys/lock.h>
40 #include <sys/mutex.h>
41 #include <sys/proc.h>
42 #include <sys/queue.h>
43 #include <sys/sched.h>
44 #include <sys/smp.h>
45 #include <sys/sysctl.h>
46 
47 #include <machine/cpu.h>
48 
49 /* Uncomment this to enable logging of critical_enter/exit. */
50 #if 0
51 #define	KTR_CRITICAL	KTR_SCHED
52 #else
53 #define	KTR_CRITICAL	0
54 #endif
55 
56 #ifdef FULL_PREEMPTION
57 #ifndef PREEMPTION
58 #error "The FULL_PREEMPTION option requires the PREEMPTION option"
59 #endif
60 #endif
61 
62 CTASSERT((RQB_BPW * RQB_LEN) == RQ_NQS);
63 
64 /*
65  * kern.sched.preemption allows user space to determine if preemption support
66  * is compiled in or not.  It is not currently a boot or runtime flag that
67  * can be changed.
68  */
69 #ifdef PREEMPTION
70 static int kern_sched_preemption = 1;
71 #else
72 static int kern_sched_preemption = 0;
73 #endif
74 SYSCTL_INT(_kern_sched, OID_AUTO, preemption, CTLFLAG_RD,
75     &kern_sched_preemption, 0, "Kernel preemption enabled");
76 
77 /*
78  * Support for scheduler stats exported via kern.sched.stats.  All stats may
79  * be reset with kern.sched.stats.reset = 1.  Stats may be defined elsewhere
80  * with SCHED_STAT_DEFINE().
81  */
82 #ifdef SCHED_STATS
83 SYSCTL_NODE(_kern_sched, OID_AUTO, stats, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
84     "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,
146     CTLTYPE_INT | CTLFLAG_WR | CTLFLAG_NEEDGIANT, NULL, 0,
147     sysctl_stats_reset, "I",
148     "Reset scheduler statistics");
149 #endif
150 
151 /************************************************************************
152  * Functions that manipulate runnability from a thread perspective.	*
153  ************************************************************************/
154 /*
155  * Select the thread that will be run next.
156  */
157 
158 static __noinline struct thread *
159 choosethread_panic(struct thread *td)
160 {
161 
162 	/*
163 	 * If we are in panic, only allow system threads,
164 	 * plus the one we are running in, to be run.
165 	 */
166 retry:
167 	if (((td->td_proc->p_flag & P_SYSTEM) == 0 &&
168 	    (td->td_flags & TDF_INPANIC) == 0)) {
169 		/* note that it is no longer on the run queue */
170 		TD_SET_CAN_RUN(td);
171 		td = sched_choose();
172 		goto retry;
173 	}
174 
175 	TD_SET_RUNNING(td);
176 	return (td);
177 }
178 
179 struct thread *
180 choosethread(void)
181 {
182 	struct thread *td;
183 
184 	td = sched_choose();
185 
186 	if (KERNEL_PANICKED())
187 		return (choosethread_panic(td));
188 
189 	TD_SET_RUNNING(td);
190 	return (td);
191 }
192 
193 /*
194  * Kernel thread preemption implementation.  Critical sections mark
195  * regions of code in which preemptions are not allowed.
196  *
197  * It might seem a good idea to inline critical_enter() but, in order
198  * to prevent instructions reordering by the compiler, a __compiler_membar()
199  * would have to be used here (the same as sched_pin()).  The performance
200  * penalty imposed by the membar could, then, produce slower code than
201  * the function call itself, for most cases.
202  */
203 void
204 critical_enter_KBI(void)
205 {
206 #ifdef KTR
207 	struct thread *td = curthread;
208 #endif
209 	critical_enter();
210 	CTR4(KTR_CRITICAL, "critical_enter by thread %p (%ld, %s) to %d", td,
211 	    (long)td->td_proc->p_pid, td->td_name, td->td_critnest);
212 }
213 
214 void __noinline
215 critical_exit_preempt(void)
216 {
217 	struct thread *td;
218 	int flags;
219 
220 	/*
221 	 * If td_critnest is 0, it is possible that we are going to get
222 	 * preempted again before reaching the code below. This happens
223 	 * rarely and is harmless. However, this means td_owepreempt may
224 	 * now be unset.
225 	 */
226 	td = curthread;
227 	if (td->td_critnest != 0)
228 		return;
229 	if (kdb_active)
230 		return;
231 
232 	/*
233 	 * Microoptimization: we committed to switch,
234 	 * disable preemption in interrupt handlers
235 	 * while spinning for the thread lock.
236 	 */
237 	td->td_critnest = 1;
238 	thread_lock(td);
239 	td->td_critnest--;
240 	flags = SW_INVOL | SW_PREEMPT;
241 	if (TD_IS_IDLETHREAD(td))
242 		flags |= SWT_IDLE;
243 	else
244 		flags |= SWT_OWEPREEMPT;
245 	mi_switch(flags);
246 }
247 
248 void
249 critical_exit_KBI(void)
250 {
251 #ifdef KTR
252 	struct thread *td = curthread;
253 #endif
254 	critical_exit();
255 	CTR4(KTR_CRITICAL, "critical_exit by thread %p (%ld, %s) to %d", td,
256 	    (long)td->td_proc->p_pid, td->td_name, td->td_critnest);
257 }
258 
259 /************************************************************************
260  * SYSTEM RUN QUEUE manipulations and tests				*
261  ************************************************************************/
262 /*
263  * Initialize a run structure.
264  */
265 void
266 runq_init(struct runq *rq)
267 {
268 	int i;
269 
270 	bzero(rq, sizeof *rq);
271 	for (i = 0; i < RQ_NQS; i++)
272 		TAILQ_INIT(&rq->rq_queues[i]);
273 }
274 
275 /*
276  * Clear the status bit of the queue corresponding to priority level pri,
277  * indicating that it is empty.
278  */
279 static __inline void
280 runq_clrbit(struct runq *rq, int pri)
281 {
282 	struct rqbits *rqb;
283 
284 	rqb = &rq->rq_status;
285 	CTR4(KTR_RUNQ, "runq_clrbit: bits=%#x %#x bit=%#x word=%d",
286 	    rqb->rqb_bits[RQB_WORD(pri)],
287 	    rqb->rqb_bits[RQB_WORD(pri)] & ~RQB_BIT(pri),
288 	    RQB_BIT(pri), RQB_WORD(pri));
289 	rqb->rqb_bits[RQB_WORD(pri)] &= ~RQB_BIT(pri);
290 }
291 
292 /*
293  * Find the index of the first non-empty run queue.  This is done by
294  * scanning the status bits, a set bit indicates a non-empty queue.
295  */
296 static __inline int
297 runq_findbit(struct runq *rq)
298 {
299 	struct rqbits *rqb;
300 	int pri;
301 	int i;
302 
303 	rqb = &rq->rq_status;
304 	for (i = 0; i < RQB_LEN; i++)
305 		if (rqb->rqb_bits[i]) {
306 			pri = RQB_FFS(rqb->rqb_bits[i]) + (i << RQB_L2BPW);
307 			CTR3(KTR_RUNQ, "runq_findbit: bits=%#x i=%d pri=%d",
308 			    rqb->rqb_bits[i], i, pri);
309 			return (pri);
310 		}
311 
312 	return (-1);
313 }
314 
315 static __inline int
316 runq_findbit_from(struct runq *rq, u_char pri)
317 {
318 	struct rqbits *rqb;
319 	rqb_word_t mask;
320 	int i;
321 
322 	/*
323 	 * Set the mask for the first word so we ignore priorities before 'pri'.
324 	 */
325 	mask = (rqb_word_t)-1 << (pri & (RQB_BPW - 1));
326 	rqb = &rq->rq_status;
327 again:
328 	for (i = RQB_WORD(pri); i < RQB_LEN; mask = -1, i++) {
329 		mask = rqb->rqb_bits[i] & mask;
330 		if (mask == 0)
331 			continue;
332 		pri = RQB_FFS(mask) + (i << RQB_L2BPW);
333 		CTR3(KTR_RUNQ, "runq_findbit_from: bits=%#x i=%d pri=%d",
334 		    mask, i, pri);
335 		return (pri);
336 	}
337 	if (pri == 0)
338 		return (-1);
339 	/*
340 	 * Wrap back around to the beginning of the list just once so we
341 	 * scan the whole thing.
342 	 */
343 	pri = 0;
344 	goto again;
345 }
346 
347 /*
348  * Set the status bit of the queue corresponding to priority level pri,
349  * indicating that it is non-empty.
350  */
351 static __inline void
352 runq_setbit(struct runq *rq, int pri)
353 {
354 	struct rqbits *rqb;
355 
356 	rqb = &rq->rq_status;
357 	CTR4(KTR_RUNQ, "runq_setbit: bits=%#x %#x bit=%#x word=%d",
358 	    rqb->rqb_bits[RQB_WORD(pri)],
359 	    rqb->rqb_bits[RQB_WORD(pri)] | RQB_BIT(pri),
360 	    RQB_BIT(pri), RQB_WORD(pri));
361 	rqb->rqb_bits[RQB_WORD(pri)] |= RQB_BIT(pri);
362 }
363 
364 /*
365  * Add the thread to the queue specified by its priority, and set the
366  * corresponding status bit.
367  */
368 void
369 runq_add(struct runq *rq, struct thread *td, int flags)
370 {
371 	struct rqhead *rqh;
372 	int pri;
373 
374 	pri = td->td_priority / RQ_PPQ;
375 	td->td_rqindex = pri;
376 	runq_setbit(rq, pri);
377 	rqh = &rq->rq_queues[pri];
378 	CTR4(KTR_RUNQ, "runq_add: td=%p pri=%d %d rqh=%p",
379 	    td, td->td_priority, pri, rqh);
380 	if (flags & SRQ_PREEMPTED) {
381 		TAILQ_INSERT_HEAD(rqh, td, td_runq);
382 	} else {
383 		TAILQ_INSERT_TAIL(rqh, td, td_runq);
384 	}
385 }
386 
387 void
388 runq_add_pri(struct runq *rq, struct thread *td, u_char pri, int flags)
389 {
390 	struct rqhead *rqh;
391 
392 	KASSERT(pri < RQ_NQS, ("runq_add_pri: %d out of range", pri));
393 	td->td_rqindex = pri;
394 	runq_setbit(rq, pri);
395 	rqh = &rq->rq_queues[pri];
396 	CTR4(KTR_RUNQ, "runq_add_pri: td=%p pri=%d idx=%d rqh=%p",
397 	    td, td->td_priority, pri, rqh);
398 	if (flags & SRQ_PREEMPTED) {
399 		TAILQ_INSERT_HEAD(rqh, td, td_runq);
400 	} else {
401 		TAILQ_INSERT_TAIL(rqh, td, td_runq);
402 	}
403 }
404 /*
405  * Return true if there are runnable processes of any priority on the run
406  * queue, false otherwise.  Has no side effects, does not modify the run
407  * queue structure.
408  */
409 int
410 runq_check(struct runq *rq)
411 {
412 	struct rqbits *rqb;
413 	int i;
414 
415 	rqb = &rq->rq_status;
416 	for (i = 0; i < RQB_LEN; i++)
417 		if (rqb->rqb_bits[i]) {
418 			CTR2(KTR_RUNQ, "runq_check: bits=%#x i=%d",
419 			    rqb->rqb_bits[i], i);
420 			return (1);
421 		}
422 	CTR0(KTR_RUNQ, "runq_check: empty");
423 
424 	return (0);
425 }
426 
427 /*
428  * Find the highest priority process on the run queue.
429  */
430 struct thread *
431 runq_choose_fuzz(struct runq *rq, int fuzz)
432 {
433 	struct rqhead *rqh;
434 	struct thread *td;
435 	int pri;
436 
437 	while ((pri = runq_findbit(rq)) != -1) {
438 		rqh = &rq->rq_queues[pri];
439 		/* fuzz == 1 is normal.. 0 or less are ignored */
440 		if (fuzz > 1) {
441 			/*
442 			 * In the first couple of entries, check if
443 			 * there is one for our CPU as a preference.
444 			 */
445 			int count = fuzz;
446 			int cpu = PCPU_GET(cpuid);
447 			struct thread *td2;
448 			td2 = td = TAILQ_FIRST(rqh);
449 
450 			while (count-- && td2) {
451 				if (td2->td_lastcpu == cpu) {
452 					td = td2;
453 					break;
454 				}
455 				td2 = TAILQ_NEXT(td2, td_runq);
456 			}
457 		} else
458 			td = TAILQ_FIRST(rqh);
459 		KASSERT(td != NULL, ("runq_choose_fuzz: no proc on busy queue"));
460 		CTR3(KTR_RUNQ,
461 		    "runq_choose_fuzz: pri=%d thread=%p rqh=%p", pri, td, rqh);
462 		return (td);
463 	}
464 	CTR1(KTR_RUNQ, "runq_choose_fuzz: idleproc pri=%d", pri);
465 
466 	return (NULL);
467 }
468 
469 /*
470  * Find the highest priority process on the run queue.
471  */
472 struct thread *
473 runq_choose(struct runq *rq)
474 {
475 	struct rqhead *rqh;
476 	struct thread *td;
477 	int pri;
478 
479 	while ((pri = runq_findbit(rq)) != -1) {
480 		rqh = &rq->rq_queues[pri];
481 		td = TAILQ_FIRST(rqh);
482 		KASSERT(td != NULL, ("runq_choose: no thread on busy queue"));
483 		CTR3(KTR_RUNQ,
484 		    "runq_choose: pri=%d thread=%p rqh=%p", pri, td, rqh);
485 		return (td);
486 	}
487 	CTR1(KTR_RUNQ, "runq_choose: idlethread pri=%d", pri);
488 
489 	return (NULL);
490 }
491 
492 struct thread *
493 runq_choose_from(struct runq *rq, u_char idx)
494 {
495 	struct rqhead *rqh;
496 	struct thread *td;
497 	int pri;
498 
499 	if ((pri = runq_findbit_from(rq, idx)) != -1) {
500 		rqh = &rq->rq_queues[pri];
501 		td = TAILQ_FIRST(rqh);
502 		KASSERT(td != NULL, ("runq_choose: no thread on busy queue"));
503 		CTR4(KTR_RUNQ,
504 		    "runq_choose_from: pri=%d thread=%p idx=%d rqh=%p",
505 		    pri, td, td->td_rqindex, rqh);
506 		return (td);
507 	}
508 	CTR1(KTR_RUNQ, "runq_choose_from: idlethread pri=%d", pri);
509 
510 	return (NULL);
511 }
512 /*
513  * Remove the thread from the queue specified by its priority, and clear the
514  * corresponding status bit if the queue becomes empty.
515  * Caller must set state afterwards.
516  */
517 void
518 runq_remove(struct runq *rq, struct thread *td)
519 {
520 
521 	runq_remove_idx(rq, td, NULL);
522 }
523 
524 void
525 runq_remove_idx(struct runq *rq, struct thread *td, u_char *idx)
526 {
527 	struct rqhead *rqh;
528 	u_char pri;
529 
530 	KASSERT(td->td_flags & TDF_INMEM,
531 		("runq_remove_idx: thread swapped out"));
532 	pri = td->td_rqindex;
533 	KASSERT(pri < RQ_NQS, ("runq_remove_idx: Invalid index %d\n", pri));
534 	rqh = &rq->rq_queues[pri];
535 	CTR4(KTR_RUNQ, "runq_remove_idx: td=%p, pri=%d %d rqh=%p",
536 	    td, td->td_priority, pri, rqh);
537 	TAILQ_REMOVE(rqh, td, td_runq);
538 	if (TAILQ_EMPTY(rqh)) {
539 		CTR0(KTR_RUNQ, "runq_remove_idx: empty");
540 		runq_clrbit(rq, pri);
541 		if (idx != NULL && *idx == pri)
542 			*idx = (pri + 1) % RQ_NQS;
543 	}
544 }
545