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/sched.h>
  44 #include <sys/sleepqueue.h>
  45 #include <sys/selinfo.h>
  46 #include <sys/turnstile.h>
  47 #include <sys/ktr.h>
  48 #include <sys/rwlock.h>
  49 #include <sys/umtx.h>
  50 #include <sys/cpuset.h>
  51 #ifdef  HWPMC_HOOKS
  52 #include <sys/pmckern.h>
  53 #endif
  54
  55 #include <security/audit/audit.h>
  56
  57 #include <vm/vm.h>
  58 #include <vm/vm_extern.h>
  59 #include <vm/uma.h>
  60 #include <sys/eventhandler.h>
  61
  62 /*
  63  * thread related storage.
  64  */
  65 static uma_zone_t thread_zone;
  66
  67 TAILQ_HEAD(, thread) zombie_threads = TAILQ_HEAD_INITIALIZER(zombie_threads);
  68 static struct mtx zombie_lock;
  69 MTX_SYSINIT(zombie_lock, &zombie_lock, "zombie lock", MTX_SPIN);
  70
  71 static void thread_zombie(struct thread *);
  72
  73 #define TID_BUFFER_SIZE 1024
  74
  75 struct mtx tid_lock;
  76 static struct unrhdr *tid_unrhdr;
  77 static lwpid_t tid_buffer[TID_BUFFER_SIZE];
  78 static int tid_head, tid_tail;
  79 static MALLOC_DEFINE(M_TIDHASH, "tidhash", "thread hash");
  80
  81 struct  tidhashhead *tidhashtbl;
  82 u_long  tidhash;
  83 struct  rwlock tidhash_lock;
  84
  85 static lwpid_t
  86 tid_alloc(void)
  87 {
  88         lwpid_t tid;
  89
  90         tid = alloc_unr(tid_unrhdr);
  91         if (tid != -1)
  92                 return (tid);
  93         mtx_lock(&tid_lock);
  94         if (tid_head == tid_tail) {
  95                 mtx_unlock(&tid_lock);
  96                 return (-1);
  97         }
  98         tid = tid_buffer[tid_head++];
  99         tid_head %= TID_BUFFER_SIZE;
 100         mtx_unlock(&tid_lock);
 101         return (tid);
 102 }
 103
 104 static void
 105 tid_free(lwpid_t tid)
 106 {
 107         lwpid_t tmp_tid = -1;
 108
 109         mtx_lock(&tid_lock);
 110         if ((tid_tail + 1) % TID_BUFFER_SIZE == tid_head) {
 111                 tmp_tid = tid_buffer[tid_head++];
 112                 tid_head = (tid_head + 1) % TID_BUFFER_SIZE;
 113         }
 114         tid_buffer[tid_tail++] = tid;
 115         tid_tail %= TID_BUFFER_SIZE;
 116         mtx_unlock(&tid_lock);
 117         if (tmp_tid != -1)
 118                 free_unr(tid_unrhdr, tmp_tid);
 119 }
 120
 121 /*
 122  * Prepare a thread for use.
 123  */
 124 static int
 125 thread_ctor(void *mem, int size, void *arg, int flags)
 126 {
 127         struct thread   *td;
 128
 129         td = (struct thread *)mem;
 130         td->td_state = TDS_INACTIVE;
 131         td->td_oncpu = NOCPU;
 132
 133         td->td_tid = tid_alloc();
 134
 135         /*
 136          * Note that td_critnest begins life as 1 because the thread is not
 137          * running and is thereby implicitly waiting to be on the receiving
 138          * end of a context switch.
 139          */
 140         td->td_critnest = 1;
 141         td->td_lend_user_pri = PRI_MAX;
 142         EVENTHANDLER_INVOKE(thread_ctor, td);
 143 #ifdef AUDIT
 144         audit_thread_alloc(td);
 145 #endif
 146         umtx_thread_alloc(td);
 147         return (0);
 148 }
 149
 150 /*
 151  * Reclaim a thread after use.
 152  */
 153 static void
 154 thread_dtor(void *mem, int size, void *arg)
 155 {
 156         struct thread *td;
 157
 158         td = (struct thread *)mem;
 159
 160 #ifdef INVARIANTS
 161         /* Verify that this thread is in a safe state to free. */
 162         switch (td->td_state) {
 163         case TDS_INHIBITED:
 164         case TDS_RUNNING:
 165         case TDS_CAN_RUN:
 166         case TDS_RUNQ:
 167                 /*
 168                  * We must never unlink a thread that is in one of
 169                  * these states, because it is currently active.
 170                  */
 171                 panic("bad state for thread unlinking");
 172                 /* NOTREACHED */
 173         case TDS_INACTIVE:
 174                 break;
 175         default:
 176                 panic("bad thread state");
 177                 /* NOTREACHED */
 178         }
 179 #endif
 180 #ifdef AUDIT
 181         audit_thread_free(td);
 182 #endif
 183         /* Free all OSD associated to this thread. */
 184         osd_thread_exit(td);
 185
 186         EVENTHANDLER_INVOKE(thread_dtor, td);
 187         tid_free(td->td_tid);
 188 }
 189
 190 /*
 191  * Initialize type-stable parts of a thread (when newly created).
 192  */
 193 static int
 194 thread_init(void *mem, int size, int flags)
 195 {
 196         struct thread *td;
 197
 198         td = (struct thread *)mem;
 199
 200         td->td_sleepqueue = sleepq_alloc();
 201         td->td_turnstile = turnstile_alloc();
 202         EVENTHANDLER_INVOKE(thread_init, td);
 203         td->td_sched = (struct td_sched *)&td[1];
 204         umtx_thread_init(td);
 205         td->td_kstack = 0;
 206         return (0);
 207 }
 208
 209 /*
 210  * Tear down type-stable parts of a thread (just before being discarded).
 211  */
 212 static void
 213 thread_fini(void *mem, int size)
 214 {
 215         struct thread *td;
 216
 217         td = (struct thread *)mem;
 218         EVENTHANDLER_INVOKE(thread_fini, td);
 219         turnstile_free(td->td_turnstile);
 220         sleepq_free(td->td_sleepqueue);
 221         umtx_thread_fini(td);
 222         seltdfini(td);
 223 }
 224
 225 /*
 226  * For a newly created process,
 227  * link up all the structures and its initial threads etc.
 228  * called from:
 229  * {arch}/{arch}/machdep.c   ia64_init(), init386() etc.
 230  * proc_dtor() (should go away)
 231  * proc_init()
 232  */
 233 void
 234 proc_linkup0(struct proc *p, struct thread *td)
 235 {
 236         TAILQ_INIT(&p->p_threads);           /* all threads in proc */
 237         proc_linkup(p, td);
 238 }
 239
 240 void
 241 proc_linkup(struct proc *p, struct thread *td)
 242 {
 243
 244         sigqueue_init(&p->p_sigqueue, p);
 245         p->p_ksi = ksiginfo_alloc(1);
 246         if (p->p_ksi != NULL) {
 247                 /* XXX p_ksi may be null if ksiginfo zone is not ready */
 248                 p->p_ksi->ksi_flags = KSI_EXT | KSI_INS;
 249         }
 250         LIST_INIT(&p->p_mqnotifier);
 251         p->p_numthreads = 0;
 252         thread_link(td, p);
 253 }
 254
 255 /*
 256  * Initialize global thread allocation resources.
 257  */
 258 void
 259 threadinit(void)
 260 {
 261
 262         mtx_init(&tid_lock, "TID lock", NULL, MTX_DEF);
 263         /* leave one number for thread0 */
 264         tid_unrhdr = new_unrhdr(PID_MAX + 2, INT_MAX, &tid_lock);
 265
 266         thread_zone = uma_zcreate("THREAD", sched_sizeof_thread(),
 267             thread_ctor, thread_dtor, thread_init, thread_fini,
 268             16 - 1, 0);
 269         tidhashtbl = hashinit(maxproc / 2, M_TIDHASH, &tidhash);
 270         rw_init(&tidhash_lock, "tidhash");
 271 }
 272
 273 /*
 274  * Place an unused thread on the zombie list.
 275  * Use the slpq as that must be unused by now.
 276  */
 277 void
 278 thread_zombie(struct thread *td)
 279 {
 280         mtx_lock_spin(&zombie_lock);
 281         TAILQ_INSERT_HEAD(&zombie_threads, td, td_slpq);
 282         mtx_unlock_spin(&zombie_lock);
 283 }
 284
 285 /*
 286  * Release a thread that has exited after cpu_throw().
 287  */
 288 void
 289 thread_stash(struct thread *td)
 290 {
 291         atomic_subtract_rel_int(&td->td_proc->p_exitthreads, 1);
 292         thread_zombie(td);
 293 }
 294
 295 /*
 296  * Reap zombie resources.
 297  */
 298 void
 299 thread_reap(void)
 300 {
 301         struct thread *td_first, *td_next;
 302
 303         /*
 304          * Don't even bother to lock if none at this instant,
 305          * we really don't care about the next instant..
 306          */
 307         if (!TAILQ_EMPTY(&zombie_threads)) {
 308                 mtx_lock_spin(&zombie_lock);
 309                 td_first = TAILQ_FIRST(&zombie_threads);
 310                 if (td_first)
 311                         TAILQ_INIT(&zombie_threads);
 312                 mtx_unlock_spin(&zombie_lock);
 313                 while (td_first) {
 314                         td_next = TAILQ_NEXT(td_first, td_slpq);
 315                         if (td_first->td_ucred)
 316                                 crfree(td_first->td_ucred);
 317                         thread_free(td_first);
 318                         td_first = td_next;
 319                 }
 320         }
 321 }
 322
 323 /*
 324  * Allocate a thread.
 325  */
 326 struct thread *
 327 thread_alloc(int pages)
 328 {
 329         struct thread *td;
 330
 331         thread_reap(); /* check if any zombies to get */
 332
 333         td = (struct thread *)uma_zalloc(thread_zone, M_WAITOK);
 334         KASSERT(td->td_kstack == 0, ("thread_alloc got thread with kstack"));
 335         if (!vm_thread_new(td, pages)) {
 336                 uma_zfree(thread_zone, td);
 337                 return (NULL);
 338         }
 339         cpu_thread_alloc(td);
 340         return (td);
 341 }
 342
 343 int
 344 thread_alloc_stack(struct thread *td, int pages)
 345 {
 346
 347         KASSERT(td->td_kstack == 0,
 348             ("thread_alloc_stack called on a thread with kstack"));
 349         if (!vm_thread_new(td, pages))
 350                 return (0);
 351         cpu_thread_alloc(td);
 352         return (1);
 353 }
 354
 355 /*
 356  * Deallocate a thread.
 357  */
 358 void
 359 thread_free(struct thread *td)
 360 {
 361
 362         lock_profile_thread_exit(td);
 363         if (td->td_cpuset)
 364                 cpuset_rel(td->td_cpuset);
 365         td->td_cpuset = NULL;
 366         cpu_thread_free(td);
 367         if (td->td_kstack != 0)
 368                 vm_thread_dispose(td);
 369         uma_zfree(thread_zone, td);
 370 }
 371
 372 /*
 373  * Discard the current thread and exit from its context.
 374  * Always called with scheduler locked.
 375  *
 376  * Because we can't free a thread while we're operating under its context,
 377  * push the current thread into our CPU's deadthread holder. This means
 378  * we needn't worry about someone else grabbing our context before we
 379  * do a cpu_throw().
 380  */
 381 void
 382 thread_exit(void)
 383 {
 384         uint64_t runtime, new_switchtime;
 385         struct thread *td;
 386         struct thread *td2;
 387         struct proc *p;
 388         int wakeup_swapper;
 389
 390         td = curthread;
 391         p = td->td_proc;
 392
 393         PROC_SLOCK_ASSERT(p, MA_OWNED);
 394         mtx_assert(&Giant, MA_NOTOWNED);
 395
 396         PROC_LOCK_ASSERT(p, MA_OWNED);
 397         KASSERT(p != NULL, ("thread exiting without a process"));
 398         CTR3(KTR_PROC, "thread_exit: thread %p (pid %ld, %s)", td,
 399             (long)p->p_pid, td->td_name);
 400         KASSERT(TAILQ_EMPTY(&td->td_sigqueue.sq_list), ("signal pending"));
 401
 402 #ifdef AUDIT
 403         AUDIT_SYSCALL_EXIT(0, td);
 404 #endif
 405         umtx_thread_exit(td);
 406         /*
 407          * drop FPU & debug register state storage, or any other
 408          * architecture specific resources that
 409          * would not be on a new untouched process.
 410          */
 411         cpu_thread_exit(td);    /* XXXSMP */
 412
 413         /*
 414          * The last thread is left attached to the process
 415          * So that the whole bundle gets recycled. Skip
 416          * all this stuff if we never had threads.
 417          * EXIT clears all sign of other threads when
 418          * it goes to single threading, so the last thread always
 419          * takes the short path.
 420          */
 421         if (p->p_flag & P_HADTHREADS) {
 422                 if (p->p_numthreads > 1) {
 423                         thread_unlink(td);
 424                         td2 = FIRST_THREAD_IN_PROC(p);
 425                         sched_exit_thread(td2, td);
 426
 427                         /*
 428                          * The test below is NOT true if we are the
 429                          * sole exiting thread. P_STOPPED_SINGLE is unset
 430                          * in exit1() after it is the only survivor.
 431                          */
 432                         if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE) {
 433                                 if (p->p_numthreads == p->p_suspcount) {
 434                                         thread_lock(p->p_singlethread);
 435                                         wakeup_swapper = thread_unsuspend_one(
 436                                                 p->p_singlethread);
 437                                         thread_unlock(p->p_singlethread);
 438                                         if (wakeup_swapper)
 439                                                 kick_proc0();
 440                                 }
 441                         }
 442
 443                         atomic_add_int(&td->td_proc->p_exitthreads, 1);
 444                         PCPU_SET(deadthread, td);
 445                 } else {
 446                         /*
 447                          * The last thread is exiting.. but not through exit()
 448                          */
 449                         panic ("thread_exit: Last thread exiting on its own");
 450                 }
 451         }
 452 #ifdef  HWPMC_HOOKS
 453         /*
 454          * If this thread is part of a process that is being tracked by hwpmc(4),
 455          * inform the module of the thread's impending exit.
 456          */
 457         if (PMC_PROC_IS_USING_PMCS(td->td_proc))
 458                 PMC_SWITCH_CONTEXT(td, PMC_FN_CSW_OUT);
 459 #endif
 460         PROC_UNLOCK(p);
 461
 462         /* Do the same timestamp bookkeeping that mi_switch() would do. */
 463         new_switchtime = cpu_ticks();
 464         runtime = new_switchtime - PCPU_GET(switchtime);
 465         td->td_runtime += runtime;
 466         td->td_incruntime += runtime;
 467         PCPU_SET(switchtime, new_switchtime);
 468         PCPU_SET(switchticks, ticks);
 469         PCPU_INC(cnt.v_swtch);
 470
 471         /* Save our resource usage in our process. */
 472         td->td_ru.ru_nvcsw++;
 473         ruxagg(p, td);
 474         rucollect(&p->p_ru, &td->td_ru);
 475
 476         thread_lock(td);
 477         PROC_SUNLOCK(p);
 478         td->td_state = TDS_INACTIVE;
 479 #ifdef WITNESS
 480         witness_thread_exit(td);
 481 #endif
 482         CTR1(KTR_PROC, "thread_exit: cpu_throw() thread %p", td);
 483         sched_throw(td);
 484         panic("I'm a teapot!");
 485         /* NOTREACHED */
 486 }
 487
 488 /*
 489  * Do any thread specific cleanups that may be needed in wait()
 490  * called with Giant, proc and schedlock not held.
 491  */
 492 void
 493 thread_wait(struct proc *p)
 494 {
 495         struct thread *td;
 496
 497         mtx_assert(&Giant, MA_NOTOWNED);
 498         KASSERT((p->p_numthreads == 1), ("Multiple threads in wait1()"));
 499         td = FIRST_THREAD_IN_PROC(p);
 500         /* Lock the last thread so we spin until it exits cpu_throw(). */
 501         thread_lock(td);
 502         thread_unlock(td);
 503         /* Wait for any remaining threads to exit cpu_throw(). */
 504         while (p->p_exitthreads)
 505                 sched_relinquish(curthread);
 506         lock_profile_thread_exit(td);
 507         cpuset_rel(td->td_cpuset);
 508         td->td_cpuset = NULL;
 509         cpu_thread_clean(td);
 510         crfree(td->td_ucred);
 511         thread_reap();  /* check for zombie threads etc. */
 512 }
 513
 514 /*
 515  * Link a thread to a process.
 516  * set up anything that needs to be initialized for it to
 517  * be used by the process.
 518  */
 519 void
 520 thread_link(struct thread *td, struct proc *p)
 521 {
 522
 523         /*
 524          * XXX This can't be enabled because it's called for proc0 before
 525          * its lock has been created.
 526          * PROC_LOCK_ASSERT(p, MA_OWNED);
 527          */
 528         td->td_state    = TDS_INACTIVE;
 529         td->td_proc     = p;
 530         td->td_flags    = TDF_INMEM;
 531
 532         LIST_INIT(&td->td_contested);
 533         LIST_INIT(&td->td_lprof[0]);
 534         LIST_INIT(&td->td_lprof[1]);
 535         sigqueue_init(&td->td_sigqueue, p);
 536         callout_init(&td->td_slpcallout, CALLOUT_MPSAFE);
 537         TAILQ_INSERT_HEAD(&p->p_threads, td, td_plist);
 538         p->p_numthreads++;
 539 }
 540
 541 /*
 542  * Convert a process with one thread to an unthreaded process.
 543  */
 544 void
 545 thread_unthread(struct thread *td)
 546 {
 547         struct proc *p = td->td_proc;
 548
 549         KASSERT((p->p_numthreads == 1), ("Unthreading with >1 threads"));
 550         p->p_flag &= ~P_HADTHREADS;
 551 }
 552
 553 /*
 554  * Called from:
 555  *  thread_exit()
 556  */
 557 void
 558 thread_unlink(struct thread *td)
 559 {
 560         struct proc *p = td->td_proc;
 561
 562         PROC_LOCK_ASSERT(p, MA_OWNED);
 563         TAILQ_REMOVE(&p->p_threads, td, td_plist);
 564         p->p_numthreads--;
 565         /* could clear a few other things here */
 566         /* Must  NOT clear links to proc! */
 567 }
 568
 569 static int
 570 calc_remaining(struct proc *p, int mode)
 571 {
 572         int remaining;
 573
 574         PROC_LOCK_ASSERT(p, MA_OWNED);
 575         PROC_SLOCK_ASSERT(p, MA_OWNED);
 576         if (mode == SINGLE_EXIT)
 577                 remaining = p->p_numthreads;
 578         else if (mode == SINGLE_BOUNDARY)
 579                 remaining = p->p_numthreads - p->p_boundary_count;
 580         else if (mode == SINGLE_NO_EXIT)
 581                 remaining = p->p_numthreads - p->p_suspcount;
 582         else
 583                 panic("calc_remaining: wrong mode %d", mode);
 584         return (remaining);
 585 }
 586
 587 /*
 588  * Enforce single-threading.
 589  *
 590  * Returns 1 if the caller must abort (another thread is waiting to
 591  * exit the process or similar). Process is locked!
 592  * Returns 0 when you are successfully the only thread running.
 593  * A process has successfully single threaded in the suspend mode when
 594  * There are no threads in user mode. Threads in the kernel must be
 595  * allowed to continue until they get to the user boundary. They may even
 596  * copy out their return values and data before suspending. They may however be
 597  * accelerated in reaching the user boundary as we will wake up
 598  * any sleeping threads that are interruptable. (PCATCH).
 599  */
 600 int
 601 thread_single(int mode)
 602 {
 603         struct thread *td;
 604         struct thread *td2;
 605         struct proc *p;
 606         int remaining, wakeup_swapper;
 607
 608         td = curthread;
 609         p = td->td_proc;
 610         mtx_assert(&Giant, MA_NOTOWNED);
 611         PROC_LOCK_ASSERT(p, MA_OWNED);
 612         KASSERT((td != NULL), ("curthread is NULL"));
 613
 614         if ((p->p_flag & P_HADTHREADS) == 0)
 615                 return (0);
 616
 617         /* Is someone already single threading? */
 618         if (p->p_singlethread != NULL && p->p_singlethread != td)
 619                 return (1);
 620
 621         if (mode == SINGLE_EXIT) {
 622                 p->p_flag |= P_SINGLE_EXIT;
 623                 p->p_flag &= ~P_SINGLE_BOUNDARY;
 624         } else {
 625                 p->p_flag &= ~P_SINGLE_EXIT;
 626                 if (mode == SINGLE_BOUNDARY)
 627                         p->p_flag |= P_SINGLE_BOUNDARY;
 628                 else
 629                         p->p_flag &= ~P_SINGLE_BOUNDARY;
 630         }
 631         p->p_flag |= P_STOPPED_SINGLE;
 632         PROC_SLOCK(p);
 633         p->p_singlethread = td;
 634         remaining = calc_remaining(p, mode);
 635         while (remaining != 1) {
 636                 if (P_SHOULDSTOP(p) != P_STOPPED_SINGLE)
 637                         goto stopme;
 638                 wakeup_swapper = 0;
 639                 FOREACH_THREAD_IN_PROC(p, td2) {
 640                         if (td2 == td)
 641                                 continue;
 642                         thread_lock(td2);
 643                         td2->td_flags |= TDF_ASTPENDING | TDF_NEEDSUSPCHK;
 644                         if (TD_IS_INHIBITED(td2)) {
 645                                 switch (mode) {
 646                                 case SINGLE_EXIT:
 647                                         if (TD_IS_SUSPENDED(td2))
 648                                                 wakeup_swapper |=
 649                                                     thread_unsuspend_one(td2);
 650                                         if (TD_ON_SLEEPQ(td2) &&
 651                                             (td2->td_flags & TDF_SINTR))
 652                                                 wakeup_swapper |=
 653                                                     sleepq_abort(td2, EINTR);
 654                                         break;
 655                                 case SINGLE_BOUNDARY:
 656                                         if (TD_IS_SUSPENDED(td2) &&
 657                                             !(td2->td_flags & TDF_BOUNDARY))
 658                                                 wakeup_swapper |=
 659                                                     thread_unsuspend_one(td2);
 660                                         if (TD_ON_SLEEPQ(td2) &&
 661                                             (td2->td_flags & TDF_SINTR))
 662                                                 wakeup_swapper |=
 663                                                     sleepq_abort(td2, ERESTART);
 664                                         break;
 665                                 case SINGLE_NO_EXIT:
 666                                         if (TD_IS_SUSPENDED(td2) &&
 667                                             !(td2->td_flags & TDF_BOUNDARY))
 668                                                 wakeup_swapper |=
 669                                                     thread_unsuspend_one(td2);
 670                                         if (TD_ON_SLEEPQ(td2) &&
 671                                             (td2->td_flags & TDF_SINTR))
 672                                                 wakeup_swapper |=
 673                                                     sleepq_abort(td2, ERESTART);
 674                                         break;
 675                                 default:
 676                                         break;
 677                                 }
 678                         }
 679 #ifdef SMP
 680                         else if (TD_IS_RUNNING(td2) && td != td2) {
 681                                 forward_signal(td2);
 682                         }
 683 #endif
 684                         thread_unlock(td2);
 685                 }
 686                 if (wakeup_swapper)
 687                         kick_proc0();
 688                 remaining = calc_remaining(p, mode);
 689
 690                 /*
 691                  * Maybe we suspended some threads.. was it enough?
 692                  */
 693                 if (remaining == 1)
 694                         break;
 695
 696 stopme:
 697                 /*
 698                  * Wake us up when everyone else has suspended.
 699                  * In the mean time we suspend as well.
 700                  */
 701                 thread_suspend_switch(td);
 702                 remaining = calc_remaining(p, mode);
 703         }
 704         if (mode == SINGLE_EXIT) {
 705                 /*
 706                  * We have gotten rid of all the other threads and we
 707                  * are about to either exit or exec. In either case,
 708                  * we try our utmost  to revert to being a non-threaded
 709                  * process.
 710                  */
 711                 p->p_singlethread = NULL;
 712                 p->p_flag &= ~(P_STOPPED_SINGLE | P_SINGLE_EXIT);
 713                 thread_unthread(td);
 714         }
 715         PROC_SUNLOCK(p);
 716         return (0);
 717 }
 718
 719 /*
 720  * Called in from locations that can safely check to see
 721  * whether we have to suspend or at least throttle for a
 722  * single-thread event (e.g. fork).
 723  *
 724  * Such locations include userret().
 725  * If the "return_instead" argument is non zero, the thread must be able to
 726  * accept 0 (caller may continue), or 1 (caller must abort) as a result.
 727  *
 728  * The 'return_instead' argument tells the function if it may do a
 729  * thread_exit() or suspend, or whether the caller must abort and back
 730  * out instead.
 731  *
 732  * If the thread that set the single_threading request has set the
 733  * P_SINGLE_EXIT bit in the process flags then this call will never return
 734  * if 'return_instead' is false, but will exit.
 735  *
 736  * P_SINGLE_EXIT | return_instead == 0| return_instead != 0
 737  *---------------+--------------------+---------------------
 738  *       0       | returns 0          |   returns 0 or 1
 739  *               | when ST ends       |   immediatly
 740  *---------------+--------------------+---------------------
 741  *       1       | thread exits       |   returns 1
 742  *               |                    |  immediatly
 743  * 0 = thread_exit() or suspension ok,
 744  * other = return error instead of stopping the thread.
 745  *
 746  * While a full suspension is under effect, even a single threading
 747  * thread would be suspended if it made this call (but it shouldn't).
 748  * This call should only be made from places where
 749  * thread_exit() would be safe as that may be the outcome unless
 750  * return_instead is set.
 751  */
 752 int
 753 thread_suspend_check(int return_instead)
 754 {
 755         struct thread *td;
 756         struct proc *p;
 757         int wakeup_swapper;
 758
 759         td = curthread;
 760         p = td->td_proc;
 761         mtx_assert(&Giant, MA_NOTOWNED);
 762         PROC_LOCK_ASSERT(p, MA_OWNED);
 763         while (P_SHOULDSTOP(p) ||
 764               ((p->p_flag & P_TRACED) && (td->td_dbgflags & TDB_SUSPEND))) {
 765                 if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE) {
 766                         KASSERT(p->p_singlethread != NULL,
 767                             ("singlethread not set"));
 768                         /*
 769                          * The only suspension in action is a
 770                          * single-threading. Single threader need not stop.
 771                          * XXX Should be safe to access unlocked
 772                          * as it can only be set to be true by us.
 773                          */
 774                         if (p->p_singlethread == td)
 775                                 return (0);     /* Exempt from stopping. */
 776                 }
 777                 if ((p->p_flag & P_SINGLE_EXIT) && return_instead)
 778                         return (EINTR);
 779
 780                 /* Should we goto user boundary if we didn't come from there? */
 781                 if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE &&
 782                     (p->p_flag & P_SINGLE_BOUNDARY) && return_instead)
 783                         return (ERESTART);
 784
 785                 /*
 786                  * If the process is waiting for us to exit,
 787                  * this thread should just suicide.
 788                  * Assumes that P_SINGLE_EXIT implies P_STOPPED_SINGLE.
 789                  */
 790                 if ((p->p_flag & P_SINGLE_EXIT) && (p->p_singlethread != td)) {
 791                         PROC_UNLOCK(p);
 792                         tidhash_remove(td);
 793                         PROC_LOCK(p);
 794                         tdsigcleanup(td);
 795                         PROC_SLOCK(p);
 796                         thread_stopped(p);
 797                         thread_exit();
 798                 }
 799
 800                 PROC_SLOCK(p);
 801                 thread_stopped(p);
 802                 if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE) {
 803                         if (p->p_numthreads == p->p_suspcount + 1) {
 804                                 thread_lock(p->p_singlethread);
 805                                 wakeup_swapper =
 806                                     thread_unsuspend_one(p->p_singlethread);
 807                                 thread_unlock(p->p_singlethread);
 808                                 if (wakeup_swapper)
 809                                         kick_proc0();
 810                         }
 811                 }
 812                 PROC_UNLOCK(p);
 813                 thread_lock(td);
 814                 /*
 815                  * When a thread suspends, it just
 816                  * gets taken off all queues.
 817                  */
 818                 thread_suspend_one(td);
 819                 if (return_instead == 0) {
 820                         p->p_boundary_count++;
 821                         td->td_flags |= TDF_BOUNDARY;
 822                 }
 823                 PROC_SUNLOCK(p);
 824                 mi_switch(SW_INVOL | SWT_SUSPEND, NULL);
 825                 if (return_instead == 0)
 826                         td->td_flags &= ~TDF_BOUNDARY;
 827                 thread_unlock(td);
 828                 PROC_LOCK(p);
 829                 if (return_instead == 0) {
 830                         PROC_SLOCK(p);
 831                         p->p_boundary_count--;
 832                         PROC_SUNLOCK(p);
 833                 }
 834         }
 835         return (0);
 836 }
 837
 838 void
 839 thread_suspend_switch(struct thread *td)
 840 {
 841         struct proc *p;
 842
 843         p = td->td_proc;
 844         KASSERT(!TD_IS_SUSPENDED(td), ("already suspended"));
 845         PROC_LOCK_ASSERT(p, MA_OWNED);
 846         PROC_SLOCK_ASSERT(p, MA_OWNED);
 847         /*
 848          * We implement thread_suspend_one in stages here to avoid
 849          * dropping the proc lock while the thread lock is owned.
 850          */
 851         thread_stopped(p);
 852         p->p_suspcount++;
 853         PROC_UNLOCK(p);
 854         thread_lock(td);
 855         td->td_flags &= ~TDF_NEEDSUSPCHK;
 856         TD_SET_SUSPENDED(td);
 857         sched_sleep(td, 0);
 858         PROC_SUNLOCK(p);
 859         DROP_GIANT();
 860         mi_switch(SW_VOL | SWT_SUSPEND, NULL);
 861         thread_unlock(td);
 862         PICKUP_GIANT();
 863         PROC_LOCK(p);
 864         PROC_SLOCK(p);
 865 }
 866
 867 void
 868 thread_suspend_one(struct thread *td)
 869 {
 870         struct proc *p = td->td_proc;
 871
 872         PROC_SLOCK_ASSERT(p, MA_OWNED);
 873         THREAD_LOCK_ASSERT(td, MA_OWNED);
 874         KASSERT(!TD_IS_SUSPENDED(td), ("already suspended"));
 875         p->p_suspcount++;
 876         td->td_flags &= ~TDF_NEEDSUSPCHK;
 877         TD_SET_SUSPENDED(td);
 878         sched_sleep(td, 0);
 879 }
 880
 881 int
 882 thread_unsuspend_one(struct thread *td)
 883 {
 884         struct proc *p = td->td_proc;
 885
 886         PROC_SLOCK_ASSERT(p, MA_OWNED);
 887         THREAD_LOCK_ASSERT(td, MA_OWNED);
 888         KASSERT(TD_IS_SUSPENDED(td), ("Thread not suspended"));
 889         TD_CLR_SUSPENDED(td);
 890         p->p_suspcount--;
 891         return (setrunnable(td));
 892 }
 893
 894 /*
 895  * Allow all threads blocked by single threading to continue running.
 896  */
 897 void
 898 thread_unsuspend(struct proc *p)
 899 {
 900         struct thread *td;
 901         int wakeup_swapper;
 902
 903         PROC_LOCK_ASSERT(p, MA_OWNED);
 904         PROC_SLOCK_ASSERT(p, MA_OWNED);
 905         wakeup_swapper = 0;
 906         if (!P_SHOULDSTOP(p)) {
 907                 FOREACH_THREAD_IN_PROC(p, td) {
 908                         thread_lock(td);
 909                         if (TD_IS_SUSPENDED(td)) {
 910                                 wakeup_swapper |= thread_unsuspend_one(td);
 911                         }
 912                         thread_unlock(td);
 913                 }
 914         } else if ((P_SHOULDSTOP(p) == P_STOPPED_SINGLE) &&
 915             (p->p_numthreads == p->p_suspcount)) {
 916                 /*
 917                  * Stopping everything also did the job for the single
 918                  * threading request. Now we've downgraded to single-threaded,
 919                  * let it continue.
 920                  */
 921                 thread_lock(p->p_singlethread);
 922                 wakeup_swapper = thread_unsuspend_one(p->p_singlethread);
 923                 thread_unlock(p->p_singlethread);
 924         }
 925         if (wakeup_swapper)
 926                 kick_proc0();
 927 }
 928
 929 /*
 930  * End the single threading mode..
 931  */
 932 void
 933 thread_single_end(void)
 934 {
 935         struct thread *td;
 936         struct proc *p;
 937         int wakeup_swapper;
 938
 939         td = curthread;
 940         p = td->td_proc;
 941         PROC_LOCK_ASSERT(p, MA_OWNED);
 942         p->p_flag &= ~(P_STOPPED_SINGLE | P_SINGLE_EXIT | P_SINGLE_BOUNDARY);
 943         PROC_SLOCK(p);
 944         p->p_singlethread = NULL;
 945         wakeup_swapper = 0;
 946         /*
 947          * If there are other threads they may now run,
 948          * unless of course there is a blanket 'stop order'
 949          * on the process. The single threader must be allowed
 950          * to continue however as this is a bad place to stop.
 951          */
 952         if ((p->p_numthreads != 1) && (!P_SHOULDSTOP(p))) {
 953                 FOREACH_THREAD_IN_PROC(p, td) {
 954                         thread_lock(td);
 955                         if (TD_IS_SUSPENDED(td)) {
 956                                 wakeup_swapper |= thread_unsuspend_one(td);
 957                         }
 958                         thread_unlock(td);
 959                 }
 960         }
 961         PROC_SUNLOCK(p);
 962         if (wakeup_swapper)
 963                 kick_proc0();
 964 }
 965
 966 struct thread *
 967 thread_find(struct proc *p, lwpid_t tid)
 968 {
 969         struct thread *td;
 970
 971         PROC_LOCK_ASSERT(p, MA_OWNED);
 972         FOREACH_THREAD_IN_PROC(p, td) {
 973                 if (td->td_tid == tid)
 974                         break;
 975         }
 976         return (td);
 977 }
 978
 979 /* Locate a thread by number; return with proc lock held. */
 980 struct thread *
 981 tdfind(lwpid_t tid, pid_t pid)
 982 {
 983 #define RUN_THRESH      16
 984         struct thread *td;
 985         int run = 0;
 986
 987         rw_rlock(&tidhash_lock);
 988         LIST_FOREACH(td, TIDHASH(tid), td_hash) {
 989                 if (td->td_tid == tid) {
 990                         if (pid != -1 && td->td_proc->p_pid != pid) {
 991                                 td = NULL;
 992                                 break;
 993                         }
 994                         PROC_LOCK(td->td_proc);
 995                         if (td->td_proc->p_state == PRS_NEW) {
 996                                 PROC_UNLOCK(td->td_proc);
 997                                 td = NULL;
 998                                 break;
 999                         }
1000                         if (run > RUN_THRESH) {
1001                                 if (rw_try_upgrade(&tidhash_lock)) {
1002                                         LIST_REMOVE(td, td_hash);
1003                                         LIST_INSERT_HEAD(TIDHASH(td->td_tid),
1004                                                 td, td_hash);
1005                                         rw_wunlock(&tidhash_lock);
1006                                         return (td);
1007                                 }
1008                         }
1009                         break;
1010                 }
1011                 run++;
1012         }
1013         rw_runlock(&tidhash_lock);
1014         return (td);
1015 }
1016
1017 void
1018 tidhash_add(struct thread *td)
1019 {
1020         rw_wlock(&tidhash_lock);
1021         LIST_INSERT_HEAD(TIDHASH(td->td_tid), td, td_hash);
1022         rw_wunlock(&tidhash_lock);
1023 }
1024
1025 void
1026 tidhash_remove(struct thread *td)
1027 {
1028         rw_wlock(&tidhash_lock);
1029         LIST_REMOVE(td, td_hash);
1030         rw_wunlock(&tidhash_lock);
1031 }