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/epoch.h>
44 #include <sys/rangelock.h>
45 #include <sys/resourcevar.h>
48 #include <sys/sched.h>
49 #include <sys/sleepqueue.h>
50 #include <sys/selinfo.h>
51 #include <sys/syscallsubr.h>
52 #include <sys/sysent.h>
53 #include <sys/turnstile.h>
55 #include <sys/rwlock.h>
57 #include <sys/vmmeter.h>
58 #include <sys/cpuset.h>
60 #include <sys/pmckern.h>
63 #include <security/audit/audit.h>
66 #include <vm/vm_extern.h>
68 #include <sys/eventhandler.h>
71 * Asserts below verify the stability of struct thread and struct proc
72 * layout, as exposed by KBI to modules. On head, the KBI is allowed
73 * to drift, change to the structures must be accompanied by the
76 * On the stable branches after KBI freeze, conditions must not be
77 * violated. Typically new fields are moved to the end of the
81 _Static_assert(offsetof(struct thread, td_flags) == 0xfc,
82 "struct thread KBI td_flags");
83 _Static_assert(offsetof(struct thread, td_pflags) == 0x104,
84 "struct thread KBI td_pflags");
85 _Static_assert(offsetof(struct thread, td_frame) == 0x478,
86 "struct thread KBI td_frame");
87 _Static_assert(offsetof(struct thread, td_emuldata) == 0x530,
88 "struct thread KBI td_emuldata");
89 _Static_assert(offsetof(struct proc, p_flag) == 0xb0,
90 "struct proc KBI p_flag");
91 _Static_assert(offsetof(struct proc, p_pid) == 0xbc,
92 "struct proc KBI p_pid");
93 _Static_assert(offsetof(struct proc, p_filemon) == 0x3c8,
94 "struct proc KBI p_filemon");
95 _Static_assert(offsetof(struct proc, p_comm) == 0x3e0,
96 "struct proc KBI p_comm");
97 _Static_assert(offsetof(struct proc, p_emuldata) == 0x4c0,
98 "struct proc KBI p_emuldata");
101 _Static_assert(offsetof(struct thread, td_flags) == 0x98,
102 "struct thread KBI td_flags");
103 _Static_assert(offsetof(struct thread, td_pflags) == 0xa0,
104 "struct thread KBI td_pflags");
105 _Static_assert(offsetof(struct thread, td_frame) == 0x2ec,
106 "struct thread KBI td_frame");
107 _Static_assert(offsetof(struct thread, td_emuldata) == 0x338,
108 "struct thread KBI td_emuldata");
109 _Static_assert(offsetof(struct proc, p_flag) == 0x68,
110 "struct proc KBI p_flag");
111 _Static_assert(offsetof(struct proc, p_pid) == 0x74,
112 "struct proc KBI p_pid");
113 _Static_assert(offsetof(struct proc, p_filemon) == 0x278,
114 "struct proc KBI p_filemon");
115 _Static_assert(offsetof(struct proc, p_comm) == 0x28c,
116 "struct proc KBI p_comm");
117 _Static_assert(offsetof(struct proc, p_emuldata) == 0x318,
118 "struct proc KBI p_emuldata");
121 SDT_PROVIDER_DECLARE(proc);
122 SDT_PROBE_DEFINE(proc, , , lwp__exit);
125 * thread related storage.
127 static uma_zone_t thread_zone;
129 TAILQ_HEAD(, thread) zombie_threads = TAILQ_HEAD_INITIALIZER(zombie_threads);
130 static struct mtx zombie_lock;
131 MTX_SYSINIT(zombie_lock, &zombie_lock, "zombie lock", MTX_SPIN);
133 static void thread_zombie(struct thread *);
134 static int thread_unsuspend_one(struct thread *td, struct proc *p,
137 #define TID_BUFFER_SIZE 1024
140 static struct unrhdr *tid_unrhdr;
141 static lwpid_t tid_buffer[TID_BUFFER_SIZE];
142 static int tid_head, tid_tail;
143 static MALLOC_DEFINE(M_TIDHASH, "tidhash", "thread hash");
145 struct tidhashhead *tidhashtbl;
147 struct rwlock tidhash_lock;
149 EVENTHANDLER_LIST_DEFINE(thread_ctor);
150 EVENTHANDLER_LIST_DEFINE(thread_dtor);
151 EVENTHANDLER_LIST_DEFINE(thread_init);
152 EVENTHANDLER_LIST_DEFINE(thread_fini);
159 tid = alloc_unr(tid_unrhdr);
163 if (tid_head == tid_tail) {
164 mtx_unlock(&tid_lock);
167 tid = tid_buffer[tid_head];
168 tid_head = (tid_head + 1) % TID_BUFFER_SIZE;
169 mtx_unlock(&tid_lock);
174 tid_free(lwpid_t tid)
176 lwpid_t tmp_tid = -1;
179 if ((tid_tail + 1) % TID_BUFFER_SIZE == tid_head) {
180 tmp_tid = tid_buffer[tid_head];
181 tid_head = (tid_head + 1) % TID_BUFFER_SIZE;
183 tid_buffer[tid_tail] = tid;
184 tid_tail = (tid_tail + 1) % TID_BUFFER_SIZE;
185 mtx_unlock(&tid_lock);
187 free_unr(tid_unrhdr, tmp_tid);
191 * Prepare a thread for use.
194 thread_ctor(void *mem, int size, void *arg, int flags)
198 td = (struct thread *)mem;
199 td->td_state = TDS_INACTIVE;
200 td->td_lastcpu = td->td_oncpu = NOCPU;
202 td->td_tid = tid_alloc();
205 * Note that td_critnest begins life as 1 because the thread is not
206 * running and is thereby implicitly waiting to be on the receiving
207 * end of a context switch.
210 td->td_lend_user_pri = PRI_MAX;
211 EVENTHANDLER_DIRECT_INVOKE(thread_ctor, td);
213 audit_thread_alloc(td);
215 umtx_thread_alloc(td);
220 * Reclaim a thread after use.
223 thread_dtor(void *mem, int size, void *arg)
227 td = (struct thread *)mem;
230 /* Verify that this thread is in a safe state to free. */
231 switch (td->td_state) {
237 * We must never unlink a thread that is in one of
238 * these states, because it is currently active.
240 panic("bad state for thread unlinking");
245 panic("bad thread state");
250 audit_thread_free(td);
252 /* Free all OSD associated to this thread. */
254 td_softdep_cleanup(td);
255 MPASS(td->td_su == NULL);
257 EVENTHANDLER_DIRECT_INVOKE(thread_dtor, td);
258 tid_free(td->td_tid);
262 * Initialize type-stable parts of a thread (when newly created).
265 thread_init(void *mem, int size, int flags)
269 td = (struct thread *)mem;
271 td->td_sleepqueue = sleepq_alloc();
272 td->td_turnstile = turnstile_alloc();
274 EVENTHANDLER_DIRECT_INVOKE(thread_init, td);
275 umtx_thread_init(td);
276 epoch_thread_init(td);
283 * Tear down type-stable parts of a thread (just before being discarded).
286 thread_fini(void *mem, int size)
290 td = (struct thread *)mem;
291 EVENTHANDLER_DIRECT_INVOKE(thread_fini, td);
292 rlqentry_free(td->td_rlqe);
293 turnstile_free(td->td_turnstile);
294 sleepq_free(td->td_sleepqueue);
295 umtx_thread_fini(td);
296 epoch_thread_fini(td);
301 * For a newly created process,
302 * link up all the structures and its initial threads etc.
304 * {arch}/{arch}/machdep.c {arch}_init(), init386() etc.
305 * proc_dtor() (should go away)
309 proc_linkup0(struct proc *p, struct thread *td)
311 TAILQ_INIT(&p->p_threads); /* all threads in proc */
316 proc_linkup(struct proc *p, struct thread *td)
319 sigqueue_init(&p->p_sigqueue, p);
320 p->p_ksi = ksiginfo_alloc(1);
321 if (p->p_ksi != NULL) {
322 /* XXX p_ksi may be null if ksiginfo zone is not ready */
323 p->p_ksi->ksi_flags = KSI_EXT | KSI_INS;
325 LIST_INIT(&p->p_mqnotifier);
331 * Initialize global thread allocation resources.
337 mtx_init(&tid_lock, "TID lock", NULL, MTX_DEF);
340 * pid_max cannot be greater than PID_MAX.
341 * leave one number for thread0.
343 tid_unrhdr = new_unrhdr(PID_MAX + 2, INT_MAX, &tid_lock);
345 thread_zone = uma_zcreate("THREAD", sched_sizeof_thread(),
346 thread_ctor, thread_dtor, thread_init, thread_fini,
347 32 - 1, UMA_ZONE_NOFREE);
348 tidhashtbl = hashinit(maxproc / 2, M_TIDHASH, &tidhash);
349 rw_init(&tidhash_lock, "tidhash");
353 * Place an unused thread on the zombie list.
354 * Use the slpq as that must be unused by now.
357 thread_zombie(struct thread *td)
359 mtx_lock_spin(&zombie_lock);
360 TAILQ_INSERT_HEAD(&zombie_threads, td, td_slpq);
361 mtx_unlock_spin(&zombie_lock);
365 * Release a thread that has exited after cpu_throw().
368 thread_stash(struct thread *td)
370 atomic_subtract_rel_int(&td->td_proc->p_exitthreads, 1);
375 * Reap zombie resources.
380 struct thread *td_first, *td_next;
383 * Don't even bother to lock if none at this instant,
384 * we really don't care about the next instant.
386 if (!TAILQ_EMPTY(&zombie_threads)) {
387 mtx_lock_spin(&zombie_lock);
388 td_first = TAILQ_FIRST(&zombie_threads);
390 TAILQ_INIT(&zombie_threads);
391 mtx_unlock_spin(&zombie_lock);
393 td_next = TAILQ_NEXT(td_first, td_slpq);
394 thread_cow_free(td_first);
395 thread_free(td_first);
405 thread_alloc(int pages)
409 thread_reap(); /* check if any zombies to get */
411 td = (struct thread *)uma_zalloc(thread_zone, M_WAITOK);
412 KASSERT(td->td_kstack == 0, ("thread_alloc got thread with kstack"));
413 if (!vm_thread_new(td, pages)) {
414 uma_zfree(thread_zone, td);
417 cpu_thread_alloc(td);
422 thread_alloc_stack(struct thread *td, int pages)
425 KASSERT(td->td_kstack == 0,
426 ("thread_alloc_stack called on a thread with kstack"));
427 if (!vm_thread_new(td, pages))
429 cpu_thread_alloc(td);
434 * Deallocate a thread.
437 thread_free(struct thread *td)
440 lock_profile_thread_exit(td);
442 cpuset_rel(td->td_cpuset);
443 td->td_cpuset = NULL;
445 if (td->td_kstack != 0)
446 vm_thread_dispose(td);
447 callout_drain(&td->td_slpcallout);
448 uma_zfree(thread_zone, td);
452 thread_cow_get_proc(struct thread *newtd, struct proc *p)
455 PROC_LOCK_ASSERT(p, MA_OWNED);
456 newtd->td_ucred = crhold(p->p_ucred);
457 newtd->td_limit = lim_hold(p->p_limit);
458 newtd->td_cowgen = p->p_cowgen;
462 thread_cow_get(struct thread *newtd, struct thread *td)
465 newtd->td_ucred = crhold(td->td_ucred);
466 newtd->td_limit = lim_hold(td->td_limit);
467 newtd->td_cowgen = td->td_cowgen;
471 thread_cow_free(struct thread *td)
474 if (td->td_ucred != NULL)
475 crfree(td->td_ucred);
476 if (td->td_limit != NULL)
477 lim_free(td->td_limit);
481 thread_cow_update(struct thread *td)
484 struct ucred *oldcred;
485 struct plimit *oldlimit;
491 if (td->td_ucred != p->p_ucred) {
492 oldcred = td->td_ucred;
493 td->td_ucred = crhold(p->p_ucred);
495 if (td->td_limit != p->p_limit) {
496 oldlimit = td->td_limit;
497 td->td_limit = lim_hold(p->p_limit);
499 td->td_cowgen = p->p_cowgen;
503 if (oldlimit != NULL)
508 * Discard the current thread and exit from its context.
509 * Always called with scheduler locked.
511 * Because we can't free a thread while we're operating under its context,
512 * push the current thread into our CPU's deadthread holder. This means
513 * we needn't worry about someone else grabbing our context before we
519 uint64_t runtime, new_switchtime;
528 PROC_SLOCK_ASSERT(p, MA_OWNED);
529 mtx_assert(&Giant, MA_NOTOWNED);
531 PROC_LOCK_ASSERT(p, MA_OWNED);
532 KASSERT(p != NULL, ("thread exiting without a process"));
533 CTR3(KTR_PROC, "thread_exit: thread %p (pid %ld, %s)", td,
534 (long)p->p_pid, td->td_name);
535 SDT_PROBE0(proc, , , lwp__exit);
536 KASSERT(TAILQ_EMPTY(&td->td_sigqueue.sq_list), ("signal pending"));
539 * drop FPU & debug register state storage, or any other
540 * architecture specific resources that
541 * would not be on a new untouched process.
546 * The last thread is left attached to the process
547 * So that the whole bundle gets recycled. Skip
548 * all this stuff if we never had threads.
549 * EXIT clears all sign of other threads when
550 * it goes to single threading, so the last thread always
551 * takes the short path.
553 if (p->p_flag & P_HADTHREADS) {
554 if (p->p_numthreads > 1) {
555 atomic_add_int(&td->td_proc->p_exitthreads, 1);
557 td2 = FIRST_THREAD_IN_PROC(p);
558 sched_exit_thread(td2, td);
561 * The test below is NOT true if we are the
562 * sole exiting thread. P_STOPPED_SINGLE is unset
563 * in exit1() after it is the only survivor.
565 if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE) {
566 if (p->p_numthreads == p->p_suspcount) {
567 thread_lock(p->p_singlethread);
568 wakeup_swapper = thread_unsuspend_one(
569 p->p_singlethread, p, false);
570 thread_unlock(p->p_singlethread);
576 PCPU_SET(deadthread, td);
579 * The last thread is exiting.. but not through exit()
581 panic ("thread_exit: Last thread exiting on its own");
586 * If this thread is part of a process that is being tracked by hwpmc(4),
587 * inform the module of the thread's impending exit.
589 if (PMC_PROC_IS_USING_PMCS(td->td_proc)) {
590 PMC_SWITCH_CONTEXT(td, PMC_FN_CSW_OUT);
591 PMC_CALL_HOOK_UNLOCKED(td, PMC_FN_THR_EXIT, NULL);
592 } else if (PMC_SYSTEM_SAMPLING_ACTIVE())
593 PMC_CALL_HOOK_UNLOCKED(td, PMC_FN_THR_EXIT_LOG, NULL);
600 /* Do the same timestamp bookkeeping that mi_switch() would do. */
601 new_switchtime = cpu_ticks();
602 runtime = new_switchtime - PCPU_GET(switchtime);
603 td->td_runtime += runtime;
604 td->td_incruntime += runtime;
605 PCPU_SET(switchtime, new_switchtime);
606 PCPU_SET(switchticks, ticks);
609 /* Save our resource usage in our process. */
610 td->td_ru.ru_nvcsw++;
612 rucollect(&p->p_ru, &td->td_ru);
615 td->td_state = TDS_INACTIVE;
617 witness_thread_exit(td);
619 CTR1(KTR_PROC, "thread_exit: cpu_throw() thread %p", td);
621 panic("I'm a teapot!");
626 * Do any thread specific cleanups that may be needed in wait()
627 * called with Giant, proc and schedlock not held.
630 thread_wait(struct proc *p)
634 mtx_assert(&Giant, MA_NOTOWNED);
635 KASSERT(p->p_numthreads == 1, ("multiple threads in thread_wait()"));
636 KASSERT(p->p_exitthreads == 0, ("p_exitthreads leaking"));
637 td = FIRST_THREAD_IN_PROC(p);
638 /* Lock the last thread so we spin until it exits cpu_throw(). */
641 lock_profile_thread_exit(td);
642 cpuset_rel(td->td_cpuset);
643 td->td_cpuset = NULL;
644 cpu_thread_clean(td);
646 callout_drain(&td->td_slpcallout);
647 thread_reap(); /* check for zombie threads etc. */
651 * Link a thread to a process.
652 * set up anything that needs to be initialized for it to
653 * be used by the process.
656 thread_link(struct thread *td, struct proc *p)
660 * XXX This can't be enabled because it's called for proc0 before
661 * its lock has been created.
662 * PROC_LOCK_ASSERT(p, MA_OWNED);
664 td->td_state = TDS_INACTIVE;
666 td->td_flags = TDF_INMEM;
668 LIST_INIT(&td->td_contested);
669 LIST_INIT(&td->td_lprof[0]);
670 LIST_INIT(&td->td_lprof[1]);
671 sigqueue_init(&td->td_sigqueue, p);
672 callout_init(&td->td_slpcallout, 1);
673 TAILQ_INSERT_TAIL(&p->p_threads, td, td_plist);
682 thread_unlink(struct thread *td)
684 struct proc *p = td->td_proc;
686 PROC_LOCK_ASSERT(p, MA_OWNED);
687 TAILQ_REMOVE(&p->p_threads, td, td_plist);
689 /* could clear a few other things here */
690 /* Must NOT clear links to proc! */
694 calc_remaining(struct proc *p, int mode)
698 PROC_LOCK_ASSERT(p, MA_OWNED);
699 PROC_SLOCK_ASSERT(p, MA_OWNED);
700 if (mode == SINGLE_EXIT)
701 remaining = p->p_numthreads;
702 else if (mode == SINGLE_BOUNDARY)
703 remaining = p->p_numthreads - p->p_boundary_count;
704 else if (mode == SINGLE_NO_EXIT || mode == SINGLE_ALLPROC)
705 remaining = p->p_numthreads - p->p_suspcount;
707 panic("calc_remaining: wrong mode %d", mode);
712 remain_for_mode(int mode)
715 return (mode == SINGLE_ALLPROC ? 0 : 1);
719 weed_inhib(int mode, struct thread *td2, struct proc *p)
723 PROC_LOCK_ASSERT(p, MA_OWNED);
724 PROC_SLOCK_ASSERT(p, MA_OWNED);
725 THREAD_LOCK_ASSERT(td2, MA_OWNED);
730 if (TD_IS_SUSPENDED(td2))
731 wakeup_swapper |= thread_unsuspend_one(td2, p, true);
732 if (TD_ON_SLEEPQ(td2) && (td2->td_flags & TDF_SINTR) != 0)
733 wakeup_swapper |= sleepq_abort(td2, EINTR);
735 case SINGLE_BOUNDARY:
737 if (TD_IS_SUSPENDED(td2) && (td2->td_flags & TDF_BOUNDARY) == 0)
738 wakeup_swapper |= thread_unsuspend_one(td2, p, false);
739 if (TD_ON_SLEEPQ(td2) && (td2->td_flags & TDF_SINTR) != 0)
740 wakeup_swapper |= sleepq_abort(td2, ERESTART);
744 * ALLPROC suspend tries to avoid spurious EINTR for
745 * threads sleeping interruptable, by suspending the
746 * thread directly, similarly to sig_suspend_threads().
747 * Since such sleep is not performed at the user
748 * boundary, TDF_BOUNDARY flag is not set, and TDF_ALLPROCSUSP
749 * is used to avoid immediate un-suspend.
751 if (TD_IS_SUSPENDED(td2) && (td2->td_flags & (TDF_BOUNDARY |
752 TDF_ALLPROCSUSP)) == 0)
753 wakeup_swapper |= thread_unsuspend_one(td2, p, false);
754 if (TD_ON_SLEEPQ(td2) && (td2->td_flags & TDF_SINTR) != 0) {
755 if ((td2->td_flags & TDF_SBDRY) == 0) {
756 thread_suspend_one(td2);
757 td2->td_flags |= TDF_ALLPROCSUSP;
759 wakeup_swapper |= sleepq_abort(td2, ERESTART);
764 return (wakeup_swapper);
768 * Enforce single-threading.
770 * Returns 1 if the caller must abort (another thread is waiting to
771 * exit the process or similar). Process is locked!
772 * Returns 0 when you are successfully the only thread running.
773 * A process has successfully single threaded in the suspend mode when
774 * There are no threads in user mode. Threads in the kernel must be
775 * allowed to continue until they get to the user boundary. They may even
776 * copy out their return values and data before suspending. They may however be
777 * accelerated in reaching the user boundary as we will wake up
778 * any sleeping threads that are interruptable. (PCATCH).
781 thread_single(struct proc *p, int mode)
785 int remaining, wakeup_swapper;
788 KASSERT(mode == SINGLE_EXIT || mode == SINGLE_BOUNDARY ||
789 mode == SINGLE_ALLPROC || mode == SINGLE_NO_EXIT,
790 ("invalid mode %d", mode));
792 * If allowing non-ALLPROC singlethreading for non-curproc
793 * callers, calc_remaining() and remain_for_mode() should be
794 * adjusted to also account for td->td_proc != p. For now
795 * this is not implemented because it is not used.
797 KASSERT((mode == SINGLE_ALLPROC && td->td_proc != p) ||
798 (mode != SINGLE_ALLPROC && td->td_proc == p),
799 ("mode %d proc %p curproc %p", mode, p, td->td_proc));
800 mtx_assert(&Giant, MA_NOTOWNED);
801 PROC_LOCK_ASSERT(p, MA_OWNED);
803 if ((p->p_flag & P_HADTHREADS) == 0 && mode != SINGLE_ALLPROC)
806 /* Is someone already single threading? */
807 if (p->p_singlethread != NULL && p->p_singlethread != td)
810 if (mode == SINGLE_EXIT) {
811 p->p_flag |= P_SINGLE_EXIT;
812 p->p_flag &= ~P_SINGLE_BOUNDARY;
814 p->p_flag &= ~P_SINGLE_EXIT;
815 if (mode == SINGLE_BOUNDARY)
816 p->p_flag |= P_SINGLE_BOUNDARY;
818 p->p_flag &= ~P_SINGLE_BOUNDARY;
820 if (mode == SINGLE_ALLPROC)
821 p->p_flag |= P_TOTAL_STOP;
822 p->p_flag |= P_STOPPED_SINGLE;
824 p->p_singlethread = td;
825 remaining = calc_remaining(p, mode);
826 while (remaining != remain_for_mode(mode)) {
827 if (P_SHOULDSTOP(p) != P_STOPPED_SINGLE)
830 FOREACH_THREAD_IN_PROC(p, td2) {
834 td2->td_flags |= TDF_ASTPENDING | TDF_NEEDSUSPCHK;
835 if (TD_IS_INHIBITED(td2)) {
836 wakeup_swapper |= weed_inhib(mode, td2, p);
838 } else if (TD_IS_RUNNING(td2) && td != td2) {
846 remaining = calc_remaining(p, mode);
849 * Maybe we suspended some threads.. was it enough?
851 if (remaining == remain_for_mode(mode))
856 * Wake us up when everyone else has suspended.
857 * In the mean time we suspend as well.
859 thread_suspend_switch(td, p);
860 remaining = calc_remaining(p, mode);
862 if (mode == SINGLE_EXIT) {
864 * Convert the process to an unthreaded process. The
865 * SINGLE_EXIT is called by exit1() or execve(), in
866 * both cases other threads must be retired.
868 KASSERT(p->p_numthreads == 1, ("Unthreading with >1 threads"));
869 p->p_singlethread = NULL;
870 p->p_flag &= ~(P_STOPPED_SINGLE | P_SINGLE_EXIT | P_HADTHREADS);
873 * Wait for any remaining threads to exit cpu_throw().
875 while (p->p_exitthreads != 0) {
878 sched_relinquish(td);
882 } else if (mode == SINGLE_BOUNDARY) {
884 * Wait until all suspended threads are removed from
885 * the processors. The thread_suspend_check()
886 * increments p_boundary_count while it is still
887 * running, which makes it possible for the execve()
888 * to destroy vmspace while our other threads are
889 * still using the address space.
891 * We lock the thread, which is only allowed to
892 * succeed after context switch code finished using
895 FOREACH_THREAD_IN_PROC(p, td2) {
899 KASSERT((td2->td_flags & TDF_BOUNDARY) != 0,
900 ("td %p not on boundary", td2));
901 KASSERT(TD_IS_SUSPENDED(td2),
902 ("td %p is not suspended", td2));
911 thread_suspend_check_needed(void)
918 PROC_LOCK_ASSERT(p, MA_OWNED);
919 return (P_SHOULDSTOP(p) || ((p->p_flag & P_TRACED) != 0 &&
920 (td->td_dbgflags & TDB_SUSPEND) != 0));
924 * Called in from locations that can safely check to see
925 * whether we have to suspend or at least throttle for a
926 * single-thread event (e.g. fork).
928 * Such locations include userret().
929 * If the "return_instead" argument is non zero, the thread must be able to
930 * accept 0 (caller may continue), or 1 (caller must abort) as a result.
932 * The 'return_instead' argument tells the function if it may do a
933 * thread_exit() or suspend, or whether the caller must abort and back
936 * If the thread that set the single_threading request has set the
937 * P_SINGLE_EXIT bit in the process flags then this call will never return
938 * if 'return_instead' is false, but will exit.
940 * P_SINGLE_EXIT | return_instead == 0| return_instead != 0
941 *---------------+--------------------+---------------------
942 * 0 | returns 0 | returns 0 or 1
943 * | when ST ends | immediately
944 *---------------+--------------------+---------------------
945 * 1 | thread exits | returns 1
947 * 0 = thread_exit() or suspension ok,
948 * other = return error instead of stopping the thread.
950 * While a full suspension is under effect, even a single threading
951 * thread would be suspended if it made this call (but it shouldn't).
952 * This call should only be made from places where
953 * thread_exit() would be safe as that may be the outcome unless
954 * return_instead is set.
957 thread_suspend_check(int return_instead)
965 mtx_assert(&Giant, MA_NOTOWNED);
966 PROC_LOCK_ASSERT(p, MA_OWNED);
967 while (thread_suspend_check_needed()) {
968 if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE) {
969 KASSERT(p->p_singlethread != NULL,
970 ("singlethread not set"));
972 * The only suspension in action is a
973 * single-threading. Single threader need not stop.
974 * It is safe to access p->p_singlethread unlocked
975 * because it can only be set to our address by us.
977 if (p->p_singlethread == td)
978 return (0); /* Exempt from stopping. */
980 if ((p->p_flag & P_SINGLE_EXIT) && return_instead)
983 /* Should we goto user boundary if we didn't come from there? */
984 if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE &&
985 (p->p_flag & P_SINGLE_BOUNDARY) && return_instead)
989 * Ignore suspend requests if they are deferred.
991 if ((td->td_flags & TDF_SBDRY) != 0) {
992 KASSERT(return_instead,
993 ("TDF_SBDRY set for unsafe thread_suspend_check"));
994 KASSERT((td->td_flags & (TDF_SEINTR | TDF_SERESTART)) !=
995 (TDF_SEINTR | TDF_SERESTART),
996 ("both TDF_SEINTR and TDF_SERESTART"));
997 return (TD_SBDRY_INTR(td) ? TD_SBDRY_ERRNO(td) : 0);
1001 * If the process is waiting for us to exit,
1002 * this thread should just suicide.
1003 * Assumes that P_SINGLE_EXIT implies P_STOPPED_SINGLE.
1005 if ((p->p_flag & P_SINGLE_EXIT) && (p->p_singlethread != td)) {
1009 * Allow Linux emulation layer to do some work
1010 * before thread suicide.
1012 if (__predict_false(p->p_sysent->sv_thread_detach != NULL))
1013 (p->p_sysent->sv_thread_detach)(td);
1014 umtx_thread_exit(td);
1016 panic("stopped thread did not exit");
1021 if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE) {
1022 if (p->p_numthreads == p->p_suspcount + 1) {
1023 thread_lock(p->p_singlethread);
1024 wakeup_swapper = thread_unsuspend_one(
1025 p->p_singlethread, p, false);
1026 thread_unlock(p->p_singlethread);
1034 * When a thread suspends, it just
1035 * gets taken off all queues.
1037 thread_suspend_one(td);
1038 if (return_instead == 0) {
1039 p->p_boundary_count++;
1040 td->td_flags |= TDF_BOUNDARY;
1043 mi_switch(SW_INVOL | SWT_SUSPEND, NULL);
1051 thread_suspend_switch(struct thread *td, struct proc *p)
1054 KASSERT(!TD_IS_SUSPENDED(td), ("already suspended"));
1055 PROC_LOCK_ASSERT(p, MA_OWNED);
1056 PROC_SLOCK_ASSERT(p, MA_OWNED);
1058 * We implement thread_suspend_one in stages here to avoid
1059 * dropping the proc lock while the thread lock is owned.
1061 if (p == td->td_proc) {
1067 td->td_flags &= ~TDF_NEEDSUSPCHK;
1068 TD_SET_SUSPENDED(td);
1072 mi_switch(SW_VOL | SWT_SUSPEND, NULL);
1080 thread_suspend_one(struct thread *td)
1085 PROC_SLOCK_ASSERT(p, MA_OWNED);
1086 THREAD_LOCK_ASSERT(td, MA_OWNED);
1087 KASSERT(!TD_IS_SUSPENDED(td), ("already suspended"));
1089 td->td_flags &= ~TDF_NEEDSUSPCHK;
1090 TD_SET_SUSPENDED(td);
1095 thread_unsuspend_one(struct thread *td, struct proc *p, bool boundary)
1098 THREAD_LOCK_ASSERT(td, MA_OWNED);
1099 KASSERT(TD_IS_SUSPENDED(td), ("Thread not suspended"));
1100 TD_CLR_SUSPENDED(td);
1101 td->td_flags &= ~TDF_ALLPROCSUSP;
1102 if (td->td_proc == p) {
1103 PROC_SLOCK_ASSERT(p, MA_OWNED);
1105 if (boundary && (td->td_flags & TDF_BOUNDARY) != 0) {
1106 td->td_flags &= ~TDF_BOUNDARY;
1107 p->p_boundary_count--;
1110 return (setrunnable(td));
1114 * Allow all threads blocked by single threading to continue running.
1117 thread_unsuspend(struct proc *p)
1122 PROC_LOCK_ASSERT(p, MA_OWNED);
1123 PROC_SLOCK_ASSERT(p, MA_OWNED);
1125 if (!P_SHOULDSTOP(p)) {
1126 FOREACH_THREAD_IN_PROC(p, td) {
1128 if (TD_IS_SUSPENDED(td)) {
1129 wakeup_swapper |= thread_unsuspend_one(td, p,
1134 } else if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE &&
1135 p->p_numthreads == p->p_suspcount) {
1137 * Stopping everything also did the job for the single
1138 * threading request. Now we've downgraded to single-threaded,
1141 if (p->p_singlethread->td_proc == p) {
1142 thread_lock(p->p_singlethread);
1143 wakeup_swapper = thread_unsuspend_one(
1144 p->p_singlethread, p, false);
1145 thread_unlock(p->p_singlethread);
1153 * End the single threading mode..
1156 thread_single_end(struct proc *p, int mode)
1161 KASSERT(mode == SINGLE_EXIT || mode == SINGLE_BOUNDARY ||
1162 mode == SINGLE_ALLPROC || mode == SINGLE_NO_EXIT,
1163 ("invalid mode %d", mode));
1164 PROC_LOCK_ASSERT(p, MA_OWNED);
1165 KASSERT((mode == SINGLE_ALLPROC && (p->p_flag & P_TOTAL_STOP) != 0) ||
1166 (mode != SINGLE_ALLPROC && (p->p_flag & P_TOTAL_STOP) == 0),
1167 ("mode %d does not match P_TOTAL_STOP", mode));
1168 KASSERT(mode == SINGLE_ALLPROC || p->p_singlethread == curthread,
1169 ("thread_single_end from other thread %p %p",
1170 curthread, p->p_singlethread));
1171 KASSERT(mode != SINGLE_BOUNDARY ||
1172 (p->p_flag & P_SINGLE_BOUNDARY) != 0,
1173 ("mis-matched SINGLE_BOUNDARY flags %x", p->p_flag));
1174 p->p_flag &= ~(P_STOPPED_SINGLE | P_SINGLE_EXIT | P_SINGLE_BOUNDARY |
1177 p->p_singlethread = NULL;
1180 * If there are other threads they may now run,
1181 * unless of course there is a blanket 'stop order'
1182 * on the process. The single threader must be allowed
1183 * to continue however as this is a bad place to stop.
1185 if (p->p_numthreads != remain_for_mode(mode) && !P_SHOULDSTOP(p)) {
1186 FOREACH_THREAD_IN_PROC(p, td) {
1188 if (TD_IS_SUSPENDED(td)) {
1189 wakeup_swapper |= thread_unsuspend_one(td, p,
1190 mode == SINGLE_BOUNDARY);
1195 KASSERT(mode != SINGLE_BOUNDARY || p->p_boundary_count == 0,
1196 ("inconsistent boundary count %d", p->p_boundary_count));
1203 thread_find(struct proc *p, lwpid_t tid)
1207 PROC_LOCK_ASSERT(p, MA_OWNED);
1208 FOREACH_THREAD_IN_PROC(p, td) {
1209 if (td->td_tid == tid)
1215 /* Locate a thread by number; return with proc lock held. */
1217 tdfind(lwpid_t tid, pid_t pid)
1219 #define RUN_THRESH 16
1223 rw_rlock(&tidhash_lock);
1224 LIST_FOREACH(td, TIDHASH(tid), td_hash) {
1225 if (td->td_tid == tid) {
1226 if (pid != -1 && td->td_proc->p_pid != pid) {
1230 PROC_LOCK(td->td_proc);
1231 if (td->td_proc->p_state == PRS_NEW) {
1232 PROC_UNLOCK(td->td_proc);
1236 if (run > RUN_THRESH) {
1237 if (rw_try_upgrade(&tidhash_lock)) {
1238 LIST_REMOVE(td, td_hash);
1239 LIST_INSERT_HEAD(TIDHASH(td->td_tid),
1241 rw_wunlock(&tidhash_lock);
1249 rw_runlock(&tidhash_lock);
1254 tidhash_add(struct thread *td)
1256 rw_wlock(&tidhash_lock);
1257 LIST_INSERT_HEAD(TIDHASH(td->td_tid), td, td_hash);
1258 rw_wunlock(&tidhash_lock);
1262 tidhash_remove(struct thread *td)
1264 rw_wlock(&tidhash_lock);
1265 LIST_REMOVE(td, td_hash);
1266 rw_wunlock(&tidhash_lock);