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_kdtrace.h"
  31 #include "opt_hwpmc_hooks.h"
  32
  33 #include <sys/cdefs.h>
  34 __FBSDID("$FreeBSD$");
  35
  36 #include <sys/param.h>
  37 #include <sys/systm.h>
  38 #include <sys/kernel.h>
  39 #include <sys/lock.h>
  40 #include <sys/mutex.h>
  41 #include <sys/proc.h>
  42 #include <sys/rangelock.h>
  43 #include <sys/resourcevar.h>
  44 #include <sys/sdt.h>
  45 #include <sys/smp.h>
  46 #include <sys/sched.h>
  47 #include <sys/sleepqueue.h>
  48 #include <sys/selinfo.h>
  49 #include <sys/turnstile.h>
  50 #include <sys/ktr.h>
  51 #include <sys/rwlock.h>
  52 #include <sys/umtx.h>
  53 #include <sys/cpuset.h>
  54 #ifdef  HWPMC_HOOKS
  55 #include <sys/pmckern.h>
  56 #endif
  57
  58 #include <security/audit/audit.h>
  59
  60 #include <vm/vm.h>
  61 #include <vm/vm_extern.h>
  62 #include <vm/uma.h>
  63 #include <sys/eventhandler.h>
  64
  65 SDT_PROVIDER_DECLARE(proc);
  66 SDT_PROBE_DEFINE(proc, , , lwp_exit, lwp-exit);
  67
  68
  69 /*
  70  * thread related storage.
  71  */
  72 static uma_zone_t thread_zone;
  73
  74 TAILQ_HEAD(, thread) zombie_threads = TAILQ_HEAD_INITIALIZER(zombie_threads);
  75 static struct mtx zombie_lock;
  76 MTX_SYSINIT(zombie_lock, &zombie_lock, "zombie lock", MTX_SPIN);
  77
  78 static void thread_zombie(struct thread *);
  79
  80 #define TID_BUFFER_SIZE 1024
  81
  82 struct mtx tid_lock;
  83 static struct unrhdr *tid_unrhdr;
  84 static lwpid_t tid_buffer[TID_BUFFER_SIZE];
  85 static int tid_head, tid_tail;
  86 static MALLOC_DEFINE(M_TIDHASH, "tidhash", "thread hash");
  87
  88 struct  tidhashhead *tidhashtbl;
  89 u_long  tidhash;
  90 struct  rwlock tidhash_lock;
  91
  92 static lwpid_t
  93 tid_alloc(void)
  94 {
  95         lwpid_t tid;
  96
  97         tid = alloc_unr(tid_unrhdr);
  98         if (tid != -1)
  99                 return (tid);
 100         mtx_lock(&tid_lock);
 101         if (tid_head == tid_tail) {
 102                 mtx_unlock(&tid_lock);
 103                 return (-1);
 104         }
 105         tid = tid_buffer[tid_head];
 106         tid_head = (tid_head + 1) % TID_BUFFER_SIZE;
 107         mtx_unlock(&tid_lock);
 108         return (tid);
 109 }
 110
 111 static void
 112 tid_free(lwpid_t tid)
 113 {
 114         lwpid_t tmp_tid = -1;
 115
 116         mtx_lock(&tid_lock);
 117         if ((tid_tail + 1) % TID_BUFFER_SIZE == tid_head) {
 118                 tmp_tid = tid_buffer[tid_head];
 119                 tid_head = (tid_head + 1) % TID_BUFFER_SIZE;
 120         }
 121         tid_buffer[tid_tail] = tid;
 122         tid_tail = (tid_tail + 1) % TID_BUFFER_SIZE;
 123         mtx_unlock(&tid_lock);
 124         if (tmp_tid != -1)
 125                 free_unr(tid_unrhdr, tmp_tid);
 126 }
 127
 128 /*
 129  * Prepare a thread for use.
 130  */
 131 static int
 132 thread_ctor(void *mem, int size, void *arg, int flags)
 133 {
 134         struct thread   *td;
 135
 136         td = (struct thread *)mem;
 137         td->td_state = TDS_INACTIVE;
 138         td->td_oncpu = NOCPU;
 139
 140         td->td_tid = tid_alloc();
 141
 142         /*
 143          * Note that td_critnest begins life as 1 because the thread is not
 144          * running and is thereby implicitly waiting to be on the receiving
 145          * end of a context switch.
 146          */
 147         td->td_critnest = 1;
 148         td->td_lend_user_pri = PRI_MAX;
 149         EVENTHANDLER_INVOKE(thread_ctor, td);
 150 #ifdef AUDIT
 151         audit_thread_alloc(td);
 152 #endif
 153         umtx_thread_alloc(td);
 154         return (0);
 155 }
 156
 157 /*
 158  * Reclaim a thread after use.
 159  */
 160 static void
 161 thread_dtor(void *mem, int size, void *arg)
 162 {
 163         struct thread *td;
 164
 165         td = (struct thread *)mem;
 166
 167 #ifdef INVARIANTS
 168         /* Verify that this thread is in a safe state to free. */
 169         switch (td->td_state) {
 170         case TDS_INHIBITED:
 171         case TDS_RUNNING:
 172         case TDS_CAN_RUN:
 173         case TDS_RUNQ:
 174                 /*
 175                  * We must never unlink a thread that is in one of
 176                  * these states, because it is currently active.
 177                  */
 178                 panic("bad state for thread unlinking");
 179                 /* NOTREACHED */
 180         case TDS_INACTIVE:
 181                 break;
 182         default:
 183                 panic("bad thread state");
 184                 /* NOTREACHED */
 185         }
 186 #endif
 187 #ifdef AUDIT
 188         audit_thread_free(td);
 189 #endif
 190         /* Free all OSD associated to this thread. */
 191         osd_thread_exit(td);
 192
 193         EVENTHANDLER_INVOKE(thread_dtor, td);
 194         tid_free(td->td_tid);
 195 }
 196
 197 /*
 198  * Initialize type-stable parts of a thread (when newly created).
 199  */
 200 static int
 201 thread_init(void *mem, int size, int flags)
 202 {
 203         struct thread *td;
 204
 205         td = (struct thread *)mem;
 206
 207         td->td_sleepqueue = sleepq_alloc();
 208         td->td_turnstile = turnstile_alloc();
 209         td->td_rlqe = NULL;
 210         EVENTHANDLER_INVOKE(thread_init, td);
 211         td->td_sched = (struct td_sched *)&td[1];
 212         umtx_thread_init(td);
 213         td->td_kstack = 0;
 214         return (0);
 215 }
 216
 217 /*
 218  * Tear down type-stable parts of a thread (just before being discarded).
 219  */
 220 static void
 221 thread_fini(void *mem, int size)
 222 {
 223         struct thread *td;
 224
 225         td = (struct thread *)mem;
 226         EVENTHANDLER_INVOKE(thread_fini, td);
 227         rlqentry_free(td->td_rlqe);
 228         turnstile_free(td->td_turnstile);
 229         sleepq_free(td->td_sleepqueue);
 230         umtx_thread_fini(td);
 231         seltdfini(td);
 232 }
 233
 234 /*
 235  * For a newly created process,
 236  * link up all the structures and its initial threads etc.
 237  * called from:
 238  * {arch}/{arch}/machdep.c   ia64_init(), init386() etc.
 239  * proc_dtor() (should go away)
 240  * proc_init()
 241  */
 242 void
 243 proc_linkup0(struct proc *p, struct thread *td)
 244 {
 245         TAILQ_INIT(&p->p_threads);           /* all threads in proc */
 246         proc_linkup(p, td);
 247 }
 248
 249 void
 250 proc_linkup(struct proc *p, struct thread *td)
 251 {
 252
 253         sigqueue_init(&p->p_sigqueue, p);
 254         p->p_ksi = ksiginfo_alloc(1);
 255         if (p->p_ksi != NULL) {
 256                 /* XXX p_ksi may be null if ksiginfo zone is not ready */
 257                 p->p_ksi->ksi_flags = KSI_EXT | KSI_INS;
 258         }
 259         LIST_INIT(&p->p_mqnotifier);
 260         p->p_numthreads = 0;
 261         thread_link(td, p);
 262 }
 263
 264 /*
 265  * Initialize global thread allocation resources.
 266  */
 267 void
 268 threadinit(void)
 269 {
 270
 271         mtx_init(&tid_lock, "TID lock", NULL, MTX_DEF);
 272
 273         /*
 274          * pid_max cannot be greater than PID_MAX.
 275          * leave one number for thread0.
 276          */
 277         tid_unrhdr = new_unrhdr(PID_MAX + 2, INT_MAX, &tid_lock);
 278
 279         thread_zone = uma_zcreate("THREAD", sched_sizeof_thread(),
 280             thread_ctor, thread_dtor, thread_init, thread_fini,
 281             16 - 1, 0);
 282         tidhashtbl = hashinit(maxproc / 2, M_TIDHASH, &tidhash);
 283         rw_init(&tidhash_lock, "tidhash");
 284 }
 285
 286 /*
 287  * Place an unused thread on the zombie list.
 288  * Use the slpq as that must be unused by now.
 289  */
 290 void
 291 thread_zombie(struct thread *td)
 292 {
 293         mtx_lock_spin(&zombie_lock);
 294         TAILQ_INSERT_HEAD(&zombie_threads, td, td_slpq);
 295         mtx_unlock_spin(&zombie_lock);
 296 }
 297
 298 /*
 299  * Release a thread that has exited after cpu_throw().
 300  */
 301 void
 302 thread_stash(struct thread *td)
 303 {
 304         atomic_subtract_rel_int(&td->td_proc->p_exitthreads, 1);
 305         thread_zombie(td);
 306 }
 307
 308 /*
 309  * Reap zombie resources.
 310  */
 311 void
 312 thread_reap(void)
 313 {
 314         struct thread *td_first, *td_next;
 315
 316         /*
 317          * Don't even bother to lock if none at this instant,
 318          * we really don't care about the next instant..
 319          */
 320         if (!TAILQ_EMPTY(&zombie_threads)) {
 321                 mtx_lock_spin(&zombie_lock);
 322                 td_first = TAILQ_FIRST(&zombie_threads);
 323                 if (td_first)
 324                         TAILQ_INIT(&zombie_threads);
 325                 mtx_unlock_spin(&zombie_lock);
 326                 while (td_first) {
 327                         td_next = TAILQ_NEXT(td_first, td_slpq);
 328                         if (td_first->td_ucred)
 329                                 crfree(td_first->td_ucred);
 330                         thread_free(td_first);
 331                         td_first = td_next;
 332                 }
 333         }
 334 }
 335
 336 /*
 337  * Allocate a thread.
 338  */
 339 struct thread *
 340 thread_alloc(int pages)
 341 {
 342         struct thread *td;
 343
 344         thread_reap(); /* check if any zombies to get */
 345
 346         td = (struct thread *)uma_zalloc(thread_zone, M_WAITOK);
 347         KASSERT(td->td_kstack == 0, ("thread_alloc got thread with kstack"));
 348         if (!vm_thread_new(td, pages)) {
 349                 uma_zfree(thread_zone, td);
 350                 return (NULL);
 351         }
 352         cpu_thread_alloc(td);
 353         return (td);
 354 }
 355
 356 int
 357 thread_alloc_stack(struct thread *td, int pages)
 358 {
 359
 360         KASSERT(td->td_kstack == 0,
 361             ("thread_alloc_stack called on a thread with kstack"));
 362         if (!vm_thread_new(td, pages))
 363                 return (0);
 364         cpu_thread_alloc(td);
 365         return (1);
 366 }
 367
 368 /*
 369  * Deallocate a thread.
 370  */
 371 void
 372 thread_free(struct thread *td)
 373 {
 374
 375         lock_profile_thread_exit(td);
 376         if (td->td_cpuset)
 377                 cpuset_rel(td->td_cpuset);
 378         td->td_cpuset = NULL;
 379         cpu_thread_free(td);
 380         if (td->td_kstack != 0)
 381                 vm_thread_dispose(td);
 382         uma_zfree(thread_zone, td);
 383 }
 384
 385 /*
 386  * Discard the current thread and exit from its context.
 387  * Always called with scheduler locked.
 388  *
 389  * Because we can't free a thread while we're operating under its context,
 390  * push the current thread into our CPU's deadthread holder. This means
 391  * we needn't worry about someone else grabbing our context before we
 392  * do a cpu_throw().
 393  */
 394 void
 395 thread_exit(void)
 396 {
 397         uint64_t runtime, new_switchtime;
 398         struct thread *td;
 399         struct thread *td2;
 400         struct proc *p;
 401         int wakeup_swapper;
 402
 403         td = curthread;
 404         p = td->td_proc;
 405
 406         PROC_SLOCK_ASSERT(p, MA_OWNED);
 407         mtx_assert(&Giant, MA_NOTOWNED);
 408
 409         PROC_LOCK_ASSERT(p, MA_OWNED);
 410         KASSERT(p != NULL, ("thread exiting without a process"));
 411         CTR3(KTR_PROC, "thread_exit: thread %p (pid %ld, %s)", td,
 412             (long)p->p_pid, td->td_name);
 413         KASSERT(TAILQ_EMPTY(&td->td_sigqueue.sq_list), ("signal pending"));
 414
 415 #ifdef AUDIT
 416         AUDIT_SYSCALL_EXIT(0, td);
 417 #endif
 418         umtx_thread_exit(td);
 419         /*
 420          * drop FPU & debug register state storage, or any other
 421          * architecture specific resources that
 422          * would not be on a new untouched process.
 423          */
 424         cpu_thread_exit(td);    /* XXXSMP */
 425
 426         /*
 427          * The last thread is left attached to the process
 428          * So that the whole bundle gets recycled. Skip
 429          * all this stuff if we never had threads.
 430          * EXIT clears all sign of other threads when
 431          * it goes to single threading, so the last thread always
 432          * takes the short path.
 433          */
 434         if (p->p_flag & P_HADTHREADS) {
 435                 if (p->p_numthreads > 1) {
 436                         thread_unlink(td);
 437                         td2 = FIRST_THREAD_IN_PROC(p);
 438                         sched_exit_thread(td2, td);
 439
 440                         /*
 441                          * The test below is NOT true if we are the
 442                          * sole exiting thread. P_STOPPED_SINGLE is unset
 443                          * in exit1() after it is the only survivor.
 444                          */
 445                         if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE) {
 446                                 if (p->p_numthreads == p->p_suspcount) {
 447                                         thread_lock(p->p_singlethread);
 448                                         wakeup_swapper = thread_unsuspend_one(
 449                                                 p->p_singlethread);
 450                                         thread_unlock(p->p_singlethread);
 451                                         if (wakeup_swapper)
 452                                                 kick_proc0();
 453                                 }
 454                         }
 455
 456                         atomic_add_int(&td->td_proc->p_exitthreads, 1);
 457                         PCPU_SET(deadthread, td);
 458                 } else {
 459                         /*
 460                          * The last thread is exiting.. but not through exit()
 461                          */
 462                         panic ("thread_exit: Last thread exiting on its own");
 463                 }
 464         }
 465 #ifdef  HWPMC_HOOKS
 466         /*
 467          * If this thread is part of a process that is being tracked by hwpmc(4),
 468          * inform the module of the thread's impending exit.
 469          */
 470         if (PMC_PROC_IS_USING_PMCS(td->td_proc))
 471                 PMC_SWITCH_CONTEXT(td, PMC_FN_CSW_OUT);
 472 #endif
 473         PROC_UNLOCK(p);
 474
 475         /* Do the same timestamp bookkeeping that mi_switch() would do. */
 476         new_switchtime = cpu_ticks();
 477         runtime = new_switchtime - PCPU_GET(switchtime);
 478         td->td_runtime += runtime;
 479         td->td_incruntime += runtime;
 480         PCPU_SET(switchtime, new_switchtime);
 481         PCPU_SET(switchticks, ticks);
 482         PCPU_INC(cnt.v_swtch);
 483
 484         /* Save our resource usage in our process. */
 485         td->td_ru.ru_nvcsw++;
 486         ruxagg(p, td);
 487         rucollect(&p->p_ru, &td->td_ru);
 488
 489         thread_lock(td);
 490         PROC_SUNLOCK(p);
 491         td->td_state = TDS_INACTIVE;
 492 #ifdef WITNESS
 493         witness_thread_exit(td);
 494 #endif
 495         CTR1(KTR_PROC, "thread_exit: cpu_throw() thread %p", td);
 496         sched_throw(td);
 497         panic("I'm a teapot!");
 498         /* NOTREACHED */
 499 }
 500
 501 /*
 502  * Do any thread specific cleanups that may be needed in wait()
 503  * called with Giant, proc and schedlock not held.
 504  */
 505 void
 506 thread_wait(struct proc *p)
 507 {
 508         struct thread *td;
 509
 510         mtx_assert(&Giant, MA_NOTOWNED);
 511         KASSERT((p->p_numthreads == 1), ("Multiple threads in wait1()"));
 512         td = FIRST_THREAD_IN_PROC(p);
 513         /* Lock the last thread so we spin until it exits cpu_throw(). */
 514         thread_lock(td);
 515         thread_unlock(td);
 516         /* Wait for any remaining threads to exit cpu_throw(). */
 517         while (p->p_exitthreads)
 518                 sched_relinquish(curthread);
 519         lock_profile_thread_exit(td);
 520         cpuset_rel(td->td_cpuset);
 521         td->td_cpuset = NULL;
 522         cpu_thread_clean(td);
 523         crfree(td->td_ucred);
 524         thread_reap();  /* check for zombie threads etc. */
 525 }
 526
 527 /*
 528  * Link a thread to a process.
 529  * set up anything that needs to be initialized for it to
 530  * be used by the process.
 531  */
 532 void
 533 thread_link(struct thread *td, struct proc *p)
 534 {
 535
 536         /*
 537          * XXX This can't be enabled because it's called for proc0 before
 538          * its lock has been created.
 539          * PROC_LOCK_ASSERT(p, MA_OWNED);
 540          */
 541         td->td_state    = TDS_INACTIVE;
 542         td->td_proc     = p;
 543         td->td_flags    = TDF_INMEM;
 544
 545         LIST_INIT(&td->td_contested);
 546         LIST_INIT(&td->td_lprof[0]);
 547         LIST_INIT(&td->td_lprof[1]);
 548         sigqueue_init(&td->td_sigqueue, p);
 549         callout_init(&td->td_slpcallout, CALLOUT_MPSAFE);
 550         TAILQ_INSERT_HEAD(&p->p_threads, td, td_plist);
 551         p->p_numthreads++;
 552 }
 553
 554 /*
 555  * Convert a process with one thread to an unthreaded process.
 556  */
 557 void
 558 thread_unthread(struct thread *td)
 559 {
 560         struct proc *p = td->td_proc;
 561
 562         KASSERT((p->p_numthreads == 1), ("Unthreading with >1 threads"));
 563         p->p_flag &= ~P_HADTHREADS;
 564 }
 565
 566 /*
 567  * Called from:
 568  *  thread_exit()
 569  */
 570 void
 571 thread_unlink(struct thread *td)
 572 {
 573         struct proc *p = td->td_proc;
 574
 575         PROC_LOCK_ASSERT(p, MA_OWNED);
 576         TAILQ_REMOVE(&p->p_threads, td, td_plist);
 577         p->p_numthreads--;
 578         /* could clear a few other things here */
 579         /* Must  NOT clear links to proc! */
 580 }
 581
 582 static int
 583 calc_remaining(struct proc *p, int mode)
 584 {
 585         int remaining;
 586
 587         PROC_LOCK_ASSERT(p, MA_OWNED);
 588         PROC_SLOCK_ASSERT(p, MA_OWNED);
 589         if (mode == SINGLE_EXIT)
 590                 remaining = p->p_numthreads;
 591         else if (mode == SINGLE_BOUNDARY)
 592                 remaining = p->p_numthreads - p->p_boundary_count;
 593         else if (mode == SINGLE_NO_EXIT)
 594                 remaining = p->p_numthreads - p->p_suspcount;
 595         else
 596                 panic("calc_remaining: wrong mode %d", mode);
 597         return (remaining);
 598 }
 599
 600 /*
 601  * Enforce single-threading.
 602  *
 603  * Returns 1 if the caller must abort (another thread is waiting to
 604  * exit the process or similar). Process is locked!
 605  * Returns 0 when you are successfully the only thread running.
 606  * A process has successfully single threaded in the suspend mode when
 607  * There are no threads in user mode. Threads in the kernel must be
 608  * allowed to continue until they get to the user boundary. They may even
 609  * copy out their return values and data before suspending. They may however be
 610  * accelerated in reaching the user boundary as we will wake up
 611  * any sleeping threads that are interruptable. (PCATCH).
 612  */
 613 int
 614 thread_single(int mode)
 615 {
 616         struct thread *td;
 617         struct thread *td2;
 618         struct proc *p;
 619         int remaining, wakeup_swapper;
 620
 621         td = curthread;
 622         p = td->td_proc;
 623         mtx_assert(&Giant, MA_NOTOWNED);
 624         PROC_LOCK_ASSERT(p, MA_OWNED);
 625
 626         if ((p->p_flag & P_HADTHREADS) == 0)
 627                 return (0);
 628
 629         /* Is someone already single threading? */
 630         if (p->p_singlethread != NULL && p->p_singlethread != td)
 631                 return (1);
 632
 633         if (mode == SINGLE_EXIT) {
 634                 p->p_flag |= P_SINGLE_EXIT;
 635                 p->p_flag &= ~P_SINGLE_BOUNDARY;
 636         } else {
 637                 p->p_flag &= ~P_SINGLE_EXIT;
 638                 if (mode == SINGLE_BOUNDARY)
 639                         p->p_flag |= P_SINGLE_BOUNDARY;
 640                 else
 641                         p->p_flag &= ~P_SINGLE_BOUNDARY;
 642         }
 643         p->p_flag |= P_STOPPED_SINGLE;
 644         PROC_SLOCK(p);
 645         p->p_singlethread = td;
 646         remaining = calc_remaining(p, mode);
 647         while (remaining != 1) {
 648                 if (P_SHOULDSTOP(p) != P_STOPPED_SINGLE)
 649                         goto stopme;
 650                 wakeup_swapper = 0;
 651                 FOREACH_THREAD_IN_PROC(p, td2) {
 652                         if (td2 == td)
 653                                 continue;
 654                         thread_lock(td2);
 655                         td2->td_flags |= TDF_ASTPENDING | TDF_NEEDSUSPCHK;
 656                         if (TD_IS_INHIBITED(td2)) {
 657                                 switch (mode) {
 658                                 case SINGLE_EXIT:
 659                                         if (TD_IS_SUSPENDED(td2))
 660                                                 wakeup_swapper |=
 661                                                     thread_unsuspend_one(td2);
 662                                         if (TD_ON_SLEEPQ(td2) &&
 663                                             (td2->td_flags & TDF_SINTR))
 664                                                 wakeup_swapper |=
 665                                                     sleepq_abort(td2, EINTR);
 666                                         break;
 667                                 case SINGLE_BOUNDARY:
 668                                         if (TD_IS_SUSPENDED(td2) &&
 669                                             !(td2->td_flags & TDF_BOUNDARY))
 670                                                 wakeup_swapper |=
 671                                                     thread_unsuspend_one(td2);
 672                                         if (TD_ON_SLEEPQ(td2) &&
 673                                             (td2->td_flags & TDF_SINTR))
 674                                                 wakeup_swapper |=
 675                                                     sleepq_abort(td2, ERESTART);
 676                                         break;
 677                                 case SINGLE_NO_EXIT:
 678                                         if (TD_IS_SUSPENDED(td2) &&
 679                                             !(td2->td_flags & TDF_BOUNDARY))
 680                                                 wakeup_swapper |=
 681                                                     thread_unsuspend_one(td2);
 682                                         if (TD_ON_SLEEPQ(td2) &&
 683                                             (td2->td_flags & TDF_SINTR))
 684                                                 wakeup_swapper |=
 685                                                     sleepq_abort(td2, ERESTART);
 686                                         break;
 687                                 default:
 688                                         break;
 689                                 }
 690                         }
 691 #ifdef SMP
 692                         else if (TD_IS_RUNNING(td2) && td != td2) {
 693                                 forward_signal(td2);
 694                         }
 695 #endif
 696                         thread_unlock(td2);
 697                 }
 698                 if (wakeup_swapper)
 699                         kick_proc0();
 700                 remaining = calc_remaining(p, mode);
 701
 702                 /*
 703                  * Maybe we suspended some threads.. was it enough?
 704                  */
 705                 if (remaining == 1)
 706                         break;
 707
 708 stopme:
 709                 /*
 710                  * Wake us up when everyone else has suspended.
 711                  * In the mean time we suspend as well.
 712                  */
 713                 thread_suspend_switch(td);
 714                 remaining = calc_remaining(p, mode);
 715         }
 716         if (mode == SINGLE_EXIT) {
 717                 /*
 718                  * We have gotten rid of all the other threads and we
 719                  * are about to either exit or exec. In either case,
 720                  * we try our utmost to revert to being a non-threaded
 721                  * process.
 722                  */
 723                 p->p_singlethread = NULL;
 724                 p->p_flag &= ~(P_STOPPED_SINGLE | P_SINGLE_EXIT);
 725                 thread_unthread(td);
 726         }
 727         PROC_SUNLOCK(p);
 728         return (0);
 729 }
 730
 731 /*
 732  * Called in from locations that can safely check to see
 733  * whether we have to suspend or at least throttle for a
 734  * single-thread event (e.g. fork).
 735  *
 736  * Such locations include userret().
 737  * If the "return_instead" argument is non zero, the thread must be able to
 738  * accept 0 (caller may continue), or 1 (caller must abort) as a result.
 739  *
 740  * The 'return_instead' argument tells the function if it may do a
 741  * thread_exit() or suspend, or whether the caller must abort and back
 742  * out instead.
 743  *
 744  * If the thread that set the single_threading request has set the
 745  * P_SINGLE_EXIT bit in the process flags then this call will never return
 746  * if 'return_instead' is false, but will exit.
 747  *
 748  * P_SINGLE_EXIT | return_instead == 0| return_instead != 0
 749  *---------------+--------------------+---------------------
 750  *       0       | returns 0          |   returns 0 or 1
 751  *               | when ST ends       |   immediately
 752  *---------------+--------------------+---------------------
 753  *       1       | thread exits       |   returns 1
 754  *               |                    |  immediately
 755  * 0 = thread_exit() or suspension ok,
 756  * other = return error instead of stopping the thread.
 757  *
 758  * While a full suspension is under effect, even a single threading
 759  * thread would be suspended if it made this call (but it shouldn't).
 760  * This call should only be made from places where
 761  * thread_exit() would be safe as that may be the outcome unless
 762  * return_instead is set.
 763  */
 764 int
 765 thread_suspend_check(int return_instead)
 766 {
 767         struct thread *td;
 768         struct proc *p;
 769         int wakeup_swapper;
 770
 771         td = curthread;
 772         p = td->td_proc;
 773         mtx_assert(&Giant, MA_NOTOWNED);
 774         PROC_LOCK_ASSERT(p, MA_OWNED);
 775         while (P_SHOULDSTOP(p) ||
 776               ((p->p_flag & P_TRACED) && (td->td_dbgflags & TDB_SUSPEND))) {
 777                 if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE) {
 778                         KASSERT(p->p_singlethread != NULL,
 779                             ("singlethread not set"));
 780                         /*
 781                          * The only suspension in action is a
 782                          * single-threading. Single threader need not stop.
 783                          * XXX Should be safe to access unlocked
 784                          * as it can only be set to be true by us.
 785                          */
 786                         if (p->p_singlethread == td)
 787                                 return (0);     /* Exempt from stopping. */
 788                 }
 789                 if ((p->p_flag & P_SINGLE_EXIT) && return_instead)
 790                         return (EINTR);
 791
 792                 /* Should we goto user boundary if we didn't come from there? */
 793                 if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE &&
 794                     (p->p_flag & P_SINGLE_BOUNDARY) && return_instead)
 795                         return (ERESTART);
 796
 797                 /*
 798                  * Ignore suspend requests for stop signals if they
 799                  * are deferred.
 800                  */
 801                 if (P_SHOULDSTOP(p) == P_STOPPED_SIG &&
 802                     td->td_flags & TDF_SBDRY) {
 803                         KASSERT(return_instead,
 804                             ("TDF_SBDRY set for unsafe thread_suspend_check"));
 805                         return (0);
 806                 }
 807
 808                 /*
 809                  * If the process is waiting for us to exit,
 810                  * this thread should just suicide.
 811                  * Assumes that P_SINGLE_EXIT implies P_STOPPED_SINGLE.
 812                  */
 813                 if ((p->p_flag & P_SINGLE_EXIT) && (p->p_singlethread != td)) {
 814                         PROC_UNLOCK(p);
 815                         tidhash_remove(td);
 816                         PROC_LOCK(p);
 817                         tdsigcleanup(td);
 818                         PROC_SLOCK(p);
 819                         thread_stopped(p);
 820                         thread_exit();
 821                 }
 822
 823                 PROC_SLOCK(p);
 824                 thread_stopped(p);
 825                 if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE) {
 826                         if (p->p_numthreads == p->p_suspcount + 1) {
 827                                 thread_lock(p->p_singlethread);
 828                                 wakeup_swapper =
 829                                     thread_unsuspend_one(p->p_singlethread);
 830                                 thread_unlock(p->p_singlethread);
 831                                 if (wakeup_swapper)
 832                                         kick_proc0();
 833                         }
 834                 }
 835                 PROC_UNLOCK(p);
 836                 thread_lock(td);
 837                 /*
 838                  * When a thread suspends, it just
 839                  * gets taken off all queues.
 840                  */
 841                 thread_suspend_one(td);
 842                 if (return_instead == 0) {
 843                         p->p_boundary_count++;
 844                         td->td_flags |= TDF_BOUNDARY;
 845                 }
 846                 PROC_SUNLOCK(p);
 847                 mi_switch(SW_INVOL | SWT_SUSPEND, NULL);
 848                 if (return_instead == 0)
 849                         td->td_flags &= ~TDF_BOUNDARY;
 850                 thread_unlock(td);
 851                 PROC_LOCK(p);
 852                 if (return_instead == 0) {
 853                         PROC_SLOCK(p);
 854                         p->p_boundary_count--;
 855                         PROC_SUNLOCK(p);
 856                 }
 857         }
 858         return (0);
 859 }
 860
 861 void
 862 thread_suspend_switch(struct thread *td)
 863 {
 864         struct proc *p;
 865
 866         p = td->td_proc;
 867         KASSERT(!TD_IS_SUSPENDED(td), ("already suspended"));
 868         PROC_LOCK_ASSERT(p, MA_OWNED);
 869         PROC_SLOCK_ASSERT(p, MA_OWNED);
 870         /*
 871          * We implement thread_suspend_one in stages here to avoid
 872          * dropping the proc lock while the thread lock is owned.
 873          */
 874         thread_stopped(p);
 875         p->p_suspcount++;
 876         PROC_UNLOCK(p);
 877         thread_lock(td);
 878         td->td_flags &= ~TDF_NEEDSUSPCHK;
 879         TD_SET_SUSPENDED(td);
 880         sched_sleep(td, 0);
 881         PROC_SUNLOCK(p);
 882         DROP_GIANT();
 883         mi_switch(SW_VOL | SWT_SUSPEND, NULL);
 884         thread_unlock(td);
 885         PICKUP_GIANT();
 886         PROC_LOCK(p);
 887         PROC_SLOCK(p);
 888 }
 889
 890 void
 891 thread_suspend_one(struct thread *td)
 892 {
 893         struct proc *p = td->td_proc;
 894
 895         PROC_SLOCK_ASSERT(p, MA_OWNED);
 896         THREAD_LOCK_ASSERT(td, MA_OWNED);
 897         KASSERT(!TD_IS_SUSPENDED(td), ("already suspended"));
 898         p->p_suspcount++;
 899         td->td_flags &= ~TDF_NEEDSUSPCHK;
 900         TD_SET_SUSPENDED(td);
 901         sched_sleep(td, 0);
 902 }
 903
 904 int
 905 thread_unsuspend_one(struct thread *td)
 906 {
 907         struct proc *p = td->td_proc;
 908
 909         PROC_SLOCK_ASSERT(p, MA_OWNED);
 910         THREAD_LOCK_ASSERT(td, MA_OWNED);
 911         KASSERT(TD_IS_SUSPENDED(td), ("Thread not suspended"));
 912         TD_CLR_SUSPENDED(td);
 913         p->p_suspcount--;
 914         return (setrunnable(td));
 915 }
 916
 917 /*
 918  * Allow all threads blocked by single threading to continue running.
 919  */
 920 void
 921 thread_unsuspend(struct proc *p)
 922 {
 923         struct thread *td;
 924         int wakeup_swapper;
 925
 926         PROC_LOCK_ASSERT(p, MA_OWNED);
 927         PROC_SLOCK_ASSERT(p, MA_OWNED);
 928         wakeup_swapper = 0;
 929         if (!P_SHOULDSTOP(p)) {
 930                 FOREACH_THREAD_IN_PROC(p, td) {
 931                         thread_lock(td);
 932                         if (TD_IS_SUSPENDED(td)) {
 933                                 wakeup_swapper |= thread_unsuspend_one(td);
 934                         }
 935                         thread_unlock(td);
 936                 }
 937         } else if ((P_SHOULDSTOP(p) == P_STOPPED_SINGLE) &&
 938             (p->p_numthreads == p->p_suspcount)) {
 939                 /*
 940                  * Stopping everything also did the job for the single
 941                  * threading request. Now we've downgraded to single-threaded,
 942                  * let it continue.
 943                  */
 944                 thread_lock(p->p_singlethread);
 945                 wakeup_swapper = thread_unsuspend_one(p->p_singlethread);
 946                 thread_unlock(p->p_singlethread);
 947         }
 948         if (wakeup_swapper)
 949                 kick_proc0();
 950 }
 951
 952 /*
 953  * End the single threading mode..
 954  */
 955 void
 956 thread_single_end(void)
 957 {
 958         struct thread *td;
 959         struct proc *p;
 960         int wakeup_swapper;
 961
 962         td = curthread;
 963         p = td->td_proc;
 964         PROC_LOCK_ASSERT(p, MA_OWNED);
 965         p->p_flag &= ~(P_STOPPED_SINGLE | P_SINGLE_EXIT | P_SINGLE_BOUNDARY);
 966         PROC_SLOCK(p);
 967         p->p_singlethread = NULL;
 968         wakeup_swapper = 0;
 969         /*
 970          * If there are other threads they may now run,
 971          * unless of course there is a blanket 'stop order'
 972          * on the process. The single threader must be allowed
 973          * to continue however as this is a bad place to stop.
 974          */
 975         if ((p->p_numthreads != 1) && (!P_SHOULDSTOP(p))) {
 976                 FOREACH_THREAD_IN_PROC(p, td) {
 977                         thread_lock(td);
 978                         if (TD_IS_SUSPENDED(td)) {
 979                                 wakeup_swapper |= thread_unsuspend_one(td);
 980                         }
 981                         thread_unlock(td);
 982                 }
 983         }
 984         PROC_SUNLOCK(p);
 985         if (wakeup_swapper)
 986                 kick_proc0();
 987 }
 988
 989 struct thread *
 990 thread_find(struct proc *p, lwpid_t tid)
 991 {
 992         struct thread *td;
 993
 994         PROC_LOCK_ASSERT(p, MA_OWNED);
 995         FOREACH_THREAD_IN_PROC(p, td) {
 996                 if (td->td_tid == tid)
 997                         break;
 998         }
 999         return (td);
1000 }
1001
1002 /* Locate a thread by number; return with proc lock held. */
1003 struct thread *
1004 tdfind(lwpid_t tid, pid_t pid)
1005 {
1006 #define RUN_THRESH      16
1007         struct thread *td;
1008         int run = 0;
1009
1010         rw_rlock(&tidhash_lock);
1011         LIST_FOREACH(td, TIDHASH(tid), td_hash) {
1012                 if (td->td_tid == tid) {
1013                         if (pid != -1 && td->td_proc->p_pid != pid) {
1014                                 td = NULL;
1015                                 break;
1016                         }
1017                         PROC_LOCK(td->td_proc);
1018                         if (td->td_proc->p_state == PRS_NEW) {
1019                                 PROC_UNLOCK(td->td_proc);
1020                                 td = NULL;
1021                                 break;
1022                         }
1023                         if (run > RUN_THRESH) {
1024                                 if (rw_try_upgrade(&tidhash_lock)) {
1025                                         LIST_REMOVE(td, td_hash);
1026                                         LIST_INSERT_HEAD(TIDHASH(td->td_tid),
1027                                                 td, td_hash);
1028                                         rw_wunlock(&tidhash_lock);
1029                                         return (td);
1030                                 }
1031                         }
1032                         break;
1033                 }
1034                 run++;
1035         }
1036         rw_runlock(&tidhash_lock);
1037         return (td);
1038 }
1039
1040 void
1041 tidhash_add(struct thread *td)
1042 {
1043         rw_wlock(&tidhash_lock);
1044         LIST_INSERT_HEAD(TIDHASH(td->td_tid), td, td_hash);
1045         rw_wunlock(&tidhash_lock);
1046 }
1047
1048 void
1049 tidhash_remove(struct thread *td)
1050 {
1051         rw_wlock(&tidhash_lock);
1052         LIST_REMOVE(td, td_hash);
1053         rw_wunlock(&tidhash_lock);
1054 }