xref: /freebsd/sys/kern/kern_thread.c (revision 891b8ed4672a213bbe6f3f10522eeadb34d01b76)
1 /*-
2  * Copyright (C) 2001 Julian Elischer <julian@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(s), this list of conditions and the following disclaimer as
10  *    the first lines of this file unmodified other than the possible
11  *    addition of one or more copyright notices.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice(s), 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 COPYRIGHT HOLDER(S) ``AS IS'' AND ANY
17  * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
18  * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
19  * DISCLAIMED.  IN NO EVENT SHALL THE COPYRIGHT HOLDER(S) BE LIABLE FOR ANY
20  * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
21  * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
22  * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
23  * 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 SUCH
26  * DAMAGE.
27  */
28 
29 #include "opt_witness.h"
30 #include "opt_hwpmc_hooks.h"
31 
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
34 
35 #include <sys/param.h>
36 #include <sys/systm.h>
37 #include <sys/kernel.h>
38 #include <sys/lock.h>
39 #include <sys/mutex.h>
40 #include <sys/proc.h>
41 #include <sys/resourcevar.h>
42 #include <sys/smp.h>
43 #include <sys/sched.h>
44 #include <sys/sleepqueue.h>
45 #include <sys/selinfo.h>
46 #include <sys/turnstile.h>
47 #include <sys/ktr.h>
48 #include <sys/rwlock.h>
49 #include <sys/umtx.h>
50 #include <sys/cpuset.h>
51 #ifdef	HWPMC_HOOKS
52 #include <sys/pmckern.h>
53 #endif
54 
55 #include <security/audit/audit.h>
56 
57 #include <vm/vm.h>
58 #include <vm/vm_extern.h>
59 #include <vm/uma.h>
60 #include <sys/eventhandler.h>
61 
62 /*
63  * thread related storage.
64  */
65 static uma_zone_t thread_zone;
66 
67 TAILQ_HEAD(, thread) zombie_threads = TAILQ_HEAD_INITIALIZER(zombie_threads);
68 static struct mtx zombie_lock;
69 MTX_SYSINIT(zombie_lock, &zombie_lock, "zombie lock", MTX_SPIN);
70 
71 static void thread_zombie(struct thread *);
72 
73 #define TID_BUFFER_SIZE	1024
74 
75 struct mtx tid_lock;
76 static struct unrhdr *tid_unrhdr;
77 static lwpid_t tid_buffer[TID_BUFFER_SIZE];
78 static int tid_head, tid_tail;
79 static MALLOC_DEFINE(M_TIDHASH, "tidhash", "thread hash");
80 
81 struct	tidhashhead *tidhashtbl;
82 u_long	tidhash;
83 struct	rwlock tidhash_lock;
84 
85 static lwpid_t
86 tid_alloc(void)
87 {
88 	lwpid_t	tid;
89 
90 	tid = alloc_unr(tid_unrhdr);
91 	if (tid != -1)
92 		return (tid);
93 	mtx_lock(&tid_lock);
94 	if (tid_head == tid_tail) {
95 		mtx_unlock(&tid_lock);
96 		return (-1);
97 	}
98 	tid = tid_buffer[tid_head++];
99 	tid_head %= TID_BUFFER_SIZE;
100 	mtx_unlock(&tid_lock);
101 	return (tid);
102 }
103 
104 static void
105 tid_free(lwpid_t tid)
106 {
107 	lwpid_t tmp_tid = -1;
108 
109 	mtx_lock(&tid_lock);
110 	if ((tid_tail + 1) % TID_BUFFER_SIZE == tid_head) {
111 		tmp_tid = tid_buffer[tid_head++];
112 		tid_head = (tid_head + 1) % TID_BUFFER_SIZE;
113 	}
114 	tid_buffer[tid_tail++] = tid;
115 	tid_tail %= TID_BUFFER_SIZE;
116 	mtx_unlock(&tid_lock);
117 	if (tmp_tid != -1)
118 		free_unr(tid_unrhdr, tmp_tid);
119 }
120 
121 /*
122  * Prepare a thread for use.
123  */
124 static int
125 thread_ctor(void *mem, int size, void *arg, int flags)
126 {
127 	struct thread	*td;
128 
129 	td = (struct thread *)mem;
130 	td->td_state = TDS_INACTIVE;
131 	td->td_oncpu = NOCPU;
132 
133 	td->td_tid = tid_alloc();
134 
135 	/*
136 	 * Note that td_critnest begins life as 1 because the thread is not
137 	 * running and is thereby implicitly waiting to be on the receiving
138 	 * end of a context switch.
139 	 */
140 	td->td_critnest = 1;
141 	td->td_lend_user_pri = PRI_MAX;
142 	EVENTHANDLER_INVOKE(thread_ctor, td);
143 #ifdef AUDIT
144 	audit_thread_alloc(td);
145 #endif
146 	umtx_thread_alloc(td);
147 	return (0);
148 }
149 
150 /*
151  * Reclaim a thread after use.
152  */
153 static void
154 thread_dtor(void *mem, int size, void *arg)
155 {
156 	struct thread *td;
157 
158 	td = (struct thread *)mem;
159 
160 #ifdef INVARIANTS
161 	/* Verify that this thread is in a safe state to free. */
162 	switch (td->td_state) {
163 	case TDS_INHIBITED:
164 	case TDS_RUNNING:
165 	case TDS_CAN_RUN:
166 	case TDS_RUNQ:
167 		/*
168 		 * We must never unlink a thread that is in one of
169 		 * these states, because it is currently active.
170 		 */
171 		panic("bad state for thread unlinking");
172 		/* NOTREACHED */
173 	case TDS_INACTIVE:
174 		break;
175 	default:
176 		panic("bad thread state");
177 		/* NOTREACHED */
178 	}
179 #endif
180 #ifdef AUDIT
181 	audit_thread_free(td);
182 #endif
183 	/* Free all OSD associated to this thread. */
184 	osd_thread_exit(td);
185 
186 	EVENTHANDLER_INVOKE(thread_dtor, td);
187 	tid_free(td->td_tid);
188 }
189 
190 /*
191  * Initialize type-stable parts of a thread (when newly created).
192  */
193 static int
194 thread_init(void *mem, int size, int flags)
195 {
196 	struct thread *td;
197 
198 	td = (struct thread *)mem;
199 
200 	td->td_sleepqueue = sleepq_alloc();
201 	td->td_turnstile = turnstile_alloc();
202 	EVENTHANDLER_INVOKE(thread_init, td);
203 	td->td_sched = (struct td_sched *)&td[1];
204 	umtx_thread_init(td);
205 	td->td_kstack = 0;
206 	return (0);
207 }
208 
209 /*
210  * Tear down type-stable parts of a thread (just before being discarded).
211  */
212 static void
213 thread_fini(void *mem, int size)
214 {
215 	struct thread *td;
216 
217 	td = (struct thread *)mem;
218 	EVENTHANDLER_INVOKE(thread_fini, td);
219 	turnstile_free(td->td_turnstile);
220 	sleepq_free(td->td_sleepqueue);
221 	umtx_thread_fini(td);
222 	seltdfini(td);
223 }
224 
225 /*
226  * For a newly created process,
227  * link up all the structures and its initial threads etc.
228  * called from:
229  * {arch}/{arch}/machdep.c   ia64_init(), init386() etc.
230  * proc_dtor() (should go away)
231  * proc_init()
232  */
233 void
234 proc_linkup0(struct proc *p, struct thread *td)
235 {
236 	TAILQ_INIT(&p->p_threads);	     /* all threads in proc */
237 	proc_linkup(p, td);
238 }
239 
240 void
241 proc_linkup(struct proc *p, struct thread *td)
242 {
243 
244 	sigqueue_init(&p->p_sigqueue, p);
245 	p->p_ksi = ksiginfo_alloc(1);
246 	if (p->p_ksi != NULL) {
247 		/* XXX p_ksi may be null if ksiginfo zone is not ready */
248 		p->p_ksi->ksi_flags = KSI_EXT | KSI_INS;
249 	}
250 	LIST_INIT(&p->p_mqnotifier);
251 	p->p_numthreads = 0;
252 	thread_link(td, p);
253 }
254 
255 /*
256  * Initialize global thread allocation resources.
257  */
258 void
259 threadinit(void)
260 {
261 
262 	mtx_init(&tid_lock, "TID lock", NULL, MTX_DEF);
263 	/* leave one number for thread0 */
264 	tid_unrhdr = new_unrhdr(PID_MAX + 2, INT_MAX, &tid_lock);
265 
266 	thread_zone = uma_zcreate("THREAD", sched_sizeof_thread(),
267 	    thread_ctor, thread_dtor, thread_init, thread_fini,
268 	    16 - 1, 0);
269 	tidhashtbl = hashinit(maxproc / 2, M_TIDHASH, &tidhash);
270 	rw_init(&tidhash_lock, "tidhash");
271 }
272 
273 /*
274  * Place an unused thread on the zombie list.
275  * Use the slpq as that must be unused by now.
276  */
277 void
278 thread_zombie(struct thread *td)
279 {
280 	mtx_lock_spin(&zombie_lock);
281 	TAILQ_INSERT_HEAD(&zombie_threads, td, td_slpq);
282 	mtx_unlock_spin(&zombie_lock);
283 }
284 
285 /*
286  * Release a thread that has exited after cpu_throw().
287  */
288 void
289 thread_stash(struct thread *td)
290 {
291 	atomic_subtract_rel_int(&td->td_proc->p_exitthreads, 1);
292 	thread_zombie(td);
293 }
294 
295 /*
296  * Reap zombie resources.
297  */
298 void
299 thread_reap(void)
300 {
301 	struct thread *td_first, *td_next;
302 
303 	/*
304 	 * Don't even bother to lock if none at this instant,
305 	 * we really don't care about the next instant..
306 	 */
307 	if (!TAILQ_EMPTY(&zombie_threads)) {
308 		mtx_lock_spin(&zombie_lock);
309 		td_first = TAILQ_FIRST(&zombie_threads);
310 		if (td_first)
311 			TAILQ_INIT(&zombie_threads);
312 		mtx_unlock_spin(&zombie_lock);
313 		while (td_first) {
314 			td_next = TAILQ_NEXT(td_first, td_slpq);
315 			if (td_first->td_ucred)
316 				crfree(td_first->td_ucred);
317 			thread_free(td_first);
318 			td_first = td_next;
319 		}
320 	}
321 }
322 
323 /*
324  * Allocate a thread.
325  */
326 struct thread *
327 thread_alloc(int pages)
328 {
329 	struct thread *td;
330 
331 	thread_reap(); /* check if any zombies to get */
332 
333 	td = (struct thread *)uma_zalloc(thread_zone, M_WAITOK);
334 	KASSERT(td->td_kstack == 0, ("thread_alloc got thread with kstack"));
335 	if (!vm_thread_new(td, pages)) {
336 		uma_zfree(thread_zone, td);
337 		return (NULL);
338 	}
339 	cpu_thread_alloc(td);
340 	return (td);
341 }
342 
343 int
344 thread_alloc_stack(struct thread *td, int pages)
345 {
346 
347 	KASSERT(td->td_kstack == 0,
348 	    ("thread_alloc_stack called on a thread with kstack"));
349 	if (!vm_thread_new(td, pages))
350 		return (0);
351 	cpu_thread_alloc(td);
352 	return (1);
353 }
354 
355 /*
356  * Deallocate a thread.
357  */
358 void
359 thread_free(struct thread *td)
360 {
361 
362 	lock_profile_thread_exit(td);
363 	if (td->td_cpuset)
364 		cpuset_rel(td->td_cpuset);
365 	td->td_cpuset = NULL;
366 	cpu_thread_free(td);
367 	if (td->td_kstack != 0)
368 		vm_thread_dispose(td);
369 	uma_zfree(thread_zone, td);
370 }
371 
372 /*
373  * Discard the current thread and exit from its context.
374  * Always called with scheduler locked.
375  *
376  * Because we can't free a thread while we're operating under its context,
377  * push the current thread into our CPU's deadthread holder. This means
378  * we needn't worry about someone else grabbing our context before we
379  * do a cpu_throw().
380  */
381 void
382 thread_exit(void)
383 {
384 	uint64_t new_switchtime;
385 	struct thread *td;
386 	struct thread *td2;
387 	struct proc *p;
388 	int wakeup_swapper;
389 
390 	td = curthread;
391 	p = td->td_proc;
392 
393 	PROC_SLOCK_ASSERT(p, MA_OWNED);
394 	mtx_assert(&Giant, MA_NOTOWNED);
395 
396 	PROC_LOCK_ASSERT(p, MA_OWNED);
397 	KASSERT(p != NULL, ("thread exiting without a process"));
398 	CTR3(KTR_PROC, "thread_exit: thread %p (pid %ld, %s)", td,
399 	    (long)p->p_pid, td->td_name);
400 	KASSERT(TAILQ_EMPTY(&td->td_sigqueue.sq_list), ("signal pending"));
401 
402 #ifdef AUDIT
403 	AUDIT_SYSCALL_EXIT(0, td);
404 #endif
405 	umtx_thread_exit(td);
406 	/*
407 	 * drop FPU & debug register state storage, or any other
408 	 * architecture specific resources that
409 	 * would not be on a new untouched process.
410 	 */
411 	cpu_thread_exit(td);	/* XXXSMP */
412 
413 	/* Do the same timestamp bookkeeping that mi_switch() would do. */
414 	new_switchtime = cpu_ticks();
415 	p->p_rux.rux_runtime += (new_switchtime - PCPU_GET(switchtime));
416 	PCPU_SET(switchtime, new_switchtime);
417 	PCPU_SET(switchticks, ticks);
418 	PCPU_INC(cnt.v_swtch);
419 	/* Save our resource usage in our process. */
420 	td->td_ru.ru_nvcsw++;
421 	rucollect(&p->p_ru, &td->td_ru);
422 	/*
423 	 * The last thread is left attached to the process
424 	 * So that the whole bundle gets recycled. Skip
425 	 * all this stuff if we never had threads.
426 	 * EXIT clears all sign of other threads when
427 	 * it goes to single threading, so the last thread always
428 	 * takes the short path.
429 	 */
430 	if (p->p_flag & P_HADTHREADS) {
431 		if (p->p_numthreads > 1) {
432 			thread_unlink(td);
433 			td2 = FIRST_THREAD_IN_PROC(p);
434 			sched_exit_thread(td2, td);
435 
436 			/*
437 			 * The test below is NOT true if we are the
438 			 * sole exiting thread. P_STOPPED_SINGLE is unset
439 			 * in exit1() after it is the only survivor.
440 			 */
441 			if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE) {
442 				if (p->p_numthreads == p->p_suspcount) {
443 					thread_lock(p->p_singlethread);
444 					wakeup_swapper = thread_unsuspend_one(
445 						p->p_singlethread);
446 					thread_unlock(p->p_singlethread);
447 					if (wakeup_swapper)
448 						kick_proc0();
449 				}
450 			}
451 
452 			atomic_add_int(&td->td_proc->p_exitthreads, 1);
453 			PCPU_SET(deadthread, td);
454 		} else {
455 			/*
456 			 * The last thread is exiting.. but not through exit()
457 			 */
458 			panic ("thread_exit: Last thread exiting on its own");
459 		}
460 	}
461 #ifdef	HWPMC_HOOKS
462 	/*
463 	 * If this thread is part of a process that is being tracked by hwpmc(4),
464 	 * inform the module of the thread's impending exit.
465 	 */
466 	if (PMC_PROC_IS_USING_PMCS(td->td_proc))
467 		PMC_SWITCH_CONTEXT(td, PMC_FN_CSW_OUT);
468 #endif
469 	PROC_UNLOCK(p);
470 	ruxagg(p, td);
471 	thread_lock(td);
472 	PROC_SUNLOCK(p);
473 	td->td_state = TDS_INACTIVE;
474 #ifdef WITNESS
475 	witness_thread_exit(td);
476 #endif
477 	CTR1(KTR_PROC, "thread_exit: cpu_throw() thread %p", td);
478 	sched_throw(td);
479 	panic("I'm a teapot!");
480 	/* NOTREACHED */
481 }
482 
483 /*
484  * Do any thread specific cleanups that may be needed in wait()
485  * called with Giant, proc and schedlock not held.
486  */
487 void
488 thread_wait(struct proc *p)
489 {
490 	struct thread *td;
491 
492 	mtx_assert(&Giant, MA_NOTOWNED);
493 	KASSERT((p->p_numthreads == 1), ("Multiple threads in wait1()"));
494 	td = FIRST_THREAD_IN_PROC(p);
495 	/* Lock the last thread so we spin until it exits cpu_throw(). */
496 	thread_lock(td);
497 	thread_unlock(td);
498 	/* Wait for any remaining threads to exit cpu_throw(). */
499 	while (p->p_exitthreads)
500 		sched_relinquish(curthread);
501 	lock_profile_thread_exit(td);
502 	cpuset_rel(td->td_cpuset);
503 	td->td_cpuset = NULL;
504 	cpu_thread_clean(td);
505 	crfree(td->td_ucred);
506 	thread_reap();	/* check for zombie threads etc. */
507 }
508 
509 /*
510  * Link a thread to a process.
511  * set up anything that needs to be initialized for it to
512  * be used by the process.
513  */
514 void
515 thread_link(struct thread *td, struct proc *p)
516 {
517 
518 	/*
519 	 * XXX This can't be enabled because it's called for proc0 before
520 	 * its lock has been created.
521 	 * PROC_LOCK_ASSERT(p, MA_OWNED);
522 	 */
523 	td->td_state    = TDS_INACTIVE;
524 	td->td_proc     = p;
525 	td->td_flags    = TDF_INMEM;
526 
527 	LIST_INIT(&td->td_contested);
528 	LIST_INIT(&td->td_lprof[0]);
529 	LIST_INIT(&td->td_lprof[1]);
530 	sigqueue_init(&td->td_sigqueue, p);
531 	callout_init(&td->td_slpcallout, CALLOUT_MPSAFE);
532 	TAILQ_INSERT_HEAD(&p->p_threads, td, td_plist);
533 	p->p_numthreads++;
534 }
535 
536 /*
537  * Convert a process with one thread to an unthreaded process.
538  */
539 void
540 thread_unthread(struct thread *td)
541 {
542 	struct proc *p = td->td_proc;
543 
544 	KASSERT((p->p_numthreads == 1), ("Unthreading with >1 threads"));
545 	p->p_flag &= ~P_HADTHREADS;
546 }
547 
548 /*
549  * Called from:
550  *  thread_exit()
551  */
552 void
553 thread_unlink(struct thread *td)
554 {
555 	struct proc *p = td->td_proc;
556 
557 	PROC_LOCK_ASSERT(p, MA_OWNED);
558 	TAILQ_REMOVE(&p->p_threads, td, td_plist);
559 	p->p_numthreads--;
560 	/* could clear a few other things here */
561 	/* Must  NOT clear links to proc! */
562 }
563 
564 static int
565 calc_remaining(struct proc *p, int mode)
566 {
567 	int remaining;
568 
569 	if (mode == SINGLE_EXIT)
570 		remaining = p->p_numthreads;
571 	else if (mode == SINGLE_BOUNDARY)
572 		remaining = p->p_numthreads - p->p_boundary_count;
573 	else if (mode == SINGLE_NO_EXIT)
574 		remaining = p->p_numthreads - p->p_suspcount;
575 	else
576 		panic("calc_remaining: wrong mode %d", mode);
577 	return (remaining);
578 }
579 
580 /*
581  * Enforce single-threading.
582  *
583  * Returns 1 if the caller must abort (another thread is waiting to
584  * exit the process or similar). Process is locked!
585  * Returns 0 when you are successfully the only thread running.
586  * A process has successfully single threaded in the suspend mode when
587  * There are no threads in user mode. Threads in the kernel must be
588  * allowed to continue until they get to the user boundary. They may even
589  * copy out their return values and data before suspending. They may however be
590  * accelerated in reaching the user boundary as we will wake up
591  * any sleeping threads that are interruptable. (PCATCH).
592  */
593 int
594 thread_single(int mode)
595 {
596 	struct thread *td;
597 	struct thread *td2;
598 	struct proc *p;
599 	int remaining, wakeup_swapper;
600 
601 	td = curthread;
602 	p = td->td_proc;
603 	mtx_assert(&Giant, MA_NOTOWNED);
604 	PROC_LOCK_ASSERT(p, MA_OWNED);
605 	KASSERT((td != NULL), ("curthread is NULL"));
606 
607 	if ((p->p_flag & P_HADTHREADS) == 0)
608 		return (0);
609 
610 	/* Is someone already single threading? */
611 	if (p->p_singlethread != NULL && p->p_singlethread != td)
612 		return (1);
613 
614 	if (mode == SINGLE_EXIT) {
615 		p->p_flag |= P_SINGLE_EXIT;
616 		p->p_flag &= ~P_SINGLE_BOUNDARY;
617 	} else {
618 		p->p_flag &= ~P_SINGLE_EXIT;
619 		if (mode == SINGLE_BOUNDARY)
620 			p->p_flag |= P_SINGLE_BOUNDARY;
621 		else
622 			p->p_flag &= ~P_SINGLE_BOUNDARY;
623 	}
624 	p->p_flag |= P_STOPPED_SINGLE;
625 	PROC_SLOCK(p);
626 	p->p_singlethread = td;
627 	remaining = calc_remaining(p, mode);
628 	while (remaining != 1) {
629 		if (P_SHOULDSTOP(p) != P_STOPPED_SINGLE)
630 			goto stopme;
631 		wakeup_swapper = 0;
632 		FOREACH_THREAD_IN_PROC(p, td2) {
633 			if (td2 == td)
634 				continue;
635 			thread_lock(td2);
636 			td2->td_flags |= TDF_ASTPENDING | TDF_NEEDSUSPCHK;
637 			if (TD_IS_INHIBITED(td2)) {
638 				switch (mode) {
639 				case SINGLE_EXIT:
640 					if (TD_IS_SUSPENDED(td2))
641 						wakeup_swapper |=
642 						    thread_unsuspend_one(td2);
643 					if (TD_ON_SLEEPQ(td2) &&
644 					    (td2->td_flags & TDF_SINTR))
645 						wakeup_swapper |=
646 						    sleepq_abort(td2, EINTR);
647 					break;
648 				case SINGLE_BOUNDARY:
649 					if (TD_IS_SUSPENDED(td2) &&
650 					    !(td2->td_flags & TDF_BOUNDARY))
651 						wakeup_swapper |=
652 						    thread_unsuspend_one(td2);
653 					if (TD_ON_SLEEPQ(td2) &&
654 					    (td2->td_flags & TDF_SINTR))
655 						wakeup_swapper |=
656 						    sleepq_abort(td2, ERESTART);
657 					break;
658 				case SINGLE_NO_EXIT:
659 					if (TD_IS_SUSPENDED(td2) &&
660 					    !(td2->td_flags & TDF_BOUNDARY))
661 						wakeup_swapper |=
662 						    thread_unsuspend_one(td2);
663 					if (TD_ON_SLEEPQ(td2) &&
664 					    (td2->td_flags & TDF_SINTR))
665 						wakeup_swapper |=
666 						    sleepq_abort(td2, ERESTART);
667 					break;
668 				default:
669 					break;
670 				}
671 			}
672 #ifdef SMP
673 			else if (TD_IS_RUNNING(td2) && td != td2) {
674 				forward_signal(td2);
675 			}
676 #endif
677 			thread_unlock(td2);
678 		}
679 		if (wakeup_swapper)
680 			kick_proc0();
681 		remaining = calc_remaining(p, mode);
682 
683 		/*
684 		 * Maybe we suspended some threads.. was it enough?
685 		 */
686 		if (remaining == 1)
687 			break;
688 
689 stopme:
690 		/*
691 		 * Wake us up when everyone else has suspended.
692 		 * In the mean time we suspend as well.
693 		 */
694 		thread_suspend_switch(td);
695 		remaining = calc_remaining(p, mode);
696 	}
697 	if (mode == SINGLE_EXIT) {
698 		/*
699 		 * We have gotten rid of all the other threads and we
700 		 * are about to either exit or exec. In either case,
701 		 * we try our utmost  to revert to being a non-threaded
702 		 * process.
703 		 */
704 		p->p_singlethread = NULL;
705 		p->p_flag &= ~(P_STOPPED_SINGLE | P_SINGLE_EXIT);
706 		thread_unthread(td);
707 	}
708 	PROC_SUNLOCK(p);
709 	return (0);
710 }
711 
712 /*
713  * Called in from locations that can safely check to see
714  * whether we have to suspend or at least throttle for a
715  * single-thread event (e.g. fork).
716  *
717  * Such locations include userret().
718  * If the "return_instead" argument is non zero, the thread must be able to
719  * accept 0 (caller may continue), or 1 (caller must abort) as a result.
720  *
721  * The 'return_instead' argument tells the function if it may do a
722  * thread_exit() or suspend, or whether the caller must abort and back
723  * out instead.
724  *
725  * If the thread that set the single_threading request has set the
726  * P_SINGLE_EXIT bit in the process flags then this call will never return
727  * if 'return_instead' is false, but will exit.
728  *
729  * P_SINGLE_EXIT | return_instead == 0| return_instead != 0
730  *---------------+--------------------+---------------------
731  *       0       | returns 0          |   returns 0 or 1
732  *               | when ST ends       |   immediatly
733  *---------------+--------------------+---------------------
734  *       1       | thread exits       |   returns 1
735  *               |                    |  immediatly
736  * 0 = thread_exit() or suspension ok,
737  * other = return error instead of stopping the thread.
738  *
739  * While a full suspension is under effect, even a single threading
740  * thread would be suspended if it made this call (but it shouldn't).
741  * This call should only be made from places where
742  * thread_exit() would be safe as that may be the outcome unless
743  * return_instead is set.
744  */
745 int
746 thread_suspend_check(int return_instead)
747 {
748 	struct thread *td;
749 	struct proc *p;
750 	int wakeup_swapper;
751 
752 	td = curthread;
753 	p = td->td_proc;
754 	mtx_assert(&Giant, MA_NOTOWNED);
755 	PROC_LOCK_ASSERT(p, MA_OWNED);
756 	while (P_SHOULDSTOP(p) ||
757 	      ((p->p_flag & P_TRACED) && (td->td_dbgflags & TDB_SUSPEND))) {
758 		if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE) {
759 			KASSERT(p->p_singlethread != NULL,
760 			    ("singlethread not set"));
761 			/*
762 			 * The only suspension in action is a
763 			 * single-threading. Single threader need not stop.
764 			 * XXX Should be safe to access unlocked
765 			 * as it can only be set to be true by us.
766 			 */
767 			if (p->p_singlethread == td)
768 				return (0);	/* Exempt from stopping. */
769 		}
770 		if ((p->p_flag & P_SINGLE_EXIT) && return_instead)
771 			return (EINTR);
772 
773 		/* Should we goto user boundary if we didn't come from there? */
774 		if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE &&
775 		    (p->p_flag & P_SINGLE_BOUNDARY) && return_instead)
776 			return (ERESTART);
777 
778 		/*
779 		 * If the process is waiting for us to exit,
780 		 * this thread should just suicide.
781 		 * Assumes that P_SINGLE_EXIT implies P_STOPPED_SINGLE.
782 		 */
783 		if ((p->p_flag & P_SINGLE_EXIT) && (p->p_singlethread != td)) {
784 			PROC_UNLOCK(p);
785 			tidhash_remove(td);
786 			PROC_LOCK(p);
787 			tdsigcleanup(td);
788 			PROC_SLOCK(p);
789 			thread_stopped(p);
790 			thread_exit();
791 		}
792 
793 		PROC_SLOCK(p);
794 		thread_stopped(p);
795 		if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE) {
796 			if (p->p_numthreads == p->p_suspcount + 1) {
797 				thread_lock(p->p_singlethread);
798 				wakeup_swapper =
799 				    thread_unsuspend_one(p->p_singlethread);
800 				thread_unlock(p->p_singlethread);
801 				if (wakeup_swapper)
802 					kick_proc0();
803 			}
804 		}
805 		PROC_UNLOCK(p);
806 		thread_lock(td);
807 		/*
808 		 * When a thread suspends, it just
809 		 * gets taken off all queues.
810 		 */
811 		thread_suspend_one(td);
812 		if (return_instead == 0) {
813 			p->p_boundary_count++;
814 			td->td_flags |= TDF_BOUNDARY;
815 		}
816 		PROC_SUNLOCK(p);
817 		mi_switch(SW_INVOL | SWT_SUSPEND, NULL);
818 		if (return_instead == 0)
819 			td->td_flags &= ~TDF_BOUNDARY;
820 		thread_unlock(td);
821 		PROC_LOCK(p);
822 		if (return_instead == 0)
823 			p->p_boundary_count--;
824 	}
825 	return (0);
826 }
827 
828 void
829 thread_suspend_switch(struct thread *td)
830 {
831 	struct proc *p;
832 
833 	p = td->td_proc;
834 	KASSERT(!TD_IS_SUSPENDED(td), ("already suspended"));
835 	PROC_LOCK_ASSERT(p, MA_OWNED);
836 	PROC_SLOCK_ASSERT(p, MA_OWNED);
837 	/*
838 	 * We implement thread_suspend_one in stages here to avoid
839 	 * dropping the proc lock while the thread lock is owned.
840 	 */
841 	thread_stopped(p);
842 	p->p_suspcount++;
843 	PROC_UNLOCK(p);
844 	thread_lock(td);
845 	td->td_flags &= ~TDF_NEEDSUSPCHK;
846 	TD_SET_SUSPENDED(td);
847 	sched_sleep(td, 0);
848 	PROC_SUNLOCK(p);
849 	DROP_GIANT();
850 	mi_switch(SW_VOL | SWT_SUSPEND, NULL);
851 	thread_unlock(td);
852 	PICKUP_GIANT();
853 	PROC_LOCK(p);
854 	PROC_SLOCK(p);
855 }
856 
857 void
858 thread_suspend_one(struct thread *td)
859 {
860 	struct proc *p = td->td_proc;
861 
862 	PROC_SLOCK_ASSERT(p, MA_OWNED);
863 	THREAD_LOCK_ASSERT(td, MA_OWNED);
864 	KASSERT(!TD_IS_SUSPENDED(td), ("already suspended"));
865 	p->p_suspcount++;
866 	td->td_flags &= ~TDF_NEEDSUSPCHK;
867 	TD_SET_SUSPENDED(td);
868 	sched_sleep(td, 0);
869 }
870 
871 int
872 thread_unsuspend_one(struct thread *td)
873 {
874 	struct proc *p = td->td_proc;
875 
876 	PROC_SLOCK_ASSERT(p, MA_OWNED);
877 	THREAD_LOCK_ASSERT(td, MA_OWNED);
878 	KASSERT(TD_IS_SUSPENDED(td), ("Thread not suspended"));
879 	TD_CLR_SUSPENDED(td);
880 	p->p_suspcount--;
881 	return (setrunnable(td));
882 }
883 
884 /*
885  * Allow all threads blocked by single threading to continue running.
886  */
887 void
888 thread_unsuspend(struct proc *p)
889 {
890 	struct thread *td;
891 	int wakeup_swapper;
892 
893 	PROC_LOCK_ASSERT(p, MA_OWNED);
894 	PROC_SLOCK_ASSERT(p, MA_OWNED);
895 	wakeup_swapper = 0;
896 	if (!P_SHOULDSTOP(p)) {
897                 FOREACH_THREAD_IN_PROC(p, td) {
898 			thread_lock(td);
899 			if (TD_IS_SUSPENDED(td)) {
900 				wakeup_swapper |= thread_unsuspend_one(td);
901 			}
902 			thread_unlock(td);
903 		}
904 	} else if ((P_SHOULDSTOP(p) == P_STOPPED_SINGLE) &&
905 	    (p->p_numthreads == p->p_suspcount)) {
906 		/*
907 		 * Stopping everything also did the job for the single
908 		 * threading request. Now we've downgraded to single-threaded,
909 		 * let it continue.
910 		 */
911 		thread_lock(p->p_singlethread);
912 		wakeup_swapper = thread_unsuspend_one(p->p_singlethread);
913 		thread_unlock(p->p_singlethread);
914 	}
915 	if (wakeup_swapper)
916 		kick_proc0();
917 }
918 
919 /*
920  * End the single threading mode..
921  */
922 void
923 thread_single_end(void)
924 {
925 	struct thread *td;
926 	struct proc *p;
927 	int wakeup_swapper;
928 
929 	td = curthread;
930 	p = td->td_proc;
931 	PROC_LOCK_ASSERT(p, MA_OWNED);
932 	p->p_flag &= ~(P_STOPPED_SINGLE | P_SINGLE_EXIT | P_SINGLE_BOUNDARY);
933 	PROC_SLOCK(p);
934 	p->p_singlethread = NULL;
935 	wakeup_swapper = 0;
936 	/*
937 	 * If there are other threads they may now run,
938 	 * unless of course there is a blanket 'stop order'
939 	 * on the process. The single threader must be allowed
940 	 * to continue however as this is a bad place to stop.
941 	 */
942 	if ((p->p_numthreads != 1) && (!P_SHOULDSTOP(p))) {
943                 FOREACH_THREAD_IN_PROC(p, td) {
944 			thread_lock(td);
945 			if (TD_IS_SUSPENDED(td)) {
946 				wakeup_swapper |= thread_unsuspend_one(td);
947 			}
948 			thread_unlock(td);
949 		}
950 	}
951 	PROC_SUNLOCK(p);
952 	if (wakeup_swapper)
953 		kick_proc0();
954 }
955 
956 struct thread *
957 thread_find(struct proc *p, lwpid_t tid)
958 {
959 	struct thread *td;
960 
961 	PROC_LOCK_ASSERT(p, MA_OWNED);
962 	FOREACH_THREAD_IN_PROC(p, td) {
963 		if (td->td_tid == tid)
964 			break;
965 	}
966 	return (td);
967 }
968 
969 /* Locate a thread by number; return with proc lock held. */
970 struct thread *
971 tdfind(lwpid_t tid, pid_t pid)
972 {
973 #define RUN_THRESH	16
974 	struct thread *td;
975 	int run = 0;
976 
977 	rw_rlock(&tidhash_lock);
978 	LIST_FOREACH(td, TIDHASH(tid), td_hash) {
979 		if (td->td_tid == tid) {
980 			if (pid != -1 && td->td_proc->p_pid != pid) {
981 				td = NULL;
982 				break;
983 			}
984 			PROC_LOCK(td->td_proc);
985 			if (td->td_proc->p_state == PRS_NEW) {
986 				PROC_UNLOCK(td->td_proc);
987 				td = NULL;
988 				break;
989 			}
990 			if (run > RUN_THRESH) {
991 				if (rw_try_upgrade(&tidhash_lock)) {
992 					LIST_REMOVE(td, td_hash);
993 					LIST_INSERT_HEAD(TIDHASH(td->td_tid),
994 						td, td_hash);
995 					rw_wunlock(&tidhash_lock);
996 					return (td);
997 				}
998 			}
999 			break;
1000 		}
1001 		run++;
1002 	}
1003 	rw_runlock(&tidhash_lock);
1004 	return (td);
1005 }
1006 
1007 void
1008 tidhash_add(struct thread *td)
1009 {
1010 	rw_wlock(&tidhash_lock);
1011 	LIST_INSERT_HEAD(TIDHASH(td->td_tid), td, td_hash);
1012 	rw_wunlock(&tidhash_lock);
1013 }
1014 
1015 void
1016 tidhash_remove(struct thread *td)
1017 {
1018 	rw_wlock(&tidhash_lock);
1019 	LIST_REMOVE(td, td_hash);
1020 	rw_wunlock(&tidhash_lock);
1021 }
1022