about summary refs log tree commit diff
path: root/malloc/arena.c
blob: afd423240a2fd9160fb651442e225425adcbe4a8 (plain) (blame)
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
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
/* Malloc implementation for multiple threads without lock contention.
   Copyright (C) 2001-2017 Free Software Foundation, Inc.
   This file is part of the GNU C Library.
   Contributed by Wolfram Gloger <wg@malloc.de>, 2001.

   The GNU C Library is free software; you can redistribute it and/or
   modify it under the terms of the GNU Lesser General Public License as
   published by the Free Software Foundation; either version 2.1 of the
   License, or (at your option) any later version.

   The GNU C Library 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
   Lesser General Public License for more details.

   You should have received a copy of the GNU Lesser General Public
   License along with the GNU C Library; see the file COPYING.LIB.  If
   not, see <http://www.gnu.org/licenses/>.  */

#include <stdbool.h>

#if HAVE_TUNABLES
# define TUNABLE_NAMESPACE malloc
#endif
#include <elf/dl-tunables.h>

/* Compile-time constants.  */

#define HEAP_MIN_SIZE (32 * 1024)
#ifndef HEAP_MAX_SIZE
# ifdef DEFAULT_MMAP_THRESHOLD_MAX
#  define HEAP_MAX_SIZE (2 * DEFAULT_MMAP_THRESHOLD_MAX)
# else
#  define HEAP_MAX_SIZE (1024 * 1024) /* must be a power of two */
# endif
#endif

/* HEAP_MIN_SIZE and HEAP_MAX_SIZE limit the size of mmap()ed heaps
   that are dynamically created for multi-threaded programs.  The
   maximum size must be a power of two, for fast determination of
   which heap belongs to a chunk.  It should be much larger than the
   mmap threshold, so that requests with a size just below that
   threshold can be fulfilled without creating too many heaps.  */

/***************************************************************************/

#define top(ar_ptr) ((ar_ptr)->top)

/* A heap is a single contiguous memory region holding (coalesceable)
   malloc_chunks.  It is allocated with mmap() and always starts at an
   address aligned to HEAP_MAX_SIZE.  */

typedef struct _heap_info
{
  mstate ar_ptr; /* Arena for this heap. */
  struct _heap_info *prev; /* Previous heap. */
  size_t size;   /* Current size in bytes. */
  size_t mprotect_size; /* Size in bytes that has been mprotected
                           PROT_READ|PROT_WRITE.  */
  /* Make sure the following data is properly aligned, particularly
     that sizeof (heap_info) + 2 * SIZE_SZ is a multiple of
     MALLOC_ALIGNMENT. */
  char pad[-6 * SIZE_SZ & MALLOC_ALIGN_MASK];
} heap_info;

/* Get a compile-time error if the heap_info padding is not correct
   to make alignment work as expected in sYSMALLOc.  */
extern int sanity_check_heap_info_alignment[(sizeof (heap_info)
                                             + 2 * SIZE_SZ) % MALLOC_ALIGNMENT
                                            ? -1 : 1];

/* Thread specific data.  */

static __thread mstate thread_arena attribute_tls_model_ie;

/* Arena free list.  free_list_lock synchronizes access to the
   free_list variable below, and the next_free and attached_threads
   members of struct malloc_state objects.  No other locks must be
   acquired after free_list_lock has been acquired.  */

__libc_lock_define_initialized (static, free_list_lock);
static size_t narenas = 1;
static mstate free_list;

/* list_lock prevents concurrent writes to the next member of struct
   malloc_state objects.

   Read access to the next member is supposed to synchronize with the
   atomic_write_barrier and the write to the next member in
   _int_new_arena.  This suffers from data races; see the FIXME
   comments in _int_new_arena and reused_arena.

   list_lock also prevents concurrent forks.  At the time list_lock is
   acquired, no arena lock must have been acquired, but it is
   permitted to acquire arena locks subsequently, while list_lock is
   acquired.  */
__libc_lock_define_initialized (static, list_lock);

/* Already initialized? */
int __malloc_initialized = -1;

/**************************************************************************/


/* arena_get() acquires an arena and locks the corresponding mutex.
   First, try the one last locked successfully by this thread.  (This
   is the common case and handled with a macro for speed.)  Then, loop
   once over the circularly linked list of arenas.  If no arena is
   readily available, create a new one.  In this latter case, `size'
   is just a hint as to how much memory will be required immediately
   in the new arena. */

#define arena_get(ptr, size) do { \
      ptr = thread_arena;						      \
      arena_lock (ptr, size);						      \
  } while (0)

#define arena_lock(ptr, size) do {					      \
      if (ptr)								      \
        __libc_lock_lock (ptr->mutex);					      \
      else								      \
        ptr = arena_get2 ((size), NULL);				      \
  } while (0)

/* find the heap and corresponding arena for a given ptr */

#define heap_for_ptr(ptr) \
  ((heap_info *) ((unsigned long) (ptr) & ~(HEAP_MAX_SIZE - 1)))
#define arena_for_chunk(ptr) \
  (chunk_main_arena (ptr) ? &main_arena : heap_for_ptr (ptr)->ar_ptr)


/**************************************************************************/

/* atfork support.  */

/* The following three functions are called around fork from a
   multi-threaded process.  We do not use the general fork handler
   mechanism to make sure that our handlers are the last ones being
   called, so that other fork handlers can use the malloc
   subsystem.  */

void
internal_function
__malloc_fork_lock_parent (void)
{
  if (__malloc_initialized < 1)
    return;

  /* We do not acquire free_list_lock here because we completely
     reconstruct free_list in __malloc_fork_unlock_child.  */

  __libc_lock_lock (list_lock);

  for (mstate ar_ptr = &main_arena;; )
    {
      __libc_lock_lock (ar_ptr->mutex);
      ar_ptr = ar_ptr->next;
      if (ar_ptr == &main_arena)
        break;
    }
}

void
internal_function
__malloc_fork_unlock_parent (void)
{
  if (__malloc_initialized < 1)
    return;

  for (mstate ar_ptr = &main_arena;; )
    {
      __libc_lock_unlock (ar_ptr->mutex);
      ar_ptr = ar_ptr->next;
      if (ar_ptr == &main_arena)
        break;
    }
  __libc_lock_unlock (list_lock);
}

void
internal_function
__malloc_fork_unlock_child (void)
{
  if (__malloc_initialized < 1)
    return;

  /* Push all arenas to the free list, except thread_arena, which is
     attached to the current thread.  */
  __libc_lock_init (free_list_lock);
  if (thread_arena != NULL)
    thread_arena->attached_threads = 1;
  free_list = NULL;
  for (mstate ar_ptr = &main_arena;; )
    {
      __libc_lock_init (ar_ptr->mutex);
      if (ar_ptr != thread_arena)
        {
	  /* This arena is no longer attached to any thread.  */
	  ar_ptr->attached_threads = 0;
          ar_ptr->next_free = free_list;
          free_list = ar_ptr;
        }
      ar_ptr = ar_ptr->next;
      if (ar_ptr == &main_arena)
        break;
    }

  __libc_lock_init (list_lock);
}

#if HAVE_TUNABLES
static inline int do_set_mallopt_check (int32_t value);
void
TUNABLE_CALLBACK (set_mallopt_check) (tunable_val_t *valp)
{
  int32_t value = (int32_t) valp->numval;
  if (value != 0)
    __malloc_check_init ();
}

# define TUNABLE_CALLBACK_FNDECL(__name, __type) \
static inline int do_ ## __name (__type value);				      \
void									      \
TUNABLE_CALLBACK (__name) (tunable_val_t *valp)				      \
{									      \
  __type value = (__type) (valp)->numval;				      \
  do_ ## __name (value);						      \
}

TUNABLE_CALLBACK_FNDECL (set_mmap_threshold, size_t)
TUNABLE_CALLBACK_FNDECL (set_mmaps_max, int32_t)
TUNABLE_CALLBACK_FNDECL (set_top_pad, size_t)
TUNABLE_CALLBACK_FNDECL (set_perturb_byte, int32_t)
TUNABLE_CALLBACK_FNDECL (set_trim_threshold, size_t)
TUNABLE_CALLBACK_FNDECL (set_arena_max, size_t)
TUNABLE_CALLBACK_FNDECL (set_arena_test, size_t)
#if USE_TCACHE
TUNABLE_CALLBACK_FNDECL (set_tcache_max, size_t)
TUNABLE_CALLBACK_FNDECL (set_tcache_count, size_t)
TUNABLE_CALLBACK_FNDECL (set_tcache_unsorted_limit, size_t)
#endif
#else
/* Initialization routine. */
#include <string.h>
extern char **_environ;

static char *
internal_function
next_env_entry (char ***position)
{
  char **current = *position;
  char *result = NULL;

  while (*current != NULL)
    {
      if (__builtin_expect ((*current)[0] == 'M', 0)
          && (*current)[1] == 'A'
          && (*current)[2] == 'L'
          && (*current)[3] == 'L'
          && (*current)[4] == 'O'
          && (*current)[5] == 'C'
          && (*current)[6] == '_')
        {
          result = &(*current)[7];

          /* Save current position for next visit.  */
          *position = ++current;

          break;
        }

      ++current;
    }

  return result;
}
#endif


#ifdef SHARED
static void *
__failing_morecore (ptrdiff_t d)
{
  return (void *) MORECORE_FAILURE;
}

extern struct dl_open_hook *_dl_open_hook;
libc_hidden_proto (_dl_open_hook);
#endif

static void
ptmalloc_init (void)
{
  if (__malloc_initialized >= 0)
    return;

  __malloc_initialized = 0;

#ifdef SHARED
  /* In case this libc copy is in a non-default namespace, never use brk.
     Likewise if dlopened from statically linked program.  */
  Dl_info di;
  struct link_map *l;

  if (_dl_open_hook != NULL
      || (_dl_addr (ptmalloc_init, &di, &l, NULL) != 0
          && l->l_ns != LM_ID_BASE))
    __morecore = __failing_morecore;
#endif

  thread_arena = &main_arena;

#if HAVE_TUNABLES
  /* Ensure initialization/consolidation and do it under a lock so that a
     thread attempting to use the arena in parallel waits on us till we
     finish.  */
  __libc_lock_lock (main_arena.mutex);
  malloc_consolidate (&main_arena);

  TUNABLE_GET (check, int32_t, TUNABLE_CALLBACK (set_mallopt_check));
  TUNABLE_GET (top_pad, size_t, TUNABLE_CALLBACK (set_top_pad));
  TUNABLE_GET (perturb, int32_t, TUNABLE_CALLBACK (set_perturb_byte));
  TUNABLE_GET (mmap_threshold, size_t, TUNABLE_CALLBACK (set_mmap_threshold));
  TUNABLE_GET (trim_threshold, size_t, TUNABLE_CALLBACK (set_trim_threshold));
  TUNABLE_GET (mmap_max, int32_t, TUNABLE_CALLBACK (set_mmaps_max));
  TUNABLE_GET (arena_max, size_t, TUNABLE_CALLBACK (set_arena_max));
  TUNABLE_GET (arena_test, size_t, TUNABLE_CALLBACK (set_arena_test));
#if USE_TCACHE
  TUNABLE_GET (tcache_max, size_t, TUNABLE_CALLBACK (set_tcache_max));
  TUNABLE_GET (tcache_count, size_t, TUNABLE_CALLBACK (set_tcache_count));
  TUNABLE_GET (tcache_unsorted_limit, size_t,
	       TUNABLE_CALLBACK (set_tcache_unsorted_limit));
#endif
  __libc_lock_unlock (main_arena.mutex);
#else
  const char *s = NULL;
  if (__glibc_likely (_environ != NULL))
    {
      char **runp = _environ;
      char *envline;

      while (__builtin_expect ((envline = next_env_entry (&runp)) != NULL,
                               0))
        {
          size_t len = strcspn (envline, "=");

          if (envline[len] != '=')
            /* This is a "MALLOC_" variable at the end of the string
               without a '=' character.  Ignore it since otherwise we
               will access invalid memory below.  */
            continue;

          switch (len)
            {
            case 6:
              if (memcmp (envline, "CHECK_", 6) == 0)
                s = &envline[7];
              break;
            case 8:
              if (!__builtin_expect (__libc_enable_secure, 0))
                {
                  if (memcmp (envline, "TOP_PAD_", 8) == 0)
                    __libc_mallopt (M_TOP_PAD, atoi (&envline[9]));
                  else if (memcmp (envline, "PERTURB_", 8) == 0)
                    __libc_mallopt (M_PERTURB, atoi (&envline[9]));
                }
              break;
            case 9:
              if (!__builtin_expect (__libc_enable_secure, 0))
                {
                  if (memcmp (envline, "MMAP_MAX_", 9) == 0)
                    __libc_mallopt (M_MMAP_MAX, atoi (&envline[10]));
                  else if (memcmp (envline, "ARENA_MAX", 9) == 0)
                    __libc_mallopt (M_ARENA_MAX, atoi (&envline[10]));
                }
              break;
            case 10:
              if (!__builtin_expect (__libc_enable_secure, 0))
                {
                  if (memcmp (envline, "ARENA_TEST", 10) == 0)
                    __libc_mallopt (M_ARENA_TEST, atoi (&envline[11]));
                }
              break;
            case 15:
              if (!__builtin_expect (__libc_enable_secure, 0))
                {
                  if (memcmp (envline, "TRIM_THRESHOLD_", 15) == 0)
                    __libc_mallopt (M_TRIM_THRESHOLD, atoi (&envline[16]));
                  else if (memcmp (envline, "MMAP_THRESHOLD_", 15) == 0)
                    __libc_mallopt (M_MMAP_THRESHOLD, atoi (&envline[16]));
                }
              break;
            default:
              break;
            }
        }
    }
  if (s && s[0] != '\0' && s[0] != '0')
    __malloc_check_init ();
#endif

#if HAVE_MALLOC_INIT_HOOK
  void (*hook) (void) = atomic_forced_read (__malloc_initialize_hook);
  if (hook != NULL)
    (*hook)();
#endif
  __malloc_initialized = 1;
}

/* Managing heaps and arenas (for concurrent threads) */

#if MALLOC_DEBUG > 1

/* Print the complete contents of a single heap to stderr. */

static void
dump_heap (heap_info *heap)
{
  char *ptr;
  mchunkptr p;

  fprintf (stderr, "Heap %p, size %10lx:\n", heap, (long) heap->size);
  ptr = (heap->ar_ptr != (mstate) (heap + 1)) ?
        (char *) (heap + 1) : (char *) (heap + 1) + sizeof (struct malloc_state);
  p = (mchunkptr) (((unsigned long) ptr + MALLOC_ALIGN_MASK) &
                   ~MALLOC_ALIGN_MASK);
  for (;; )
    {
      fprintf (stderr, "chunk %p size %10lx", p, (long) p->size);
      if (p == top (heap->ar_ptr))
        {
          fprintf (stderr, " (top)\n");
          break;
        }
      else if (p->size == (0 | PREV_INUSE))
        {
          fprintf (stderr, " (fence)\n");
          break;
        }
      fprintf (stderr, "\n");
      p = next_chunk (p);
    }
}
#endif /* MALLOC_DEBUG > 1 */

/* If consecutive mmap (0, HEAP_MAX_SIZE << 1, ...) calls return decreasing
   addresses as opposed to increasing, new_heap would badly fragment the
   address space.  In that case remember the second HEAP_MAX_SIZE part
   aligned to HEAP_MAX_SIZE from last mmap (0, HEAP_MAX_SIZE << 1, ...)
   call (if it is already aligned) and try to reuse it next time.  We need
   no locking for it, as kernel ensures the atomicity for us - worst case
   we'll call mmap (addr, HEAP_MAX_SIZE, ...) for some value of addr in
   multiple threads, but only one will succeed.  */
static char *aligned_heap_area;

/* Create a new heap.  size is automatically rounded up to a multiple
   of the page size. */

static heap_info *
internal_function
new_heap (size_t size, size_t top_pad)
{
  size_t pagesize = GLRO (dl_pagesize);
  char *p1, *p2;
  unsigned long ul;
  heap_info *h;

  if (size + top_pad < HEAP_MIN_SIZE)
    size = HEAP_MIN_SIZE;
  else if (size + top_pad <= HEAP_MAX_SIZE)
    size += top_pad;
  else if (size > HEAP_MAX_SIZE)
    return 0;
  else
    size = HEAP_MAX_SIZE;
  size = ALIGN_UP (size, pagesize);

  /* A memory region aligned to a multiple of HEAP_MAX_SIZE is needed.
     No swap space needs to be reserved for the following large
     mapping (on Linux, this is the case for all non-writable mappings
     anyway). */
  p2 = MAP_FAILED;
  if (aligned_heap_area)
    {
      p2 = (char *) MMAP (aligned_heap_area, HEAP_MAX_SIZE, PROT_NONE,
                          MAP_NORESERVE);
      aligned_heap_area = NULL;
      if (p2 != MAP_FAILED && ((unsigned long) p2 & (HEAP_MAX_SIZE - 1)))
        {
          __munmap (p2, HEAP_MAX_SIZE);
          p2 = MAP_FAILED;
        }
    }
  if (p2 == MAP_FAILED)
    {
      p1 = (char *) MMAP (0, HEAP_MAX_SIZE << 1, PROT_NONE, MAP_NORESERVE);
      if (p1 != MAP_FAILED)
        {
          p2 = (char *) (((unsigned long) p1 + (HEAP_MAX_SIZE - 1))
                         & ~(HEAP_MAX_SIZE - 1));
          ul = p2 - p1;
          if (ul)
            __munmap (p1, ul);
          else
            aligned_heap_area = p2 + HEAP_MAX_SIZE;
          __munmap (p2 + HEAP_MAX_SIZE, HEAP_MAX_SIZE - ul);
        }
      else
        {
          /* Try to take the chance that an allocation of only HEAP_MAX_SIZE
             is already aligned. */
          p2 = (char *) MMAP (0, HEAP_MAX_SIZE, PROT_NONE, MAP_NORESERVE);
          if (p2 == MAP_FAILED)
            return 0;

          if ((unsigned long) p2 & (HEAP_MAX_SIZE - 1))
            {
              __munmap (p2, HEAP_MAX_SIZE);
              return 0;
            }
        }
    }
  if (__mprotect (p2, size, PROT_READ | PROT_WRITE) != 0)
    {
      __munmap (p2, HEAP_MAX_SIZE);
      return 0;
    }
  h = (heap_info *) p2;
  h->size = size;
  h->mprotect_size = size;
  LIBC_PROBE (memory_heap_new, 2, h, h->size);
  return h;
}

/* Grow a heap.  size is automatically rounded up to a
   multiple of the page size. */

static int
grow_heap (heap_info *h, long diff)
{
  size_t pagesize = GLRO (dl_pagesize);
  long new_size;

  diff = ALIGN_UP (diff, pagesize);
  new_size = (long) h->size + diff;
  if ((unsigned long) new_size > (unsigned long) HEAP_MAX_SIZE)
    return -1;

  if ((unsigned long) new_size > h->mprotect_size)
    {
      if (__mprotect ((char *) h + h->mprotect_size,
                      (unsigned long) new_size - h->mprotect_size,
                      PROT_READ | PROT_WRITE) != 0)
        return -2;

      h->mprotect_size = new_size;
    }

  h->size = new_size;
  LIBC_PROBE (memory_heap_more, 2, h, h->size);
  return 0;
}

/* Shrink a heap.  */

static int
shrink_heap (heap_info *h, long diff)
{
  long new_size;

  new_size = (long) h->size - diff;
  if (new_size < (long) sizeof (*h))
    return -1;

  /* Try to re-map the extra heap space freshly to save memory, and make it
     inaccessible.  See malloc-sysdep.h to know when this is true.  */
  if (__glibc_unlikely (check_may_shrink_heap ()))
    {
      if ((char *) MMAP ((char *) h + new_size, diff, PROT_NONE,
                         MAP_FIXED) == (char *) MAP_FAILED)
        return -2;

      h->mprotect_size = new_size;
    }
  else
    __madvise ((char *) h + new_size, diff, MADV_DONTNEED);
  /*fprintf(stderr, "shrink %p %08lx\n", h, new_size);*/

  h->size = new_size;
  LIBC_PROBE (memory_heap_less, 2, h, h->size);
  return 0;
}

/* Delete a heap. */

#define delete_heap(heap) \
  do {									      \
      if ((char *) (heap) + HEAP_MAX_SIZE == aligned_heap_area)		      \
        aligned_heap_area = NULL;					      \
      __munmap ((char *) (heap), HEAP_MAX_SIZE);			      \
    } while (0)

static int
internal_function
heap_trim (heap_info *heap, size_t pad)
{
  mstate ar_ptr = heap->ar_ptr;
  unsigned long pagesz = GLRO (dl_pagesize);
  mchunkptr top_chunk = top (ar_ptr), p, bck, fwd;
  heap_info *prev_heap;
  long new_size, top_size, top_area, extra, prev_size, misalign;

  /* Can this heap go away completely? */
  while (top_chunk == chunk_at_offset (heap, sizeof (*heap)))
    {
      prev_heap = heap->prev;
      prev_size = prev_heap->size - (MINSIZE - 2 * SIZE_SZ);
      p = chunk_at_offset (prev_heap, prev_size);
      /* fencepost must be properly aligned.  */
      misalign = ((long) p) & MALLOC_ALIGN_MASK;
      p = chunk_at_offset (prev_heap, prev_size - misalign);
      assert (chunksize_nomask (p) == (0 | PREV_INUSE)); /* must be fencepost */
      p = prev_chunk (p);
      new_size = chunksize (p) + (MINSIZE - 2 * SIZE_SZ) + misalign;
      assert (new_size > 0 && new_size < (long) (2 * MINSIZE));
      if (!prev_inuse (p))
        new_size += prev_size (p);
      assert (new_size > 0 && new_size < HEAP_MAX_SIZE);
      if (new_size + (HEAP_MAX_SIZE - prev_heap->size) < pad + MINSIZE + pagesz)
        break;
      ar_ptr->system_mem -= heap->size;
      LIBC_PROBE (memory_heap_free, 2, heap, heap->size);
      delete_heap (heap);
      heap = prev_heap;
      if (!prev_inuse (p)) /* consolidate backward */
        {
          p = prev_chunk (p);
          unlink (ar_ptr, p, bck, fwd);
        }
      assert (((unsigned long) ((char *) p + new_size) & (pagesz - 1)) == 0);
      assert (((char *) p + new_size) == ((char *) heap + heap->size));
      top (ar_ptr) = top_chunk = p;
      set_head (top_chunk, new_size | PREV_INUSE);
      /*check_chunk(ar_ptr, top_chunk);*/
    }

  /* Uses similar logic for per-thread arenas as the main arena with systrim
     and _int_free by preserving the top pad and rounding down to the nearest
     page.  */
  top_size = chunksize (top_chunk);
  if ((unsigned long)(top_size) <
      (unsigned long)(mp_.trim_threshold))
    return 0;

  top_area = top_size - MINSIZE - 1;
  if (top_area < 0 || (size_t) top_area <= pad)
    return 0;

  /* Release in pagesize units and round down to the nearest page.  */
  extra = ALIGN_DOWN(top_area - pad, pagesz);
  if (extra == 0)
    return 0;

  /* Try to shrink. */
  if (shrink_heap (heap, extra) != 0)
    return 0;

  ar_ptr->system_mem -= extra;

  /* Success. Adjust top accordingly. */
  set_head (top_chunk, (top_size - extra) | PREV_INUSE);
  /*check_chunk(ar_ptr, top_chunk);*/
  return 1;
}

/* Create a new arena with initial size "size".  */

/* If REPLACED_ARENA is not NULL, detach it from this thread.  Must be
   called while free_list_lock is held.  */
static void
detach_arena (mstate replaced_arena)
{
  if (replaced_arena != NULL)
    {
      assert (replaced_arena->attached_threads > 0);
      /* The current implementation only detaches from main_arena in
	 case of allocation failure.  This means that it is likely not
	 beneficial to put the arena on free_list even if the
	 reference count reaches zero.  */
      --replaced_arena->attached_threads;
    }
}

static mstate
_int_new_arena (size_t size)
{
  mstate a;
  heap_info *h;
  char *ptr;
  unsigned long misalign;

  h = new_heap (size + (sizeof (*h) + sizeof (*a) + MALLOC_ALIGNMENT),
                mp_.top_pad);
  if (!h)
    {
      /* Maybe size is too large to fit in a single heap.  So, just try
         to create a minimally-sized arena and let _int_malloc() attempt
         to deal with the large request via mmap_chunk().  */
      h = new_heap (sizeof (*h) + sizeof (*a) + MALLOC_ALIGNMENT, mp_.top_pad);
      if (!h)
        return 0;
    }
  a = h->ar_ptr = (mstate) (h + 1);
  malloc_init_state (a);
  a->attached_threads = 1;
  /*a->next = NULL;*/
  a->system_mem = a->max_system_mem = h->size;

  /* Set up the top chunk, with proper alignment. */
  ptr = (char *) (a + 1);
  misalign = (unsigned long) chunk2mem (ptr) & MALLOC_ALIGN_MASK;
  if (misalign > 0)
    ptr += MALLOC_ALIGNMENT - misalign;
  top (a) = (mchunkptr) ptr;
  set_head (top (a), (((char *) h + h->size) - ptr) | PREV_INUSE);

  LIBC_PROBE (memory_arena_new, 2, a, size);
  mstate replaced_arena = thread_arena;
  thread_arena = a;
  __libc_lock_init (a->mutex);

  __libc_lock_lock (list_lock);

  /* Add the new arena to the global list.  */
  a->next = main_arena.next;
  /* FIXME: The barrier is an attempt to synchronize with read access
     in reused_arena, which does not acquire list_lock while
     traversing the list.  */
  atomic_write_barrier ();
  main_arena.next = a;

  __libc_lock_unlock (list_lock);

  __libc_lock_lock (free_list_lock);
  detach_arena (replaced_arena);
  __libc_lock_unlock (free_list_lock);

  /* Lock this arena.  NB: Another thread may have been attached to
     this arena because the arena is now accessible from the
     main_arena.next list and could have been picked by reused_arena.
     This can only happen for the last arena created (before the arena
     limit is reached).  At this point, some arena has to be attached
     to two threads.  We could acquire the arena lock before list_lock
     to make it less likely that reused_arena picks this new arena,
     but this could result in a deadlock with
     __malloc_fork_lock_parent.  */

  __libc_lock_lock (a->mutex);

  return a;
}


/* Remove an arena from free_list.  */
static mstate
get_free_list (void)
{
  mstate replaced_arena = thread_arena;
  mstate result = free_list;
  if (result != NULL)
    {
      __libc_lock_lock (free_list_lock);
      result = free_list;
      if (result != NULL)
	{
	  free_list = result->next_free;

	  /* The arena will be attached to this thread.  */
	  assert (result->attached_threads == 0);
	  result->attached_threads = 1;

	  detach_arena (replaced_arena);
	}
      __libc_lock_unlock (free_list_lock);

      if (result != NULL)
        {
          LIBC_PROBE (memory_arena_reuse_free_list, 1, result);
          __libc_lock_lock (result->mutex);
	  thread_arena = result;
        }
    }

  return result;
}

/* Remove the arena from the free list (if it is present).
   free_list_lock must have been acquired by the caller.  */
static void
remove_from_free_list (mstate arena)
{
  mstate *previous = &free_list;
  for (mstate p = free_list; p != NULL; p = p->next_free)
    {
      assert (p->attached_threads == 0);
      if (p == arena)
	{
	  /* Remove the requested arena from the list.  */
	  *previous = p->next_free;
	  break;
	}
      else
	previous = &p->next_free;
    }
}

/* Lock and return an arena that can be reused for memory allocation.
   Avoid AVOID_ARENA as we have already failed to allocate memory in
   it and it is currently locked.  */
static mstate
reused_arena (mstate avoid_arena)
{
  mstate result;
  /* FIXME: Access to next_to_use suffers from data races.  */
  static mstate next_to_use;
  if (next_to_use == NULL)
    next_to_use = &main_arena;

  /* Iterate over all arenas (including those linked from
     free_list).  */
  result = next_to_use;
  do
    {
      if (!__libc_lock_trylock (result->mutex))
        goto out;

      /* FIXME: This is a data race, see _int_new_arena.  */
      result = result->next;
    }
  while (result != next_to_use);

  /* Avoid AVOID_ARENA as we have already failed to allocate memory
     in that arena and it is currently locked.   */
  if (result == avoid_arena)
    result = result->next;

  /* No arena available without contention.  Wait for the next in line.  */
  LIBC_PROBE (memory_arena_reuse_wait, 3, &result->mutex, result, avoid_arena);
  __libc_lock_lock (result->mutex);

out:
  /* Attach the arena to the current thread.  */
  {
    /* Update the arena thread attachment counters.   */
    mstate replaced_arena = thread_arena;
    __libc_lock_lock (free_list_lock);
    detach_arena (replaced_arena);

    /* We may have picked up an arena on the free list.  We need to
       preserve the invariant that no arena on the free list has a
       positive attached_threads counter (otherwise,
       arena_thread_freeres cannot use the counter to determine if the
       arena needs to be put on the free list).  We unconditionally
       remove the selected arena from the free list.  The caller of
       reused_arena checked the free list and observed it to be empty,
       so the list is very short.  */
    remove_from_free_list (result);

    ++result->attached_threads;

    __libc_lock_unlock (free_list_lock);
  }

  LIBC_PROBE (memory_arena_reuse, 2, result, avoid_arena);
  thread_arena = result;
  next_to_use = result->next;

  return result;
}

static mstate
internal_function
arena_get2 (size_t size, mstate avoid_arena)
{
  mstate a;

  static size_t narenas_limit;

  a = get_free_list ();
  if (a == NULL)
    {
      /* Nothing immediately available, so generate a new arena.  */
      if (narenas_limit == 0)
        {
          if (mp_.arena_max != 0)
            narenas_limit = mp_.arena_max;
          else if (narenas > mp_.arena_test)
            {
              int n = __get_nprocs ();

              if (n >= 1)
                narenas_limit = NARENAS_FROM_NCORES (n);
              else
                /* We have no information about the system.  Assume two
                   cores.  */
                narenas_limit = NARENAS_FROM_NCORES (2);
            }
        }
    repeat:;
      size_t n = narenas;
      /* NB: the following depends on the fact that (size_t)0 - 1 is a
         very large number and that the underflow is OK.  If arena_max
         is set the value of arena_test is irrelevant.  If arena_test
         is set but narenas is not yet larger or equal to arena_test
         narenas_limit is 0.  There is no possibility for narenas to
         be too big for the test to always fail since there is not
         enough address space to create that many arenas.  */
      if (__glibc_unlikely (n <= narenas_limit - 1))
        {
          if (catomic_compare_and_exchange_bool_acq (&narenas, n + 1, n))
            goto repeat;
          a = _int_new_arena (size);
	  if (__glibc_unlikely (a == NULL))
            catomic_decrement (&narenas);
        }
      else
        a = reused_arena (avoid_arena);
    }
  return a;
}

/* If we don't have the main arena, then maybe the failure is due to running
   out of mmapped areas, so we can try allocating on the main arena.
   Otherwise, it is likely that sbrk() has failed and there is still a chance
   to mmap(), so try one of the other arenas.  */
static mstate
arena_get_retry (mstate ar_ptr, size_t bytes)
{
  LIBC_PROBE (memory_arena_retry, 2, bytes, ar_ptr);
  if (ar_ptr != &main_arena)
    {
      __libc_lock_unlock (ar_ptr->mutex);
      ar_ptr = &main_arena;
      __libc_lock_lock (ar_ptr->mutex);
    }
  else
    {
      __libc_lock_unlock (ar_ptr->mutex);
      ar_ptr = arena_get2 (bytes, ar_ptr);
    }

  return ar_ptr;
}

static void __attribute__ ((section ("__libc_thread_freeres_fn")))
arena_thread_freeres (void)
{
  mstate a = thread_arena;
  thread_arena = NULL;

  if (a != NULL)
    {
      __libc_lock_lock (free_list_lock);
      /* If this was the last attached thread for this arena, put the
	 arena on the free list.  */
      assert (a->attached_threads > 0);
      if (--a->attached_threads == 0)
	{
	  a->next_free = free_list;
	  free_list = a;
	}
      __libc_lock_unlock (free_list_lock);
    }
}
text_set_element (__libc_thread_subfreeres, arena_thread_freeres);

/*
 * Local variables:
 * c-basic-offset: 2
 * End:
 */