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 <sys/cdefs.h>
30 __FBSDID("$FreeBSD: releng/6.2/sys/kern/kern_thread.c 164286 2006-11-14 20:42:41Z cvs2svn $");
31
32 #include <sys/param.h>
33 #include <sys/systm.h>
34 #include <sys/kernel.h>
35 #include <sys/lock.h>
36 #include <sys/mutex.h>
37 #include <sys/proc.h>
38 #include <sys/resourcevar.h>
39 #include <sys/smp.h>
40 #include <sys/sysctl.h>
41 #include <sys/sched.h>
42 #include <sys/sleepqueue.h>
43 #include <sys/turnstile.h>
44 #include <sys/ktr.h>
45 #include <sys/umtx.h>
46
47 #include <security/audit/audit.h>
48
49 #include <vm/vm.h>
50 #include <vm/vm_extern.h>
51 #include <vm/uma.h>
52
53 /*
54 * KSEGRP related storage.
55 */
56 static uma_zone_t ksegrp_zone;
57 static uma_zone_t thread_zone;
58
59 /* DEBUG ONLY */
60 SYSCTL_NODE(_kern, OID_AUTO, threads, CTLFLAG_RW, 0, "thread allocation");
61 static int thread_debug = 0;
62 SYSCTL_INT(_kern_threads, OID_AUTO, debug, CTLFLAG_RW,
63 &thread_debug, 0, "thread debug");
64
65 int max_threads_per_proc = 1500;
66 SYSCTL_INT(_kern_threads, OID_AUTO, max_threads_per_proc, CTLFLAG_RW,
67 &max_threads_per_proc, 0, "Limit on threads per proc");
68
69 int max_groups_per_proc = 1500;
70 SYSCTL_INT(_kern_threads, OID_AUTO, max_groups_per_proc, CTLFLAG_RW,
71 &max_groups_per_proc, 0, "Limit on thread groups per proc");
72
73 int max_threads_hits;
74 SYSCTL_INT(_kern_threads, OID_AUTO, max_threads_hits, CTLFLAG_RD,
75 &max_threads_hits, 0, "");
76
77 int virtual_cpu;
78
79 TAILQ_HEAD(, thread) zombie_threads = TAILQ_HEAD_INITIALIZER(zombie_threads);
80 TAILQ_HEAD(, ksegrp) zombie_ksegrps = TAILQ_HEAD_INITIALIZER(zombie_ksegrps);
81 struct mtx kse_zombie_lock;
82 MTX_SYSINIT(kse_zombie_lock, &kse_zombie_lock, "kse zombie lock", MTX_SPIN);
83
84 static int
85 sysctl_kse_virtual_cpu(SYSCTL_HANDLER_ARGS)
86 {
87 int error, new_val;
88 int def_val;
89
90 def_val = mp_ncpus;
91 if (virtual_cpu == 0)
92 new_val = def_val;
93 else
94 new_val = virtual_cpu;
95 error = sysctl_handle_int(oidp, &new_val, 0, req);
96 if (error != 0 || req->newptr == NULL)
97 return (error);
98 if (new_val < 0)
99 return (EINVAL);
100 virtual_cpu = new_val;
101 return (0);
102 }
103
104 /* DEBUG ONLY */
105 SYSCTL_PROC(_kern_threads, OID_AUTO, virtual_cpu, CTLTYPE_INT|CTLFLAG_RW,
106 0, sizeof(virtual_cpu), sysctl_kse_virtual_cpu, "I",
107 "debug virtual cpus");
108
109 struct mtx tid_lock;
110 static struct unrhdr *tid_unrhdr;
111
112 /*
113 * Prepare a thread for use.
114 */
115 static int
116 thread_ctor(void *mem, int size, void *arg, int flags)
117 {
118 struct thread *td;
119
120 td = (struct thread *)mem;
121 td->td_state = TDS_INACTIVE;
122 td->td_oncpu = NOCPU;
123
124 td->td_tid = alloc_unr(tid_unrhdr);
125
126 /*
127 * Note that td_critnest begins life as 1 because the thread is not
128 * running and is thereby implicitly waiting to be on the receiving
129 * end of a context switch. A context switch must occur inside a
130 * critical section, and in fact, includes hand-off of the sched_lock.
131 * After a context switch to a newly created thread, it will release
132 * sched_lock for the first time, and its td_critnest will hit 0 for
133 * the first time. This happens on the far end of a context switch,
134 * and when it context switches away from itself, it will in fact go
135 * back into a critical section, and hand off the sched lock to the
136 * next thread.
137 */
138 td->td_critnest = 1;
139
140 #ifdef AUDIT
141 audit_thread_alloc(td);
142 #endif
143 return (0);
144 }
145
146 /*
147 * Reclaim a thread after use.
148 */
149 static void
150 thread_dtor(void *mem, int size, void *arg)
151 {
152 struct thread *td;
153
154 td = (struct thread *)mem;
155
156 #ifdef INVARIANTS
157 /* Verify that this thread is in a safe state to free. */
158 switch (td->td_state) {
159 case TDS_INHIBITED:
160 case TDS_RUNNING:
161 case TDS_CAN_RUN:
162 case TDS_RUNQ:
163 /*
164 * We must never unlink a thread that is in one of
165 * these states, because it is currently active.
166 */
167 panic("bad state for thread unlinking");
168 /* NOTREACHED */
169 case TDS_INACTIVE:
170 break;
171 default:
172 panic("bad thread state");
173 /* NOTREACHED */
174 }
175 #endif
176 #ifdef AUDIT
177 audit_thread_free(td);
178 #endif
179 free_unr(tid_unrhdr, td->td_tid);
180 sched_newthread(td);
181 }
182
183 /*
184 * Initialize type-stable parts of a thread (when newly created).
185 */
186 static int
187 thread_init(void *mem, int size, int flags)
188 {
189 struct thread *td;
190
191 td = (struct thread *)mem;
192
193 vm_thread_new(td, 0);
194 cpu_thread_setup(td);
195 td->td_sleepqueue = sleepq_alloc();
196 td->td_turnstile = turnstile_alloc();
197 td->td_umtxq = umtxq_alloc();
198 td->td_sched = (struct td_sched *)&td[1];
199 sched_newthread(td);
200 return (0);
201 }
202
203 /*
204 * Tear down type-stable parts of a thread (just before being discarded).
205 */
206 static void
207 thread_fini(void *mem, int size)
208 {
209 struct thread *td;
210
211 td = (struct thread *)mem;
212 turnstile_free(td->td_turnstile);
213 sleepq_free(td->td_sleepqueue);
214 umtxq_free(td->td_umtxq);
215 vm_thread_dispose(td);
216 }
217
218 /*
219 * Initialize type-stable parts of a ksegrp (when newly created).
220 */
221 static int
222 ksegrp_ctor(void *mem, int size, void *arg, int flags)
223 {
224 struct ksegrp *kg;
225
226 kg = (struct ksegrp *)mem;
227 bzero(mem, size);
228 kg->kg_sched = (struct kg_sched *)&kg[1];
229 return (0);
230 }
231
232 void
233 ksegrp_link(struct ksegrp *kg, struct proc *p)
234 {
235
236 TAILQ_INIT(&kg->kg_threads);
237 TAILQ_INIT(&kg->kg_runq); /* links with td_runq */
238 TAILQ_INIT(&kg->kg_upcalls); /* all upcall structure in ksegrp */
239 kg->kg_proc = p;
240 /*
241 * the following counters are in the -zero- section
242 * and may not need clearing
243 */
244 kg->kg_numthreads = 0;
245 kg->kg_numupcalls = 0;
246 /* link it in now that it's consistent */
247 p->p_numksegrps++;
248 TAILQ_INSERT_HEAD(&p->p_ksegrps, kg, kg_ksegrp);
249 }
250
251 /*
252 * Called from:
253 * thread-exit()
254 */
255 void
256 ksegrp_unlink(struct ksegrp *kg)
257 {
258 struct proc *p;
259
260 mtx_assert(&sched_lock, MA_OWNED);
261 KASSERT((kg->kg_numthreads == 0), ("ksegrp_unlink: residual threads"));
262 KASSERT((kg->kg_numupcalls == 0), ("ksegrp_unlink: residual upcalls"));
263
264 p = kg->kg_proc;
265 TAILQ_REMOVE(&p->p_ksegrps, kg, kg_ksegrp);
266 p->p_numksegrps--;
267 /*
268 * Aggregate stats from the KSE
269 */
270 if (p->p_procscopegrp == kg)
271 p->p_procscopegrp = NULL;
272 }
273
274 /*
275 * For a newly created process,
276 * link up all the structures and its initial threads etc.
277 * called from:
278 * {arch}/{arch}/machdep.c ia64_init(), init386() etc.
279 * proc_dtor() (should go away)
280 * proc_init()
281 */
282 void
283 proc_linkup(struct proc *p, struct ksegrp *kg, struct thread *td)
284 {
285
286 TAILQ_INIT(&p->p_ksegrps); /* all ksegrps in proc */
287 TAILQ_INIT(&p->p_threads); /* all threads in proc */
288 TAILQ_INIT(&p->p_suspended); /* Threads suspended */
289 p->p_numksegrps = 0;
290 p->p_numthreads = 0;
291
292 ksegrp_link(kg, p);
293 thread_link(td, kg);
294 }
295
296 /*
297 * Initialize global thread allocation resources.
298 */
299 void
300 threadinit(void)
301 {
302
303 mtx_init(&tid_lock, "TID lock", NULL, MTX_DEF);
304 tid_unrhdr = new_unrhdr(PID_MAX + 1, INT_MAX, &tid_lock);
305
306 thread_zone = uma_zcreate("THREAD", sched_sizeof_thread(),
307 thread_ctor, thread_dtor, thread_init, thread_fini,
308 UMA_ALIGN_CACHE, 0);
309 ksegrp_zone = uma_zcreate("KSEGRP", sched_sizeof_ksegrp(),
310 ksegrp_ctor, NULL, NULL, NULL,
311 UMA_ALIGN_CACHE, 0);
312 kseinit(); /* set up kse specific stuff e.g. upcall zone*/
313 }
314
315 /*
316 * Stash an embarasingly extra thread into the zombie thread queue.
317 */
318 void
319 thread_stash(struct thread *td)
320 {
321 mtx_lock_spin(&kse_zombie_lock);
322 TAILQ_INSERT_HEAD(&zombie_threads, td, td_runq);
323 mtx_unlock_spin(&kse_zombie_lock);
324 }
325
326 /*
327 * Stash an embarasingly extra ksegrp into the zombie ksegrp queue.
328 */
329 void
330 ksegrp_stash(struct ksegrp *kg)
331 {
332 mtx_lock_spin(&kse_zombie_lock);
333 TAILQ_INSERT_HEAD(&zombie_ksegrps, kg, kg_ksegrp);
334 mtx_unlock_spin(&kse_zombie_lock);
335 }
336
337 /*
338 * Reap zombie kse resource.
339 */
340 void
341 thread_reap(void)
342 {
343 struct thread *td_first, *td_next;
344 struct ksegrp *kg_first, * kg_next;
345
346 /*
347 * Don't even bother to lock if none at this instant,
348 * we really don't care about the next instant..
349 */
350 if ((!TAILQ_EMPTY(&zombie_threads))
351 || (!TAILQ_EMPTY(&zombie_ksegrps))) {
352 mtx_lock_spin(&kse_zombie_lock);
353 td_first = TAILQ_FIRST(&zombie_threads);
354 kg_first = TAILQ_FIRST(&zombie_ksegrps);
355 if (td_first)
356 TAILQ_INIT(&zombie_threads);
357 if (kg_first)
358 TAILQ_INIT(&zombie_ksegrps);
359 mtx_unlock_spin(&kse_zombie_lock);
360 while (td_first) {
361 td_next = TAILQ_NEXT(td_first, td_runq);
362 if (td_first->td_ucred)
363 crfree(td_first->td_ucred);
364 thread_free(td_first);
365 td_first = td_next;
366 }
367 while (kg_first) {
368 kg_next = TAILQ_NEXT(kg_first, kg_ksegrp);
369 ksegrp_free(kg_first);
370 kg_first = kg_next;
371 }
372 /*
373 * there will always be a thread on the list if one of these
374 * is there.
375 */
376 kse_GC();
377 }
378 }
379
380 /*
381 * Allocate a ksegrp.
382 */
383 struct ksegrp *
384 ksegrp_alloc(void)
385 {
386 return (uma_zalloc(ksegrp_zone, M_WAITOK));
387 }
388
389 /*
390 * Allocate a thread.
391 */
392 struct thread *
393 thread_alloc(void)
394 {
395 thread_reap(); /* check if any zombies to get */
396 return (uma_zalloc(thread_zone, M_WAITOK));
397 }
398
399 /*
400 * Deallocate a ksegrp.
401 */
402 void
403 ksegrp_free(struct ksegrp *td)
404 {
405 uma_zfree(ksegrp_zone, td);
406 }
407
408 /*
409 * Deallocate a thread.
410 */
411 void
412 thread_free(struct thread *td)
413 {
414
415 cpu_thread_clean(td);
416 uma_zfree(thread_zone, td);
417 }
418
419 /*
420 * Discard the current thread and exit from its context.
421 * Always called with scheduler locked.
422 *
423 * Because we can't free a thread while we're operating under its context,
424 * push the current thread into our CPU's deadthread holder. This means
425 * we needn't worry about someone else grabbing our context before we
426 * do a cpu_throw(). This may not be needed now as we are under schedlock.
427 * Maybe we can just do a thread_stash() as thr_exit1 does.
428 */
429 /* XXX
430 * libthr expects its thread exit to return for the last
431 * thread, meaning that the program is back to non-threaded
432 * mode I guess. Because we do this (cpu_throw) unconditionally
433 * here, they have their own version of it. (thr_exit1())
434 * that doesn't do it all if this was the last thread.
435 * It is also called from thread_suspend_check().
436 * Of course in the end, they end up coming here through exit1
437 * anyhow.. After fixing 'thr' to play by the rules we should be able
438 * to merge these two functions together.
439 *
440 * called from:
441 * exit1()
442 * kse_exit()
443 * thr_exit()
444 * thread_user_enter()
445 * thread_userret()
446 * thread_suspend_check()
447 */
448 void
449 thread_exit(void)
450 {
451 struct bintime new_switchtime;
452 struct thread *td;
453 struct proc *p;
454 struct ksegrp *kg;
455
456 td = curthread;
457 kg = td->td_ksegrp;
458 p = td->td_proc;
459
460 mtx_assert(&sched_lock, MA_OWNED);
461 mtx_assert(&Giant, MA_NOTOWNED);
462 PROC_LOCK_ASSERT(p, MA_OWNED);
463 KASSERT(p != NULL, ("thread exiting without a process"));
464 KASSERT(kg != NULL, ("thread exiting without a kse group"));
465 CTR3(KTR_PROC, "thread_exit: thread %p (pid %ld, %s)", td,
466 (long)p->p_pid, p->p_comm);
467
468 #ifdef AUDIT
469 AUDIT_SYSCALL_EXIT(0, td);
470 #endif
471
472 if (td->td_standin != NULL) {
473 /*
474 * Note that we don't need to free the cred here as it
475 * is done in thread_reap().
476 */
477 thread_stash(td->td_standin);
478 td->td_standin = NULL;
479 }
480
481 /*
482 * drop FPU & debug register state storage, or any other
483 * architecture specific resources that
484 * would not be on a new untouched process.
485 */
486 cpu_thread_exit(td); /* XXXSMP */
487
488 /*
489 * The thread is exiting. scheduler can release its stuff
490 * and collect stats etc.
491 */
492 sched_thread_exit(td);
493
494 /* Do the same timestamp bookkeeping that mi_switch() would do. */
495 binuptime(&new_switchtime);
496 bintime_add(&p->p_rux.rux_runtime, &new_switchtime);
497 bintime_sub(&p->p_rux.rux_runtime, PCPU_PTR(switchtime));
498 PCPU_SET(switchtime, new_switchtime);
499 PCPU_SET(switchticks, ticks);
500 cnt.v_swtch++;
501
502 /* Add our usage into the usage of all our children. */
503 if (p->p_numthreads == 1)
504 ruadd(p->p_ru, &p->p_rux, &p->p_stats->p_cru, &p->p_crux);
505
506 /*
507 * The last thread is left attached to the process
508 * So that the whole bundle gets recycled. Skip
509 * all this stuff if we never had threads.
510 * EXIT clears all sign of other threads when
511 * it goes to single threading, so the last thread always
512 * takes the short path.
513 */
514 if (p->p_flag & P_HADTHREADS) {
515 if (p->p_numthreads > 1) {
516 thread_unlink(td);
517
518 /* XXX first arg not used in 4BSD or ULE */
519 sched_exit_thread(FIRST_THREAD_IN_PROC(p), td);
520
521 /*
522 * The test below is NOT true if we are the
523 * sole exiting thread. P_STOPPED_SNGL is unset
524 * in exit1() after it is the only survivor.
525 */
526 if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE) {
527 if (p->p_numthreads == p->p_suspcount) {
528 thread_unsuspend_one(p->p_singlethread);
529 }
530 }
531
532 /*
533 * Because each upcall structure has an owner thread,
534 * owner thread exits only when process is in exiting
535 * state, so upcall to userland is no longer needed,
536 * deleting upcall structure is safe here.
537 * So when all threads in a group is exited, all upcalls
538 * in the group should be automatically freed.
539 * XXXKSE This is a KSE thing and should be exported
540 * there somehow.
541 */
542 upcall_remove(td);
543
544 /*
545 * If the thread we unlinked above was the last one,
546 * then this ksegrp should go away too.
547 */
548 if (kg->kg_numthreads == 0) {
549 /*
550 * let the scheduler know about this in case
551 * it needs to recover stats or resources.
552 * Theoretically we could let
553 * sched_exit_ksegrp() do the equivalent of
554 * setting the concurrency to 0
555 * but don't do it yet to avoid changing
556 * the existing scheduler code until we
557 * are ready.
558 * We supply a random other ksegrp
559 * as the recipient of any built up
560 * cpu usage etc. (If the scheduler wants it).
561 * XXXKSE
562 * This is probably not fair so think of
563 * a better answer.
564 */
565 sched_exit_ksegrp(FIRST_KSEGRP_IN_PROC(p), td);
566 sched_set_concurrency(kg, 0); /* XXX TEMP */
567 ksegrp_unlink(kg);
568 ksegrp_stash(kg);
569 }
570 PROC_UNLOCK(p);
571 td->td_ksegrp = NULL;
572 PCPU_SET(deadthread, td);
573 } else {
574 /*
575 * The last thread is exiting.. but not through exit()
576 * what should we do?
577 * Theoretically this can't happen
578 * exit1() - clears threading flags before coming here
579 * kse_exit() - treats last thread specially
580 * thr_exit() - treats last thread specially
581 * thread_user_enter() - only if more exist
582 * thread_userret() - only if more exist
583 * thread_suspend_check() - only if more exist
584 */
585 panic ("thread_exit: Last thread exiting on its own");
586 }
587 } else {
588 /*
589 * non threaded process comes here.
590 * This includes an EX threaded process that is coming
591 * here via exit1(). (exit1 dethreads the proc first).
592 */
593 PROC_UNLOCK(p);
594 }
595 td->td_state = TDS_INACTIVE;
596 CTR1(KTR_PROC, "thread_exit: cpu_throw() thread %p", td);
597 cpu_throw(td, choosethread());
598 panic("I'm a teapot!");
599 /* NOTREACHED */
600 }
601
602 /*
603 * Do any thread specific cleanups that may be needed in wait()
604 * called with Giant, proc and schedlock not held.
605 */
606 void
607 thread_wait(struct proc *p)
608 {
609 struct thread *td;
610
611 mtx_assert(&Giant, MA_NOTOWNED);
612 KASSERT((p->p_numthreads == 1), ("Multiple threads in wait1()"));
613 KASSERT((p->p_numksegrps == 1), ("Multiple ksegrps in wait1()"));
614 FOREACH_THREAD_IN_PROC(p, td) {
615 if (td->td_standin != NULL) {
616 if (td->td_standin->td_ucred != NULL) {
617 crfree(td->td_standin->td_ucred);
618 td->td_standin->td_ucred = NULL;
619 }
620 thread_free(td->td_standin);
621 td->td_standin = NULL;
622 }
623 cpu_thread_clean(td);
624 crfree(td->td_ucred);
625 }
626 thread_reap(); /* check for zombie threads etc. */
627 }
628
629 /*
630 * Link a thread to a process.
631 * set up anything that needs to be initialized for it to
632 * be used by the process.
633 *
634 * Note that we do not link to the proc's ucred here.
635 * The thread is linked as if running but no KSE assigned.
636 * Called from:
637 * proc_linkup()
638 * thread_schedule_upcall()
639 * thr_create()
640 */
641 void
642 thread_link(struct thread *td, struct ksegrp *kg)
643 {
644 struct proc *p;
645
646 p = kg->kg_proc;
647 td->td_state = TDS_INACTIVE;
648 td->td_proc = p;
649 td->td_ksegrp = kg;
650 td->td_flags = 0;
651 td->td_kflags = 0;
652
653 LIST_INIT(&td->td_contested);
654 callout_init(&td->td_slpcallout, CALLOUT_MPSAFE);
655 TAILQ_INSERT_HEAD(&p->p_threads, td, td_plist);
656 TAILQ_INSERT_HEAD(&kg->kg_threads, td, td_kglist);
657 p->p_numthreads++;
658 kg->kg_numthreads++;
659 }
660
661 /*
662 * Convert a process with one thread to an unthreaded process.
663 * Called from:
664 * thread_single(exit) (called from execve and exit)
665 * kse_exit() XXX may need cleaning up wrt KSE stuff
666 */
667 void
668 thread_unthread(struct thread *td)
669 {
670 struct proc *p = td->td_proc;
671
672 KASSERT((p->p_numthreads == 1), ("Unthreading with >1 threads"));
673 upcall_remove(td);
674 p->p_flag &= ~(P_SA|P_HADTHREADS);
675 td->td_mailbox = NULL;
676 td->td_pflags &= ~(TDP_SA | TDP_CAN_UNBIND);
677 if (td->td_standin != NULL) {
678 thread_stash(td->td_standin);
679 td->td_standin = NULL;
680 }
681 sched_set_concurrency(td->td_ksegrp, 1);
682 }
683
684 /*
685 * Called from:
686 * thread_exit()
687 */
688 void
689 thread_unlink(struct thread *td)
690 {
691 struct proc *p = td->td_proc;
692 struct ksegrp *kg = td->td_ksegrp;
693
694 mtx_assert(&sched_lock, MA_OWNED);
695 TAILQ_REMOVE(&p->p_threads, td, td_plist);
696 p->p_numthreads--;
697 TAILQ_REMOVE(&kg->kg_threads, td, td_kglist);
698 kg->kg_numthreads--;
699 /* could clear a few other things here */
700 /* Must NOT clear links to proc and ksegrp! */
701 }
702
703 /*
704 * Enforce single-threading.
705 *
706 * Returns 1 if the caller must abort (another thread is waiting to
707 * exit the process or similar). Process is locked!
708 * Returns 0 when you are successfully the only thread running.
709 * A process has successfully single threaded in the suspend mode when
710 * There are no threads in user mode. Threads in the kernel must be
711 * allowed to continue until they get to the user boundary. They may even
712 * copy out their return values and data before suspending. They may however be
713 * accellerated in reaching the user boundary as we will wake up
714 * any sleeping threads that are interruptable. (PCATCH).
715 */
716 int
717 thread_single(int mode)
718 {
719 struct thread *td;
720 struct thread *td2;
721 struct proc *p;
722 int remaining;
723
724 td = curthread;
725 p = td->td_proc;
726 mtx_assert(&Giant, MA_NOTOWNED);
727 PROC_LOCK_ASSERT(p, MA_OWNED);
728 KASSERT((td != NULL), ("curthread is NULL"));
729
730 if ((p->p_flag & P_HADTHREADS) == 0)
731 return (0);
732
733 /* Is someone already single threading? */
734 if (p->p_singlethread != NULL && p->p_singlethread != td)
735 return (1);
736
737 if (mode == SINGLE_EXIT) {
738 p->p_flag |= P_SINGLE_EXIT;
739 p->p_flag &= ~P_SINGLE_BOUNDARY;
740 } else {
741 p->p_flag &= ~P_SINGLE_EXIT;
742 if (mode == SINGLE_BOUNDARY)
743 p->p_flag |= P_SINGLE_BOUNDARY;
744 else
745 p->p_flag &= ~P_SINGLE_BOUNDARY;
746 }
747 p->p_flag |= P_STOPPED_SINGLE;
748 mtx_lock_spin(&sched_lock);
749 p->p_singlethread = td;
750 if (mode == SINGLE_EXIT)
751 remaining = p->p_numthreads;
752 else if (mode == SINGLE_BOUNDARY)
753 remaining = p->p_numthreads - p->p_boundary_count;
754 else
755 remaining = p->p_numthreads - p->p_suspcount;
756 while (remaining != 1) {
757 if (P_SHOULDSTOP(p) != P_STOPPED_SINGLE)
758 goto stopme;
759 FOREACH_THREAD_IN_PROC(p, td2) {
760 if (td2 == td)
761 continue;
762 td2->td_flags |= TDF_ASTPENDING;
763 if (TD_IS_INHIBITED(td2)) {
764 switch (mode) {
765 case SINGLE_EXIT:
766 if (td->td_flags & TDF_DBSUSPEND)
767 td->td_flags &= ~TDF_DBSUSPEND;
768 if (TD_IS_SUSPENDED(td2))
769 thread_unsuspend_one(td2);
770 if (TD_ON_SLEEPQ(td2) &&
771 (td2->td_flags & TDF_SINTR))
772 sleepq_abort(td2, EINTR);
773 break;
774 case SINGLE_BOUNDARY:
775 if (TD_IS_SUSPENDED(td2) &&
776 !(td2->td_flags & TDF_BOUNDARY))
777 thread_unsuspend_one(td2);
778 if (TD_ON_SLEEPQ(td2) &&
779 (td2->td_flags & TDF_SINTR))
780 sleepq_abort(td2, ERESTART);
781 break;
782 default:
783 if (TD_IS_SUSPENDED(td2))
784 continue;
785 /*
786 * maybe other inhibitted states too?
787 */
788 if ((td2->td_flags & TDF_SINTR) &&
789 (td2->td_inhibitors &
790 (TDI_SLEEPING | TDI_SWAPPED)))
791 thread_suspend_one(td2);
792 break;
793 }
794 }
795 #ifdef SMP
796 else if (TD_IS_RUNNING(td2) && td != td2) {
797 forward_signal(td2);
798 }
799 #endif
800 }
801 if (mode == SINGLE_EXIT)
802 remaining = p->p_numthreads;
803 else if (mode == SINGLE_BOUNDARY)
804 remaining = p->p_numthreads - p->p_boundary_count;
805 else
806 remaining = p->p_numthreads - p->p_suspcount;
807
808 /*
809 * Maybe we suspended some threads.. was it enough?
810 */
811 if (remaining == 1)
812 break;
813
814 stopme:
815 /*
816 * Wake us up when everyone else has suspended.
817 * In the mean time we suspend as well.
818 */
819 thread_stopped(p);
820 thread_suspend_one(td);
821 PROC_UNLOCK(p);
822 mi_switch(SW_VOL, NULL);
823 mtx_unlock_spin(&sched_lock);
824 PROC_LOCK(p);
825 mtx_lock_spin(&sched_lock);
826 if (mode == SINGLE_EXIT)
827 remaining = p->p_numthreads;
828 else if (mode == SINGLE_BOUNDARY)
829 remaining = p->p_numthreads - p->p_boundary_count;
830 else
831 remaining = p->p_numthreads - p->p_suspcount;
832 }
833 if (mode == SINGLE_EXIT) {
834 /*
835 * We have gotten rid of all the other threads and we
836 * are about to either exit or exec. In either case,
837 * we try our utmost to revert to being a non-threaded
838 * process.
839 */
840 p->p_singlethread = NULL;
841 p->p_flag &= ~(P_STOPPED_SINGLE | P_SINGLE_EXIT);
842 thread_unthread(td);
843 }
844 mtx_unlock_spin(&sched_lock);
845 return (0);
846 }
847
848 /*
849 * Called in from locations that can safely check to see
850 * whether we have to suspend or at least throttle for a
851 * single-thread event (e.g. fork).
852 *
853 * Such locations include userret().
854 * If the "return_instead" argument is non zero, the thread must be able to
855 * accept 0 (caller may continue), or 1 (caller must abort) as a result.
856 *
857 * The 'return_instead' argument tells the function if it may do a
858 * thread_exit() or suspend, or whether the caller must abort and back
859 * out instead.
860 *
861 * If the thread that set the single_threading request has set the
862 * P_SINGLE_EXIT bit in the process flags then this call will never return
863 * if 'return_instead' is false, but will exit.
864 *
865 * P_SINGLE_EXIT | return_instead == 0| return_instead != 0
866 *---------------+--------------------+---------------------
867 * 0 | returns 0 | returns 0 or 1
868 * | when ST ends | immediatly
869 *---------------+--------------------+---------------------
870 * 1 | thread exits | returns 1
871 * | | immediatly
872 * 0 = thread_exit() or suspension ok,
873 * other = return error instead of stopping the thread.
874 *
875 * While a full suspension is under effect, even a single threading
876 * thread would be suspended if it made this call (but it shouldn't).
877 * This call should only be made from places where
878 * thread_exit() would be safe as that may be the outcome unless
879 * return_instead is set.
880 */
881 int
882 thread_suspend_check(int return_instead)
883 {
884 struct thread *td;
885 struct proc *p;
886
887 td = curthread;
888 p = td->td_proc;
889 mtx_assert(&Giant, MA_NOTOWNED);
890 PROC_LOCK_ASSERT(p, MA_OWNED);
891 while (P_SHOULDSTOP(p) ||
892 ((p->p_flag & P_TRACED) && (td->td_flags & TDF_DBSUSPEND))) {
893 if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE) {
894 KASSERT(p->p_singlethread != NULL,
895 ("singlethread not set"));
896 /*
897 * The only suspension in action is a
898 * single-threading. Single threader need not stop.
899 * XXX Should be safe to access unlocked
900 * as it can only be set to be true by us.
901 */
902 if (p->p_singlethread == td)
903 return (0); /* Exempt from stopping. */
904 }
905 if ((p->p_flag & P_SINGLE_EXIT) && return_instead)
906 return (EINTR);
907
908 /* Should we goto user boundary if we didn't come from there? */
909 if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE &&
910 (p->p_flag & P_SINGLE_BOUNDARY) && return_instead)
911 return (ERESTART);
912
913 mtx_lock_spin(&sched_lock);
914 thread_stopped(p);
915 /*
916 * If the process is waiting for us to exit,
917 * this thread should just suicide.
918 * Assumes that P_SINGLE_EXIT implies P_STOPPED_SINGLE.
919 */
920 if ((p->p_flag & P_SINGLE_EXIT) && (p->p_singlethread != td))
921 thread_exit();
922
923 /*
924 * When a thread suspends, it just
925 * moves to the processes's suspend queue
926 * and stays there.
927 */
928 thread_suspend_one(td);
929 if (return_instead == 0) {
930 p->p_boundary_count++;
931 td->td_flags |= TDF_BOUNDARY;
932 }
933 if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE) {
934 if (p->p_numthreads == p->p_suspcount)
935 thread_unsuspend_one(p->p_singlethread);
936 }
937 PROC_UNLOCK(p);
938 mi_switch(SW_INVOL, NULL);
939 if (return_instead == 0) {
940 p->p_boundary_count--;
941 td->td_flags &= ~TDF_BOUNDARY;
942 }
943 mtx_unlock_spin(&sched_lock);
944 PROC_LOCK(p);
945 }
946 return (0);
947 }
948
949 void
950 thread_suspend_one(struct thread *td)
951 {
952 struct proc *p = td->td_proc;
953
954 mtx_assert(&sched_lock, MA_OWNED);
955 PROC_LOCK_ASSERT(p, MA_OWNED);
956 KASSERT(!TD_IS_SUSPENDED(td), ("already suspended"));
957 p->p_suspcount++;
958 TD_SET_SUSPENDED(td);
959 TAILQ_INSERT_TAIL(&p->p_suspended, td, td_runq);
960 }
961
962 void
963 thread_unsuspend_one(struct thread *td)
964 {
965 struct proc *p = td->td_proc;
966
967 mtx_assert(&sched_lock, MA_OWNED);
968 PROC_LOCK_ASSERT(p, MA_OWNED);
969 TAILQ_REMOVE(&p->p_suspended, td, td_runq);
970 TD_CLR_SUSPENDED(td);
971 p->p_suspcount--;
972 setrunnable(td);
973 }
974
975 /*
976 * Allow all threads blocked by single threading to continue running.
977 */
978 void
979 thread_unsuspend(struct proc *p)
980 {
981 struct thread *td;
982
983 mtx_assert(&sched_lock, MA_OWNED);
984 PROC_LOCK_ASSERT(p, MA_OWNED);
985 if (!P_SHOULDSTOP(p)) {
986 while ((td = TAILQ_FIRST(&p->p_suspended))) {
987 thread_unsuspend_one(td);
988 }
989 } else if ((P_SHOULDSTOP(p) == P_STOPPED_SINGLE) &&
990 (p->p_numthreads == p->p_suspcount)) {
991 /*
992 * Stopping everything also did the job for the single
993 * threading request. Now we've downgraded to single-threaded,
994 * let it continue.
995 */
996 thread_unsuspend_one(p->p_singlethread);
997 }
998 }
999
1000 /*
1001 * End the single threading mode..
1002 */
1003 void
1004 thread_single_end(void)
1005 {
1006 struct thread *td;
1007 struct proc *p;
1008
1009 td = curthread;
1010 p = td->td_proc;
1011 PROC_LOCK_ASSERT(p, MA_OWNED);
1012 p->p_flag &= ~(P_STOPPED_SINGLE | P_SINGLE_EXIT | P_SINGLE_BOUNDARY);
1013 mtx_lock_spin(&sched_lock);
1014 p->p_singlethread = NULL;
1015 p->p_procscopegrp = NULL;
1016 /*
1017 * If there are other threads they mey now run,
1018 * unless of course there is a blanket 'stop order'
1019 * on the process. The single threader must be allowed
1020 * to continue however as this is a bad place to stop.
1021 */
1022 if ((p->p_numthreads != 1) && (!P_SHOULDSTOP(p))) {
1023 while ((td = TAILQ_FIRST(&p->p_suspended))) {
1024 thread_unsuspend_one(td);
1025 }
1026 }
1027 mtx_unlock_spin(&sched_lock);
1028 }
1029
1030 /*
1031 * Called before going into an interruptible sleep to see if we have been
1032 * interrupted or requested to exit.
1033 */
1034 int
1035 thread_sleep_check(struct thread *td)
1036 {
1037 struct proc *p;
1038
1039 p = td->td_proc;
1040 mtx_assert(&sched_lock, MA_OWNED);
1041 if (p->p_flag & P_HADTHREADS) {
1042 if (p->p_singlethread != td) {
1043 if (p->p_flag & P_SINGLE_EXIT)
1044 return (EINTR);
1045 if (p->p_flag & P_SINGLE_BOUNDARY)
1046 return (ERESTART);
1047 }
1048 if (td->td_flags & TDF_INTERRUPT)
1049 return (td->td_intrval);
1050 }
1051 return (0);
1052 }
Cache object: 7f59e7c2ccde6f9fb10cd6ca3480772b
|