sys/kern/kern_thread.c

   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/sysctl.h>
  44 #include <sys/sched.h>
  45 #include <sys/sleepqueue.h>
  46 #include <sys/selinfo.h>
  47 #include <sys/turnstile.h>
  48 #include <sys/ktr.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 SYSCTL_NODE(_kern, OID_AUTO, threads, CTLFLAG_RW, 0, "thread allocation");
  68
  69 int max_threads_per_proc = 1500;
  70 SYSCTL_INT(_kern_threads, OID_AUTO, max_threads_per_proc, CTLFLAG_RW,
  71         &max_threads_per_proc, 0, "Limit on threads 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 TAILQ_HEAD(, thread) zombie_threads = TAILQ_HEAD_INITIALIZER(zombie_threads);
  78 static struct mtx zombie_lock;
  79 MTX_SYSINIT(zombie_lock, &zombie_lock, "zombie lock", MTX_SPIN);
  80
  81 static void thread_zombie(struct thread *);
  82
  83 struct mtx tid_lock;
  84 static struct unrhdr *tid_unrhdr;
  85
  86 /*
  87  * Prepare a thread for use.
  88  */
  89 static int
  90 thread_ctor(void *mem, int size, void *arg, int flags)
  91 {
  92         struct thread   *td;
  93
  94         td = (struct thread *)mem;
  95         td->td_state = TDS_INACTIVE;
  96         td->td_oncpu = NOCPU;
  97
  98         td->td_tid = alloc_unr(tid_unrhdr);
  99         td->td_syscalls = 0;
 100
 101         /*
 102          * Note that td_critnest begins life as 1 because the thread is not
 103          * running and is thereby implicitly waiting to be on the receiving
 104          * end of a context switch.
 105          */
 106         td->td_critnest = 1;
 107         EVENTHANDLER_INVOKE(thread_ctor, td);
 108 #ifdef AUDIT
 109         audit_thread_alloc(td);
 110 #endif
 111         umtx_thread_alloc(td);
 112         return (0);
 113 }
 114
 115 /*
 116  * Reclaim a thread after use.
 117  */
 118 static void
 119 thread_dtor(void *mem, int size, void *arg)
 120 {
 121         struct thread *td;
 122
 123         td = (struct thread *)mem;
 124
 125 #ifdef INVARIANTS
 126         /* Verify that this thread is in a safe state to free. */
 127         switch (td->td_state) {
 128         case TDS_INHIBITED:
 129         case TDS_RUNNING:
 130         case TDS_CAN_RUN:
 131         case TDS_RUNQ:
 132                 /*
 133                  * We must never unlink a thread that is in one of
 134                  * these states, because it is currently active.
 135                  */
 136                 panic("bad state for thread unlinking");
 137                 /* NOTREACHED */
 138         case TDS_INACTIVE:
 139                 break;
 140         default:
 141                 panic("bad thread state");
 142                 /* NOTREACHED */
 143         }
 144 #endif
 145 #ifdef AUDIT
 146         audit_thread_free(td);
 147 #endif
 148         /* Free all OSD associated to this thread. */
 149         osd_thread_exit(td);
 150
 151         EVENTHANDLER_INVOKE(thread_dtor, td);
 152         free_unr(tid_unrhdr, td->td_tid);
 153 }
 154
 155 /*
 156  * Initialize type-stable parts of a thread (when newly created).
 157  */
 158 static int
 159 thread_init(void *mem, int size, int flags)
 160 {
 161         struct thread *td;
 162
 163         td = (struct thread *)mem;
 164
 165         td->td_sleepqueue = sleepq_alloc();
 166         td->td_turnstile = turnstile_alloc();
 167         EVENTHANDLER_INVOKE(thread_init, td);
 168         td->td_sched = (struct td_sched *)&td[1];
 169         umtx_thread_init(td);
 170         td->td_kstack = 0;
 171         return (0);
 172 }
 173
 174 /*
 175  * Tear down type-stable parts of a thread (just before being discarded).
 176  */
 177 static void
 178 thread_fini(void *mem, int size)
 179 {
 180         struct thread *td;
 181
 182         td = (struct thread *)mem;
 183         EVENTHANDLER_INVOKE(thread_fini, td);
 184         turnstile_free(td->td_turnstile);
 185         sleepq_free(td->td_sleepqueue);
 186         umtx_thread_fini(td);
 187         seltdfini(td);
 188 }
 189
 190 /*
 191  * For a newly created process,
 192  * link up all the structures and its initial threads etc.
 193  * called from:
 194  * {arch}/{arch}/machdep.c   ia64_init(), init386() etc.
 195  * proc_dtor() (should go away)
 196  * proc_init()
 197  */
 198 void
 199 proc_linkup0(struct proc *p, struct thread *td)
 200 {
 201         TAILQ_INIT(&p->p_threads);           /* all threads in proc */
 202         proc_linkup(p, td);
 203 }
 204
 205 void
 206 proc_linkup(struct proc *p, struct thread *td)
 207 {
 208
 209         sigqueue_init(&p->p_sigqueue, p);
 210         p->p_ksi = ksiginfo_alloc(1);
 211         if (p->p_ksi != NULL) {
 212                 /* XXX p_ksi may be null if ksiginfo zone is not ready */
 213                 p->p_ksi->ksi_flags = KSI_EXT | KSI_INS;
 214         }
 215         LIST_INIT(&p->p_mqnotifier);
 216         p->p_numthreads = 0;
 217         thread_link(td, p);
 218 }
 219
 220 /*
 221  * Initialize global thread allocation resources.
 222  */
 223 void
 224 threadinit(void)
 225 {
 226
 227         mtx_init(&tid_lock, "TID lock", NULL, MTX_DEF);
 228         /* leave one number for thread0 */
 229         tid_unrhdr = new_unrhdr(PID_MAX + 2, INT_MAX, &tid_lock);
 230
 231         thread_zone = uma_zcreate("THREAD", sched_sizeof_thread(),
 232             thread_ctor, thread_dtor, thread_init, thread_fini,
 233             16 - 1, 0);
 234 }
 235
 236 /*
 237  * Place an unused thread on the zombie list.
 238  * Use the slpq as that must be unused by now.
 239  */
 240 void
 241 thread_zombie(struct thread *td)
 242 {
 243         mtx_lock_spin(&zombie_lock);
 244         TAILQ_INSERT_HEAD(&zombie_threads, td, td_slpq);
 245         mtx_unlock_spin(&zombie_lock);
 246 }
 247
 248 /*
 249  * Release a thread that has exited after cpu_throw().
 250  */
 251 void
 252 thread_stash(struct thread *td)
 253 {
 254         atomic_subtract_rel_int(&td->td_proc->p_exitthreads, 1);
 255         thread_zombie(td);
 256 }
 257
 258 /*
 259  * Reap zombie resources.
 260  */
 261 void
 262 thread_reap(void)
 263 {
 264         struct thread *td_first, *td_next;
 265
 266         /*
 267          * Don't even bother to lock if none at this instant,
 268          * we really don't care about the next instant..
 269          */
 270         if (!TAILQ_EMPTY(&zombie_threads)) {
 271                 mtx_lock_spin(&zombie_lock);
 272                 td_first = TAILQ_FIRST(&zombie_threads);
 273                 if (td_first)
 274                         TAILQ_INIT(&zombie_threads);
 275                 mtx_unlock_spin(&zombie_lock);
 276                 while (td_first) {
 277                         td_next = TAILQ_NEXT(td_first, td_slpq);
 278                         if (td_first->td_ucred)
 279                                 crfree(td_first->td_ucred);
 280                         thread_free(td_first);
 281                         td_first = td_next;
 282                 }
 283         }
 284 }
 285
 286 /*
 287  * Allocate a thread.
 288  */
 289 struct thread *
 290 thread_alloc(int pages)
 291 {
 292         struct thread *td;
 293
 294         thread_reap(); /* check if any zombies to get */
 295
 296         td = (struct thread *)uma_zalloc(thread_zone, M_WAITOK);
 297         KASSERT(td->td_kstack == 0, ("thread_alloc got thread with kstack"));
 298         if (!vm_thread_new(td, pages)) {
 299                 uma_zfree(thread_zone, td);
 300                 return (NULL);
 301         }
 302         cpu_thread_alloc(td);
 303         return (td);
 304 }
 305
 306 int
 307 thread_alloc_stack(struct thread *td, int pages)
 308 {
 309
 310         KASSERT(td->td_kstack == 0,
 311             ("thread_alloc_stack called on a thread with kstack"));
 312         if (!vm_thread_new(td, pages))
 313                 return (0);
 314         cpu_thread_alloc(td);
 315         return (1);
 316 }
 317
 318 /*
 319  * Deallocate a thread.
 320  */
 321 void
 322 thread_free(struct thread *td)
 323 {
 324
 325         lock_profile_thread_exit(td);
 326         if (td->td_cpuset)
 327                 cpuset_rel(td->td_cpuset);
 328         td->td_cpuset = NULL;
 329         cpu_thread_free(td);
 330         if (td->td_kstack != 0)
 331                 vm_thread_dispose(td);
 332         uma_zfree(thread_zone, td);
 333 }
 334
 335 /*
 336  * Discard the current thread and exit from its context.
 337  * Always called with scheduler locked.
 338  *
 339  * Because we can't free a thread while we're operating under its context,
 340  * push the current thread into our CPU's deadthread holder. This means
 341  * we needn't worry about someone else grabbing our context before we
 342  * do a cpu_throw().
 343  */
 344 void
 345 thread_exit(void)
 346 {
 347         uint64_t new_switchtime;
 348         struct thread *td;
 349         struct thread *td2;
 350         struct proc *p;
 351         int wakeup_swapper;
 352
 353         td = curthread;
 354         p = td->td_proc;
 355
 356         PROC_SLOCK_ASSERT(p, MA_OWNED);
 357         mtx_assert(&Giant, MA_NOTOWNED);
 358
 359         PROC_LOCK_ASSERT(p, MA_OWNED);
 360         KASSERT(p != NULL, ("thread exiting without a process"));
 361         CTR3(KTR_PROC, "thread_exit: thread %p (pid %ld, %s)", td,
 362             (long)p->p_pid, td->td_name);
 363         KASSERT(TAILQ_EMPTY(&td->td_sigqueue.sq_list), ("signal pending"));
 364
 365 #ifdef AUDIT
 366         AUDIT_SYSCALL_EXIT(0, td);
 367 #endif
 368         umtx_thread_exit(td);
 369         /*
 370          * drop FPU & debug register state storage, or any other
 371          * architecture specific resources that
 372          * would not be on a new untouched process.
 373          */
 374         cpu_thread_exit(td);    /* XXXSMP */
 375
 376         /* Do the same timestamp bookkeeping that mi_switch() would do. */
 377         new_switchtime = cpu_ticks();
 378         p->p_rux.rux_runtime += (new_switchtime - PCPU_GET(switchtime));
 379         PCPU_SET(switchtime, new_switchtime);
 380         PCPU_SET(switchticks, ticks);
 381         PCPU_INC(cnt.v_swtch);
 382         /* Save our resource usage in our process. */
 383         td->td_ru.ru_nvcsw++;
 384         rucollect(&p->p_ru, &td->td_ru);
 385         /*
 386          * The last thread is left attached to the process
 387          * So that the whole bundle gets recycled. Skip
 388          * all this stuff if we never had threads.
 389          * EXIT clears all sign of other threads when
 390          * it goes to single threading, so the last thread always
 391          * takes the short path.
 392          */
 393         if (p->p_flag & P_HADTHREADS) {
 394                 if (p->p_numthreads > 1) {
 395                         thread_unlink(td);
 396                         td2 = FIRST_THREAD_IN_PROC(p);
 397                         sched_exit_thread(td2, td);
 398
 399                         /*
 400                          * The test below is NOT true if we are the
 401                          * sole exiting thread. P_STOPPED_SINGLE is unset
 402                          * in exit1() after it is the only survivor.
 403                          */
 404                         if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE) {
 405                                 if (p->p_numthreads == p->p_suspcount) {
 406                                         thread_lock(p->p_singlethread);
 407                                         wakeup_swapper = thread_unsuspend_one(
 408                                                 p->p_singlethread);
 409                                         thread_unlock(p->p_singlethread);
 410                                         if (wakeup_swapper)
 411                                                 kick_proc0();
 412                                 }
 413                         }
 414
 415                         atomic_add_int(&td->td_proc->p_exitthreads, 1);
 416                         PCPU_SET(deadthread, td);
 417                 } else {
 418                         /*
 419                          * The last thread is exiting.. but not through exit()
 420                          */
 421                         panic ("thread_exit: Last thread exiting on its own");
 422                 }
 423         }
 424 #ifdef  HWPMC_HOOKS
 425         /*
 426          * If this thread is part of a process that is being tracked by hwpmc(4),
 427          * inform the module of the thread's impending exit.
 428          */
 429         if (PMC_PROC_IS_USING_PMCS(td->td_proc))
 430                 PMC_SWITCH_CONTEXT(td, PMC_FN_CSW_OUT);
 431 #endif
 432         PROC_UNLOCK(p);
 433         ruxagg(p, td);
 434         thread_lock(td);
 435         PROC_SUNLOCK(p);
 436         td->td_state = TDS_INACTIVE;
 437 #ifdef WITNESS
 438         witness_thread_exit(td);
 439 #endif
 440         CTR1(KTR_PROC, "thread_exit: cpu_throw() thread %p", td);
 441         sched_throw(td);
 442         panic("I'm a teapot!");
 443         /* NOTREACHED */
 444 }
 445
 446 /*
 447  * Do any thread specific cleanups that may be needed in wait()
 448  * called with Giant, proc and schedlock not held.
 449  */
 450 void
 451 thread_wait(struct proc *p)
 452 {
 453         struct thread *td;
 454
 455         mtx_assert(&Giant, MA_NOTOWNED);
 456         KASSERT((p->p_numthreads == 1), ("Multiple threads in wait1()"));
 457         td = FIRST_THREAD_IN_PROC(p);
 458         /* Lock the last thread so we spin until it exits cpu_throw(). */
 459         thread_lock(td);
 460         thread_unlock(td);
 461         /* Wait for any remaining threads to exit cpu_throw(). */
 462         while (p->p_exitthreads)
 463                 sched_relinquish(curthread);
 464         lock_profile_thread_exit(td);
 465         cpuset_rel(td->td_cpuset);
 466         td->td_cpuset = NULL;
 467         cpu_thread_clean(td);
 468         crfree(td->td_ucred);
 469         thread_reap();  /* check for zombie threads etc. */
 470 }
 471
 472 /*
 473  * Link a thread to a process.
 474  * set up anything that needs to be initialized for it to
 475  * be used by the process.
 476  */
 477 void
 478 thread_link(struct thread *td, struct proc *p)
 479 {
 480
 481         /*
 482          * XXX This can't be enabled because it's called for proc0 before
 483          * its lock has been created.
 484          * PROC_LOCK_ASSERT(p, MA_OWNED);
 485          */
 486         td->td_state    = TDS_INACTIVE;
 487         td->td_proc     = p;
 488         td->td_flags    = TDF_INMEM;
 489
 490         LIST_INIT(&td->td_contested);
 491         LIST_INIT(&td->td_lprof[0]);
 492         LIST_INIT(&td->td_lprof[1]);
 493         sigqueue_init(&td->td_sigqueue, p);
 494         callout_init(&td->td_slpcallout, CALLOUT_MPSAFE);
 495         TAILQ_INSERT_HEAD(&p->p_threads, td, td_plist);
 496         p->p_numthreads++;
 497 }
 498
 499 /*
 500  * Convert a process with one thread to an unthreaded process.
 501  */
 502 void
 503 thread_unthread(struct thread *td)
 504 {
 505         struct proc *p = td->td_proc;
 506
 507         KASSERT((p->p_numthreads == 1), ("Unthreading with >1 threads"));
 508         p->p_flag &= ~P_HADTHREADS;
 509 }
 510
 511 /*
 512  * Called from:
 513  *  thread_exit()
 514  */
 515 void
 516 thread_unlink(struct thread *td)
 517 {
 518         struct proc *p = td->td_proc;
 519
 520         PROC_LOCK_ASSERT(p, MA_OWNED);
 521         TAILQ_REMOVE(&p->p_threads, td, td_plist);
 522         p->p_numthreads--;
 523         /* could clear a few other things here */
 524         /* Must  NOT clear links to proc! */
 525 }
 526
 527 static int
 528 calc_remaining(struct proc *p, int mode)
 529 {
 530         int remaining;
 531
 532         if (mode == SINGLE_EXIT)
 533                 remaining = p->p_numthreads;
 534         else if (mode == SINGLE_BOUNDARY)
 535                 remaining = p->p_numthreads - p->p_boundary_count;
 536         else if (mode == SINGLE_NO_EXIT)
 537                 remaining = p->p_numthreads - p->p_suspcount;
 538         else
 539                 panic("calc_remaining: wrong mode %d", mode);
 540         return (remaining);
 541 }
 542
 543 /*
 544  * Enforce single-threading.
 545  *
 546  * Returns 1 if the caller must abort (another thread is waiting to
 547  * exit the process or similar). Process is locked!
 548  * Returns 0 when you are successfully the only thread running.
 549  * A process has successfully single threaded in the suspend mode when
 550  * There are no threads in user mode. Threads in the kernel must be
 551  * allowed to continue until they get to the user boundary. They may even
 552  * copy out their return values and data before suspending. They may however be
 553  * accelerated in reaching the user boundary as we will wake up
 554  * any sleeping threads that are interruptable. (PCATCH).
 555  */
 556 int
 557 thread_single(int mode)
 558 {
 559         struct thread *td;
 560         struct thread *td2;
 561         struct proc *p;
 562         int remaining, wakeup_swapper;
 563
 564         td = curthread;
 565         p = td->td_proc;
 566         mtx_assert(&Giant, MA_NOTOWNED);
 567         PROC_LOCK_ASSERT(p, MA_OWNED);
 568         KASSERT((td != NULL), ("curthread is NULL"));
 569
 570         if ((p->p_flag & P_HADTHREADS) == 0)
 571                 return (0);
 572
 573         /* Is someone already single threading? */
 574         if (p->p_singlethread != NULL && p->p_singlethread != td)
 575                 return (1);
 576
 577         if (mode == SINGLE_EXIT) {
 578                 p->p_flag |= P_SINGLE_EXIT;
 579                 p->p_flag &= ~P_SINGLE_BOUNDARY;
 580         } else {
 581                 p->p_flag &= ~P_SINGLE_EXIT;
 582                 if (mode == SINGLE_BOUNDARY)
 583                         p->p_flag |= P_SINGLE_BOUNDARY;
 584                 else
 585                         p->p_flag &= ~P_SINGLE_BOUNDARY;
 586         }
 587         p->p_flag |= P_STOPPED_SINGLE;
 588         PROC_SLOCK(p);
 589         p->p_singlethread = td;
 590         remaining = calc_remaining(p, mode);
 591         while (remaining != 1) {
 592                 if (P_SHOULDSTOP(p) != P_STOPPED_SINGLE)
 593                         goto stopme;
 594                 wakeup_swapper = 0;
 595                 FOREACH_THREAD_IN_PROC(p, td2) {
 596                         if (td2 == td)
 597                                 continue;
 598                         thread_lock(td2);
 599                         td2->td_flags |= TDF_ASTPENDING | TDF_NEEDSUSPCHK;
 600                         if (TD_IS_INHIBITED(td2)) {
 601                                 switch (mode) {
 602                                 case SINGLE_EXIT:
 603                                         if (TD_IS_SUSPENDED(td2))
 604                                                 wakeup_swapper |=
 605                                                     thread_unsuspend_one(td2);
 606                                         if (TD_ON_SLEEPQ(td2) &&
 607                                             (td2->td_flags & TDF_SINTR))
 608                                                 wakeup_swapper |=
 609                                                     sleepq_abort(td2, EINTR);
 610                                         break;
 611                                 case SINGLE_BOUNDARY:
 612                                         if (TD_IS_SUSPENDED(td2) &&
 613                                             !(td2->td_flags & TDF_BOUNDARY))
 614                                                 wakeup_swapper |=
 615                                                     thread_unsuspend_one(td2);
 616                                         if (TD_ON_SLEEPQ(td2) &&
 617                                             (td2->td_flags & TDF_SINTR))
 618                                                 wakeup_swapper |=
 619                                                     sleepq_abort(td2, ERESTART);
 620                                         break;
 621                                 case SINGLE_NO_EXIT:
 622                                         if (TD_IS_SUSPENDED(td2) &&
 623                                             !(td2->td_flags & TDF_BOUNDARY))
 624                                                 wakeup_swapper |=
 625                                                     thread_unsuspend_one(td2);
 626                                         if (TD_ON_SLEEPQ(td2) &&
 627                                             (td2->td_flags & TDF_SINTR))
 628                                                 wakeup_swapper |=
 629                                                     sleepq_abort(td2, ERESTART);
 630                                         break;
 631                                 default:
 632                                         break;
 633                                 }
 634                         }
 635 #ifdef SMP
 636                         else if (TD_IS_RUNNING(td2) && td != td2) {
 637                                 forward_signal(td2);
 638                         }
 639 #endif
 640                         thread_unlock(td2);
 641                 }
 642                 if (wakeup_swapper)
 643                         kick_proc0();
 644                 remaining = calc_remaining(p, mode);
 645
 646                 /*
 647                  * Maybe we suspended some threads.. was it enough?
 648                  */
 649                 if (remaining == 1)
 650                         break;
 651
 652 stopme:
 653                 /*
 654                  * Wake us up when everyone else has suspended.
 655                  * In the mean time we suspend as well.
 656                  */
 657                 thread_suspend_switch(td);
 658                 remaining = calc_remaining(p, mode);
 659         }
 660         if (mode == SINGLE_EXIT) {
 661                 /*
 662                  * We have gotten rid of all the other threads and we
 663                  * are about to either exit or exec. In either case,
 664                  * we try our utmost  to revert to being a non-threaded
 665                  * process.
 666                  */
 667                 p->p_singlethread = NULL;
 668                 p->p_flag &= ~(P_STOPPED_SINGLE | P_SINGLE_EXIT);
 669                 thread_unthread(td);
 670         }
 671         PROC_SUNLOCK(p);
 672         return (0);
 673 }
 674
 675 /*
 676  * Called in from locations that can safely check to see
 677  * whether we have to suspend or at least throttle for a
 678  * single-thread event (e.g. fork).
 679  *
 680  * Such locations include userret().
 681  * If the "return_instead" argument is non zero, the thread must be able to
 682  * accept 0 (caller may continue), or 1 (caller must abort) as a result.
 683  *
 684  * The 'return_instead' argument tells the function if it may do a
 685  * thread_exit() or suspend, or whether the caller must abort and back
 686  * out instead.
 687  *
 688  * If the thread that set the single_threading request has set the
 689  * P_SINGLE_EXIT bit in the process flags then this call will never return
 690  * if 'return_instead' is false, but will exit.
 691  *
 692  * P_SINGLE_EXIT | return_instead == 0| return_instead != 0
 693  *---------------+--------------------+---------------------
 694  *       0       | returns 0          |   returns 0 or 1
 695  *               | when ST ends       |   immediatly
 696  *---------------+--------------------+---------------------
 697  *       1       | thread exits       |   returns 1
 698  *               |                    |  immediatly
 699  * 0 = thread_exit() or suspension ok,
 700  * other = return error instead of stopping the thread.
 701  *
 702  * While a full suspension is under effect, even a single threading
 703  * thread would be suspended if it made this call (but it shouldn't).
 704  * This call should only be made from places where
 705  * thread_exit() would be safe as that may be the outcome unless
 706  * return_instead is set.
 707  */
 708 int
 709 thread_suspend_check(int return_instead)
 710 {
 711         struct thread *td;
 712         struct proc *p;
 713         int wakeup_swapper;
 714
 715         td = curthread;
 716         p = td->td_proc;
 717         mtx_assert(&Giant, MA_NOTOWNED);
 718         PROC_LOCK_ASSERT(p, MA_OWNED);
 719         while (P_SHOULDSTOP(p) ||
 720               ((p->p_flag & P_TRACED) && (td->td_dbgflags & TDB_SUSPEND))) {
 721                 if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE) {
 722                         KASSERT(p->p_singlethread != NULL,
 723                             ("singlethread not set"));
 724                         /*
 725                          * The only suspension in action is a
 726                          * single-threading. Single threader need not stop.
 727                          * XXX Should be safe to access unlocked
 728                          * as it can only be set to be true by us.
 729                          */
 730                         if (p->p_singlethread == td)
 731                                 return (0);     /* Exempt from stopping. */
 732                 }
 733                 if ((p->p_flag & P_SINGLE_EXIT) && return_instead)
 734                         return (EINTR);
 735
 736                 /* Should we goto user boundary if we didn't come from there? */
 737                 if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE &&
 738                     (p->p_flag & P_SINGLE_BOUNDARY) && return_instead)
 739                         return (ERESTART);
 740
 741                 /* If thread will exit, flush its pending signals */
 742                 if ((p->p_flag & P_SINGLE_EXIT) && (p->p_singlethread != td))
 743                         sigqueue_flush(&td->td_sigqueue);
 744
 745                 PROC_SLOCK(p);
 746                 thread_stopped(p);
 747                 /*
 748                  * If the process is waiting for us to exit,
 749                  * this thread should just suicide.
 750                  * Assumes that P_SINGLE_EXIT implies P_STOPPED_SINGLE.
 751                  */
 752                 if ((p->p_flag & P_SINGLE_EXIT) && (p->p_singlethread != td))
 753                         thread_exit();
 754                 if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE) {
 755                         if (p->p_numthreads == p->p_suspcount + 1) {
 756                                 thread_lock(p->p_singlethread);
 757                                 wakeup_swapper =
 758                                     thread_unsuspend_one(p->p_singlethread);
 759                                 thread_unlock(p->p_singlethread);
 760                                 if (wakeup_swapper)
 761                                         kick_proc0();
 762                         }
 763                 }
 764                 PROC_UNLOCK(p);
 765                 thread_lock(td);
 766                 /*
 767                  * When a thread suspends, it just
 768                  * gets taken off all queues.
 769                  */
 770                 thread_suspend_one(td);
 771                 if (return_instead == 0) {
 772                         p->p_boundary_count++;
 773                         td->td_flags |= TDF_BOUNDARY;
 774                 }
 775                 PROC_SUNLOCK(p);
 776                 mi_switch(SW_INVOL | SWT_SUSPEND, NULL);
 777                 if (return_instead == 0)
 778                         td->td_flags &= ~TDF_BOUNDARY;
 779                 thread_unlock(td);
 780                 PROC_LOCK(p);
 781                 if (return_instead == 0)
 782                         p->p_boundary_count--;
 783         }
 784         return (0);
 785 }
 786
 787 void
 788 thread_suspend_switch(struct thread *td)
 789 {
 790         struct proc *p;
 791
 792         p = td->td_proc;
 793         KASSERT(!TD_IS_SUSPENDED(td), ("already suspended"));
 794         PROC_LOCK_ASSERT(p, MA_OWNED);
 795         PROC_SLOCK_ASSERT(p, MA_OWNED);
 796         /*
 797          * We implement thread_suspend_one in stages here to avoid
 798          * dropping the proc lock while the thread lock is owned.
 799          */
 800         thread_stopped(p);
 801         p->p_suspcount++;
 802         PROC_UNLOCK(p);
 803         thread_lock(td);
 804         td->td_flags &= ~TDF_NEEDSUSPCHK;
 805         TD_SET_SUSPENDED(td);
 806         sched_sleep(td, 0);
 807         PROC_SUNLOCK(p);
 808         DROP_GIANT();
 809         mi_switch(SW_VOL | SWT_SUSPEND, NULL);
 810         thread_unlock(td);
 811         PICKUP_GIANT();
 812         PROC_LOCK(p);
 813         PROC_SLOCK(p);
 814 }
 815
 816 void
 817 thread_suspend_one(struct thread *td)
 818 {
 819         struct proc *p = td->td_proc;
 820
 821         PROC_SLOCK_ASSERT(p, MA_OWNED);
 822         THREAD_LOCK_ASSERT(td, MA_OWNED);
 823         KASSERT(!TD_IS_SUSPENDED(td), ("already suspended"));
 824         p->p_suspcount++;
 825         td->td_flags &= ~TDF_NEEDSUSPCHK;
 826         TD_SET_SUSPENDED(td);
 827         sched_sleep(td, 0);
 828 }
 829
 830 int
 831 thread_unsuspend_one(struct thread *td)
 832 {
 833         struct proc *p = td->td_proc;
 834
 835         PROC_SLOCK_ASSERT(p, MA_OWNED);
 836         THREAD_LOCK_ASSERT(td, MA_OWNED);
 837         KASSERT(TD_IS_SUSPENDED(td), ("Thread not suspended"));
 838         TD_CLR_SUSPENDED(td);
 839         p->p_suspcount--;
 840         return (setrunnable(td));
 841 }
 842
 843 /*
 844  * Allow all threads blocked by single threading to continue running.
 845  */
 846 void
 847 thread_unsuspend(struct proc *p)
 848 {
 849         struct thread *td;
 850         int wakeup_swapper;
 851
 852         PROC_LOCK_ASSERT(p, MA_OWNED);
 853         PROC_SLOCK_ASSERT(p, MA_OWNED);
 854         wakeup_swapper = 0;
 855         if (!P_SHOULDSTOP(p)) {
 856                 FOREACH_THREAD_IN_PROC(p, td) {
 857                         thread_lock(td);
 858                         if (TD_IS_SUSPENDED(td)) {
 859                                 wakeup_swapper |= thread_unsuspend_one(td);
 860                         }
 861                         thread_unlock(td);
 862                 }
 863         } else if ((P_SHOULDSTOP(p) == P_STOPPED_SINGLE) &&
 864             (p->p_numthreads == p->p_suspcount)) {
 865                 /*
 866                  * Stopping everything also did the job for the single
 867                  * threading request. Now we've downgraded to single-threaded,
 868                  * let it continue.
 869                  */
 870                 thread_lock(p->p_singlethread);
 871                 wakeup_swapper = thread_unsuspend_one(p->p_singlethread);
 872                 thread_unlock(p->p_singlethread);
 873         }
 874         if (wakeup_swapper)
 875                 kick_proc0();
 876 }
 877
 878 /*
 879  * End the single threading mode..
 880  */
 881 void
 882 thread_single_end(void)
 883 {
 884         struct thread *td;
 885         struct proc *p;
 886         int wakeup_swapper;
 887
 888         td = curthread;
 889         p = td->td_proc;
 890         PROC_LOCK_ASSERT(p, MA_OWNED);
 891         p->p_flag &= ~(P_STOPPED_SINGLE | P_SINGLE_EXIT | P_SINGLE_BOUNDARY);
 892         PROC_SLOCK(p);
 893         p->p_singlethread = NULL;
 894         wakeup_swapper = 0;
 895         /*
 896          * If there are other threads they may now run,
 897          * unless of course there is a blanket 'stop order'
 898          * on the process. The single threader must be allowed
 899          * to continue however as this is a bad place to stop.
 900          */
 901         if ((p->p_numthreads != 1) && (!P_SHOULDSTOP(p))) {
 902                 FOREACH_THREAD_IN_PROC(p, td) {
 903                         thread_lock(td);
 904                         if (TD_IS_SUSPENDED(td)) {
 905                                 wakeup_swapper |= thread_unsuspend_one(td);
 906                         }
 907                         thread_unlock(td);
 908                 }
 909         }
 910         PROC_SUNLOCK(p);
 911         if (wakeup_swapper)
 912                 kick_proc0();
 913 }
 914
 915 struct thread *
 916 thread_find(struct proc *p, lwpid_t tid)
 917 {
 918         struct thread *td;
 919
 920         PROC_LOCK_ASSERT(p, MA_OWNED);
 921         FOREACH_THREAD_IN_PROC(p, td) {
 922                 if (td->td_tid == tid)
 923                         break;
 924         }
 925         return (td);
 926 }