sys/kern/kern_thread.c

   1 /*-
   2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
   3  *
   4  * Copyright (C) 2001 Julian Elischer <julian@freebsd.org>.
   5  *  All rights reserved.
   6  *
   7  * Redistribution and use in source and binary forms, with or without
   8  * modification, are permitted provided that the following conditions
   9  * are met:
  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.
  17  *
  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
  28  * DAMAGE.
  29  */
  30
  31 #include "opt_witness.h"
  32 #include "opt_hwpmc_hooks.h"
  33
  34 #include <sys/cdefs.h>
  35 __FBSDID("$FreeBSD$");
  36
  37 #include <sys/param.h>
  38 #include <sys/systm.h>
  39 #include <sys/kernel.h>
  40 #include <sys/lock.h>
  41 #include <sys/mutex.h>
  42 #include <sys/proc.h>
  43 #include <sys/epoch.h>
  44 #include <sys/rangelock.h>
  45 #include <sys/resourcevar.h>
  46 #include <sys/sdt.h>
  47 #include <sys/smp.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>
  54 #include <sys/ktr.h>
  55 #include <sys/rwlock.h>
  56 #include <sys/umtx.h>
  57 #include <sys/vmmeter.h>
  58 #include <sys/cpuset.h>
  59 #ifdef  HWPMC_HOOKS
  60 #include <sys/pmckern.h>
  61 #endif
  62
  63 #include <security/audit/audit.h>
  64
  65 #include <vm/vm.h>
  66 #include <vm/vm_extern.h>
  67 #include <vm/uma.h>
  68 #include <sys/eventhandler.h>
  69
  70 /*
  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
  74  * assert update.
  75  *
  76  * On the stable branches after KBI freeze, conditions must not be
  77  * violated.  Typically new fields are moved to the end of the
  78  * structures.
  79  */
  80 #ifdef __amd64__
  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) == 0x548,
  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");
  99 #endif
 100 #ifdef __i386__
 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");
 119 #endif
 120
 121 SDT_PROVIDER_DECLARE(proc);
 122 SDT_PROBE_DEFINE(proc, , , lwp__exit);
 123
 124 /*
 125  * thread related storage.
 126  */
 127 static uma_zone_t thread_zone;
 128
 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);
 132
 133 static void thread_zombie(struct thread *);
 134 static int thread_unsuspend_one(struct thread *td, struct proc *p,
 135     bool boundary);
 136
 137 #define TID_BUFFER_SIZE 1024
 138
 139 struct mtx tid_lock;
 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");
 144
 145 struct  tidhashhead *tidhashtbl;
 146 u_long  tidhash;
 147 struct  rwlock tidhash_lock;
 148
 149 EVENTHANDLER_LIST_DEFINE(thread_ctor);
 150 EVENTHANDLER_LIST_DEFINE(thread_dtor);
 151 EVENTHANDLER_LIST_DEFINE(thread_init);
 152 EVENTHANDLER_LIST_DEFINE(thread_fini);
 153
 154 static lwpid_t
 155 tid_alloc(void)
 156 {
 157         lwpid_t tid;
 158
 159         tid = alloc_unr(tid_unrhdr);
 160         if (tid != -1)
 161                 return (tid);
 162         mtx_lock(&tid_lock);
 163         if (tid_head == tid_tail) {
 164                 mtx_unlock(&tid_lock);
 165                 return (-1);
 166         }
 167         tid = tid_buffer[tid_head];
 168         tid_head = (tid_head + 1) % TID_BUFFER_SIZE;
 169         mtx_unlock(&tid_lock);
 170         return (tid);
 171 }
 172
 173 static void
 174 tid_free(lwpid_t tid)
 175 {
 176         lwpid_t tmp_tid = -1;
 177
 178         mtx_lock(&tid_lock);
 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;
 182         }
 183         tid_buffer[tid_tail] = tid;
 184         tid_tail = (tid_tail + 1) % TID_BUFFER_SIZE;
 185         mtx_unlock(&tid_lock);
 186         if (tmp_tid != -1)
 187                 free_unr(tid_unrhdr, tmp_tid);
 188 }
 189
 190 /*
 191  * Prepare a thread for use.
 192  */
 193 static int
 194 thread_ctor(void *mem, int size, void *arg, int flags)
 195 {
 196         struct thread   *td;
 197
 198         td = (struct thread *)mem;
 199         td->td_state = TDS_INACTIVE;
 200         td->td_lastcpu = td->td_oncpu = NOCPU;
 201
 202         td->td_tid = tid_alloc();
 203
 204         /*
 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.
 208          */
 209         td->td_critnest = 1;
 210         td->td_lend_user_pri = PRI_MAX;
 211         EVENTHANDLER_DIRECT_INVOKE(thread_ctor, td);
 212 #ifdef AUDIT
 213         audit_thread_alloc(td);
 214 #endif
 215         umtx_thread_alloc(td);
 216         return (0);
 217 }
 218
 219 /*
 220  * Reclaim a thread after use.
 221  */
 222 static void
 223 thread_dtor(void *mem, int size, void *arg)
 224 {
 225         struct thread *td;
 226
 227         td = (struct thread *)mem;
 228
 229 #ifdef INVARIANTS
 230         /* Verify that this thread is in a safe state to free. */
 231         switch (td->td_state) {
 232         case TDS_INHIBITED:
 233         case TDS_RUNNING:
 234         case TDS_CAN_RUN:
 235         case TDS_RUNQ:
 236                 /*
 237                  * We must never unlink a thread that is in one of
 238                  * these states, because it is currently active.
 239                  */
 240                 panic("bad state for thread unlinking");
 241                 /* NOTREACHED */
 242         case TDS_INACTIVE:
 243                 break;
 244         default:
 245                 panic("bad thread state");
 246                 /* NOTREACHED */
 247         }
 248 #endif
 249 #ifdef AUDIT
 250         audit_thread_free(td);
 251 #endif
 252         /* Free all OSD associated to this thread. */
 253         osd_thread_exit(td);
 254         td_softdep_cleanup(td);
 255         MPASS(td->td_su == NULL);
 256
 257         EVENTHANDLER_DIRECT_INVOKE(thread_dtor, td);
 258         tid_free(td->td_tid);
 259 }
 260
 261 /*
 262  * Initialize type-stable parts of a thread (when newly created).
 263  */
 264 static int
 265 thread_init(void *mem, int size, int flags)
 266 {
 267         struct thread *td;
 268
 269         td = (struct thread *)mem;
 270
 271         td->td_sleepqueue = sleepq_alloc();
 272         td->td_turnstile = turnstile_alloc();
 273         td->td_rlqe = NULL;
 274         EVENTHANDLER_DIRECT_INVOKE(thread_init, td);
 275         umtx_thread_init(td);
 276         epoch_thread_init(td);
 277         td->td_kstack = 0;
 278         td->td_sel = NULL;
 279         return (0);
 280 }
 281
 282 /*
 283  * Tear down type-stable parts of a thread (just before being discarded).
 284  */
 285 static void
 286 thread_fini(void *mem, int size)
 287 {
 288         struct thread *td;
 289
 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);
 297         seltdfini(td);
 298 }
 299
 300 /*
 301  * For a newly created process,
 302  * link up all the structures and its initial threads etc.
 303  * called from:
 304  * {arch}/{arch}/machdep.c   {arch}_init(), init386() etc.
 305  * proc_dtor() (should go away)
 306  * proc_init()
 307  */
 308 void
 309 proc_linkup0(struct proc *p, struct thread *td)
 310 {
 311         TAILQ_INIT(&p->p_threads);           /* all threads in proc */
 312         proc_linkup(p, td);
 313 }
 314
 315 void
 316 proc_linkup(struct proc *p, struct thread *td)
 317 {
 318
 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;
 324         }
 325         LIST_INIT(&p->p_mqnotifier);
 326         p->p_numthreads = 0;
 327         thread_link(td, p);
 328 }
 329
 330 /*
 331  * Initialize global thread allocation resources.
 332  */
 333 void
 334 threadinit(void)
 335 {
 336
 337         mtx_init(&tid_lock, "TID lock", NULL, MTX_DEF);
 338
 339         /*
 340          * pid_max cannot be greater than PID_MAX.
 341          * leave one number for thread0.
 342          */
 343         tid_unrhdr = new_unrhdr(PID_MAX + 2, INT_MAX, &tid_lock);
 344
 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");
 350 }
 351
 352 /*
 353  * Place an unused thread on the zombie list.
 354  * Use the slpq as that must be unused by now.
 355  */
 356 void
 357 thread_zombie(struct thread *td)
 358 {
 359         mtx_lock_spin(&zombie_lock);
 360         TAILQ_INSERT_HEAD(&zombie_threads, td, td_slpq);
 361         mtx_unlock_spin(&zombie_lock);
 362 }
 363
 364 /*
 365  * Release a thread that has exited after cpu_throw().
 366  */
 367 void
 368 thread_stash(struct thread *td)
 369 {
 370         atomic_subtract_rel_int(&td->td_proc->p_exitthreads, 1);
 371         thread_zombie(td);
 372 }
 373
 374 /*
 375  * Reap zombie resources.
 376  */
 377 void
 378 thread_reap(void)
 379 {
 380         struct thread *td_first, *td_next;
 381
 382         /*
 383          * Don't even bother to lock if none at this instant,
 384          * we really don't care about the next instant.
 385          */
 386         if (!TAILQ_EMPTY(&zombie_threads)) {
 387                 mtx_lock_spin(&zombie_lock);
 388                 td_first = TAILQ_FIRST(&zombie_threads);
 389                 if (td_first)
 390                         TAILQ_INIT(&zombie_threads);
 391                 mtx_unlock_spin(&zombie_lock);
 392                 while (td_first) {
 393                         td_next = TAILQ_NEXT(td_first, td_slpq);
 394                         thread_cow_free(td_first);
 395                         thread_free(td_first);
 396                         td_first = td_next;
 397                 }
 398         }
 399 }
 400
 401 /*
 402  * Allocate a thread.
 403  */
 404 struct thread *
 405 thread_alloc(int pages)
 406 {
 407         struct thread *td;
 408
 409         thread_reap(); /* check if any zombies to get */
 410
 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);
 415                 return (NULL);
 416         }
 417         cpu_thread_alloc(td);
 418         return (td);
 419 }
 420
 421 int
 422 thread_alloc_stack(struct thread *td, int pages)
 423 {
 424
 425         KASSERT(td->td_kstack == 0,
 426             ("thread_alloc_stack called on a thread with kstack"));
 427         if (!vm_thread_new(td, pages))
 428                 return (0);
 429         cpu_thread_alloc(td);
 430         return (1);
 431 }
 432
 433 /*
 434  * Deallocate a thread.
 435  */
 436 void
 437 thread_free(struct thread *td)
 438 {
 439
 440         lock_profile_thread_exit(td);
 441         if (td->td_cpuset)
 442                 cpuset_rel(td->td_cpuset);
 443         td->td_cpuset = NULL;
 444         cpu_thread_free(td);
 445         if (td->td_kstack != 0)
 446                 vm_thread_dispose(td);
 447         callout_drain(&td->td_slpcallout);
 448         uma_zfree(thread_zone, td);
 449 }
 450
 451 void
 452 thread_cow_get_proc(struct thread *newtd, struct proc *p)
 453 {
 454
 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;
 459 }
 460
 461 void
 462 thread_cow_get(struct thread *newtd, struct thread *td)
 463 {
 464
 465         newtd->td_ucred = crhold(td->td_ucred);
 466         newtd->td_limit = lim_hold(td->td_limit);
 467         newtd->td_cowgen = td->td_cowgen;
 468 }
 469
 470 void
 471 thread_cow_free(struct thread *td)
 472 {
 473
 474         if (td->td_ucred != NULL)
 475                 crfree(td->td_ucred);
 476         if (td->td_limit != NULL)
 477                 lim_free(td->td_limit);
 478 }
 479
 480 void
 481 thread_cow_update(struct thread *td)
 482 {
 483         struct proc *p;
 484         struct ucred *oldcred;
 485         struct plimit *oldlimit;
 486
 487         p = td->td_proc;
 488         oldcred = NULL;
 489         oldlimit = NULL;
 490         PROC_LOCK(p);
 491         if (td->td_ucred != p->p_ucred) {
 492                 oldcred = td->td_ucred;
 493                 td->td_ucred = crhold(p->p_ucred);
 494         }
 495         if (td->td_limit != p->p_limit) {
 496                 oldlimit = td->td_limit;
 497                 td->td_limit = lim_hold(p->p_limit);
 498         }
 499         td->td_cowgen = p->p_cowgen;
 500         PROC_UNLOCK(p);
 501         if (oldcred != NULL)
 502                 crfree(oldcred);
 503         if (oldlimit != NULL)
 504                 lim_free(oldlimit);
 505 }
 506
 507 /*
 508  * Discard the current thread and exit from its context.
 509  * Always called with scheduler locked.
 510  *
 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
 514  * do a cpu_throw().
 515  */
 516 void
 517 thread_exit(void)
 518 {
 519         uint64_t runtime, new_switchtime;
 520         struct thread *td;
 521         struct thread *td2;
 522         struct proc *p;
 523         int wakeup_swapper;
 524
 525         td = curthread;
 526         p = td->td_proc;
 527
 528         PROC_SLOCK_ASSERT(p, MA_OWNED);
 529         mtx_assert(&Giant, MA_NOTOWNED);
 530
 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"));
 537
 538         /*
 539          * drop FPU & debug register state storage, or any other
 540          * architecture specific resources that
 541          * would not be on a new untouched process.
 542          */
 543         cpu_thread_exit(td);
 544
 545         /*
 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.
 552          */
 553         if (p->p_flag & P_HADTHREADS) {
 554                 if (p->p_numthreads > 1) {
 555                         atomic_add_int(&td->td_proc->p_exitthreads, 1);
 556                         thread_unlink(td);
 557                         td2 = FIRST_THREAD_IN_PROC(p);
 558                         sched_exit_thread(td2, td);
 559
 560                         /*
 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.
 564                          */
 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);
 571                                         if (wakeup_swapper)
 572                                                 kick_proc0();
 573                                 }
 574                         }
 575
 576                         PCPU_SET(deadthread, td);
 577                 } else {
 578                         /*
 579                          * The last thread is exiting.. but not through exit()
 580                          */
 581                         panic ("thread_exit: Last thread exiting on its own");
 582                 }
 583         }
 584 #ifdef  HWPMC_HOOKS
 585         /*
 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.
 588          */
 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);
 594 #endif
 595         PROC_UNLOCK(p);
 596         PROC_STATLOCK(p);
 597         thread_lock(td);
 598         PROC_SUNLOCK(p);
 599
 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);
 607         VM_CNT_INC(v_swtch);
 608
 609         /* Save our resource usage in our process. */
 610         td->td_ru.ru_nvcsw++;
 611         ruxagg(p, td);
 612         rucollect(&p->p_ru, &td->td_ru);
 613         PROC_STATUNLOCK(p);
 614
 615         td->td_state = TDS_INACTIVE;
 616 #ifdef WITNESS
 617         witness_thread_exit(td);
 618 #endif
 619         CTR1(KTR_PROC, "thread_exit: cpu_throw() thread %p", td);
 620         sched_throw(td);
 621         panic("I'm a teapot!");
 622         /* NOTREACHED */
 623 }
 624
 625 /*
 626  * Do any thread specific cleanups that may be needed in wait()
 627  * called with Giant, proc and schedlock not held.
 628  */
 629 void
 630 thread_wait(struct proc *p)
 631 {
 632         struct thread *td;
 633
 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(). */
 639         thread_lock(td);
 640         thread_unlock(td);
 641         lock_profile_thread_exit(td);
 642         cpuset_rel(td->td_cpuset);
 643         td->td_cpuset = NULL;
 644         cpu_thread_clean(td);
 645         thread_cow_free(td);
 646         callout_drain(&td->td_slpcallout);
 647         thread_reap();  /* check for zombie threads etc. */
 648 }
 649
 650 /*
 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.
 654  */
 655 void
 656 thread_link(struct thread *td, struct proc *p)
 657 {
 658
 659         /*
 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);
 663          */
 664         td->td_state    = TDS_INACTIVE;
 665         td->td_proc     = p;
 666         td->td_flags    = TDF_INMEM;
 667
 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);
 674         p->p_numthreads++;
 675 }
 676
 677 /*
 678  * Called from:
 679  *  thread_exit()
 680  */
 681 void
 682 thread_unlink(struct thread *td)
 683 {
 684         struct proc *p = td->td_proc;
 685
 686         PROC_LOCK_ASSERT(p, MA_OWNED);
 687         TAILQ_REMOVE(&p->p_threads, td, td_plist);
 688         p->p_numthreads--;
 689         /* could clear a few other things here */
 690         /* Must  NOT clear links to proc! */
 691 }
 692
 693 static int
 694 calc_remaining(struct proc *p, int mode)
 695 {
 696         int remaining;
 697
 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;
 706         else
 707                 panic("calc_remaining: wrong mode %d", mode);
 708         return (remaining);
 709 }
 710
 711 static int
 712 remain_for_mode(int mode)
 713 {
 714
 715         return (mode == SINGLE_ALLPROC ? 0 : 1);
 716 }
 717
 718 static int
 719 weed_inhib(int mode, struct thread *td2, struct proc *p)
 720 {
 721         int wakeup_swapper;
 722
 723         PROC_LOCK_ASSERT(p, MA_OWNED);
 724         PROC_SLOCK_ASSERT(p, MA_OWNED);
 725         THREAD_LOCK_ASSERT(td2, MA_OWNED);
 726
 727         wakeup_swapper = 0;
 728         switch (mode) {
 729         case SINGLE_EXIT:
 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);
 734                 break;
 735         case SINGLE_BOUNDARY:
 736         case SINGLE_NO_EXIT:
 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);
 741                 break;
 742         case SINGLE_ALLPROC:
 743                 /*
 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.
 750                  */
 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;
 758                         } else {
 759                                 wakeup_swapper |= sleepq_abort(td2, ERESTART);
 760                         }
 761                 }
 762                 break;
 763         }
 764         return (wakeup_swapper);
 765 }
 766
 767 /*
 768  * Enforce single-threading.
 769  *
 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).
 779  */
 780 int
 781 thread_single(struct proc *p, int mode)
 782 {
 783         struct thread *td;
 784         struct thread *td2;
 785         int remaining, wakeup_swapper;
 786
 787         td = curthread;
 788         KASSERT(mode == SINGLE_EXIT || mode == SINGLE_BOUNDARY ||
 789             mode == SINGLE_ALLPROC || mode == SINGLE_NO_EXIT,
 790             ("invalid mode %d", mode));
 791         /*
 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.
 796          */
 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);
 802
 803         if ((p->p_flag & P_HADTHREADS) == 0 && mode != SINGLE_ALLPROC)
 804                 return (0);
 805
 806         /* Is someone already single threading? */
 807         if (p->p_singlethread != NULL && p->p_singlethread != td)
 808                 return (1);
 809
 810         if (mode == SINGLE_EXIT) {
 811                 p->p_flag |= P_SINGLE_EXIT;
 812                 p->p_flag &= ~P_SINGLE_BOUNDARY;
 813         } else {
 814                 p->p_flag &= ~P_SINGLE_EXIT;
 815                 if (mode == SINGLE_BOUNDARY)
 816                         p->p_flag |= P_SINGLE_BOUNDARY;
 817                 else
 818                         p->p_flag &= ~P_SINGLE_BOUNDARY;
 819         }
 820         if (mode == SINGLE_ALLPROC)
 821                 p->p_flag |= P_TOTAL_STOP;
 822         p->p_flag |= P_STOPPED_SINGLE;
 823         PROC_SLOCK(p);
 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)
 828                         goto stopme;
 829                 wakeup_swapper = 0;
 830                 FOREACH_THREAD_IN_PROC(p, td2) {
 831                         if (td2 == td)
 832                                 continue;
 833                         thread_lock(td2);
 834                         td2->td_flags |= TDF_ASTPENDING | TDF_NEEDSUSPCHK;
 835                         if (TD_IS_INHIBITED(td2)) {
 836                                 wakeup_swapper |= weed_inhib(mode, td2, p);
 837 #ifdef SMP
 838                         } else if (TD_IS_RUNNING(td2) && td != td2) {
 839                                 forward_signal(td2);
 840 #endif
 841                         }
 842                         thread_unlock(td2);
 843                 }
 844                 if (wakeup_swapper)
 845                         kick_proc0();
 846                 remaining = calc_remaining(p, mode);
 847
 848                 /*
 849                  * Maybe we suspended some threads.. was it enough?
 850                  */
 851                 if (remaining == remain_for_mode(mode))
 852                         break;
 853
 854 stopme:
 855                 /*
 856                  * Wake us up when everyone else has suspended.
 857                  * In the mean time we suspend as well.
 858                  */
 859                 thread_suspend_switch(td, p);
 860                 remaining = calc_remaining(p, mode);
 861         }
 862         if (mode == SINGLE_EXIT) {
 863                 /*
 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.
 867                  */
 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);
 871
 872                 /*
 873                  * Wait for any remaining threads to exit cpu_throw().
 874                  */
 875                 while (p->p_exitthreads != 0) {
 876                         PROC_SUNLOCK(p);
 877                         PROC_UNLOCK(p);
 878                         sched_relinquish(td);
 879                         PROC_LOCK(p);
 880                         PROC_SLOCK(p);
 881                 }
 882         } else if (mode == SINGLE_BOUNDARY) {
 883                 /*
 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.
 890                  *
 891                  * We lock the thread, which is only allowed to
 892                  * succeed after context switch code finished using
 893                  * the address space.
 894                  */
 895                 FOREACH_THREAD_IN_PROC(p, td2) {
 896                         if (td2 == td)
 897                                 continue;
 898                         thread_lock(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));
 903                         thread_unlock(td2);
 904                 }
 905         }
 906         PROC_SUNLOCK(p);
 907         return (0);
 908 }
 909
 910 bool
 911 thread_suspend_check_needed(void)
 912 {
 913         struct proc *p;
 914         struct thread *td;
 915
 916         td = curthread;
 917         p = td->td_proc;
 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));
 921 }
 922
 923 /*
 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).
 927  *
 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.
 931  *
 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
 934  * out instead.
 935  *
 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.
 939  *
 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
 946  *               |                    |  immediately
 947  * 0 = thread_exit() or suspension ok,
 948  * other = return error instead of stopping the thread.
 949  *
 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.
 955  */
 956 int
 957 thread_suspend_check(int return_instead)
 958 {
 959         struct thread *td;
 960         struct proc *p;
 961         int wakeup_swapper;
 962
 963         td = curthread;
 964         p = td->td_proc;
 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"));
 971                         /*
 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.
 976                          */
 977                         if (p->p_singlethread == td)
 978                                 return (0);     /* Exempt from stopping. */
 979                 }
 980                 if ((p->p_flag & P_SINGLE_EXIT) && return_instead)
 981                         return (EINTR);
 982
 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)
 986                         return (ERESTART);
 987
 988                 /*
 989                  * Ignore suspend requests if they are deferred.
 990                  */
 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);
 998                 }
 999
1000                 /*
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.
1004                  */
1005                 if ((p->p_flag & P_SINGLE_EXIT) && (p->p_singlethread != td)) {
1006                         PROC_UNLOCK(p);
1007
1008                         /*
1009                          * Allow Linux emulation layer to do some work
1010                          * before thread suicide.
1011                          */
1012                         if (__predict_false(p->p_sysent->sv_thread_detach != NULL))
1013                                 (p->p_sysent->sv_thread_detach)(td);
1014                         umtx_thread_exit(td);
1015                         kern_thr_exit(td);
1016                         panic("stopped thread did not exit");
1017                 }
1018
1019                 PROC_SLOCK(p);
1020                 thread_stopped(p);
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);
1027                                 if (wakeup_swapper)
1028                                         kick_proc0();
1029                         }
1030                 }
1031                 PROC_UNLOCK(p);
1032                 thread_lock(td);
1033                 /*
1034                  * When a thread suspends, it just
1035                  * gets taken off all queues.
1036                  */
1037                 thread_suspend_one(td);
1038                 if (return_instead == 0) {
1039                         p->p_boundary_count++;
1040                         td->td_flags |= TDF_BOUNDARY;
1041                 }
1042                 PROC_SUNLOCK(p);
1043                 mi_switch(SW_INVOL | SWT_SUSPEND, NULL);
1044                 thread_unlock(td);
1045                 PROC_LOCK(p);
1046         }
1047         return (0);
1048 }
1049
1050 void
1051 thread_suspend_switch(struct thread *td, struct proc *p)
1052 {
1053
1054         KASSERT(!TD_IS_SUSPENDED(td), ("already suspended"));
1055         PROC_LOCK_ASSERT(p, MA_OWNED);
1056         PROC_SLOCK_ASSERT(p, MA_OWNED);
1057         /*
1058          * We implement thread_suspend_one in stages here to avoid
1059          * dropping the proc lock while the thread lock is owned.
1060          */
1061         if (p == td->td_proc) {
1062                 thread_stopped(p);
1063                 p->p_suspcount++;
1064         }
1065         PROC_UNLOCK(p);
1066         thread_lock(td);
1067         td->td_flags &= ~TDF_NEEDSUSPCHK;
1068         TD_SET_SUSPENDED(td);
1069         sched_sleep(td, 0);
1070         PROC_SUNLOCK(p);
1071         DROP_GIANT();
1072         mi_switch(SW_VOL | SWT_SUSPEND, NULL);
1073         thread_unlock(td);
1074         PICKUP_GIANT();
1075         PROC_LOCK(p);
1076         PROC_SLOCK(p);
1077 }
1078
1079 void
1080 thread_suspend_one(struct thread *td)
1081 {
1082         struct proc *p;
1083
1084         p = td->td_proc;
1085         PROC_SLOCK_ASSERT(p, MA_OWNED);
1086         THREAD_LOCK_ASSERT(td, MA_OWNED);
1087         KASSERT(!TD_IS_SUSPENDED(td), ("already suspended"));
1088         p->p_suspcount++;
1089         td->td_flags &= ~TDF_NEEDSUSPCHK;
1090         TD_SET_SUSPENDED(td);
1091         sched_sleep(td, 0);
1092 }
1093
1094 static int
1095 thread_unsuspend_one(struct thread *td, struct proc *p, bool boundary)
1096 {
1097
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);
1104                 p->p_suspcount--;
1105                 if (boundary && (td->td_flags & TDF_BOUNDARY) != 0) {
1106                         td->td_flags &= ~TDF_BOUNDARY;
1107                         p->p_boundary_count--;
1108                 }
1109         }
1110         return (setrunnable(td));
1111 }
1112
1113 /*
1114  * Allow all threads blocked by single threading to continue running.
1115  */
1116 void
1117 thread_unsuspend(struct proc *p)
1118 {
1119         struct thread *td;
1120         int wakeup_swapper;
1121
1122         PROC_LOCK_ASSERT(p, MA_OWNED);
1123         PROC_SLOCK_ASSERT(p, MA_OWNED);
1124         wakeup_swapper = 0;
1125         if (!P_SHOULDSTOP(p)) {
1126                 FOREACH_THREAD_IN_PROC(p, td) {
1127                         thread_lock(td);
1128                         if (TD_IS_SUSPENDED(td)) {
1129                                 wakeup_swapper |= thread_unsuspend_one(td, p,
1130                                     true);
1131                         }
1132                         thread_unlock(td);
1133                 }
1134         } else if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE &&
1135             p->p_numthreads == p->p_suspcount) {
1136                 /*
1137                  * Stopping everything also did the job for the single
1138                  * threading request. Now we've downgraded to single-threaded,
1139                  * let it continue.
1140                  */
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);
1146                 }
1147         }
1148         if (wakeup_swapper)
1149                 kick_proc0();
1150 }
1151
1152 /*
1153  * End the single threading mode..
1154  */
1155 void
1156 thread_single_end(struct proc *p, int mode)
1157 {
1158         struct thread *td;
1159         int wakeup_swapper;
1160
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 |
1175             P_TOTAL_STOP);
1176         PROC_SLOCK(p);
1177         p->p_singlethread = NULL;
1178         wakeup_swapper = 0;
1179         /*
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.
1184          */
1185         if (p->p_numthreads != remain_for_mode(mode) && !P_SHOULDSTOP(p)) {
1186                 FOREACH_THREAD_IN_PROC(p, td) {
1187                         thread_lock(td);
1188                         if (TD_IS_SUSPENDED(td)) {
1189                                 wakeup_swapper |= thread_unsuspend_one(td, p,
1190                                     mode == SINGLE_BOUNDARY);
1191                         }
1192                         thread_unlock(td);
1193                 }
1194         }
1195         KASSERT(mode != SINGLE_BOUNDARY || p->p_boundary_count == 0,
1196             ("inconsistent boundary count %d", p->p_boundary_count));
1197         PROC_SUNLOCK(p);
1198         if (wakeup_swapper)
1199                 kick_proc0();
1200 }
1201
1202 struct thread *
1203 thread_find(struct proc *p, lwpid_t tid)
1204 {
1205         struct thread *td;
1206
1207         PROC_LOCK_ASSERT(p, MA_OWNED);
1208         FOREACH_THREAD_IN_PROC(p, td) {
1209                 if (td->td_tid == tid)
1210                         break;
1211         }
1212         return (td);
1213 }
1214
1215 /* Locate a thread by number; return with proc lock held. */
1216 struct thread *
1217 tdfind(lwpid_t tid, pid_t pid)
1218 {
1219 #define RUN_THRESH      16
1220         struct thread *td;
1221         int run = 0;
1222
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) {
1227                                 td = NULL;
1228                                 break;
1229                         }
1230                         PROC_LOCK(td->td_proc);
1231                         if (td->td_proc->p_state == PRS_NEW) {
1232                                 PROC_UNLOCK(td->td_proc);
1233                                 td = NULL;
1234                                 break;
1235                         }
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),
1240                                                 td, td_hash);
1241                                         rw_wunlock(&tidhash_lock);
1242                                         return (td);
1243                                 }
1244                         }
1245                         break;
1246                 }
1247                 run++;
1248         }
1249         rw_runlock(&tidhash_lock);
1250         return (td);
1251 }
1252
1253 void
1254 tidhash_add(struct thread *td)
1255 {
1256         rw_wlock(&tidhash_lock);
1257         LIST_INSERT_HEAD(TIDHASH(td->td_tid), td, td_hash);
1258         rw_wunlock(&tidhash_lock);
1259 }
1260
1261 void
1262 tidhash_remove(struct thread *td)
1263 {
1264         rw_wlock(&tidhash_lock);
1265         LIST_REMOVE(td, td_hash);
1266         rw_wunlock(&tidhash_lock);
1267 }