aboutsummaryrefslogtreecommitdiff
path: root/linuxthreads/manager.c
blob: 76ef6cf9fb6580b8a5fbfabd5a230138d5aca10f (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
/* Linuxthreads - a simple clone()-based implementation of Posix        */
/* threads for Linux.                                                   */
/* Copyright (C) 1996 Xavier Leroy (Xavier.Leroy@inria.fr)              */
/*                                                                      */
/* This program is free software; you can redistribute it and/or        */
/* modify it under the terms of the GNU Library General Public License  */
/* as published by the Free Software Foundation; either version 2       */
/* of the License, or (at your option) any later version.               */
/*                                                                      */
/* This program is distributed in the hope that it will be useful,      */
/* but WITHOUT ANY WARRANTY; without even the implied warranty of       */
/* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the        */
/* GNU Library General Public License for more details.                 */

/* The "thread manager" thread: manages creation and termination of threads */

#include <errno.h>
#include <sched.h>
#include <stddef.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <unistd.h>
#include <sys/poll.h>		/* for poll */
#include <sys/mman.h>           /* for mmap */
#include <sys/param.h>
#include <sys/time.h>
#include <sys/wait.h>           /* for waitpid macros */

#include "pthread.h"
#include "internals.h"
#include "spinlock.h"
#include "restart.h"
#include "semaphore.h"

/* Array of active threads. Entry 0 is reserved for the initial thread. */
struct pthread_handle_struct __pthread_handles[PTHREAD_THREADS_MAX] =
{ { LOCK_INITIALIZER, &__pthread_initial_thread, 0},
  { LOCK_INITIALIZER, &__pthread_manager_thread, 0}, /* All NULLs */ };

/* For debugging purposes put the maximum number of threads in a variable.  */
const int __linuxthreads_pthread_threads_max = PTHREAD_THREADS_MAX;

/* Indicate whether at least one thread has a user-defined stack (if 1),
   or if all threads have stacks supplied by LinuxThreads (if 0). */
int __pthread_nonstandard_stacks;

/* Number of active entries in __pthread_handles (used by gdb) */
volatile int __pthread_handles_num = 2;

/* Whether to use debugger additional actions for thread creation
   (set to 1 by gdb) */
volatile int __pthread_threads_debug;

/* Globally enabled events.  */
volatile td_thr_events_t __pthread_threads_events;

/* Pointer to thread descriptor with last event.  */
volatile pthread_descr __pthread_last_event;

/* Mapping from stack segment to thread descriptor. */
/* Stack segment numbers are also indices into the __pthread_handles array. */
/* Stack segment number 0 is reserved for the initial thread. */

static inline pthread_descr thread_segment(int seg)
{
  return (pthread_descr)(THREAD_STACK_START_ADDRESS - (seg - 1) * STACK_SIZE)
         - 1;
}

/* Flag set in signal handler to record child termination */

static volatile int terminated_children = 0;

/* Flag set when the initial thread is blocked on pthread_exit waiting
   for all other threads to terminate */

static int main_thread_exiting = 0;

/* Counter used to generate unique thread identifier.
   Thread identifier is pthread_threads_counter + segment. */

static pthread_t pthread_threads_counter = 0;

#ifdef NEED_SEPARATE_REGISTER_STACK
/* Signal masks for the manager.  These have to be global only when clone2
   is used since it's currently borken wrt signals in the child.  */
static sigset_t manager_mask;		/* Manager normal signal mask	*/
static sigset_t manager_mask_all;	/* All bits set.	*/
#endif

/* Forward declarations */

static int pthread_handle_create(pthread_t *thread, const pthread_attr_t *attr,
                                 void * (*start_routine)(void *), void *arg,
                                 sigset_t *mask, int father_pid,
				 int report_events,
				 td_thr_events_t *event_maskp);
static void pthread_handle_free(pthread_t th_id);
static void pthread_handle_exit(pthread_descr issuing_thread, int exitcode);
static void pthread_reap_children(void);
static void pthread_kill_all_threads(int sig, int main_thread_also);

/* The server thread managing requests for thread creation and termination */

int __pthread_manager(void *arg)
{
  int reqfd = (int) (long int) arg;
  struct pollfd ufd;
#ifndef NEED_SEPARATE_REGISTER_STACK
  sigset_t manager_mask;
#endif
  int n;
  struct pthread_request request;

  /* If we have special thread_self processing, initialize it.  */
#ifdef INIT_THREAD_SELF
  INIT_THREAD_SELF(&__pthread_manager_thread, 1);
#endif
  /* Set the error variable.  */
  __pthread_manager_thread.p_errnop = &__pthread_manager_thread.p_errno;
  __pthread_manager_thread.p_h_errnop = &__pthread_manager_thread.p_h_errno;
  /* Block all signals except __pthread_sig_cancel and SIGTRAP */
  sigfillset(&manager_mask);
  sigdelset(&manager_mask, __pthread_sig_cancel); /* for thread termination */
  sigdelset(&manager_mask, SIGTRAP);            /* for debugging purposes */
  if (__pthread_threads_debug && __pthread_sig_debug > 0)
    sigdelset(&manager_mask, __pthread_sig_debug);
  sigprocmask(SIG_SETMASK, &manager_mask, NULL);
#ifdef NEED_SEPARATE_REGISTER_STACK
  sigfillset(&manager_mask_all);
#endif
  /* Raise our priority to match that of main thread */
  __pthread_manager_adjust_prio(__pthread_main_thread->p_priority);
  /* Synchronize debugging of the thread manager */
  n = __libc_read(reqfd, (char *)&request, sizeof(request));
  ASSERT(n == sizeof(request) && request.req_kind == REQ_DEBUG);
  ufd.fd = reqfd;
  ufd.events = POLLIN;
  /* Enter server loop */
  while(1) {
    n = __poll(&ufd, 1, 2000);

    /* Check for termination of the main thread */
    if (getppid() == 1) {
      pthread_kill_all_threads(SIGKILL, 0);
      _exit(0);
    }
    /* Check for dead children */
    if (terminated_children) {
      terminated_children = 0;
      pthread_reap_children();
    }
    /* Read and execute request */
    if (n == 1 && (ufd.revents & POLLIN)) {
      n = __libc_read(reqfd, (char *)&request, sizeof(request));
      ASSERT(n == sizeof(request));
      switch(request.req_kind) {
      case REQ_CREATE:
        request.req_thread->p_retcode =
          pthread_handle_create((pthread_t *) &request.req_thread->p_retval,
                                request.req_args.create.attr,
                                request.req_args.create.fn,
                                request.req_args.create.arg,
                                &request.req_args.create.mask,
                                request.req_thread->p_pid,
				request.req_thread->p_report_events,
				&request.req_thread->p_eventbuf.eventmask);
        restart(request.req_thread);
        break;
      case REQ_FREE:
	pthread_handle_free(request.req_args.free.thread_id);
        break;
      case REQ_PROCESS_EXIT:
        pthread_handle_exit(request.req_thread,
                            request.req_args.exit.code);
	/* NOTREACHED */
        break;
      case REQ_MAIN_THREAD_EXIT:
        main_thread_exiting = 1;
	/* Reap children in case all other threads died and the signal handler
	   went off before we set main_thread_exiting to 1, and therefore did
	   not do REQ_KICK. */
	pthread_reap_children();

        if (__pthread_main_thread->p_nextlive == __pthread_main_thread) {
          restart(__pthread_main_thread);
	  /* The main thread will now call exit() which will trigger an
	     __on_exit handler, which in turn will send REQ_PROCESS_EXIT
	     to the thread manager. In case you are wondering how the
	     manager terminates from its loop here. */
	}
        break;
      case REQ_POST:
        __new_sem_post(request.req_args.post);
        break;
      case REQ_DEBUG:
	/* Make gdb aware of new thread and gdb will restart the
	   new thread when it is ready to handle the new thread. */
	if (__pthread_threads_debug && __pthread_sig_debug > 0)
	  raise(__pthread_sig_debug);
        break;
      case REQ_KICK:
	/* This is just a prod to get the manager to reap some
	   threads right away, avoiding a potential delay at shutdown. */
	break;
      }
    }
  }
}

int __pthread_manager_event(void *arg)
{
  /* If we have special thread_self processing, initialize it.  */
#ifdef INIT_THREAD_SELF
  INIT_THREAD_SELF(&__pthread_manager_thread, 1);
#endif

  /* Get the lock the manager will free once all is correctly set up.  */
  __pthread_lock (THREAD_GETMEM((&__pthread_manager_thread), p_lock), NULL);
  /* Free it immediately.  */
  __pthread_unlock (THREAD_GETMEM((&__pthread_manager_thread), p_lock));

  return __pthread_manager(arg);
}

/* Process creation */

static int pthread_start_thread(void *arg)
{
  pthread_descr self = (pthread_descr) arg;
  struct pthread_request request;
  void * outcome;
  /* Initialize special thread_self processing, if any.  */
#ifdef INIT_THREAD_SELF
  INIT_THREAD_SELF(self, self->p_nr);
#endif
  /* Make sure our pid field is initialized, just in case we get there
     before our father has initialized it. */
  THREAD_SETMEM(self, p_pid, __getpid());
  /* Initial signal mask is that of the creating thread. (Otherwise,
     we'd just inherit the mask of the thread manager.) */
  sigprocmask(SIG_SETMASK, &self->p_start_args.mask, NULL);
  /* Set the scheduling policy and priority for the new thread, if needed */
  if (THREAD_GETMEM(self, p_start_args.schedpolicy) >= 0)
    /* Explicit scheduling attributes were provided: apply them */
    __sched_setscheduler(THREAD_GETMEM(self, p_pid),
			 THREAD_GETMEM(self, p_start_args.schedpolicy),
                         &self->p_start_args.schedparam);
  else if (__pthread_manager_thread.p_priority > 0)
    /* Default scheduling required, but thread manager runs in realtime
       scheduling: switch new thread to SCHED_OTHER policy */
    {
      struct sched_param default_params;
      default_params.sched_priority = 0;
      __sched_setscheduler(THREAD_GETMEM(self, p_pid),
                           SCHED_OTHER, &default_params);
    }
  /* Make gdb aware of new thread */
  if (__pthread_threads_debug && __pthread_sig_debug > 0) {
    request.req_thread = self;
    request.req_kind = REQ_DEBUG;
    __libc_write(__pthread_manager_request,
                 (char *) &request, sizeof(request));
    suspend(self);
  }
  /* Run the thread code */
  outcome = self->p_start_args.start_routine(THREAD_GETMEM(self,
							   p_start_args.arg));
  /* Exit with the given return value */
  pthread_exit(outcome);
  return 0;
}

static int pthread_start_thread_event(void *arg)
{
  pthread_descr self = (pthread_descr) arg;

#ifdef INIT_THREAD_SELF
  INIT_THREAD_SELF(self, self->p_nr);
#endif
  /* Make sure our pid field is initialized, just in case we get there
     before our father has initialized it. */
  THREAD_SETMEM(self, p_pid, __getpid());
  /* Get the lock the manager will free once all is correctly set up.  */
  __pthread_lock (THREAD_GETMEM(self, p_lock), NULL);
  /* Free it immediately.  */
  __pthread_unlock (THREAD_GETMEM(self, p_lock));

  /* Continue with the real function.  */
  return pthread_start_thread (arg);
}

static int pthread_allocate_stack(const pthread_attr_t *attr,
                                  pthread_descr default_new_thread,
                                  int pagesize,
                                  pthread_descr * out_new_thread,
                                  char ** out_new_thread_bottom,
                                  char ** out_guardaddr,
                                  size_t * out_guardsize)
{
  pthread_descr new_thread;
  char * new_thread_bottom;
  char * guardaddr;
  size_t stacksize, guardsize;

  if (attr != NULL && attr->__stackaddr_set)
    {
      /* The user provided a stack.  For now we interpret the supplied
	 address as 1 + the highest addr. in the stack segment.  If a
	 separate register stack is needed, we place it at the low end
	 of the segment, relying on the associated stacksize to
	 determine the low end of the segment.  This differs from many
	 (but not all) other pthreads implementations.  The intent is
	 that on machines with a single stack growing toward higher
	 addresses, stackaddr would be the lowest address in the stack
	 segment, so that it is consistently close to the initial sp
	 value. */
      new_thread =
        (pthread_descr) ((long)(attr->__stackaddr) & -sizeof(void *)) - 1;
      new_thread_bottom = (char *) attr->__stackaddr - attr->__stacksize;
      guardaddr = NULL;
      guardsize = 0;
      __pthread_nonstandard_stacks = 1;
    }
  else
    {
#ifdef NEED_SEPARATE_REGISTER_STACK
      size_t granularity = 2 * pagesize;
      /* Try to make stacksize/2 a multiple of pagesize */
#else
      size_t granularity = pagesize;
#endif
      /* Allocate space for stack and thread descriptor at default address */
      if (attr != NULL)
	{
	  guardsize = page_roundup (attr->__guardsize, granularity);
	  stacksize = STACK_SIZE - guardsize;
	  stacksize = MIN (stacksize,
			   page_roundup (attr->__stacksize, granularity));
	}
      else
	{
	  guardsize = granularity;
	  stacksize = STACK_SIZE - granularity;
	}
      new_thread = default_new_thread;
#ifdef NEED_SEPARATE_REGISTER_STACK
      new_thread_bottom = (char *) (new_thread + 1) - stacksize - guardsize;
      /* Includes guard area, unlike the normal case.  Use the bottom
       end of the segment as backing store for the register stack.
       Needed on IA64.  In this case, we also map the entire stack at
       once.  According to David Mosberger, that's cheaper.  It also
       avoids the risk of intermittent failures due to other mappings
       in the same region.  The cost is that we might be able to map
       slightly fewer stacks.  */

      /* First the main stack: */
      if (mmap((caddr_t)((char *)(new_thread + 1) - stacksize / 2),
	       stacksize / 2, PROT_READ | PROT_WRITE | PROT_EXEC,
	       MAP_PRIVATE | MAP_ANONYMOUS | MAP_FIXED, -1, 0)
	  == MAP_FAILED)
	/* Bad luck, this segment is already mapped. */
	return -1;
      /* Then the register stack:	*/
      if (mmap((caddr_t)new_thread_bottom, stacksize/2,
	       PROT_READ | PROT_WRITE | PROT_EXEC,
	       MAP_PRIVATE | MAP_ANONYMOUS | MAP_FIXED, -1, 0)
	  == MAP_FAILED)
	{
	  munmap((caddr_t)((char *)(new_thread + 1) - stacksize/2),
		 stacksize/2);
	  return -1;
	}

      guardaddr = new_thread_bottom + stacksize/2;
      /* We leave the guard area in the middle unmapped.	*/
#else  /* !NEED_SEPARATE_REGISTER_STACK */
      new_thread_bottom = (char *) (new_thread + 1) - stacksize;
      if (mmap((caddr_t)((char *)(new_thread + 1) - INITIAL_STACK_SIZE),
               INITIAL_STACK_SIZE, PROT_READ | PROT_WRITE | PROT_EXEC,
               MAP_PRIVATE | MAP_ANONYMOUS | MAP_FIXED | MAP_GROWSDOWN,
               -1, 0) == MAP_FAILED)
        /* Bad luck, this segment is already mapped. */
        return -1;
      /* We manage to get a stack.  Now see whether we need a guard
	 and allocate it if necessary.  Notice that the default
	 attributes (stack_size = STACK_SIZE - pagesize and guardsize
	 = pagesize) do not need a guard page, since the RLIMIT_STACK
	 soft limit prevents stacks from running into one another. */
      if (stacksize == STACK_SIZE - pagesize)
        {
          /* We don't need a guard page. */
          guardaddr = NULL;
          guardsize = 0;
        }
      else
        {
          /* Put a bad page at the bottom of the stack */
	  guardaddr = (void *)new_thread_bottom - guardsize;
          if (mmap ((caddr_t) guardaddr, guardsize, 0, MAP_FIXED, -1, 0)
              == MAP_FAILED)
            {
              /* We don't make this an error.  */
              guardaddr = NULL;
              guardsize = 0;
            }
        }
#endif /* !NEED_SEPARATE_REGISTER_STACK */
    }
  /* Clear the thread data structure.  */
  memset (new_thread, '\0', sizeof (*new_thread));
  *out_new_thread = new_thread;
  *out_new_thread_bottom = new_thread_bottom;
  *out_guardaddr = guardaddr;
  *out_guardsize = guardsize;
  return 0;
}

static int pthread_handle_create(pthread_t *thread, const pthread_attr_t *attr,
				 void * (*start_routine)(void *), void *arg,
				 sigset_t * mask, int father_pid,
				 int report_events,
				 td_thr_events_t *event_maskp)
{
  size_t sseg;
  int pid;
  pthread_descr new_thread;
  char * new_thread_bottom;
  pthread_t new_thread_id;
  char *guardaddr = NULL;
  size_t guardsize = 0;
  int pagesize = __getpagesize();

  /* First check whether we have to change the policy and if yes, whether
     we can  do this.  Normally this should be done by examining the
     return value of the __sched_setscheduler call in pthread_start_thread
     but this is hard to implement.  FIXME  */
  if (attr != NULL && attr->__schedpolicy != SCHED_OTHER && geteuid () != 0)
    return EPERM;
  /* Find a free segment for the thread, and allocate a stack if needed */
  for (sseg = 2; ; sseg++)
    {
      if (sseg >= PTHREAD_THREADS_MAX)
	return EAGAIN;
      if (__pthread_handles[sseg].h_descr != NULL)
	continue;
      if (pthread_allocate_stack(attr, thread_segment(sseg), pagesize,
                                 &new_thread, &new_thread_bottom,
                                 &guardaddr, &guardsize) == 0)
        break;
    }
  __pthread_handles_num++;
  /* Allocate new thread identifier */
  pthread_threads_counter += PTHREAD_THREADS_MAX;
  new_thread_id = sseg + pthread_threads_counter;
  /* Initialize the thread descriptor.  Elements which have to be
     initialized to zero already have this value.  */
  new_thread->p_tid = new_thread_id;
  new_thread->p_lock = &(__pthread_handles[sseg].h_lock);
  new_thread->p_cancelstate = PTHREAD_CANCEL_ENABLE;
  new_thread->p_canceltype = PTHREAD_CANCEL_DEFERRED;
  new_thread->p_errnop = &new_thread->p_errno;
  new_thread->p_h_errnop = &new_thread->p_h_errno;
  new_thread->p_res._sock = -1;
  new_thread->p_resp = &new_thread->p_res;
  new_thread->p_guardaddr = guardaddr;
  new_thread->p_guardsize = guardsize;
  new_thread->p_header.data.self = new_thread;
  new_thread->p_nr = sseg;
  /* Initialize the thread handle */
  __pthread_init_lock(&__pthread_handles[sseg].h_lock);
  __pthread_handles[sseg].h_descr = new_thread;
  __pthread_handles[sseg].h_bottom = new_thread_bottom;
  /* Determine scheduling parameters for the thread */
  new_thread->p_start_args.schedpolicy = -1;
  if (attr != NULL) {
    new_thread->p_detached = attr->__detachstate;
    new_thread->p_userstack = attr->__stackaddr_set;

    switch(attr->__inheritsched) {
    case PTHREAD_EXPLICIT_SCHED:
      new_thread->p_start_args.schedpolicy = attr->__schedpolicy;
      memcpy (&new_thread->p_start_args.schedparam, &attr->__schedparam,
	      sizeof (struct sched_param));
      break;
    case PTHREAD_INHERIT_SCHED:
      new_thread->p_start_args.schedpolicy = __sched_getscheduler(father_pid);
      __sched_getparam(father_pid, &new_thread->p_start_args.schedparam);
      break;
    }
    new_thread->p_priority =
      new_thread->p_start_args.schedparam.sched_priority;
  }
  /* Finish setting up arguments to pthread_start_thread */
  new_thread->p_start_args.start_routine = start_routine;
  new_thread->p_start_args.arg = arg;
  new_thread->p_start_args.mask = *mask;
  /* Make the new thread ID available already now.  If any of the later
     functions fail we return an error value and the caller must not use
     the stored thread ID.  */
  *thread = new_thread_id;
  /* Raise priority of thread manager if needed */
  __pthread_manager_adjust_prio(new_thread->p_priority);
  /* Do the cloning.  We have to use two different functions depending
     on whether we are debugging or not.  */
  pid = 0;	/* Note that the thread never can have PID zero.  */
  if (report_events)
    {
      /* See whether the TD_CREATE event bit is set in any of the
         masks.  */
      int idx = __td_eventword (TD_CREATE);
      uint32_t mask = __td_eventmask (TD_CREATE);

      if ((mask & (__pthread_threads_events.event_bits[idx]
		   | event_maskp->event_bits[idx])) != 0)
	{
	  /* Lock the mutex the child will use now so that it will stop.  */
	  __pthread_lock(new_thread->p_lock, NULL);

	  /* We have to report this event.  */
#ifdef NEED_SEPARATE_REGISTER_STACK
	  /* Perhaps this version should be used on all platforms. But
	   this requires that __clone2 be uniformly supported
	   everywhere.

	   And there is some argument for changing the __clone2
	   interface to pass sp and bsp instead, making it more IA64
	   specific, but allowing stacks to grow outward from each
	   other, to get less paging and fewer mmaps.  Clone2
	   currently can't take signals in the child right after
	   process creation.  Mask them in the child.  It resets the
	   mask once it starts up.  */
	  sigprocmask(SIG_SETMASK, &manager_mask_all, NULL);
	  pid = __clone2(pthread_start_thread_event,
  		 (void **)new_thread_bottom,
			 (char *)new_thread - new_thread_bottom,
			 CLONE_VM | CLONE_FS | CLONE_FILES | CLONE_SIGHAND |
			 __pthread_sig_cancel, new_thread);
	  sigprocmask(SIG_SETMASK, &manager_mask, NULL);
#else
	  pid = __clone(pthread_start_thread_event, (void **) new_thread,
			CLONE_VM | CLONE_FS | CLONE_FILES | CLONE_SIGHAND |
			__pthread_sig_cancel, new_thread);
#endif
	  if (pid != -1)
	    {
	      /* Now fill in the information about the new thread in
		 the newly created thread's data structure.  We cannot let
		 the new thread do this since we don't know whether it was
		 already scheduled when we send the event.  */
	      new_thread->p_eventbuf.eventdata = new_thread;
	      new_thread->p_eventbuf.eventnum = TD_CREATE;
	      __pthread_last_event = new_thread;

	      /* We have to set the PID here since the callback function
		 in the debug library will need it and we cannot guarantee
		 the child got scheduled before the debugger.  */
	      new_thread->p_pid = pid;

	      /* Now call the function which signals the event.  */
	      __linuxthreads_create_event ();

	      /* Now restart the thread.  */
	      __pthread_unlock(new_thread->p_lock);
	    }
	}
    }
  if (pid == 0)
    {
#ifdef NEED_SEPARATE_REGISTER_STACK
      sigprocmask(SIG_SETMASK, &manager_mask_all, NULL);
      pid = __clone2(pthread_start_thread,
		     (void **)new_thread_bottom,
                     (char *)new_thread - new_thread_bottom,
		     CLONE_VM | CLONE_FS | CLONE_FILES | CLONE_SIGHAND |
		     __pthread_sig_cancel, new_thread);
      sigprocmask(SIG_SETMASK, &manager_mask, NULL);
#else
      pid = __clone(pthread_start_thread, (void **) new_thread,
		    CLONE_VM | CLONE_FS | CLONE_FILES | CLONE_SIGHAND |
		    __pthread_sig_cancel, new_thread);
#endif /* !NEED_SEPARATE_REGISTER_STACK */
    }
  /* Check if cloning succeeded */
  if (pid == -1) {
    /* Free the stack if we allocated it */
    if (attr == NULL || !attr->__stackaddr_set)
      {
#ifdef NEED_SEPARATE_REGISTER_STACK
	size_t stacksize = ((char *)(new_thread->p_guardaddr)
			    - new_thread_bottom);
	munmap((caddr_t)new_thread_bottom, stacksize);
	munmap((caddr_t)new_thread_bottom + stacksize
	       + new_thread->p_guardsize, stacksize);
#else
	if (new_thread->p_guardsize != 0)
	  munmap(new_thread->p_guardaddr, new_thread->p_guardsize);
	munmap((caddr_t)((char *)(new_thread+1) - INITIAL_STACK_SIZE),
	       INITIAL_STACK_SIZE);
#endif
      }
    __pthread_handles[sseg].h_descr = NULL;
    __pthread_handles[sseg].h_bottom = NULL;
    __pthread_handles_num--;
    return errno;
  }
  /* Insert new thread in doubly linked list of active threads */
  new_thread->p_prevlive = __pthread_main_thread;
  new_thread->p_nextlive = __pthread_main_thread->p_nextlive;
  __pthread_main_thread->p_nextlive->p_prevlive = new_thread;
  __pthread_main_thread->p_nextlive = new_thread;
  /* Set pid field of the new thread, in case we get there before the
     child starts. */
  new_thread->p_pid = pid;
  return 0;
}


/* Try to free the resources of a thread when requested by pthread_join
   or pthread_detach on a terminated thread. */

static void pthread_free(pthread_descr th)
{
  pthread_handle handle;
  pthread_readlock_info *iter, *next;

  ASSERT(th->p_exited);
  /* Make the handle invalid */
  handle =  thread_handle(th->p_tid);
  __pthread_lock(&handle->h_lock, NULL);
  handle->h_descr = NULL;
  handle->h_bottom = (char *)(-1L);
  __pthread_unlock(&handle->h_lock);
#ifdef FREE_THREAD
  FREE_THREAD(th, th->p_nr);
#endif
  /* One fewer threads in __pthread_handles */
  __pthread_handles_num--;

  /* Destroy read lock list, and list of free read lock structures.
     If the former is not empty, it means the thread exited while
     holding read locks! */

  for (iter = th->p_readlock_list; iter != NULL; iter = next)
    {
      next = iter->pr_next;
      free(iter);
    }

  for (iter = th->p_readlock_free; iter != NULL; iter = next)
    {
      next = iter->pr_next;
      free(iter);
    }

  /* If initial thread, nothing to free */
  if (th == &__pthread_initial_thread) return;
  if (!th->p_userstack)
    {
      size_t guardsize = th->p_guardsize;
      /* Free the stack and thread descriptor area */
#ifdef NEED_SEPARATE_REGISTER_STACK
      char *guardaddr = th->p_guardaddr;
      /* We unmap exactly what we mapped, in case there was something
	 else in the same region.  Guardaddr is always set, eve if
	 guardsize is 0.  This allows us to compute everything else.  */
      size_t stacksize = (char *)(th+1) - guardaddr - guardsize;
      /* Unmap the register stack, which is below guardaddr.  */
      munmap((caddr_t)(guardaddr-stacksize), stacksize);
      /* Unmap the main stack.	*/
      munmap((caddr_t)(guardaddr+guardsize), stacksize);
#else
      /* The following assumes that we only allocate stacks of one
	 size.  That's currently true but probably shouldn't be.  This
	 looks like it fails for growing stacks if there was something
	 else mapped just below the stack?  */
      if (guardsize != 0)
	munmap(th->p_guardaddr, guardsize);
      munmap((caddr_t) ((char *)(th+1) - STACK_SIZE), STACK_SIZE);
#endif
    }
}

/* Handle threads that have exited */

static void pthread_exited(pid_t pid)
{
  pthread_descr th;
  int detached;
  /* Find thread with that pid */
  for (th = __pthread_main_thread->p_nextlive;
       th != __pthread_main_thread;
       th = th->p_nextlive) {
    if (th->p_pid == pid) {
      /* Remove thread from list of active threads */
      th->p_nextlive->p_prevlive = th->p_prevlive;
      th->p_prevlive->p_nextlive = th->p_nextlive;
      /* Mark thread as exited, and if detached, free its resources */
      __pthread_lock(th->p_lock, NULL);
      th->p_exited = 1;
      /* If we have to signal this event do it now.  */
      if (th->p_report_events)
	{
	  /* See whether TD_DEATH is in any of the mask.  */
	  int idx = __td_eventword (TD_REAP);
	  uint32_t mask = __td_eventmask (TD_REAP);

	  if ((mask & (__pthread_threads_events.event_bits[idx]
		       | th->p_eventbuf.eventmask.event_bits[idx])) != 0)
	    {
	      /* Yep, we have to signal the death.  */
	      th->p_eventbuf.eventnum = TD_DEATH;
	      th->p_eventbuf.eventdata = th;
	      __pthread_last_event = th;

	      /* Now call the function to signal the event.  */
	      __linuxthreads_reap_event();
	    }
	}
      detached = th->p_detached;
      __pthread_unlock(th->p_lock);
      if (detached)
	pthread_free(th);
      break;
    }
  }
  /* If all threads have exited and the main thread is pending on a
     pthread_exit, wake up the main thread and terminate ourselves. */
  if (main_thread_exiting &&
      __pthread_main_thread->p_nextlive == __pthread_main_thread) {
    restart(__pthread_main_thread);
    /* Same logic as REQ_MAIN_THREAD_EXIT. */
  }
}

static void pthread_reap_children(void)
{
  pid_t pid;
  int status;

  while ((pid = __libc_waitpid(-1, &status, WNOHANG | __WCLONE)) > 0) {
    pthread_exited(pid);
    if (WIFSIGNALED(status)) {
      /* If a thread died due to a signal, send the same signal to
         all other threads, including the main thread. */
      pthread_kill_all_threads(WTERMSIG(status), 1);
      _exit(0);
    }
  }
}

/* Try to free the resources of a thread when requested by pthread_join
   or pthread_detach on a terminated thread. */

static void pthread_handle_free(pthread_t th_id)
{
  pthread_handle handle = thread_handle(th_id);
  pthread_descr th;

  __pthread_lock(&handle->h_lock, NULL);
  if (nonexisting_handle(handle, th_id)) {
    /* pthread_reap_children has deallocated the thread already,
       nothing needs to be done */
    __pthread_unlock(&handle->h_lock);
    return;
  }
  th = handle->h_descr;
  if (th->p_exited) {
    __pthread_unlock(&handle->h_lock);
    pthread_free(th);
  } else {
    /* The Unix process of the thread is still running.
       Mark the thread as detached so that the thread manager will
       deallocate its resources when the Unix process exits. */
    th->p_detached = 1;
    __pthread_unlock(&handle->h_lock);
  }
}

/* Send a signal to all running threads */

static void pthread_kill_all_threads(int sig, int main_thread_also)
{
  pthread_descr th;
  for (th = __pthread_main_thread->p_nextlive;
       th != __pthread_main_thread;
       th = th->p_nextlive) {
    kill(th->p_pid, sig);
  }
  if (main_thread_also) {
    kill(__pthread_main_thread->p_pid, sig);
  }
}

/* Process-wide exit() */

static void pthread_handle_exit(pthread_descr issuing_thread, int exitcode)
{
  pthread_descr th;
  __pthread_exit_requested = 1;
  __pthread_exit_code = exitcode;
  /* Send the CANCEL signal to all running threads, including the main
     thread, but excluding the thread from which the exit request originated
     (that thread must complete the exit, e.g. calling atexit functions
     and flushing stdio buffers). */
  for (th = issuing_thread->p_nextlive;
       th != issuing_thread;
       th = th->p_nextlive) {
    kill(th->p_pid, __pthread_sig_cancel);
  }
  /* Now, wait for all these threads, so that they don't become zombies
     and their times are properly added to the thread manager's times. */
  for (th = issuing_thread->p_nextlive;
       th != issuing_thread;
       th = th->p_nextlive) {
    waitpid(th->p_pid, NULL, __WCLONE);
  }
  restart(issuing_thread);
  _exit(0);
}

/* Handler for __pthread_sig_cancel in thread manager thread */

void __pthread_manager_sighandler(int sig)
{
  int kick_manager = terminated_children == 0 && main_thread_exiting;
  terminated_children = 1;

  /* If the main thread is terminating, kick the thread manager loop
     each time some threads terminate. This eliminates a two second
     shutdown delay caused by the thread manager sleeping in the
     call to __poll(). Instead, the thread manager is kicked into
     action, reaps the outstanding threads and resumes the main thread
     so that it can complete the shutdown. */

  if (kick_manager) {
    struct pthread_request request;
    request.req_thread = 0;
    request.req_kind = REQ_KICK;
    __libc_write(__pthread_manager_request, (char *) &request, sizeof(request));
  }
}

/* Adjust priority of thread manager so that it always run at a priority
   higher than all threads */

void __pthread_manager_adjust_prio(int thread_prio)
{
  struct sched_param param;

  if (thread_prio <= __pthread_manager_thread.p_priority) return;
  param.sched_priority =
    thread_prio < __sched_get_priority_max(SCHED_FIFO)
    ? thread_prio + 1 : thread_prio;
  __sched_setscheduler(__pthread_manager_thread.p_pid, SCHED_FIFO, &param);
  __pthread_manager_thread.p_priority = thread_prio;
}