2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
4 * Copyright (C) 2001 Julian Elischer <julian@freebsd.org>.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
10 * 1. Redistributions of source code must retain the above copyright
11 * notice(s), this list of conditions and the following disclaimer as
12 * the first lines of this file unmodified other than the possible
13 * addition of one or more copyright notices.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice(s), this list of conditions and the following disclaimer in the
16 * documentation and/or other materials provided with the distribution.
18 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDER(S) ``AS IS'' AND ANY
19 * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
20 * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
21 * DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER(S) BE LIABLE FOR ANY
22 * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
23 * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
24 * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
25 * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH
31 #include "opt_witness.h"
32 #include "opt_hwpmc_hooks.h"
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/kernel.h>
41 #include <sys/mutex.h>
43 #include <sys/rangelock.h>
44 #include <sys/resourcevar.h>
47 #include <sys/sched.h>
48 #include <sys/sleepqueue.h>
49 #include <sys/selinfo.h>
50 #include <sys/syscallsubr.h>
51 #include <sys/sysent.h>
52 #include <sys/turnstile.h>
54 #include <sys/rwlock.h>
56 #include <sys/vmmeter.h>
57 #include <sys/cpuset.h>
59 #include <sys/pmckern.h>
62 #include <security/audit/audit.h>
65 #include <vm/vm_extern.h>
67 #include <sys/eventhandler.h>
70 * Asserts below verify the stability of struct thread and struct proc
71 * layout, as exposed by KBI to modules. On head, the KBI is allowed
72 * to drift, change to the structures must be accompanied by the
75 * On the stable branches after KBI freeze, conditions must not be
76 * violated. Typically new fields are moved to the end of the
80 _Static_assert(offsetof(struct thread, td_flags) == 0xfc,
81 "struct thread KBI td_flags");
82 _Static_assert(offsetof(struct thread, td_pflags) == 0x104,
83 "struct thread KBI td_pflags");
84 _Static_assert(offsetof(struct thread, td_frame) == 0x470,
85 "struct thread KBI td_frame");
86 _Static_assert(offsetof(struct thread, td_emuldata) == 0x528,
87 "struct thread KBI td_emuldata");
88 _Static_assert(offsetof(struct proc, p_flag) == 0xb0,
89 "struct proc KBI p_flag");
90 _Static_assert(offsetof(struct proc, p_pid) == 0xbc,
91 "struct proc KBI p_pid");
92 _Static_assert(offsetof(struct proc, p_filemon) == 0x3d0,
93 "struct proc KBI p_filemon");
94 _Static_assert(offsetof(struct proc, p_comm) == 0x3e4,
95 "struct proc KBI p_comm");
96 _Static_assert(offsetof(struct proc, p_emuldata) == 0x4b8,
97 "struct proc KBI p_emuldata");
100 _Static_assert(offsetof(struct thread, td_flags) == 0x98,
101 "struct thread KBI td_flags");
102 _Static_assert(offsetof(struct thread, td_pflags) == 0xa0,
103 "struct thread KBI td_pflags");
104 _Static_assert(offsetof(struct thread, td_frame) == 0x2e8,
105 "struct thread KBI td_frame");
106 _Static_assert(offsetof(struct thread, td_emuldata) == 0x334,
107 "struct thread KBI td_emuldata");
108 _Static_assert(offsetof(struct proc, p_flag) == 0x68,
109 "struct proc KBI p_flag");
110 _Static_assert(offsetof(struct proc, p_pid) == 0x74,
111 "struct proc KBI p_pid");
112 _Static_assert(offsetof(struct proc, p_filemon) == 0x27c,
113 "struct proc KBI p_filemon");
114 _Static_assert(offsetof(struct proc, p_comm) == 0x28c,
115 "struct proc KBI p_comm");
116 _Static_assert(offsetof(struct proc, p_emuldata) == 0x318,
117 "struct proc KBI p_emuldata");
120 SDT_PROVIDER_DECLARE(proc);
121 SDT_PROBE_DEFINE(proc, , , lwp__exit);
124 * thread related storage.
126 static uma_zone_t thread_zone;
128 TAILQ_HEAD(, thread) zombie_threads = TAILQ_HEAD_INITIALIZER(zombie_threads);
129 static struct mtx zombie_lock;
130 MTX_SYSINIT(zombie_lock, &zombie_lock, "zombie lock", MTX_SPIN);
132 static void thread_zombie(struct thread *);
133 static int thread_unsuspend_one(struct thread *td, struct proc *p,
136 #define TID_BUFFER_SIZE 1024
139 static struct unrhdr *tid_unrhdr;
140 static lwpid_t tid_buffer[TID_BUFFER_SIZE];
141 static int tid_head, tid_tail;
142 static MALLOC_DEFINE(M_TIDHASH, "tidhash", "thread hash");
144 struct tidhashhead *tidhashtbl;
146 struct rwlock tidhash_lock;
148 EVENTHANDLER_LIST_DEFINE(thread_ctor);
149 EVENTHANDLER_LIST_DEFINE(thread_dtor);
150 EVENTHANDLER_LIST_DEFINE(thread_init);
151 EVENTHANDLER_LIST_DEFINE(thread_fini);
158 tid = alloc_unr(tid_unrhdr);
162 if (tid_head == tid_tail) {
163 mtx_unlock(&tid_lock);
166 tid = tid_buffer[tid_head];
167 tid_head = (tid_head + 1) % TID_BUFFER_SIZE;
168 mtx_unlock(&tid_lock);
173 tid_free(lwpid_t tid)
175 lwpid_t tmp_tid = -1;
178 if ((tid_tail + 1) % TID_BUFFER_SIZE == tid_head) {
179 tmp_tid = tid_buffer[tid_head];
180 tid_head = (tid_head + 1) % TID_BUFFER_SIZE;
182 tid_buffer[tid_tail] = tid;
183 tid_tail = (tid_tail + 1) % TID_BUFFER_SIZE;
184 mtx_unlock(&tid_lock);
186 free_unr(tid_unrhdr, tmp_tid);
190 * Prepare a thread for use.
193 thread_ctor(void *mem, int size, void *arg, int flags)
197 td = (struct thread *)mem;
198 td->td_state = TDS_INACTIVE;
199 td->td_lastcpu = td->td_oncpu = NOCPU;
201 td->td_tid = tid_alloc();
204 * Note that td_critnest begins life as 1 because the thread is not
205 * running and is thereby implicitly waiting to be on the receiving
206 * end of a context switch.
209 td->td_lend_user_pri = PRI_MAX;
210 EVENTHANDLER_DIRECT_INVOKE(thread_ctor, td);
212 audit_thread_alloc(td);
214 umtx_thread_alloc(td);
219 * Reclaim a thread after use.
222 thread_dtor(void *mem, int size, void *arg)
226 td = (struct thread *)mem;
229 /* Verify that this thread is in a safe state to free. */
230 switch (td->td_state) {
236 * We must never unlink a thread that is in one of
237 * these states, because it is currently active.
239 panic("bad state for thread unlinking");
244 panic("bad thread state");
249 audit_thread_free(td);
251 /* Free all OSD associated to this thread. */
253 td_softdep_cleanup(td);
254 MPASS(td->td_su == NULL);
256 EVENTHANDLER_DIRECT_INVOKE(thread_dtor, td);
257 tid_free(td->td_tid);
261 * Initialize type-stable parts of a thread (when newly created).
264 thread_init(void *mem, int size, int flags)
268 td = (struct thread *)mem;
270 td->td_sleepqueue = sleepq_alloc();
271 td->td_turnstile = turnstile_alloc();
273 EVENTHANDLER_DIRECT_INVOKE(thread_init, td);
274 umtx_thread_init(td);
281 * Tear down type-stable parts of a thread (just before being discarded).
284 thread_fini(void *mem, int size)
288 td = (struct thread *)mem;
289 EVENTHANDLER_DIRECT_INVOKE(thread_fini, td);
290 rlqentry_free(td->td_rlqe);
291 turnstile_free(td->td_turnstile);
292 sleepq_free(td->td_sleepqueue);
293 umtx_thread_fini(td);
298 * For a newly created process,
299 * link up all the structures and its initial threads etc.
301 * {arch}/{arch}/machdep.c {arch}_init(), init386() etc.
302 * proc_dtor() (should go away)
306 proc_linkup0(struct proc *p, struct thread *td)
308 TAILQ_INIT(&p->p_threads); /* all threads in proc */
313 proc_linkup(struct proc *p, struct thread *td)
316 sigqueue_init(&p->p_sigqueue, p);
317 p->p_ksi = ksiginfo_alloc(1);
318 if (p->p_ksi != NULL) {
319 /* XXX p_ksi may be null if ksiginfo zone is not ready */
320 p->p_ksi->ksi_flags = KSI_EXT | KSI_INS;
322 LIST_INIT(&p->p_mqnotifier);
328 * Initialize global thread allocation resources.
334 mtx_init(&tid_lock, "TID lock", NULL, MTX_DEF);
337 * pid_max cannot be greater than PID_MAX.
338 * leave one number for thread0.
340 tid_unrhdr = new_unrhdr(PID_MAX + 2, INT_MAX, &tid_lock);
342 thread_zone = uma_zcreate("THREAD", sched_sizeof_thread(),
343 thread_ctor, thread_dtor, thread_init, thread_fini,
344 32 - 1, UMA_ZONE_NOFREE);
345 tidhashtbl = hashinit(maxproc / 2, M_TIDHASH, &tidhash);
346 rw_init(&tidhash_lock, "tidhash");
350 * Place an unused thread on the zombie list.
351 * Use the slpq as that must be unused by now.
354 thread_zombie(struct thread *td)
356 mtx_lock_spin(&zombie_lock);
357 TAILQ_INSERT_HEAD(&zombie_threads, td, td_slpq);
358 mtx_unlock_spin(&zombie_lock);
362 * Release a thread that has exited after cpu_throw().
365 thread_stash(struct thread *td)
367 atomic_subtract_rel_int(&td->td_proc->p_exitthreads, 1);
372 * Reap zombie resources.
377 struct thread *td_first, *td_next;
380 * Don't even bother to lock if none at this instant,
381 * we really don't care about the next instant.
383 if (!TAILQ_EMPTY(&zombie_threads)) {
384 mtx_lock_spin(&zombie_lock);
385 td_first = TAILQ_FIRST(&zombie_threads);
387 TAILQ_INIT(&zombie_threads);
388 mtx_unlock_spin(&zombie_lock);
390 td_next = TAILQ_NEXT(td_first, td_slpq);
391 thread_cow_free(td_first);
392 thread_free(td_first);
402 thread_alloc(int pages)
406 thread_reap(); /* check if any zombies to get */
408 td = (struct thread *)uma_zalloc(thread_zone, M_WAITOK);
409 KASSERT(td->td_kstack == 0, ("thread_alloc got thread with kstack"));
410 if (!vm_thread_new(td, pages)) {
411 uma_zfree(thread_zone, td);
414 cpu_thread_alloc(td);
419 thread_alloc_stack(struct thread *td, int pages)
422 KASSERT(td->td_kstack == 0,
423 ("thread_alloc_stack called on a thread with kstack"));
424 if (!vm_thread_new(td, pages))
426 cpu_thread_alloc(td);
431 * Deallocate a thread.
434 thread_free(struct thread *td)
437 lock_profile_thread_exit(td);
439 cpuset_rel(td->td_cpuset);
440 td->td_cpuset = NULL;
442 if (td->td_kstack != 0)
443 vm_thread_dispose(td);
444 callout_drain(&td->td_slpcallout);
445 uma_zfree(thread_zone, td);
449 thread_cow_get_proc(struct thread *newtd, struct proc *p)
452 PROC_LOCK_ASSERT(p, MA_OWNED);
453 newtd->td_ucred = crhold(p->p_ucred);
454 newtd->td_limit = lim_hold(p->p_limit);
455 newtd->td_cowgen = p->p_cowgen;
459 thread_cow_get(struct thread *newtd, struct thread *td)
462 newtd->td_ucred = crhold(td->td_ucred);
463 newtd->td_limit = lim_hold(td->td_limit);
464 newtd->td_cowgen = td->td_cowgen;
468 thread_cow_free(struct thread *td)
471 if (td->td_ucred != NULL)
472 crfree(td->td_ucred);
473 if (td->td_limit != NULL)
474 lim_free(td->td_limit);
478 thread_cow_update(struct thread *td)
481 struct ucred *oldcred;
482 struct plimit *oldlimit;
488 if (td->td_ucred != p->p_ucred) {
489 oldcred = td->td_ucred;
490 td->td_ucred = crhold(p->p_ucred);
492 if (td->td_limit != p->p_limit) {
493 oldlimit = td->td_limit;
494 td->td_limit = lim_hold(p->p_limit);
496 td->td_cowgen = p->p_cowgen;
500 if (oldlimit != NULL)
505 * Discard the current thread and exit from its context.
506 * Always called with scheduler locked.
508 * Because we can't free a thread while we're operating under its context,
509 * push the current thread into our CPU's deadthread holder. This means
510 * we needn't worry about someone else grabbing our context before we
516 uint64_t runtime, new_switchtime;
525 PROC_SLOCK_ASSERT(p, MA_OWNED);
526 mtx_assert(&Giant, MA_NOTOWNED);
528 PROC_LOCK_ASSERT(p, MA_OWNED);
529 KASSERT(p != NULL, ("thread exiting without a process"));
530 CTR3(KTR_PROC, "thread_exit: thread %p (pid %ld, %s)", td,
531 (long)p->p_pid, td->td_name);
532 SDT_PROBE0(proc, , , lwp__exit);
533 KASSERT(TAILQ_EMPTY(&td->td_sigqueue.sq_list), ("signal pending"));
536 * drop FPU & debug register state storage, or any other
537 * architecture specific resources that
538 * would not be on a new untouched process.
543 * The last thread is left attached to the process
544 * So that the whole bundle gets recycled. Skip
545 * all this stuff if we never had threads.
546 * EXIT clears all sign of other threads when
547 * it goes to single threading, so the last thread always
548 * takes the short path.
550 if (p->p_flag & P_HADTHREADS) {
551 if (p->p_numthreads > 1) {
552 atomic_add_int(&td->td_proc->p_exitthreads, 1);
554 td2 = FIRST_THREAD_IN_PROC(p);
555 sched_exit_thread(td2, td);
558 * The test below is NOT true if we are the
559 * sole exiting thread. P_STOPPED_SINGLE is unset
560 * in exit1() after it is the only survivor.
562 if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE) {
563 if (p->p_numthreads == p->p_suspcount) {
564 thread_lock(p->p_singlethread);
565 wakeup_swapper = thread_unsuspend_one(
566 p->p_singlethread, p, false);
567 thread_unlock(p->p_singlethread);
573 PCPU_SET(deadthread, td);
576 * The last thread is exiting.. but not through exit()
578 panic ("thread_exit: Last thread exiting on its own");
583 * If this thread is part of a process that is being tracked by hwpmc(4),
584 * inform the module of the thread's impending exit.
586 if (PMC_PROC_IS_USING_PMCS(td->td_proc)) {
587 PMC_SWITCH_CONTEXT(td, PMC_FN_CSW_OUT);
588 PMC_CALL_HOOK_UNLOCKED(td, PMC_FN_THR_EXIT, NULL);
589 } else if (PMC_SYSTEM_SAMPLING_ACTIVE())
590 PMC_CALL_HOOK_UNLOCKED(td, PMC_FN_THR_EXIT_LOG, NULL);
597 /* Do the same timestamp bookkeeping that mi_switch() would do. */
598 new_switchtime = cpu_ticks();
599 runtime = new_switchtime - PCPU_GET(switchtime);
600 td->td_runtime += runtime;
601 td->td_incruntime += runtime;
602 PCPU_SET(switchtime, new_switchtime);
603 PCPU_SET(switchticks, ticks);
606 /* Save our resource usage in our process. */
607 td->td_ru.ru_nvcsw++;
609 rucollect(&p->p_ru, &td->td_ru);
612 td->td_state = TDS_INACTIVE;
614 witness_thread_exit(td);
616 CTR1(KTR_PROC, "thread_exit: cpu_throw() thread %p", td);
618 panic("I'm a teapot!");
623 * Do any thread specific cleanups that may be needed in wait()
624 * called with Giant, proc and schedlock not held.
627 thread_wait(struct proc *p)
631 mtx_assert(&Giant, MA_NOTOWNED);
632 KASSERT(p->p_numthreads == 1, ("multiple threads in thread_wait()"));
633 KASSERT(p->p_exitthreads == 0, ("p_exitthreads leaking"));
634 td = FIRST_THREAD_IN_PROC(p);
635 /* Lock the last thread so we spin until it exits cpu_throw(). */
638 lock_profile_thread_exit(td);
639 cpuset_rel(td->td_cpuset);
640 td->td_cpuset = NULL;
641 cpu_thread_clean(td);
643 callout_drain(&td->td_slpcallout);
644 thread_reap(); /* check for zombie threads etc. */
648 * Link a thread to a process.
649 * set up anything that needs to be initialized for it to
650 * be used by the process.
653 thread_link(struct thread *td, struct proc *p)
657 * XXX This can't be enabled because it's called for proc0 before
658 * its lock has been created.
659 * PROC_LOCK_ASSERT(p, MA_OWNED);
661 td->td_state = TDS_INACTIVE;
663 td->td_flags = TDF_INMEM;
665 LIST_INIT(&td->td_contested);
666 LIST_INIT(&td->td_lprof[0]);
667 LIST_INIT(&td->td_lprof[1]);
668 sigqueue_init(&td->td_sigqueue, p);
669 callout_init(&td->td_slpcallout, 1);
670 TAILQ_INSERT_TAIL(&p->p_threads, td, td_plist);
679 thread_unlink(struct thread *td)
681 struct proc *p = td->td_proc;
683 PROC_LOCK_ASSERT(p, MA_OWNED);
684 TAILQ_REMOVE(&p->p_threads, td, td_plist);
686 /* could clear a few other things here */
687 /* Must NOT clear links to proc! */
691 calc_remaining(struct proc *p, int mode)
695 PROC_LOCK_ASSERT(p, MA_OWNED);
696 PROC_SLOCK_ASSERT(p, MA_OWNED);
697 if (mode == SINGLE_EXIT)
698 remaining = p->p_numthreads;
699 else if (mode == SINGLE_BOUNDARY)
700 remaining = p->p_numthreads - p->p_boundary_count;
701 else if (mode == SINGLE_NO_EXIT || mode == SINGLE_ALLPROC)
702 remaining = p->p_numthreads - p->p_suspcount;
704 panic("calc_remaining: wrong mode %d", mode);
709 remain_for_mode(int mode)
712 return (mode == SINGLE_ALLPROC ? 0 : 1);
716 weed_inhib(int mode, struct thread *td2, struct proc *p)
720 PROC_LOCK_ASSERT(p, MA_OWNED);
721 PROC_SLOCK_ASSERT(p, MA_OWNED);
722 THREAD_LOCK_ASSERT(td2, MA_OWNED);
727 if (TD_IS_SUSPENDED(td2))
728 wakeup_swapper |= thread_unsuspend_one(td2, p, true);
729 if (TD_ON_SLEEPQ(td2) && (td2->td_flags & TDF_SINTR) != 0)
730 wakeup_swapper |= sleepq_abort(td2, EINTR);
732 case SINGLE_BOUNDARY:
734 if (TD_IS_SUSPENDED(td2) && (td2->td_flags & TDF_BOUNDARY) == 0)
735 wakeup_swapper |= thread_unsuspend_one(td2, p, false);
736 if (TD_ON_SLEEPQ(td2) && (td2->td_flags & TDF_SINTR) != 0)
737 wakeup_swapper |= sleepq_abort(td2, ERESTART);
741 * ALLPROC suspend tries to avoid spurious EINTR for
742 * threads sleeping interruptable, by suspending the
743 * thread directly, similarly to sig_suspend_threads().
744 * Since such sleep is not performed at the user
745 * boundary, TDF_BOUNDARY flag is not set, and TDF_ALLPROCSUSP
746 * is used to avoid immediate un-suspend.
748 if (TD_IS_SUSPENDED(td2) && (td2->td_flags & (TDF_BOUNDARY |
749 TDF_ALLPROCSUSP)) == 0)
750 wakeup_swapper |= thread_unsuspend_one(td2, p, false);
751 if (TD_ON_SLEEPQ(td2) && (td2->td_flags & TDF_SINTR) != 0) {
752 if ((td2->td_flags & TDF_SBDRY) == 0) {
753 thread_suspend_one(td2);
754 td2->td_flags |= TDF_ALLPROCSUSP;
756 wakeup_swapper |= sleepq_abort(td2, ERESTART);
761 return (wakeup_swapper);
765 * Enforce single-threading.
767 * Returns 1 if the caller must abort (another thread is waiting to
768 * exit the process or similar). Process is locked!
769 * Returns 0 when you are successfully the only thread running.
770 * A process has successfully single threaded in the suspend mode when
771 * There are no threads in user mode. Threads in the kernel must be
772 * allowed to continue until they get to the user boundary. They may even
773 * copy out their return values and data before suspending. They may however be
774 * accelerated in reaching the user boundary as we will wake up
775 * any sleeping threads that are interruptable. (PCATCH).
778 thread_single(struct proc *p, int mode)
782 int remaining, wakeup_swapper;
785 KASSERT(mode == SINGLE_EXIT || mode == SINGLE_BOUNDARY ||
786 mode == SINGLE_ALLPROC || mode == SINGLE_NO_EXIT,
787 ("invalid mode %d", mode));
789 * If allowing non-ALLPROC singlethreading for non-curproc
790 * callers, calc_remaining() and remain_for_mode() should be
791 * adjusted to also account for td->td_proc != p. For now
792 * this is not implemented because it is not used.
794 KASSERT((mode == SINGLE_ALLPROC && td->td_proc != p) ||
795 (mode != SINGLE_ALLPROC && td->td_proc == p),
796 ("mode %d proc %p curproc %p", mode, p, td->td_proc));
797 mtx_assert(&Giant, MA_NOTOWNED);
798 PROC_LOCK_ASSERT(p, MA_OWNED);
800 if ((p->p_flag & P_HADTHREADS) == 0 && mode != SINGLE_ALLPROC)
803 /* Is someone already single threading? */
804 if (p->p_singlethread != NULL && p->p_singlethread != td)
807 if (mode == SINGLE_EXIT) {
808 p->p_flag |= P_SINGLE_EXIT;
809 p->p_flag &= ~P_SINGLE_BOUNDARY;
811 p->p_flag &= ~P_SINGLE_EXIT;
812 if (mode == SINGLE_BOUNDARY)
813 p->p_flag |= P_SINGLE_BOUNDARY;
815 p->p_flag &= ~P_SINGLE_BOUNDARY;
817 if (mode == SINGLE_ALLPROC)
818 p->p_flag |= P_TOTAL_STOP;
819 p->p_flag |= P_STOPPED_SINGLE;
821 p->p_singlethread = td;
822 remaining = calc_remaining(p, mode);
823 while (remaining != remain_for_mode(mode)) {
824 if (P_SHOULDSTOP(p) != P_STOPPED_SINGLE)
827 FOREACH_THREAD_IN_PROC(p, td2) {
831 td2->td_flags |= TDF_ASTPENDING | TDF_NEEDSUSPCHK;
832 if (TD_IS_INHIBITED(td2)) {
833 wakeup_swapper |= weed_inhib(mode, td2, p);
835 } else if (TD_IS_RUNNING(td2) && td != td2) {
843 remaining = calc_remaining(p, mode);
846 * Maybe we suspended some threads.. was it enough?
848 if (remaining == remain_for_mode(mode))
853 * Wake us up when everyone else has suspended.
854 * In the mean time we suspend as well.
856 thread_suspend_switch(td, p);
857 remaining = calc_remaining(p, mode);
859 if (mode == SINGLE_EXIT) {
861 * Convert the process to an unthreaded process. The
862 * SINGLE_EXIT is called by exit1() or execve(), in
863 * both cases other threads must be retired.
865 KASSERT(p->p_numthreads == 1, ("Unthreading with >1 threads"));
866 p->p_singlethread = NULL;
867 p->p_flag &= ~(P_STOPPED_SINGLE | P_SINGLE_EXIT | P_HADTHREADS);
870 * Wait for any remaining threads to exit cpu_throw().
872 while (p->p_exitthreads != 0) {
875 sched_relinquish(td);
879 } else if (mode == SINGLE_BOUNDARY) {
881 * Wait until all suspended threads are removed from
882 * the processors. The thread_suspend_check()
883 * increments p_boundary_count while it is still
884 * running, which makes it possible for the execve()
885 * to destroy vmspace while our other threads are
886 * still using the address space.
888 * We lock the thread, which is only allowed to
889 * succeed after context switch code finished using
892 FOREACH_THREAD_IN_PROC(p, td2) {
896 KASSERT((td2->td_flags & TDF_BOUNDARY) != 0,
897 ("td %p not on boundary", td2));
898 KASSERT(TD_IS_SUSPENDED(td2),
899 ("td %p is not suspended", td2));
908 thread_suspend_check_needed(void)
915 PROC_LOCK_ASSERT(p, MA_OWNED);
916 return (P_SHOULDSTOP(p) || ((p->p_flag & P_TRACED) != 0 &&
917 (td->td_dbgflags & TDB_SUSPEND) != 0));
921 * Called in from locations that can safely check to see
922 * whether we have to suspend or at least throttle for a
923 * single-thread event (e.g. fork).
925 * Such locations include userret().
926 * If the "return_instead" argument is non zero, the thread must be able to
927 * accept 0 (caller may continue), or 1 (caller must abort) as a result.
929 * The 'return_instead' argument tells the function if it may do a
930 * thread_exit() or suspend, or whether the caller must abort and back
933 * If the thread that set the single_threading request has set the
934 * P_SINGLE_EXIT bit in the process flags then this call will never return
935 * if 'return_instead' is false, but will exit.
937 * P_SINGLE_EXIT | return_instead == 0| return_instead != 0
938 *---------------+--------------------+---------------------
939 * 0 | returns 0 | returns 0 or 1
940 * | when ST ends | immediately
941 *---------------+--------------------+---------------------
942 * 1 | thread exits | returns 1
944 * 0 = thread_exit() or suspension ok,
945 * other = return error instead of stopping the thread.
947 * While a full suspension is under effect, even a single threading
948 * thread would be suspended if it made this call (but it shouldn't).
949 * This call should only be made from places where
950 * thread_exit() would be safe as that may be the outcome unless
951 * return_instead is set.
954 thread_suspend_check(int return_instead)
962 mtx_assert(&Giant, MA_NOTOWNED);
963 PROC_LOCK_ASSERT(p, MA_OWNED);
964 while (thread_suspend_check_needed()) {
965 if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE) {
966 KASSERT(p->p_singlethread != NULL,
967 ("singlethread not set"));
969 * The only suspension in action is a
970 * single-threading. Single threader need not stop.
971 * It is safe to access p->p_singlethread unlocked
972 * because it can only be set to our address by us.
974 if (p->p_singlethread == td)
975 return (0); /* Exempt from stopping. */
977 if ((p->p_flag & P_SINGLE_EXIT) && return_instead)
980 /* Should we goto user boundary if we didn't come from there? */
981 if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE &&
982 (p->p_flag & P_SINGLE_BOUNDARY) && return_instead)
986 * Ignore suspend requests if they are deferred.
988 if ((td->td_flags & TDF_SBDRY) != 0) {
989 KASSERT(return_instead,
990 ("TDF_SBDRY set for unsafe thread_suspend_check"));
991 KASSERT((td->td_flags & (TDF_SEINTR | TDF_SERESTART)) !=
992 (TDF_SEINTR | TDF_SERESTART),
993 ("both TDF_SEINTR and TDF_SERESTART"));
994 return (TD_SBDRY_INTR(td) ? TD_SBDRY_ERRNO(td) : 0);
998 * If the process is waiting for us to exit,
999 * this thread should just suicide.
1000 * Assumes that P_SINGLE_EXIT implies P_STOPPED_SINGLE.
1002 if ((p->p_flag & P_SINGLE_EXIT) && (p->p_singlethread != td)) {
1006 * Allow Linux emulation layer to do some work
1007 * before thread suicide.
1009 if (__predict_false(p->p_sysent->sv_thread_detach != NULL))
1010 (p->p_sysent->sv_thread_detach)(td);
1011 umtx_thread_exit(td);
1013 panic("stopped thread did not exit");
1018 if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE) {
1019 if (p->p_numthreads == p->p_suspcount + 1) {
1020 thread_lock(p->p_singlethread);
1021 wakeup_swapper = thread_unsuspend_one(
1022 p->p_singlethread, p, false);
1023 thread_unlock(p->p_singlethread);
1031 * When a thread suspends, it just
1032 * gets taken off all queues.
1034 thread_suspend_one(td);
1035 if (return_instead == 0) {
1036 p->p_boundary_count++;
1037 td->td_flags |= TDF_BOUNDARY;
1040 mi_switch(SW_INVOL | SWT_SUSPEND, NULL);
1048 thread_suspend_switch(struct thread *td, struct proc *p)
1051 KASSERT(!TD_IS_SUSPENDED(td), ("already suspended"));
1052 PROC_LOCK_ASSERT(p, MA_OWNED);
1053 PROC_SLOCK_ASSERT(p, MA_OWNED);
1055 * We implement thread_suspend_one in stages here to avoid
1056 * dropping the proc lock while the thread lock is owned.
1058 if (p == td->td_proc) {
1064 td->td_flags &= ~TDF_NEEDSUSPCHK;
1065 TD_SET_SUSPENDED(td);
1069 mi_switch(SW_VOL | SWT_SUSPEND, NULL);
1077 thread_suspend_one(struct thread *td)
1082 PROC_SLOCK_ASSERT(p, MA_OWNED);
1083 THREAD_LOCK_ASSERT(td, MA_OWNED);
1084 KASSERT(!TD_IS_SUSPENDED(td), ("already suspended"));
1086 td->td_flags &= ~TDF_NEEDSUSPCHK;
1087 TD_SET_SUSPENDED(td);
1092 thread_unsuspend_one(struct thread *td, struct proc *p, bool boundary)
1095 THREAD_LOCK_ASSERT(td, MA_OWNED);
1096 KASSERT(TD_IS_SUSPENDED(td), ("Thread not suspended"));
1097 TD_CLR_SUSPENDED(td);
1098 td->td_flags &= ~TDF_ALLPROCSUSP;
1099 if (td->td_proc == p) {
1100 PROC_SLOCK_ASSERT(p, MA_OWNED);
1102 if (boundary && (td->td_flags & TDF_BOUNDARY) != 0) {
1103 td->td_flags &= ~TDF_BOUNDARY;
1104 p->p_boundary_count--;
1107 return (setrunnable(td));
1111 * Allow all threads blocked by single threading to continue running.
1114 thread_unsuspend(struct proc *p)
1119 PROC_LOCK_ASSERT(p, MA_OWNED);
1120 PROC_SLOCK_ASSERT(p, MA_OWNED);
1122 if (!P_SHOULDSTOP(p)) {
1123 FOREACH_THREAD_IN_PROC(p, td) {
1125 if (TD_IS_SUSPENDED(td)) {
1126 wakeup_swapper |= thread_unsuspend_one(td, p,
1131 } else if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE &&
1132 p->p_numthreads == p->p_suspcount) {
1134 * Stopping everything also did the job for the single
1135 * threading request. Now we've downgraded to single-threaded,
1138 if (p->p_singlethread->td_proc == p) {
1139 thread_lock(p->p_singlethread);
1140 wakeup_swapper = thread_unsuspend_one(
1141 p->p_singlethread, p, false);
1142 thread_unlock(p->p_singlethread);
1150 * End the single threading mode..
1153 thread_single_end(struct proc *p, int mode)
1158 KASSERT(mode == SINGLE_EXIT || mode == SINGLE_BOUNDARY ||
1159 mode == SINGLE_ALLPROC || mode == SINGLE_NO_EXIT,
1160 ("invalid mode %d", mode));
1161 PROC_LOCK_ASSERT(p, MA_OWNED);
1162 KASSERT((mode == SINGLE_ALLPROC && (p->p_flag & P_TOTAL_STOP) != 0) ||
1163 (mode != SINGLE_ALLPROC && (p->p_flag & P_TOTAL_STOP) == 0),
1164 ("mode %d does not match P_TOTAL_STOP", mode));
1165 KASSERT(mode == SINGLE_ALLPROC || p->p_singlethread == curthread,
1166 ("thread_single_end from other thread %p %p",
1167 curthread, p->p_singlethread));
1168 KASSERT(mode != SINGLE_BOUNDARY ||
1169 (p->p_flag & P_SINGLE_BOUNDARY) != 0,
1170 ("mis-matched SINGLE_BOUNDARY flags %x", p->p_flag));
1171 p->p_flag &= ~(P_STOPPED_SINGLE | P_SINGLE_EXIT | P_SINGLE_BOUNDARY |
1174 p->p_singlethread = NULL;
1177 * If there are other threads they may now run,
1178 * unless of course there is a blanket 'stop order'
1179 * on the process. The single threader must be allowed
1180 * to continue however as this is a bad place to stop.
1182 if (p->p_numthreads != remain_for_mode(mode) && !P_SHOULDSTOP(p)) {
1183 FOREACH_THREAD_IN_PROC(p, td) {
1185 if (TD_IS_SUSPENDED(td)) {
1186 wakeup_swapper |= thread_unsuspend_one(td, p,
1187 mode == SINGLE_BOUNDARY);
1192 KASSERT(mode != SINGLE_BOUNDARY || p->p_boundary_count == 0,
1193 ("inconsistent boundary count %d", p->p_boundary_count));
1200 thread_find(struct proc *p, lwpid_t tid)
1204 PROC_LOCK_ASSERT(p, MA_OWNED);
1205 FOREACH_THREAD_IN_PROC(p, td) {
1206 if (td->td_tid == tid)
1212 /* Locate a thread by number; return with proc lock held. */
1214 tdfind(lwpid_t tid, pid_t pid)
1216 #define RUN_THRESH 16
1220 rw_rlock(&tidhash_lock);
1221 LIST_FOREACH(td, TIDHASH(tid), td_hash) {
1222 if (td->td_tid == tid) {
1223 if (pid != -1 && td->td_proc->p_pid != pid) {
1227 PROC_LOCK(td->td_proc);
1228 if (td->td_proc->p_state == PRS_NEW) {
1229 PROC_UNLOCK(td->td_proc);
1233 if (run > RUN_THRESH) {
1234 if (rw_try_upgrade(&tidhash_lock)) {
1235 LIST_REMOVE(td, td_hash);
1236 LIST_INSERT_HEAD(TIDHASH(td->td_tid),
1238 rw_wunlock(&tidhash_lock);
1246 rw_runlock(&tidhash_lock);
1251 tidhash_add(struct thread *td)
1253 rw_wlock(&tidhash_lock);
1254 LIST_INSERT_HEAD(TIDHASH(td->td_tid), td, td_hash);
1255 rw_wunlock(&tidhash_lock);
1259 tidhash_remove(struct thread *td)
1261 rw_wlock(&tidhash_lock);
1262 LIST_REMOVE(td, td_hash);
1263 rw_wunlock(&tidhash_lock);