summaryrefslogtreecommitdiff
path: root/src/tests/tcmalloc_unittest.cc
blob: 236bab6fda18791a095fc0d1503214cd4d6462a7 (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
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
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*-
// Copyright (c) 2005, Google Inc.
// All rights reserved.
//
// Redistribution and use in source and binary forms, with or without
// modification, are permitted provided that the following conditions are
// met:
//
//     * Redistributions of source code must retain the above copyright
// notice, this list of conditions and the following disclaimer.
//     * Redistributions in binary form must reproduce the above
// copyright notice, this list of conditions and the following disclaimer
// in the documentation and/or other materials provided with the
// distribution.
//     * Neither the name of Google Inc. nor the names of its
// contributors may be used to endorse or promote products derived from
// this software without specific prior written permission.
//
// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

// ---
// Unittest for the TCMalloc implementation.
//
// * The test consists of a set of threads.
// * Each thread maintains a set of allocated objects, with
//   a bound on the total amount of data in the set.
// * Each allocated object's contents are generated by
//   hashing the object pointer, and a generation count
//   in the object.  This allows us to easily check for
//   data corruption.
// * At any given step, the thread can do any of the following:
//     a. Allocate an object
//     b. Increment an object's generation count and update
//        its contents.
//     c. Pass the object to another thread
//     d. Free an object
//   Also, at the end of every step, object(s) are freed to maintain
//   the memory upper-bound.
//
// If this test is compiled with -DDEBUGALLOCATION, then we don't
// run some tests that test the inner workings of tcmalloc and
// break on debugallocation: that certain allocations are aligned
// in a certain way (even though no standard requires it), and that
// realloc() tries to minimize copying (which debug allocators don't
// care about).

#include "config_for_unittests.h"
// Complicated ordering requirements.  tcmalloc.h defines (indirectly)
// _POSIX_C_SOURCE, which it needs so stdlib.h defines posix_memalign.
// unistd.h, on the other hand, requires _POSIX_C_SOURCE to be unset,
// at least on FreeBSD, in order to define sbrk.  The solution
// is to #include unistd.h first.  This is safe because unistd.h
// doesn't sub-include stdlib.h, so we'll still get posix_memalign
// when we #include stdlib.h.  Blah.
#ifdef HAVE_UNISTD_H
#include <unistd.h>        // for testing sbrk hooks
#endif
#include "tcmalloc.h"      // must come early, to pick up posix_memalign
#include <stdlib.h>
#include <string.h>
#include <stdio.h>
#if defined HAVE_STDINT_H
#include <stdint.h>        // for intptr_t
#endif
#include <sys/types.h>     // for size_t
#ifdef HAVE_FCNTL_H
#include <fcntl.h>         // for open; used with mmap-hook test
#endif
#ifdef HAVE_MMAP
#include <sys/mman.h>      // for testing mmap hooks
#endif
#ifdef HAVE_MALLOC_H
#include <malloc.h>        // defines pvalloc/etc on cygwin
#endif
#include <assert.h>
#include <vector>
#include <algorithm>
#include <string>
#include <new>
#include "base/logging.h"
#include "base/simple_mutex.h"
#include "gperftools/malloc_hook.h"
#include "gperftools/malloc_extension.h"
#include "gperftools/tcmalloc.h"
#include "thread_cache.h"
#include "system-alloc.h"
#include "tests/testutil.h"

// Windows doesn't define pvalloc and a few other obsolete unix
// functions; nor does it define posix_memalign (which is not obsolete).
#if defined(_WIN32)
# define cfree free         // don't bother to try to test these obsolete fns
# define valloc malloc
# define pvalloc malloc
// I'd like to map posix_memalign to _aligned_malloc, but _aligned_malloc
// must be paired with _aligned_free (not normal free), which is too
// invasive a change to how we allocate memory here.  So just bail
static bool kOSSupportsMemalign = false;
static inline void* Memalign(size_t align, size_t size) {
  //LOG(FATAL) << "memalign not supported on windows";
  exit(1);
  return NULL;
}
static inline int PosixMemalign(void** ptr, size_t align, size_t size) {
  //LOG(FATAL) << "posix_memalign not supported on windows";
  exit(1);
  return -1;
}

// OS X defines posix_memalign in some OS versions but not others;
// it's confusing enough to check that it's easiest to just not to test.
#elif defined(__APPLE__)
static bool kOSSupportsMemalign = false;
static inline void* Memalign(size_t align, size_t size) {
  //LOG(FATAL) << "memalign not supported on OS X";
  exit(1);
  return NULL;
}
static inline int PosixMemalign(void** ptr, size_t align, size_t size) {
  //LOG(FATAL) << "posix_memalign not supported on OS X";
  exit(1);
  return -1;
}

#else
static bool kOSSupportsMemalign = true;
static inline void* Memalign(size_t align, size_t size) {
  return memalign(align, size);
}
static inline int PosixMemalign(void** ptr, size_t align, size_t size) {
  return posix_memalign(ptr, align, size);
}

#endif

// On systems (like freebsd) that don't define MAP_ANONYMOUS, use the old
// form of the name instead.
#ifndef MAP_ANONYMOUS
# define MAP_ANONYMOUS MAP_ANON
#endif

#define LOGSTREAM   stdout

using std::vector;
using std::string;

DECLARE_double(tcmalloc_release_rate);
DECLARE_int32(max_free_queue_size);     // in debugallocation.cc
DECLARE_int64(tcmalloc_sample_parameter);

namespace testing {

static const int FLAGS_numtests = 50000;
static const int FLAGS_log_every_n_tests = 50000; // log exactly once

// Testing parameters
static const int FLAGS_lgmaxsize = 16;   // lg() of the max size object to alloc
static const int FLAGS_numthreads = 10;  // Number of threads
static const int FLAGS_threadmb = 4;     // Max memory size allocated by thread
static const int FLAGS_lg_max_memalign = 18; // lg of max alignment for memalign

static const double FLAGS_memalign_min_fraction = 0;    // min expected%
static const double FLAGS_memalign_max_fraction = 0.4;  // max expected%
static const double FLAGS_memalign_max_alignment_ratio = 6;  // alignment/size

// Weights of different operations
static const int FLAGS_allocweight = 50;    // Weight for picking allocation
static const int FLAGS_freeweight = 50;     // Weight for picking free
static const int FLAGS_updateweight = 10;   // Weight for picking update
static const int FLAGS_passweight = 1;      // Weight for passing object

static const int kSizeBits = 8 * sizeof(size_t);
static const size_t kMaxSize = ~static_cast<size_t>(0);
static const size_t kMaxSignedSize = ((size_t(1) << (kSizeBits-1)) - 1);

static const size_t kNotTooBig = 100000;
// We want an allocation that is definitely more than main memory.  OS
// X has special logic to discard very big allocs before even passing
// the request along to the user-defined memory allocator; we're not
// interested in testing their logic, so we have to make sure we're
// not *too* big.
static const size_t kTooBig = kMaxSize - 100000;

static int news_handled = 0;

// Global array of threads
class TesterThread;
static TesterThread** threads;

// To help with generating random numbers
class TestHarness {
 private:
  // Information kept per type
  struct Type {
    string      name;
    int         type;
    int         weight;
  };

 public:
  TestHarness(int seed)
      : types_(new vector<Type>), total_weight_(0), num_tests_(0) {
    srandom(seed);
  }
  ~TestHarness() {
    delete types_;
  }

  // Add operation type with specified weight.  When starting a new
  // iteration, an operation type is picked with probability
  // proportional to its weight.
  //
  // "type" must be non-negative.
  // "weight" must be non-negative.
  void AddType(int type, int weight, const char* name);

  // Call this to get the type of operation for the next iteration.
  // It returns a random operation type from the set of registered
  // operations.  Returns -1 if tests should finish.
  int PickType();

  // If n == 0, returns the next pseudo-random number in the range [0 .. 0]
  // If n != 0, returns the next pseudo-random number in the range [0 .. n)
  int Uniform(int n) {
    if (n == 0) {
      return random() * 0;
    } else {
      return random() % n;
    }
  }
  // Pick "base" uniformly from range [0,max_log] and then return
  // "base" random bits.  The effect is to pick a number in the range
  // [0,2^max_log-1] with bias towards smaller numbers.
  int Skewed(int max_log) {
    const int base = random() % (max_log+1);
    return random() % (1 << base);
  }

 private:
  vector<Type>*         types_;         // Registered types
  int                   total_weight_;  // Total weight of all types
  int                   num_tests_;     // Num tests run so far
};

void TestHarness::AddType(int type, int weight, const char* name) {
  Type t;
  t.name = name;
  t.type = type;
  t.weight = weight;
  types_->push_back(t);
  total_weight_ += weight;
}

int TestHarness::PickType() {
  if (num_tests_ >= FLAGS_numtests) return -1;
  num_tests_++;

  assert(total_weight_ > 0);
  // This is a little skewed if total_weight_ doesn't divide 2^31, but it's close
  int v = Uniform(total_weight_);
  int i;
  for (i = 0; i < types_->size(); i++) {
    v -= (*types_)[i].weight;
    if (v < 0) {
      break;
    }
  }

  assert(i < types_->size());
  if ((num_tests_ % FLAGS_log_every_n_tests) == 0) {
    fprintf(LOGSTREAM, "  Test %d out of %d: %s\n",
            num_tests_, FLAGS_numtests, (*types_)[i].name.c_str());
  }
  return (*types_)[i].type;
}

class AllocatorState : public TestHarness {
 public:
  explicit AllocatorState(int seed) : TestHarness(seed), memalign_fraction_(0) {
    if (kOSSupportsMemalign) {
      CHECK_GE(FLAGS_memalign_max_fraction, 0);
      CHECK_LE(FLAGS_memalign_max_fraction, 1);
      CHECK_GE(FLAGS_memalign_min_fraction, 0);
      CHECK_LE(FLAGS_memalign_min_fraction, 1);
      double delta = FLAGS_memalign_max_fraction - FLAGS_memalign_min_fraction;
      CHECK_GE(delta, 0);
      memalign_fraction_ = (Uniform(10000)/10000.0 * delta +
                            FLAGS_memalign_min_fraction);
      //fprintf(LOGSTREAM, "memalign fraction: %f\n", memalign_fraction_);
    }
  }
  virtual ~AllocatorState() {}

  // Allocate memory.  Randomly choose between malloc() or posix_memalign().
  void* alloc(size_t size) {
    if (Uniform(100) < memalign_fraction_ * 100) {
      // Try a few times to find a reasonable alignment, or fall back on malloc.
      for (int i = 0; i < 5; i++) {
        size_t alignment = 1 << Uniform(FLAGS_lg_max_memalign);
        if (alignment >= sizeof(intptr_t) &&
            (size < sizeof(intptr_t) ||
             alignment < FLAGS_memalign_max_alignment_ratio * size)) {
          void *result = reinterpret_cast<void*>(static_cast<intptr_t>(0x1234));
          int err = PosixMemalign(&result, alignment, size);
          if (err != 0) {
            CHECK_EQ(err, ENOMEM);
          }
          return err == 0 ? result : NULL;
        }
      }
    }
    return malloc(size);
  }

 private:
  double memalign_fraction_;
};


// Info kept per thread
class TesterThread {
 private:
  // Info kept per allocated object
  struct Object {
    char*       ptr;                    // Allocated pointer
    int         size;                   // Allocated size
    int         generation;             // Generation counter of object contents
  };

  Mutex                 lock_;          // For passing in another thread's obj
  int                   id_;            // My thread id
  AllocatorState        rnd_;           // For generating random numbers
  vector<Object>        heap_;          // This thread's heap
  vector<Object>        passed_;        // Pending objects passed from others
  size_t                heap_size_;     // Current heap size
  int                   locks_ok_;      // Number of OK TryLock() ops
  int                   locks_failed_;  // Number of failed TryLock() ops

  // Type of operations
  enum Type { ALLOC, FREE, UPDATE, PASS };

  // ACM minimal standard random number generator.  (re-entrant.)
  class ACMRandom {
    int32 seed_;
   public:
    explicit ACMRandom(int32 seed) { seed_ = seed; }
    int32 Next() {
      const int32 M = 2147483647L;   // 2^31-1
      const int32 A = 16807;
      // In effect, we are computing seed_ = (seed_ * A) % M, where M = 2^31-1
      uint32 lo = A * (int32)(seed_ & 0xFFFF);
      uint32 hi = A * (int32)((uint32)seed_ >> 16);
      lo += (hi & 0x7FFF) << 16;
      if (lo > M) {
        lo &= M;
        ++lo;
      }
      lo += hi >> 15;
      if (lo > M) {
        lo &= M;
        ++lo;
      }
      return (seed_ = (int32) lo);
    }
  };

 public:
  TesterThread(int id)
    : id_(id),
      rnd_(id+1),
      heap_size_(0),
      locks_ok_(0),
      locks_failed_(0) {
  }

  virtual ~TesterThread() {
    if (FLAGS_verbose)
      fprintf(LOGSTREAM, "Thread %2d: locks %6d ok; %6d trylocks failed\n",
              id_, locks_ok_, locks_failed_);
    if (locks_ok_ + locks_failed_ >= 1000) {
      CHECK_LE(locks_failed_, locks_ok_ / 2);
    }
  }

  virtual void Run() {
    rnd_.AddType(ALLOC,  FLAGS_allocweight,   "allocate");
    rnd_.AddType(FREE,   FLAGS_freeweight,    "free");
    rnd_.AddType(UPDATE, FLAGS_updateweight,  "update");
    rnd_.AddType(PASS,   FLAGS_passweight,    "pass");

    while (true) {
      AcquirePassedObjects();

      switch (rnd_.PickType()) {
        case ALLOC:   AllocateObject(); break;
        case FREE:    FreeObject();     break;
        case UPDATE:  UpdateObject();   break;
        case PASS:    PassObject();     break;
        case -1:      goto done;
        default:      assert(NULL == "Unknown type");
      }

      ShrinkHeap();
    }

 done:
    DeleteHeap();
  }

  // Allocate a new object
  void AllocateObject() {
    Object object;
    object.size = rnd_.Skewed(FLAGS_lgmaxsize);
    object.ptr = static_cast<char*>(rnd_.alloc(object.size));
    CHECK(object.ptr);
    object.generation = 0;
    FillContents(&object);
    heap_.push_back(object);
    heap_size_ += object.size;
  }

  // Mutate a random object
  void UpdateObject() {
    if (heap_.empty()) return;
    const int index = rnd_.Uniform(heap_.size());
    CheckContents(heap_[index]);
    heap_[index].generation++;
    FillContents(&heap_[index]);
  }

  // Free a random object
  void FreeObject() {
    if (heap_.empty()) return;
    const int index = rnd_.Uniform(heap_.size());
    Object object = heap_[index];
    CheckContents(object);
    free(object.ptr);
    heap_size_ -= object.size;
    heap_[index] = heap_[heap_.size()-1];
    heap_.pop_back();
  }

  // Delete all objects in the heap
  void DeleteHeap() {
    while (!heap_.empty()) {
      FreeObject();
    }
  }

  // Free objects until our heap is small enough
  void ShrinkHeap() {
    while (heap_size_ > FLAGS_threadmb << 20) {
      assert(!heap_.empty());
      FreeObject();
    }
  }

  // Pass a random object to another thread
  void PassObject() {
    // Pick object to pass
    if (heap_.empty()) return;
    const int index = rnd_.Uniform(heap_.size());
    Object object = heap_[index];
    CheckContents(object);

    // Pick thread to pass
    const int tid = rnd_.Uniform(FLAGS_numthreads);
    TesterThread* thread = threads[tid];

    if (thread->lock_.TryLock()) {
      // Pass the object
      locks_ok_++;
      thread->passed_.push_back(object);
      thread->lock_.Unlock();
      heap_size_ -= object.size;
      heap_[index] = heap_[heap_.size()-1];
      heap_.pop_back();
    } else {
      locks_failed_++;
    }
  }

  // Grab any objects passed to this thread by another thread
  void AcquirePassedObjects() {
    // We do not create unnecessary contention by always using
    // TryLock().  Plus we unlock immediately after swapping passed
    // objects into a local vector.
    vector<Object> copy;
    { // Locking scope
      if (!lock_.TryLock()) {
        locks_failed_++;
        return;
      }
      locks_ok_++;
      swap(copy, passed_);
      lock_.Unlock();
    }

    for (int i = 0; i < copy.size(); ++i) {
      const Object& object = copy[i];
      CheckContents(object);
      heap_.push_back(object);
      heap_size_ += object.size;
    }
  }

  // Fill object contents according to ptr/generation
  void FillContents(Object* object) {
    ACMRandom r(reinterpret_cast<intptr_t>(object->ptr) & 0x7fffffff);
    for (int i = 0; i < object->generation; ++i) {
      r.Next();
    }
    const char c = static_cast<char>(r.Next());
    memset(object->ptr, c, object->size);
  }

  // Check object contents
  void CheckContents(const Object& object) {
    ACMRandom r(reinterpret_cast<intptr_t>(object.ptr) & 0x7fffffff);
    for (int i = 0; i < object.generation; ++i) {
      r.Next();
    }

    // For large objects, we just check a prefix/suffix
    const char expected = static_cast<char>(r.Next());
    const int limit1 = object.size < 32 ? object.size : 32;
    const int start2 = limit1 > object.size - 32 ? limit1 : object.size - 32;
    for (int i = 0; i < limit1; ++i) {
      CHECK_EQ(object.ptr[i], expected);
    }
    for (int i = start2; i < object.size; ++i) {
      CHECK_EQ(object.ptr[i], expected);
    }
  }
};

static void RunThread(int thread_id) {
  threads[thread_id]->Run();
}

static void TryHugeAllocation(size_t s, AllocatorState* rnd) {
  void* p = rnd->alloc(s);
  CHECK(p == NULL);   // huge allocation s should fail!
}

static void TestHugeAllocations(AllocatorState* rnd) {
  // Check that asking for stuff tiny bit smaller than largest possible
  // size returns NULL.
  for (size_t i = 0; i < 70000; i += rnd->Uniform(20)) {
    TryHugeAllocation(kMaxSize - i, rnd);
  }
  // Asking for memory sizes near signed/unsigned boundary (kMaxSignedSize)
  // might work or not, depending on the amount of virtual memory.
#ifndef DEBUGALLOCATION    // debug allocation takes forever for huge allocs
  for (size_t i = 0; i < 100; i++) {
    void* p = NULL;
    p = rnd->alloc(kMaxSignedSize + i);
    if (p) free(p);    // if: free(NULL) is not necessarily defined
    p = rnd->alloc(kMaxSignedSize - i);
    if (p) free(p);
  }
#endif

  // Check that ReleaseFreeMemory has no visible effect (aka, does not
  // crash the test):
  MallocExtension* inst = MallocExtension::instance();
  CHECK(inst);
  inst->ReleaseFreeMemory();
}

static void TestCalloc(size_t n, size_t s, bool ok) {
  char* p = reinterpret_cast<char*>(calloc(n, s));
  if (FLAGS_verbose)
    fprintf(LOGSTREAM, "calloc(%" PRIxS ", %" PRIxS "): %p\n", n, s, p);
  if (!ok) {
    CHECK(p == NULL);  // calloc(n, s) should not succeed
  } else {
    CHECK(p != NULL);  // calloc(n, s) should succeed
    for (int i = 0; i < n*s; i++) {
      CHECK(p[i] == '\0');
    }
    free(p);
  }
}

// This makes sure that reallocing a small number of bytes in either
// direction doesn't cause us to allocate new memory.
static void TestRealloc() {
#ifndef DEBUGALLOCATION  // debug alloc doesn't try to minimize reallocs
  // When sampling, we always allocate in units of page-size, which
  // makes reallocs of small sizes do extra work (thus, failing these
  // checks).  Since sampling is random, we turn off sampling to make
  // sure that doesn't happen to us here.
  const int64 old_sample_parameter = FLAGS_tcmalloc_sample_parameter;
  FLAGS_tcmalloc_sample_parameter = 0;   // turn off sampling

  int start_sizes[] = { 100, 1000, 10000, 100000 };
  int deltas[] = { 1, -2, 4, -8, 16, -32, 64, -128 };

  for (int s = 0; s < sizeof(start_sizes)/sizeof(*start_sizes); ++s) {
    void* p = malloc(start_sizes[s]);
    CHECK(p);
    // The larger the start-size, the larger the non-reallocing delta.
    for (int d = 0; d < (s+1) * 2; ++d) {
      void* new_p = realloc(p, start_sizes[s] + deltas[d]);
      CHECK(p == new_p);  // realloc should not allocate new memory
    }
    // Test again, but this time reallocing smaller first.
    for (int d = 0; d < s*2; ++d) {
      void* new_p = realloc(p, start_sizes[s] - deltas[d]);
      CHECK(p == new_p);  // realloc should not allocate new memory
    }
    free(p);
  }
  FLAGS_tcmalloc_sample_parameter = old_sample_parameter;
#endif
}

static void TestNewHandler() throw (std::bad_alloc) {
  ++news_handled;
  throw std::bad_alloc();
}

static void TestOneNew(void* (*func)(size_t)) {
  // success test
  try {
    void* ptr = (*func)(kNotTooBig);
    if (0 == ptr) {
      fprintf(LOGSTREAM, "allocation should not have failed.\n");
      abort();
    }
  } catch (...) {
    fprintf(LOGSTREAM, "allocation threw unexpected exception.\n");
    abort();
  }

  // failure test
  // we should always receive a bad_alloc exception
  try {
    (*func)(kTooBig);
    fprintf(LOGSTREAM, "allocation should have failed.\n");
    abort();
  } catch (const std::bad_alloc&) {
    // correct
  } catch (...) {
    fprintf(LOGSTREAM, "allocation threw unexpected exception.\n");
    abort();
  }
}

static void TestNew(void* (*func)(size_t)) {
  news_handled = 0;

  // test without new_handler:
  std::new_handler saved_handler = std::set_new_handler(0);
  TestOneNew(func);

  // test with new_handler:
  std::set_new_handler(TestNewHandler);
  TestOneNew(func);
  if (news_handled != 1) {
    fprintf(LOGSTREAM, "new_handler was not called.\n");
    abort();
  }
  std::set_new_handler(saved_handler);
}

static void TestOneNothrowNew(void* (*func)(size_t, const std::nothrow_t&)) {
  // success test
  try {
    void* ptr = (*func)(kNotTooBig, std::nothrow);
    if (0 == ptr) {
      fprintf(LOGSTREAM, "allocation should not have failed.\n");
      abort();
    }
  } catch (...) {
    fprintf(LOGSTREAM, "allocation threw unexpected exception.\n");
    abort();
  }

  // failure test
  // we should always receive a bad_alloc exception
  try {
    if ((*func)(kTooBig, std::nothrow) != 0) {
      fprintf(LOGSTREAM, "allocation should have failed.\n");
      abort();
    }
  } catch (...) {
    fprintf(LOGSTREAM, "nothrow allocation threw unexpected exception.\n");
    abort();
  }
}

static void TestNothrowNew(void* (*func)(size_t, const std::nothrow_t&)) {
  news_handled = 0;

  // test without new_handler:
  std::new_handler saved_handler = std::set_new_handler(0);
  TestOneNothrowNew(func);

  // test with new_handler:
  std::set_new_handler(TestNewHandler);
  TestOneNothrowNew(func);
  if (news_handled != 1) {
    fprintf(LOGSTREAM, "nothrow new_handler was not called.\n");
    abort();
  }
  std::set_new_handler(saved_handler);
}


// These are used as callbacks by the sanity-check.  Set* and Reset*
// register the hook that counts how many times the associated memory
// function is called.  After each such call, call Verify* to verify
// that we used the tcmalloc version of the call, and not the libc.
// Note the ... in the hook signature: we don't care what arguments
// the hook takes.
#define MAKE_HOOK_CALLBACK(hook_type)                                   \
  static volatile int g_##hook_type##_calls = 0;                                 \
  static void IncrementCallsTo##hook_type(...) {                        \
    g_##hook_type##_calls++;                                            \
  }                                                                     \
  static void Verify##hook_type##WasCalled() {                          \
    CHECK_GT(g_##hook_type##_calls, 0);                                 \
    g_##hook_type##_calls = 0;  /* reset for next call */               \
  }                                                                     \
  static void Set##hook_type() {                                        \
    CHECK(MallocHook::Add##hook_type(                                   \
        (MallocHook::hook_type)&IncrementCallsTo##hook_type));          \
  }                                                                     \
  static void Reset##hook_type() {                                      \
    CHECK(MallocHook::Remove##hook_type(                                \
        (MallocHook::hook_type)&IncrementCallsTo##hook_type));          \
  }

// We do one for each hook typedef in malloc_hook.h
MAKE_HOOK_CALLBACK(NewHook);
MAKE_HOOK_CALLBACK(DeleteHook);
MAKE_HOOK_CALLBACK(MmapHook);
MAKE_HOOK_CALLBACK(MremapHook);
MAKE_HOOK_CALLBACK(MunmapHook);
MAKE_HOOK_CALLBACK(SbrkHook);

static void TestAlignmentForSize(int size) {
  fprintf(LOGSTREAM, "Testing alignment of malloc(%d)\n", size);
  static const int kNum = 100;
  void* ptrs[kNum];
  for (int i = 0; i < kNum; i++) {
    ptrs[i] = malloc(size);
    uintptr_t p = reinterpret_cast<uintptr_t>(ptrs[i]);
    CHECK((p % sizeof(void*)) == 0);
    CHECK((p % sizeof(double)) == 0);

    // Must have 16-byte (or 8-byte in case of -DTCMALLOC_ALIGN_8BYTES)
    // alignment for large enough objects
    if (size >= kMinAlign) {
      CHECK((p % kMinAlign) == 0);
    }
  }
  for (int i = 0; i < kNum; i++) {
    free(ptrs[i]);
  }
}

static void TestMallocAlignment() {
  for (int lg = 0; lg < 16; lg++) {
    TestAlignmentForSize((1<<lg) - 1);
    TestAlignmentForSize((1<<lg) + 0);
    TestAlignmentForSize((1<<lg) + 1);
  }
}

static void TestHugeThreadCache() {
  fprintf(LOGSTREAM, "==== Testing huge thread cache\n");
  // More than 2^16 to cause integer overflow of 16 bit counters.
  static const int kNum = 70000;
  char** array = new char*[kNum];
  for (int i = 0; i < kNum; ++i) {
    array[i] = new char[10];
  }
  for (int i = 0; i < kNum; ++i) {
    delete[] array[i];
  }
  delete[] array;
}

namespace {

struct RangeCallbackState {
  uintptr_t ptr;
  base::MallocRange::Type expected_type;
  size_t min_size;
  bool matched;
};

static void RangeCallback(void* arg, const base::MallocRange* r) {
  RangeCallbackState* state = reinterpret_cast<RangeCallbackState*>(arg);
  if (state->ptr >= r->address &&
      state->ptr < r->address + r->length) {
    if (state->expected_type == base::MallocRange::FREE) {
      // We are expecting r->type == FREE, but ReleaseMemory
      // may have already moved us to UNMAPPED state instead (this happens in
      // approximately 0.1% of executions). Accept either state.
      CHECK(r->type == base::MallocRange::FREE ||
            r->type == base::MallocRange::UNMAPPED);
    } else {
      CHECK_EQ(r->type, state->expected_type);
    }
    CHECK_GE(r->length, state->min_size);
    state->matched = true;
  }
}

// Check that at least one of the callbacks from Ranges() contains
// the specified address with the specified type, and has size
// >= min_size.
static void CheckRangeCallback(void* ptr, base::MallocRange::Type type,
                               size_t min_size) {
  RangeCallbackState state;
  state.ptr = reinterpret_cast<uintptr_t>(ptr);
  state.expected_type = type;
  state.min_size = min_size;
  state.matched = false;
  MallocExtension::instance()->Ranges(&state, RangeCallback);
  CHECK(state.matched);
}

}

static bool HaveSystemRelease =
    TCMalloc_SystemRelease(TCMalloc_SystemAlloc(kPageSize, NULL, 0), kPageSize);

static void TestRanges() {
  static const int MB = 1048576;
  void* a = malloc(MB);
  void* b = malloc(MB);
  base::MallocRange::Type releasedType =
      HaveSystemRelease ? base::MallocRange::UNMAPPED : base::MallocRange::FREE;

  CheckRangeCallback(a, base::MallocRange::INUSE, MB);
  CheckRangeCallback(b, base::MallocRange::INUSE, MB);
  free(a);
  CheckRangeCallback(a, base::MallocRange::FREE, MB);
  CheckRangeCallback(b, base::MallocRange::INUSE, MB);
  MallocExtension::instance()->ReleaseFreeMemory();
  CheckRangeCallback(a, releasedType, MB);
  CheckRangeCallback(b, base::MallocRange::INUSE, MB);
  free(b);
  CheckRangeCallback(a, releasedType, MB);
  CheckRangeCallback(b, base::MallocRange::FREE, MB);
}

#ifndef DEBUGALLOCATION
static size_t GetUnmappedBytes() {
  size_t bytes;
  CHECK(MallocExtension::instance()->GetNumericProperty(
      "tcmalloc.pageheap_unmapped_bytes", &bytes));
  return bytes;
}
#endif

class AggressiveDecommitChanger {
  size_t old_value_;
public:
  AggressiveDecommitChanger(size_t new_value) {
    MallocExtension *inst = MallocExtension::instance();
    bool rv = inst->GetNumericProperty("tcmalloc.aggressive_memory_decommit", &old_value_);
    CHECK_CONDITION(rv);
    rv = inst->SetNumericProperty("tcmalloc.aggressive_memory_decommit", new_value);
    CHECK_CONDITION(rv);
  }
  ~AggressiveDecommitChanger() {
    MallocExtension *inst = MallocExtension::instance();
    bool rv = inst->SetNumericProperty("tcmalloc.aggressive_memory_decommit", old_value_);
    CHECK_CONDITION(rv);
  }
};

static void TestReleaseToSystem() {
  // Debug allocation mode adds overhead to each allocation which
  // messes up all the equality tests here.  I just disable the
  // teset in this mode.  TODO(csilvers): get it to work for debugalloc?
#ifndef DEBUGALLOCATION

  if(!HaveSystemRelease) return;

  const double old_tcmalloc_release_rate = FLAGS_tcmalloc_release_rate;
  FLAGS_tcmalloc_release_rate = 0;

  AggressiveDecommitChanger disabler(0);

  static const int MB = 1048576;
  void* a = malloc(MB);
  void* b = malloc(MB);
  MallocExtension::instance()->ReleaseFreeMemory();
  size_t starting_bytes = GetUnmappedBytes();

  // Calling ReleaseFreeMemory() a second time shouldn't do anything.
  MallocExtension::instance()->ReleaseFreeMemory();
  EXPECT_EQ(starting_bytes, GetUnmappedBytes());

  // ReleaseToSystem shouldn't do anything either.
  MallocExtension::instance()->ReleaseToSystem(MB);
  EXPECT_EQ(starting_bytes, GetUnmappedBytes());

  free(a);

  // The span to release should be 1MB.
  MallocExtension::instance()->ReleaseToSystem(MB/2);
  EXPECT_EQ(starting_bytes + MB, GetUnmappedBytes());

  // Should do nothing since the previous call released too much.
  MallocExtension::instance()->ReleaseToSystem(MB/4);
  EXPECT_EQ(starting_bytes + MB, GetUnmappedBytes());

  free(b);

  // Use up the extra MB/4 bytes from 'a' and also release 'b'.
  MallocExtension::instance()->ReleaseToSystem(MB/2);
  EXPECT_EQ(starting_bytes + 2*MB, GetUnmappedBytes());

  // Should do nothing since the previous call released too much.
  MallocExtension::instance()->ReleaseToSystem(MB/2);
  EXPECT_EQ(starting_bytes + 2*MB, GetUnmappedBytes());

  // Nothing else to release.
  MallocExtension::instance()->ReleaseFreeMemory();
  EXPECT_EQ(starting_bytes + 2*MB, GetUnmappedBytes());

  a = malloc(MB);
  free(a);
  EXPECT_EQ(starting_bytes + MB, GetUnmappedBytes());

  // Releasing less than a page should still trigger a release.
  MallocExtension::instance()->ReleaseToSystem(1);
  EXPECT_EQ(starting_bytes + 2*MB, GetUnmappedBytes());

  FLAGS_tcmalloc_release_rate = old_tcmalloc_release_rate;
#endif   // #ifndef DEBUGALLOCATION
}

static void TestAggressiveDecommit() {
  // Debug allocation mode adds overhead to each allocation which
  // messes up all the equality tests here.  I just disable the
  // teset in this mode.
#ifndef DEBUGALLOCATION

  if(!HaveSystemRelease) return;

  fprintf(LOGSTREAM, "Testing aggressive de-commit\n");

  AggressiveDecommitChanger enabler(1);

  static const int MB = 1048576;
  void* a = malloc(MB);
  void* b = malloc(MB);

  size_t starting_bytes = GetUnmappedBytes();

  // ReleaseToSystem shouldn't do anything either.
  MallocExtension::instance()->ReleaseToSystem(MB);
  EXPECT_EQ(starting_bytes, GetUnmappedBytes());

  free(a);

  // The span to release should be 1MB.
  EXPECT_EQ(starting_bytes + MB, GetUnmappedBytes());

  free(b);

  EXPECT_EQ(starting_bytes + 2*MB, GetUnmappedBytes());

  // Nothing else to release.
  MallocExtension::instance()->ReleaseFreeMemory();
  EXPECT_EQ(starting_bytes + 2*MB, GetUnmappedBytes());

  a = malloc(MB);
  free(a);

  EXPECT_EQ(starting_bytes + 2*MB, GetUnmappedBytes());

  fprintf(LOGSTREAM, "Done testing aggressive de-commit\n");

#endif   // #ifndef DEBUGALLOCATION
}

// On MSVC10, in release mode, the optimizer convinces itself
// g_no_memory is never changed (I guess it doesn't realize OnNoMemory
// might be called).  Work around this by setting the var volatile.
volatile bool g_no_memory = false;
std::new_handler g_old_handler = NULL;
static void OnNoMemory() {
  g_no_memory = true;
  std::set_new_handler(g_old_handler);
}

static void TestSetNewMode() {
  int old_mode = tc_set_new_mode(1);

  g_old_handler = std::set_new_handler(&OnNoMemory);
  g_no_memory = false;
  void* ret = malloc(kTooBig);
  EXPECT_EQ(NULL, ret);
  EXPECT_TRUE(g_no_memory);

  g_old_handler = std::set_new_handler(&OnNoMemory);
  g_no_memory = false;
  ret = calloc(1, kTooBig);
  EXPECT_EQ(NULL, ret);
  EXPECT_TRUE(g_no_memory);

  g_old_handler = std::set_new_handler(&OnNoMemory);
  g_no_memory = false;
  ret = realloc(NULL, kTooBig);
  EXPECT_EQ(NULL, ret);
  EXPECT_TRUE(g_no_memory);

  if (kOSSupportsMemalign) {
    // Not really important, but must be small enough such that
    // kAlignment + kTooBig does not overflow.
    const int kAlignment = 1 << 5;

    g_old_handler = std::set_new_handler(&OnNoMemory);
    g_no_memory = false;
    ret = Memalign(kAlignment, kTooBig);
    EXPECT_EQ(NULL, ret);
    EXPECT_TRUE(g_no_memory);

    g_old_handler = std::set_new_handler(&OnNoMemory);
    g_no_memory = false;
    EXPECT_EQ(ENOMEM,
              PosixMemalign(&ret, kAlignment, kTooBig));
    EXPECT_EQ(NULL, ret);
    EXPECT_TRUE(g_no_memory);
  }

  tc_set_new_mode(old_mode);
}

static int RunAllTests(int argc, char** argv) {
  // Optional argv[1] is the seed
  AllocatorState rnd(argc > 1 ? atoi(argv[1]) : 100);

  SetTestResourceLimit();

  // TODO(odo):  This test has been disabled because it is only by luck that it
  // does not result in fragmentation.  When tcmalloc makes an allocation which
  // spans previously unused leaves of the pagemap it will allocate and fill in
  // the leaves to cover the new allocation.  The leaves happen to be 256MiB in
  // the 64-bit build, and with the sbrk allocator these allocations just
  // happen to fit in one leaf by luck.  With other allocators (mmap,
  // memfs_malloc when used with small pages) the allocations generally span
  // two leaves and this results in a very bad fragmentation pattern with this
  // code.  The same failure can be forced with the sbrk allocator just by
  // allocating something on the order of 128MiB prior to starting this test so
  // that the test allocations straddle a 256MiB boundary.

  // TODO(csilvers): port MemoryUsage() over so the test can use that
#if 0
# include <unistd.h>      // for getpid()
  // Allocate and deallocate blocks of increasing sizes to check if the alloc
  // metadata fragments the memory. (Do not put other allocations/deallocations
  // before this test, it may break).
  {
    size_t memory_usage = MemoryUsage(getpid());
    fprintf(LOGSTREAM, "Testing fragmentation\n");
    for ( int i = 200; i < 240; ++i ) {
      int size = i << 20;
      void *test1 = rnd.alloc(size);
      CHECK(test1);
      for ( int j = 0; j < size; j += (1 << 12) ) {
        static_cast<char*>(test1)[j] = 1;
      }
      free(test1);
    }
    // There may still be a bit of fragmentation at the beginning, until we
    // reach kPageMapBigAllocationThreshold bytes so we check for
    // 200 + 240 + margin.
    CHECK_LT(MemoryUsage(getpid()), memory_usage + (450 << 20) );
  }
#endif

  // Check that empty allocation works
  fprintf(LOGSTREAM, "Testing empty allocation\n");
  {
    void* p1 = rnd.alloc(0);
    CHECK(p1 != NULL);
    void* p2 = rnd.alloc(0);
    CHECK(p2 != NULL);
    CHECK(p1 != p2);
    free(p1);
    free(p2);
  }

  // This code stresses some of the memory allocation via STL.
  // It may call operator delete(void*, nothrow_t).
  fprintf(LOGSTREAM, "Testing STL use\n");
  {
    std::vector<int> v;
    v.push_back(1);
    v.push_back(2);
    v.push_back(3);
    v.push_back(0);
    std::stable_sort(v.begin(), v.end());
  }

  // Test each of the memory-allocation functions once, just as a sanity-check
  fprintf(LOGSTREAM, "Sanity-testing all the memory allocation functions\n");
  {
    // We use new-hook and delete-hook to verify we actually called the
    // tcmalloc version of these routines, and not the libc version.
    SetNewHook();      // defined as part of MAKE_HOOK_CALLBACK, above
    SetDeleteHook();   // ditto

    void* p1 = malloc(10);
    CHECK(p1 != NULL);    // force use of this variable
    VerifyNewHookWasCalled();
    // Also test the non-standard tc_malloc_size
    size_t actual_p1_size = tc_malloc_size(p1);
    CHECK_GE(actual_p1_size, 10);
    CHECK_LT(actual_p1_size, 100000);   // a reasonable upper-bound, I think
    free(p1);
    VerifyDeleteHookWasCalled();


    p1 = calloc(10, 2);
    CHECK(p1 != NULL);
    VerifyNewHookWasCalled();
    // We make sure we realloc to a big size, since some systems (OS
    // X) will notice if the realloced size continues to fit into the
    // malloc-block and make this a noop if so.
    p1 = realloc(p1, 30000);
    CHECK(p1 != NULL);
    VerifyNewHookWasCalled();
    VerifyDeleteHookWasCalled();
    cfree(p1);  // synonym for free
    VerifyDeleteHookWasCalled();

    if (kOSSupportsMemalign) {
      CHECK_EQ(PosixMemalign(&p1, sizeof(p1), 40), 0);
      CHECK(p1 != NULL);
      VerifyNewHookWasCalled();
      free(p1);
      VerifyDeleteHookWasCalled();

      p1 = Memalign(sizeof(p1) * 2, 50);
      CHECK(p1 != NULL);
      VerifyNewHookWasCalled();
      free(p1);
      VerifyDeleteHookWasCalled();
    }

    // Windows has _aligned_malloc.  Let's test that that's captured too.
#if (defined(_MSC_VER) || defined(__MINGW32__)) && !defined(PERFTOOLS_NO_ALIGNED_MALLOC)
    p1 = _aligned_malloc(sizeof(p1) * 2, 64);
    CHECK(p1 != NULL);
    VerifyNewHookWasCalled();
    _aligned_free(p1);
    VerifyDeleteHookWasCalled();
#endif

    p1 = valloc(60);
    CHECK(p1 != NULL);
    VerifyNewHookWasCalled();
    free(p1);
    VerifyDeleteHookWasCalled();

    p1 = pvalloc(70);
    CHECK(p1 != NULL);
    VerifyNewHookWasCalled();
    free(p1);
    VerifyDeleteHookWasCalled();

    char* p2 = new char;
    CHECK(p2 != NULL);
    VerifyNewHookWasCalled();
    delete p2;
    VerifyDeleteHookWasCalled();

    p2 = new char[100];
    CHECK(p2 != NULL);
    VerifyNewHookWasCalled();
    delete[] p2;
    VerifyDeleteHookWasCalled();

    p2 = new(std::nothrow) char;
    CHECK(p2 != NULL);
    VerifyNewHookWasCalled();
    delete p2;
    VerifyDeleteHookWasCalled();

    p2 = new(std::nothrow) char[100];
    CHECK(p2 != NULL);
    VerifyNewHookWasCalled();
    delete[] p2;
    VerifyDeleteHookWasCalled();

    // Another way of calling operator new
    p2 = static_cast<char*>(::operator new(100));
    CHECK(p2 != NULL);
    VerifyNewHookWasCalled();
    ::operator delete(p2);
    VerifyDeleteHookWasCalled();

    // Try to call nothrow's delete too.  Compilers use this.
    p2 = static_cast<char*>(::operator new(100, std::nothrow));
    CHECK(p2 != NULL);
    VerifyNewHookWasCalled();
    ::operator delete(p2, std::nothrow);
    VerifyDeleteHookWasCalled();

    // Try strdup(), which the system allocates but we must free.  If
    // all goes well, libc will use our malloc!
    p2 = strdup("test");
    CHECK(p2 != NULL);
    VerifyNewHookWasCalled();
    free(p2);
    VerifyDeleteHookWasCalled();


    // Test mmap too: both anonymous mmap and mmap of a file
    // Note that for right now we only override mmap on linux
    // systems, so those are the only ones for which we check.
    SetMmapHook();
    SetMremapHook();
    SetMunmapHook();
#if defined(HAVE_MMAP) && defined(__linux) && \
       (defined(__i386__) || defined(__x86_64__))
    int size = 8192*2;
    p1 = mmap(NULL, size, PROT_WRITE|PROT_READ, MAP_ANONYMOUS|MAP_PRIVATE,
              -1, 0);
    CHECK(p1 != NULL);
    VerifyMmapHookWasCalled();
    p1 = mremap(p1, size, size/2, 0);
    CHECK(p1 != NULL);
    VerifyMremapHookWasCalled();
    size /= 2;
    munmap(p1, size);
    VerifyMunmapHookWasCalled();

    int fd = open("/dev/zero", O_RDONLY);
    CHECK_GE(fd, 0);   // make sure the open succeeded
    p1 = mmap(NULL, 8192, PROT_READ, MAP_SHARED, fd, 0);
    CHECK(p1 != NULL);
    VerifyMmapHookWasCalled();
    munmap(p1, 8192);
    VerifyMunmapHookWasCalled();
    close(fd);
#else   // this is just to quiet the compiler: make sure all fns are called
    IncrementCallsToMmapHook();
    IncrementCallsToMunmapHook();
    IncrementCallsToMremapHook();
    VerifyMmapHookWasCalled();
    VerifyMremapHookWasCalled();
    VerifyMunmapHookWasCalled();
#endif

    // Test sbrk
    SetSbrkHook();
#if defined(HAVE_SBRK) && defined(__linux) && \
       (defined(__i386__) || defined(__x86_64__))
    p1 = sbrk(8192);
    CHECK(p1 != NULL);
    VerifySbrkHookWasCalled();
    p1 = sbrk(-8192);
    CHECK(p1 != NULL);
    VerifySbrkHookWasCalled();
    // However, sbrk hook should *not* be called with sbrk(0)
    p1 = sbrk(0);
    CHECK(p1 != NULL);
    CHECK_EQ(g_SbrkHook_calls, 0);
#else   // this is just to quiet the compiler: make sure all fns are called
    IncrementCallsToSbrkHook();
    VerifySbrkHookWasCalled();
#endif

    // Reset the hooks to what they used to be.  These are all
    // defined as part of MAKE_HOOK_CALLBACK, above.
    ResetNewHook();
    ResetDeleteHook();
    ResetMmapHook();
    ResetMremapHook();
    ResetMunmapHook();
    ResetSbrkHook();
  }

  // Check that "lots" of memory can be allocated
  fprintf(LOGSTREAM, "Testing large allocation\n");
  {
    const int mb_to_allocate = 100;
    void* p = rnd.alloc(mb_to_allocate << 20);
    CHECK(p != NULL);  // could not allocate
    free(p);
  }

  TestMallocAlignment();

  // Check calloc() with various arguments
  fprintf(LOGSTREAM, "Testing calloc\n");
  TestCalloc(0, 0, true);
  TestCalloc(0, 1, true);
  TestCalloc(1, 1, true);
  TestCalloc(1<<10, 0, true);
  TestCalloc(1<<20, 0, true);
  TestCalloc(0, 1<<10, true);
  TestCalloc(0, 1<<20, true);
  TestCalloc(1<<20, 2, true);
  TestCalloc(2, 1<<20, true);
  TestCalloc(1000, 1000, true);

  TestCalloc(kMaxSize, 2, false);
  TestCalloc(2, kMaxSize, false);
  TestCalloc(kMaxSize, kMaxSize, false);

  TestCalloc(kMaxSignedSize, 3, false);
  TestCalloc(3, kMaxSignedSize, false);
  TestCalloc(kMaxSignedSize, kMaxSignedSize, false);

  // Test that realloc doesn't always reallocate and copy memory.
  fprintf(LOGSTREAM, "Testing realloc\n");
  TestRealloc();

  fprintf(LOGSTREAM, "Testing operator new(nothrow).\n");
  TestNothrowNew(&::operator new);
  fprintf(LOGSTREAM, "Testing operator new[](nothrow).\n");
  TestNothrowNew(&::operator new[]);
  fprintf(LOGSTREAM, "Testing operator new.\n");
  TestNew(&::operator new);
  fprintf(LOGSTREAM, "Testing operator new[].\n");
  TestNew(&::operator new[]);

  // Create threads
  fprintf(LOGSTREAM, "Testing threaded allocation/deallocation (%d threads)\n",
          FLAGS_numthreads);
  threads = new TesterThread*[FLAGS_numthreads];
  for (int i = 0; i < FLAGS_numthreads; ++i) {
    threads[i] = new TesterThread(i);
  }

  // This runs all the tests at the same time, with a 1M stack size each
  RunManyThreadsWithId(RunThread, FLAGS_numthreads, 1<<20);

  for (int i = 0; i < FLAGS_numthreads; ++i) delete threads[i];    // Cleanup

  // Do the memory intensive tests after threads are done, since exhausting
  // the available address space can make pthread_create to fail.

  // Check that huge allocations fail with NULL instead of crashing
  fprintf(LOGSTREAM, "Testing huge allocations\n");
  TestHugeAllocations(&rnd);

  // Check that large allocations fail with NULL instead of crashing
#ifndef DEBUGALLOCATION    // debug allocation takes forever for huge allocs
  fprintf(LOGSTREAM, "Testing out of memory\n");
  for (int s = 0; ; s += (10<<20)) {
    void* large_object = rnd.alloc(s);
    if (large_object == NULL) break;
    free(large_object);
  }
#endif

  TestHugeThreadCache();
  TestRanges();
  TestReleaseToSystem();
  TestAggressiveDecommit();
  TestSetNewMode();

  return 0;
}

}

using testing::RunAllTests;

int main(int argc, char** argv) {
#ifdef DEBUGALLOCATION    // debug allocation takes forever for huge allocs
  FLAGS_max_free_queue_size = 0;  // return freed blocks to tcmalloc immediately
#endif

  RunAllTests(argc, argv);

  // Test tc_version()
  fprintf(LOGSTREAM, "Testing tc_version()\n");
  int major;
  int minor;
  const char* patch;
  char mmp[64];
  const char* human_version = tc_version(&major, &minor, &patch);
  snprintf(mmp, sizeof(mmp), "%d.%d%s", major, minor, patch);
  CHECK(!strcmp(PACKAGE_STRING, human_version));
  CHECK(!strcmp(PACKAGE_VERSION, mmp));

  fprintf(LOGSTREAM, "PASS\n");
}