xref: /freebsd/sys/kern/kern_switch.c (revision 10f0bcab61ef441cb5af32fb706688d8cbd55dc0)
1 /*-
2  * Copyright (c) 2001 Jake Burkholder <jake@FreeBSD.org>
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  *
14  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24  * SUCH DAMAGE.
25  */
26 
27 
28 #include <sys/cdefs.h>
29 __FBSDID("$FreeBSD$");
30 
31 #include "opt_sched.h"
32 
33 #include <sys/param.h>
34 #include <sys/systm.h>
35 #include <sys/kdb.h>
36 #include <sys/kernel.h>
37 #include <sys/ktr.h>
38 #include <sys/lock.h>
39 #include <sys/mutex.h>
40 #include <sys/proc.h>
41 #include <sys/queue.h>
42 #include <sys/sched.h>
43 #include <sys/smp.h>
44 #include <sys/sysctl.h>
45 
46 #include <machine/cpu.h>
47 
48 /* Uncomment this to enable logging of critical_enter/exit. */
49 #if 0
50 #define	KTR_CRITICAL	KTR_SCHED
51 #else
52 #define	KTR_CRITICAL	0
53 #endif
54 
55 #ifdef FULL_PREEMPTION
56 #ifndef PREEMPTION
57 #error "The FULL_PREEMPTION option requires the PREEMPTION option"
58 #endif
59 #endif
60 
61 CTASSERT((RQB_BPW * RQB_LEN) == RQ_NQS);
62 
63 /*
64  * kern.sched.preemption allows user space to determine if preemption support
65  * is compiled in or not.  It is not currently a boot or runtime flag that
66  * can be changed.
67  */
68 #ifdef PREEMPTION
69 static int kern_sched_preemption = 1;
70 #else
71 static int kern_sched_preemption = 0;
72 #endif
73 SYSCTL_INT(_kern_sched, OID_AUTO, preemption, CTLFLAG_RD,
74     &kern_sched_preemption, 0, "Kernel preemption enabled");
75 
76 #ifdef SCHED_STATS
77 long switch_preempt;
78 long switch_owepreempt;
79 long switch_turnstile;
80 long switch_sleepq;
81 long switch_sleepqtimo;
82 long switch_relinquish;
83 long switch_needresched;
84 static SYSCTL_NODE(_kern_sched, OID_AUTO, stats, CTLFLAG_RW, 0, "switch stats");
85 SYSCTL_INT(_kern_sched_stats, OID_AUTO, preempt, CTLFLAG_RD, &switch_preempt, 0, "");
86 SYSCTL_INT(_kern_sched_stats, OID_AUTO, owepreempt, CTLFLAG_RD, &switch_owepreempt, 0, "");
87 SYSCTL_INT(_kern_sched_stats, OID_AUTO, turnstile, CTLFLAG_RD, &switch_turnstile, 0, "");
88 SYSCTL_INT(_kern_sched_stats, OID_AUTO, sleepq, CTLFLAG_RD, &switch_sleepq, 0, "");
89 SYSCTL_INT(_kern_sched_stats, OID_AUTO, sleepqtimo, CTLFLAG_RD, &switch_sleepqtimo, 0, "");
90 SYSCTL_INT(_kern_sched_stats, OID_AUTO, relinquish, CTLFLAG_RD, &switch_relinquish, 0, "");
91 SYSCTL_INT(_kern_sched_stats, OID_AUTO, needresched, CTLFLAG_RD, &switch_needresched, 0, "");
92 static int
93 sysctl_stats_reset(SYSCTL_HANDLER_ARGS)
94 {
95         int error;
96 	int val;
97 
98         val = 0;
99         error = sysctl_handle_int(oidp, &val, 0, req);
100         if (error != 0 || req->newptr == NULL)
101                 return (error);
102         if (val == 0)
103                 return (0);
104 	switch_preempt = 0;
105 	switch_owepreempt = 0;
106 	switch_turnstile = 0;
107 	switch_sleepq = 0;
108 	switch_sleepqtimo = 0;
109 	switch_relinquish = 0;
110 	switch_needresched = 0;
111 
112 	return (0);
113 }
114 
115 SYSCTL_PROC(_kern_sched_stats, OID_AUTO, reset, CTLTYPE_INT | CTLFLAG_WR, NULL,
116     0, sysctl_stats_reset, "I", "Reset scheduler statistics");
117 #endif
118 
119 /************************************************************************
120  * Functions that manipulate runnability from a thread perspective.	*
121  ************************************************************************/
122 /*
123  * Select the thread that will be run next.
124  */
125 struct thread *
126 choosethread(void)
127 {
128 	struct thread *td;
129 
130 retry:
131 	td = sched_choose();
132 
133 	/*
134 	 * If we are in panic, only allow system threads,
135 	 * plus the one we are running in, to be run.
136 	 */
137 	if (panicstr && ((td->td_proc->p_flag & P_SYSTEM) == 0 &&
138 	    (td->td_flags & TDF_INPANIC) == 0)) {
139 		/* note that it is no longer on the run queue */
140 		TD_SET_CAN_RUN(td);
141 		goto retry;
142 	}
143 
144 	TD_SET_RUNNING(td);
145 	return (td);
146 }
147 
148 /*
149  * Kernel thread preemption implementation.  Critical sections mark
150  * regions of code in which preemptions are not allowed.
151  */
152 void
153 critical_enter(void)
154 {
155 	struct thread *td;
156 
157 	td = curthread;
158 	td->td_critnest++;
159 	CTR4(KTR_CRITICAL, "critical_enter by thread %p (%ld, %s) to %d", td,
160 	    (long)td->td_proc->p_pid, td->td_name, td->td_critnest);
161 }
162 
163 void
164 critical_exit(void)
165 {
166 	struct thread *td;
167 
168 	td = curthread;
169 	KASSERT(td->td_critnest != 0,
170 	    ("critical_exit: td_critnest == 0"));
171 
172 	if (td->td_critnest == 1) {
173 		td->td_critnest = 0;
174 		if (td->td_owepreempt) {
175 			td->td_critnest = 1;
176 			thread_lock(td);
177 			td->td_critnest--;
178 			SCHED_STAT_INC(switch_owepreempt);
179 			mi_switch(SW_INVOL|SW_PREEMPT, NULL);
180 			thread_unlock(td);
181 		}
182 	} else
183 		td->td_critnest--;
184 
185 	CTR4(KTR_CRITICAL, "critical_exit by thread %p (%ld, %s) to %d", td,
186 	    (long)td->td_proc->p_pid, td->td_name, td->td_critnest);
187 }
188 
189 /************************************************************************
190  * SYSTEM RUN QUEUE manipulations and tests				*
191  ************************************************************************/
192 /*
193  * Initialize a run structure.
194  */
195 void
196 runq_init(struct runq *rq)
197 {
198 	int i;
199 
200 	bzero(rq, sizeof *rq);
201 	for (i = 0; i < RQ_NQS; i++)
202 		TAILQ_INIT(&rq->rq_queues[i]);
203 }
204 
205 /*
206  * Clear the status bit of the queue corresponding to priority level pri,
207  * indicating that it is empty.
208  */
209 static __inline void
210 runq_clrbit(struct runq *rq, int pri)
211 {
212 	struct rqbits *rqb;
213 
214 	rqb = &rq->rq_status;
215 	CTR4(KTR_RUNQ, "runq_clrbit: bits=%#x %#x bit=%#x word=%d",
216 	    rqb->rqb_bits[RQB_WORD(pri)],
217 	    rqb->rqb_bits[RQB_WORD(pri)] & ~RQB_BIT(pri),
218 	    RQB_BIT(pri), RQB_WORD(pri));
219 	rqb->rqb_bits[RQB_WORD(pri)] &= ~RQB_BIT(pri);
220 }
221 
222 /*
223  * Find the index of the first non-empty run queue.  This is done by
224  * scanning the status bits, a set bit indicates a non-empty queue.
225  */
226 static __inline int
227 runq_findbit(struct runq *rq)
228 {
229 	struct rqbits *rqb;
230 	int pri;
231 	int i;
232 
233 	rqb = &rq->rq_status;
234 	for (i = 0; i < RQB_LEN; i++)
235 		if (rqb->rqb_bits[i]) {
236 			pri = RQB_FFS(rqb->rqb_bits[i]) + (i << RQB_L2BPW);
237 			CTR3(KTR_RUNQ, "runq_findbit: bits=%#x i=%d pri=%d",
238 			    rqb->rqb_bits[i], i, pri);
239 			return (pri);
240 		}
241 
242 	return (-1);
243 }
244 
245 static __inline int
246 runq_findbit_from(struct runq *rq, u_char pri)
247 {
248 	struct rqbits *rqb;
249 	rqb_word_t mask;
250 	int i;
251 
252 	/*
253 	 * Set the mask for the first word so we ignore priorities before 'pri'.
254 	 */
255 	mask = (rqb_word_t)-1 << (pri & (RQB_BPW - 1));
256 	rqb = &rq->rq_status;
257 again:
258 	for (i = RQB_WORD(pri); i < RQB_LEN; mask = -1, i++) {
259 		mask = rqb->rqb_bits[i] & mask;
260 		if (mask == 0)
261 			continue;
262 		pri = RQB_FFS(mask) + (i << RQB_L2BPW);
263 		CTR3(KTR_RUNQ, "runq_findbit_from: bits=%#x i=%d pri=%d",
264 		    mask, i, pri);
265 		return (pri);
266 	}
267 	if (pri == 0)
268 		return (-1);
269 	/*
270 	 * Wrap back around to the beginning of the list just once so we
271 	 * scan the whole thing.
272 	 */
273 	pri = 0;
274 	goto again;
275 }
276 
277 /*
278  * Set the status bit of the queue corresponding to priority level pri,
279  * indicating that it is non-empty.
280  */
281 static __inline void
282 runq_setbit(struct runq *rq, int pri)
283 {
284 	struct rqbits *rqb;
285 
286 	rqb = &rq->rq_status;
287 	CTR4(KTR_RUNQ, "runq_setbit: bits=%#x %#x bit=%#x word=%d",
288 	    rqb->rqb_bits[RQB_WORD(pri)],
289 	    rqb->rqb_bits[RQB_WORD(pri)] | RQB_BIT(pri),
290 	    RQB_BIT(pri), RQB_WORD(pri));
291 	rqb->rqb_bits[RQB_WORD(pri)] |= RQB_BIT(pri);
292 }
293 
294 /*
295  * Add the thread to the queue specified by its priority, and set the
296  * corresponding status bit.
297  */
298 void
299 runq_add(struct runq *rq, struct thread *td, int flags)
300 {
301 	struct rqhead *rqh;
302 	int pri;
303 
304 	pri = td->td_priority / RQ_PPQ;
305 	td->td_rqindex = pri;
306 	runq_setbit(rq, pri);
307 	rqh = &rq->rq_queues[pri];
308 	CTR4(KTR_RUNQ, "runq_add: td=%p pri=%d %d rqh=%p",
309 	    td, td->td_priority, pri, rqh);
310 	if (flags & SRQ_PREEMPTED) {
311 		TAILQ_INSERT_HEAD(rqh, td, td_runq);
312 	} else {
313 		TAILQ_INSERT_TAIL(rqh, td, td_runq);
314 	}
315 }
316 
317 void
318 runq_add_pri(struct runq *rq, struct thread *td, u_char pri, int flags)
319 {
320 	struct rqhead *rqh;
321 
322 	KASSERT(pri < RQ_NQS, ("runq_add_pri: %d out of range", pri));
323 	td->td_rqindex = pri;
324 	runq_setbit(rq, pri);
325 	rqh = &rq->rq_queues[pri];
326 	CTR4(KTR_RUNQ, "runq_add_pri: td=%p pri=%d idx=%d rqh=%p",
327 	    td, td->td_priority, pri, rqh);
328 	if (flags & SRQ_PREEMPTED) {
329 		TAILQ_INSERT_HEAD(rqh, td, td_runq);
330 	} else {
331 		TAILQ_INSERT_TAIL(rqh, td, td_runq);
332 	}
333 }
334 /*
335  * Return true if there are runnable processes of any priority on the run
336  * queue, false otherwise.  Has no side effects, does not modify the run
337  * queue structure.
338  */
339 int
340 runq_check(struct runq *rq)
341 {
342 	struct rqbits *rqb;
343 	int i;
344 
345 	rqb = &rq->rq_status;
346 	for (i = 0; i < RQB_LEN; i++)
347 		if (rqb->rqb_bits[i]) {
348 			CTR2(KTR_RUNQ, "runq_check: bits=%#x i=%d",
349 			    rqb->rqb_bits[i], i);
350 			return (1);
351 		}
352 	CTR0(KTR_RUNQ, "runq_check: empty");
353 
354 	return (0);
355 }
356 
357 /*
358  * Find the highest priority process on the run queue.
359  */
360 struct thread *
361 runq_choose_fuzz(struct runq *rq, int fuzz)
362 {
363 	struct rqhead *rqh;
364 	struct thread *td;
365 	int pri;
366 
367 	while ((pri = runq_findbit(rq)) != -1) {
368 		rqh = &rq->rq_queues[pri];
369 		/* fuzz == 1 is normal.. 0 or less are ignored */
370 		if (fuzz > 1) {
371 			/*
372 			 * In the first couple of entries, check if
373 			 * there is one for our CPU as a preference.
374 			 */
375 			int count = fuzz;
376 			int cpu = PCPU_GET(cpuid);
377 			struct thread *td2;
378 			td2 = td = TAILQ_FIRST(rqh);
379 
380 			while (count-- && td2) {
381 				if (td->td_lastcpu == cpu) {
382 					td = td2;
383 					break;
384 				}
385 				td2 = TAILQ_NEXT(td2, td_runq);
386 			}
387 		} else
388 			td = TAILQ_FIRST(rqh);
389 		KASSERT(td != NULL, ("runq_choose_fuzz: no proc on busy queue"));
390 		CTR3(KTR_RUNQ,
391 		    "runq_choose_fuzz: pri=%d thread=%p rqh=%p", pri, td, rqh);
392 		return (td);
393 	}
394 	CTR1(KTR_RUNQ, "runq_choose_fuzz: idleproc pri=%d", pri);
395 
396 	return (NULL);
397 }
398 
399 /*
400  * Find the highest priority process on the run queue.
401  */
402 struct thread *
403 runq_choose(struct runq *rq)
404 {
405 	struct rqhead *rqh;
406 	struct thread *td;
407 	int pri;
408 
409 	while ((pri = runq_findbit(rq)) != -1) {
410 		rqh = &rq->rq_queues[pri];
411 		td = TAILQ_FIRST(rqh);
412 		KASSERT(td != NULL, ("runq_choose: no thread on busy queue"));
413 		CTR3(KTR_RUNQ,
414 		    "runq_choose: pri=%d thread=%p rqh=%p", pri, td, rqh);
415 		return (td);
416 	}
417 	CTR1(KTR_RUNQ, "runq_choose: idlethread pri=%d", pri);
418 
419 	return (NULL);
420 }
421 
422 struct thread *
423 runq_choose_from(struct runq *rq, u_char idx)
424 {
425 	struct rqhead *rqh;
426 	struct thread *td;
427 	int pri;
428 
429 	if ((pri = runq_findbit_from(rq, idx)) != -1) {
430 		rqh = &rq->rq_queues[pri];
431 		td = TAILQ_FIRST(rqh);
432 		KASSERT(td != NULL, ("runq_choose: no thread on busy queue"));
433 		CTR4(KTR_RUNQ,
434 		    "runq_choose_from: pri=%d thread=%p idx=%d rqh=%p",
435 		    pri, td, td->td_rqindex, rqh);
436 		return (td);
437 	}
438 	CTR1(KTR_RUNQ, "runq_choose_from: idlethread pri=%d", pri);
439 
440 	return (NULL);
441 }
442 /*
443  * Remove the thread from the queue specified by its priority, and clear the
444  * corresponding status bit if the queue becomes empty.
445  * Caller must set state afterwards.
446  */
447 void
448 runq_remove(struct runq *rq, struct thread *td)
449 {
450 
451 	runq_remove_idx(rq, td, NULL);
452 }
453 
454 void
455 runq_remove_idx(struct runq *rq, struct thread *td, u_char *idx)
456 {
457 	struct rqhead *rqh;
458 	u_char pri;
459 
460 	KASSERT(td->td_flags & TDF_INMEM,
461 		("runq_remove_idx: thread swapped out"));
462 	pri = td->td_rqindex;
463 	KASSERT(pri < RQ_NQS, ("runq_remove_idx: Invalid index %d\n", pri));
464 	rqh = &rq->rq_queues[pri];
465 	CTR4(KTR_RUNQ, "runq_remove_idx: td=%p, pri=%d %d rqh=%p",
466 	    td, td->td_priority, pri, rqh);
467 	TAILQ_REMOVE(rqh, td, td_runq);
468 	if (TAILQ_EMPTY(rqh)) {
469 		CTR0(KTR_RUNQ, "runq_remove_idx: empty");
470 		runq_clrbit(rq, pri);
471 		if (idx != NULL && *idx == pri)
472 			*idx = (pri + 1) % RQ_NQS;
473 	}
474 }
475