• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 
2 /*--------------------------------------------------------------------*/
3 /*--- An implementation of malloc/free which doesn't use sbrk.     ---*/
4 /*---                                               m_mallocfree.c ---*/
5 /*--------------------------------------------------------------------*/
6 
7 /*
8    This file is part of Valgrind, a dynamic binary instrumentation
9    framework.
10 
11    Copyright (C) 2000-2017 Julian Seward
12       jseward@acm.org
13 
14    This program is free software; you can redistribute it and/or
15    modify it under the terms of the GNU General Public License as
16    published by the Free Software Foundation; either version 2 of the
17    License, or (at your option) any later version.
18 
19    This program is distributed in the hope that it will be useful, but
20    WITHOUT ANY WARRANTY; without even the implied warranty of
21    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
22    General Public License for more details.
23 
24    You should have received a copy of the GNU General Public License
25    along with this program; if not, write to the Free Software
26    Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
27    02111-1307, USA.
28 
29    The GNU General Public License is contained in the file COPYING.
30 */
31 
32 #include "pub_core_basics.h"
33 #include "pub_core_vki.h"
34 #include "pub_core_debuglog.h"
35 #include "pub_core_libcbase.h"
36 #include "pub_core_aspacemgr.h"
37 #include "pub_core_libcassert.h"
38 #include "pub_core_libcprint.h"
39 #include "pub_core_mallocfree.h"
40 #include "pub_core_options.h"
41 #include "pub_core_threadstate.h"   // For VG_INVALID_THREADID
42 #include "pub_core_gdbserver.h"
43 #include "pub_core_transtab.h"
44 #include "pub_core_tooliface.h"
45 
46 #include "pub_core_inner.h"
47 #if defined(ENABLE_INNER_CLIENT_REQUEST)
48 #include "memcheck/memcheck.h"
49 #endif
50 
51 // #define DEBUG_MALLOC      // turn on heavyweight debugging machinery
52 // #define VERBOSE_MALLOC    // make verbose, esp. in debugging machinery
53 
54 /* Number and total size of blocks in free queue. Used by mallinfo(). */
55 Long VG_(free_queue_volume) = 0;
56 Long VG_(free_queue_length) = 0;
57 
58 static void cc_analyse_alloc_arena ( ArenaId aid ); /* fwds */
59 
60 /*------------------------------------------------------------*/
61 /*--- Main types                                           ---*/
62 /*------------------------------------------------------------*/
63 
64 #define N_MALLOC_LISTS     112    // do not change this
65 
66 // The amount you can ask for is limited only by sizeof(SizeT)...
67 #define MAX_PSZB              (~((SizeT)0x0))
68 
69 // Each arena has a sorted array of superblocks, which expands
70 // dynamically.  This is its initial size.
71 #define SBLOCKS_SIZE_INITIAL 50
72 
73 typedef UChar UByte;
74 
75 /* Layout of an in-use block:
76 
77       cost center (OPTIONAL)   (VG_MIN_MALLOC_SZB bytes, only when h-p enabled)
78       this block total szB     (sizeof(SizeT) bytes)
79       red zone bytes           (depends on Arena.rz_szB, but >= sizeof(void*))
80       (payload bytes)
81       red zone bytes           (depends on Arena.rz_szB, but >= sizeof(void*))
82       this block total szB     (sizeof(SizeT) bytes)
83 
84    Layout of a block on the free list:
85 
86       cost center (OPTIONAL)   (VG_MIN_MALLOC_SZB bytes, only when h-p enabled)
87       this block total szB     (sizeof(SizeT) bytes)
88       freelist previous ptr    (sizeof(void*) bytes)
89       excess red zone bytes    (if Arena.rz_szB > sizeof(void*))
90       (payload bytes)
91       excess red zone bytes    (if Arena.rz_szB > sizeof(void*))
92       freelist next ptr        (sizeof(void*) bytes)
93       this block total szB     (sizeof(SizeT) bytes)
94 
95    Total size in bytes (bszB) and payload size in bytes (pszB)
96    are related by:
97 
98       bszB == pszB + 2*sizeof(SizeT) + 2*a->rz_szB
99 
100    when heap profiling is not enabled, and
101 
102       bszB == pszB + 2*sizeof(SizeT) + 2*a->rz_szB + VG_MIN_MALLOC_SZB
103 
104    when it is enabled.  It follows that the minimum overhead per heap
105    block for arenas used by the core is:
106 
107       32-bit platforms:  2*4 + 2*4 == 16 bytes
108       64-bit platforms:  2*8 + 2*8 == 32 bytes
109 
110    when heap profiling is not enabled, and
111 
112       32-bit platforms:  2*4 + 2*4 + 8  == 24 bytes
113       64-bit platforms:  2*8 + 2*8 + 16 == 48 bytes
114 
115    when it is enabled.  In all cases, extra overhead may be incurred
116    when rounding the payload size up to VG_MIN_MALLOC_SZB.
117 
118    Furthermore, both size fields in the block have their least-significant
119    bit set if the block is not in use, and unset if it is in use.
120    (The bottom 3 or so bits are always free for this because of alignment.)
121    A block size of zero is not possible, because a block always has at
122    least two SizeTs and two pointers of overhead.
123 
124    Nb: All Block payloads must be VG_MIN_MALLOC_SZB-aligned.  This is
125    achieved by ensuring that Superblocks are VG_MIN_MALLOC_SZB-aligned
126    (see newSuperblock() for how), and that the lengths of the following
127    things are a multiple of VG_MIN_MALLOC_SZB:
128    - Superblock admin section lengths (due to elastic padding)
129    - Block admin section (low and high) lengths (due to elastic redzones)
130    - Block payload lengths (due to req_pszB rounding up)
131 
132    The heap-profile cost-center field is 8 bytes even on 32 bit
133    platforms.  This is so as to keep the payload field 8-aligned.  On
134    a 64-bit platform, this cc-field contains a pointer to a const
135    HChar*, which is the cost center name.  On 32-bit platforms, the
136    pointer lives in the lower-addressed half of the field, regardless
137    of the endianness of the host.
138 */
139 typedef
140    struct {
141       // No fields are actually used in this struct, because a Block has
142       // many variable sized fields and so can't be accessed
143       // meaningfully with normal fields.  So we use access functions all
144       // the time.  This struct gives us a type to use, though.  Also, we
145       // make sizeof(Block) 1 byte so that we can do arithmetic with the
146       // Block* type in increments of 1!
147       UByte dummy;
148    }
149    Block;
150 
151 /* Ensure that Block payloads can be safely cast to various pointers below. */
152 STATIC_ASSERT(VG_MIN_MALLOC_SZB % sizeof(void *) == 0);
153 
154 // A superblock.  'padding' is never used, it just ensures that if the
155 // entire Superblock is aligned to VG_MIN_MALLOC_SZB, then payload_bytes[]
156 // will be too.  It can add small amounts of padding unnecessarily -- eg.
157 // 8-bytes on 32-bit machines with an 8-byte VG_MIN_MALLOC_SZB -- because
158 // it's too hard to make a constant expression that works perfectly in all
159 // cases.
160 // 'unsplittable' is set to NULL if superblock can be split, otherwise
161 // it is set to the address of the superblock. An unsplittable superblock
162 // will contain only one allocated block. An unsplittable superblock will
163 // be unmapped when its (only) allocated block is freed.
164 // The free space at the end of an unsplittable superblock is not used to
165 // make a free block. Note that this means that an unsplittable superblock can
166 // have up to slightly less than 1 page of unused bytes at the end of the
167 // superblock.
168 // 'unsplittable' is used to avoid quadratic memory usage for linear
169 // reallocation of big structures
170 // (see http://bugs.kde.org/show_bug.cgi?id=250101).
171 // ??? unsplittable replaces 'void *padding2'. Choosed this
172 // ??? to avoid changing the alignment logic. Maybe something cleaner
173 // ??? can be done.
174 // A splittable block can be reclaimed when all its blocks are freed :
175 // the reclaim of such a block is deferred till either another superblock
176 // of the same arena can be reclaimed or till a new superblock is needed
177 // in any arena.
178 // payload_bytes[] is made a single big Block when the Superblock is
179 // created, and then can be split and the splittings remerged, but Blocks
180 // always cover its entire length -- there's never any unused bytes at the
181 // end, for example.
182 typedef
183    struct _Superblock {
184       SizeT n_payload_bytes;
185       struct _Superblock* unsplittable;
186       UByte padding[ VG_MIN_MALLOC_SZB -
187                         ((sizeof(struct _Superblock*) + sizeof(SizeT)) %
188                          VG_MIN_MALLOC_SZB) ];
189       UByte payload_bytes[0];
190    }
191    Superblock;
192 
193 // An arena. 'freelist' is a circular, doubly-linked list.  'rz_szB' is
194 // elastic, in that it can be bigger than asked-for to ensure alignment.
195 typedef
196    struct {
197       const HChar* name;
198       Bool         clientmem;        // Allocates in the client address space?
199       SizeT        rz_szB;           // Red zone size in bytes
200       SizeT        min_sblock_szB;   // Minimum superblock size in bytes
201       SizeT        min_unsplittable_sblock_szB;
202       // Minimum unsplittable superblock size in bytes. To be marked as
203       // unsplittable, a superblock must have a
204       // size >= min_unsplittable_sblock_szB and cannot be split.
205       // So, to avoid big overhead, superblocks used to provide aligned
206       // blocks on big alignments are splittable.
207       // Unsplittable superblocks will be reclaimed when their (only)
208       // allocated block is freed.
209       // Smaller size superblocks are splittable and can be reclaimed when all
210       // their blocks are freed.
211       Block*       freelist[N_MALLOC_LISTS];
212       // A dynamically expanding, ordered array of (pointers to)
213       // superblocks in the arena.  If this array is expanded, which
214       // is rare, the previous space it occupies is simply abandoned.
215       // To avoid having to get yet another block from m_aspacemgr for
216       // the first incarnation of this array, the first allocation of
217       // it is within this struct.  If it has to be expanded then the
218       // new space is acquired from m_aspacemgr as you would expect.
219       Superblock** sblocks;
220       SizeT        sblocks_size;
221       SizeT        sblocks_used;
222       Superblock*  sblocks_initial[SBLOCKS_SIZE_INITIAL];
223       Superblock*  deferred_reclaimed_sb;
224 
225       // VG_(arena_perm_malloc) returns memory from superblocks
226       // only used for permanent blocks. No overhead. These superblocks
227       // are not stored in sblocks array above.
228       Addr         perm_malloc_current; // first byte free in perm_malloc sb.
229       Addr         perm_malloc_limit; // maximum usable byte in perm_malloc sb.
230 
231       // Stats only
232       SizeT        stats__perm_bytes_on_loan;
233       SizeT        stats__perm_blocks;
234 
235       ULong        stats__nreclaim_unsplit;
236       ULong        stats__nreclaim_split;
237       /* total # of reclaim executed for unsplittable/splittable superblocks */
238       SizeT        stats__bytes_on_loan;
239       SizeT        stats__bytes_mmaped;
240       SizeT        stats__bytes_on_loan_max;
241       ULong        stats__tot_blocks; /* total # blocks alloc'd */
242       ULong        stats__tot_bytes; /* total # bytes alloc'd */
243       ULong        stats__nsearches; /* total # freelist checks */
244       // If profiling, when should the next profile happen at
245       // (in terms of stats__bytes_on_loan_max) ?
246       SizeT        next_profile_at;
247       SizeT        stats__bytes_mmaped_max;
248    }
249    Arena;
250 
251 
252 /*------------------------------------------------------------*/
253 /*--- Low-level functions for working with Blocks.         ---*/
254 /*------------------------------------------------------------*/
255 
256 #define SIZE_T_0x1      ((SizeT)0x1)
257 
258 static const char* probably_your_fault =
259    "This is probably caused by your program erroneously writing past the\n"
260    "end of a heap block and corrupting heap metadata.  If you fix any\n"
261    "invalid writes reported by Memcheck, this assertion failure will\n"
262    "probably go away.  Please try that before reporting this as a bug.\n";
263 
264 // Mark a bszB as in-use, and not in-use, and remove the in-use attribute.
265 static __inline__
mk_inuse_bszB(SizeT bszB)266 SizeT mk_inuse_bszB ( SizeT bszB )
267 {
268    vg_assert2(bszB != 0, probably_your_fault);
269    return bszB & (~SIZE_T_0x1);
270 }
271 static __inline__
mk_free_bszB(SizeT bszB)272 SizeT mk_free_bszB ( SizeT bszB )
273 {
274    vg_assert2(bszB != 0, probably_your_fault);
275    return bszB | SIZE_T_0x1;
276 }
277 static __inline__
mk_plain_bszB(SizeT bszB)278 SizeT mk_plain_bszB ( SizeT bszB )
279 {
280    vg_assert2(bszB != 0, probably_your_fault);
281    return bszB & (~SIZE_T_0x1);
282 }
283 
284 // Forward definition.
285 static
286 void ensure_mm_init ( ArenaId aid );
287 
288 // return either 0 or sizeof(ULong) depending on whether or not
289 // heap profiling is engaged
290 #define hp_overhead_szB() set_at_init_hp_overhead_szB
291 static SizeT set_at_init_hp_overhead_szB = -1000000;
292 // startup value chosen to very likely cause a problem if used before
293 // a proper value is given by ensure_mm_init.
294 
295 //---------------------------------------------------------------------------
296 
297 // Get a block's size as stored, ie with the in-use/free attribute.
298 static __inline__
get_bszB_as_is(Block * b)299 SizeT get_bszB_as_is ( Block* b )
300 {
301    UByte* b2     = (UByte*)b;
302    SizeT bszB_lo = *ASSUME_ALIGNED(SizeT*, &b2[0 + hp_overhead_szB()]);
303    SizeT bszB_hi = *ASSUME_ALIGNED(SizeT*,
304                                    &b2[mk_plain_bszB(bszB_lo) - sizeof(SizeT)]);
305    vg_assert2(bszB_lo == bszB_hi,
306       "Heap block lo/hi size mismatch: lo = %llu, hi = %llu.\n%s",
307       (ULong)bszB_lo, (ULong)bszB_hi, probably_your_fault);
308    return bszB_lo;
309 }
310 
311 // Get a block's plain size, ie. remove the in-use/free attribute.
312 static __inline__
get_bszB(Block * b)313 SizeT get_bszB ( Block* b )
314 {
315    return mk_plain_bszB(get_bszB_as_is(b));
316 }
317 
318 // Set the size fields of a block.  bszB may have the in-use/free attribute.
319 static __inline__
set_bszB(Block * b,SizeT bszB)320 void set_bszB ( Block* b, SizeT bszB )
321 {
322    UByte* b2 = (UByte*)b;
323    *ASSUME_ALIGNED(SizeT*, &b2[0 + hp_overhead_szB()])               = bszB;
324    *ASSUME_ALIGNED(SizeT*, &b2[mk_plain_bszB(bszB) - sizeof(SizeT)]) = bszB;
325 }
326 
327 //---------------------------------------------------------------------------
328 
329 // Does this block have the in-use attribute?
330 static __inline__
is_inuse_block(Block * b)331 Bool is_inuse_block ( Block* b )
332 {
333    SizeT bszB = get_bszB_as_is(b);
334    vg_assert2(bszB != 0, probably_your_fault);
335    return (0 != (bszB & SIZE_T_0x1)) ? False : True;
336 }
337 
338 //---------------------------------------------------------------------------
339 
340 // Return the lower, upper and total overhead in bytes for a block.
341 // These are determined purely by which arena the block lives in.
342 static __inline__
overhead_szB_lo(Arena * a)343 SizeT overhead_szB_lo ( Arena* a )
344 {
345    return hp_overhead_szB() + sizeof(SizeT) + a->rz_szB;
346 }
347 static __inline__
overhead_szB_hi(Arena * a)348 SizeT overhead_szB_hi ( Arena* a )
349 {
350    return a->rz_szB + sizeof(SizeT);
351 }
352 static __inline__
overhead_szB(Arena * a)353 SizeT overhead_szB ( Arena* a )
354 {
355    return overhead_szB_lo(a) + overhead_szB_hi(a);
356 }
357 
358 //---------------------------------------------------------------------------
359 
360 // Return the minimum bszB for a block in this arena.  Can have zero-length
361 // payloads, so it's the size of the admin bytes.
362 static __inline__
min_useful_bszB(Arena * a)363 SizeT min_useful_bszB ( Arena* a )
364 {
365    return overhead_szB(a);
366 }
367 
368 //---------------------------------------------------------------------------
369 
370 // Convert payload size <--> block size (both in bytes).
371 static __inline__
pszB_to_bszB(Arena * a,SizeT pszB)372 SizeT pszB_to_bszB ( Arena* a, SizeT pszB )
373 {
374    return pszB + overhead_szB(a);
375 }
376 static __inline__
bszB_to_pszB(Arena * a,SizeT bszB)377 SizeT bszB_to_pszB ( Arena* a, SizeT bszB )
378 {
379    vg_assert2(bszB >= overhead_szB(a), probably_your_fault);
380    return bszB - overhead_szB(a);
381 }
382 
383 //---------------------------------------------------------------------------
384 
385 // Get a block's payload size.
386 static __inline__
get_pszB(Arena * a,Block * b)387 SizeT get_pszB ( Arena* a, Block* b )
388 {
389    return bszB_to_pszB(a, get_bszB(b));
390 }
391 
392 //---------------------------------------------------------------------------
393 
394 // Given the addr of a block, return the addr of its payload, and vice versa.
395 static __inline__
get_block_payload(Arena * a,Block * b)396 UByte* get_block_payload ( Arena* a, Block* b )
397 {
398    UByte* b2 = (UByte*)b;
399    return & b2[ overhead_szB_lo(a) ];
400 }
401 // Given the addr of a block's payload, return the addr of the block itself.
402 static __inline__
get_payload_block(Arena * a,UByte * payload)403 Block* get_payload_block ( Arena* a, UByte* payload )
404 {
405    return (Block*)&payload[ -overhead_szB_lo(a) ];
406 }
407 
408 //---------------------------------------------------------------------------
409 
410 // Set and get the next and previous link fields of a block.
411 static __inline__
set_prev_b(Block * b,Block * prev_p)412 void set_prev_b ( Block* b, Block* prev_p )
413 {
414    UByte* b2 = (UByte*)b;
415    *ASSUME_ALIGNED(Block**, &b2[hp_overhead_szB() + sizeof(SizeT)]) = prev_p;
416 }
417 static __inline__
set_next_b(Block * b,Block * next_p)418 void set_next_b ( Block* b, Block* next_p )
419 {
420    UByte* b2 = (UByte*)b;
421    *ASSUME_ALIGNED(Block**,
422                    &b2[get_bszB(b) - sizeof(SizeT) - sizeof(void*)]) = next_p;
423 }
424 static __inline__
get_prev_b(Block * b)425 Block* get_prev_b ( Block* b )
426 {
427    UByte* b2 = (UByte*)b;
428    return *ASSUME_ALIGNED(Block**, &b2[hp_overhead_szB() + sizeof(SizeT)]);
429 }
430 static __inline__
get_next_b(Block * b)431 Block* get_next_b ( Block* b )
432 {
433    UByte* b2 = (UByte*)b;
434    return *ASSUME_ALIGNED(Block**,
435                           &b2[get_bszB(b) - sizeof(SizeT) - sizeof(void*)]);
436 }
437 
438 //---------------------------------------------------------------------------
439 
440 // Set and get the cost-center field of a block.
441 static __inline__
set_cc(Block * b,const HChar * cc)442 void set_cc ( Block* b, const HChar* cc )
443 {
444    UByte* b2 = (UByte*)b;
445    vg_assert( VG_(clo_profile_heap) );
446    *ASSUME_ALIGNED(const HChar**, &b2[0]) = cc;
447 }
448 static __inline__
get_cc(Block * b)449 const HChar* get_cc ( Block* b )
450 {
451    UByte* b2 = (UByte*)b;
452    vg_assert( VG_(clo_profile_heap) );
453    return *ASSUME_ALIGNED(const HChar**, &b2[0]);
454 }
455 
456 //---------------------------------------------------------------------------
457 
458 // Get the block immediately preceding this one in the Superblock.
459 static __inline__
get_predecessor_block(Block * b)460 Block* get_predecessor_block ( Block* b )
461 {
462    UByte* b2 = (UByte*)b;
463    SizeT  bszB = mk_plain_bszB(*ASSUME_ALIGNED(SizeT*, &b2[-sizeof(SizeT)]));
464    return (Block*)&b2[-bszB];
465 }
466 
467 //---------------------------------------------------------------------------
468 
469 // Read and write the lower and upper red-zone bytes of a block.
470 static __inline__
set_rz_lo_byte(Block * b,UInt rz_byteno,UByte v)471 void set_rz_lo_byte ( Block* b, UInt rz_byteno, UByte v )
472 {
473    UByte* b2 = (UByte*)b;
474    b2[hp_overhead_szB() + sizeof(SizeT) + rz_byteno] = v;
475 }
476 static __inline__
set_rz_hi_byte(Block * b,UInt rz_byteno,UByte v)477 void set_rz_hi_byte ( Block* b, UInt rz_byteno, UByte v )
478 {
479    UByte* b2 = (UByte*)b;
480    b2[get_bszB(b) - sizeof(SizeT) - rz_byteno - 1] = v;
481 }
482 static __inline__
get_rz_lo_byte(Block * b,UInt rz_byteno)483 UByte get_rz_lo_byte ( Block* b, UInt rz_byteno )
484 {
485    UByte* b2 = (UByte*)b;
486    return b2[hp_overhead_szB() + sizeof(SizeT) + rz_byteno];
487 }
488 static __inline__
get_rz_hi_byte(Block * b,UInt rz_byteno)489 UByte get_rz_hi_byte ( Block* b, UInt rz_byteno )
490 {
491    UByte* b2 = (UByte*)b;
492    return b2[get_bszB(b) - sizeof(SizeT) - rz_byteno - 1];
493 }
494 
495 #if defined(ENABLE_INNER_CLIENT_REQUEST)
496 /* When running as an inner, the block headers before and after
497    (see 'Layout of an in-use block:' above) are made non accessible
498    by VALGRIND_MALLOCLIKE_BLOCK/VALGRIND_FREELIKE_BLOCK
499    to allow the outer to detect block overrun.
500    The below two functions are used when these headers must be
501    temporarily accessed. */
mkBhdrAccess(Arena * a,Block * b)502 static void mkBhdrAccess( Arena* a, Block* b )
503 {
504    VALGRIND_MAKE_MEM_DEFINED (b,
505                               hp_overhead_szB() + sizeof(SizeT) + a->rz_szB);
506    VALGRIND_MAKE_MEM_DEFINED (b + get_bszB(b) - a->rz_szB - sizeof(SizeT),
507                               a->rz_szB + sizeof(SizeT));
508 }
509 
510 /* Mark block hdr as not accessible.
511    !!! Currently, we do not mark the cost center and szB fields unaccessible
512    as these are accessed at too many places. */
mkBhdrNoAccess(Arena * a,Block * b)513 static void mkBhdrNoAccess( Arena* a, Block* b )
514 {
515    VALGRIND_MAKE_MEM_NOACCESS (b + hp_overhead_szB() + sizeof(SizeT),
516                                a->rz_szB);
517    VALGRIND_MAKE_MEM_NOACCESS (b + get_bszB(b) - sizeof(SizeT) - a->rz_szB,
518                                a->rz_szB);
519 }
520 
521 /* Make the cc+szB fields accessible. */
mkBhdrSzAccess(Arena * a,Block * b)522 static void mkBhdrSzAccess( Arena* a, Block* b )
523 {
524    VALGRIND_MAKE_MEM_DEFINED (b,
525                               hp_overhead_szB() + sizeof(SizeT));
526    /* We cannot use  get_bszB(b), as this reads the 'hi' szB we want
527       to mark accessible. So, we only access the 'lo' szB. */
528    SizeT bszB_lo = mk_plain_bszB(*(SizeT*)&b[0 + hp_overhead_szB()]);
529    VALGRIND_MAKE_MEM_DEFINED (b + bszB_lo - sizeof(SizeT),
530                               sizeof(SizeT));
531 }
532 #endif
533 
534 /*------------------------------------------------------------*/
535 /*--- Arena management                                     ---*/
536 /*------------------------------------------------------------*/
537 
538 #define CORE_ARENA_MIN_SZB    1048576
539 
540 // The arena structures themselves.
541 static Arena vg_arena[VG_N_ARENAS];
542 
543 // Functions external to this module identify arenas using ArenaIds,
544 // not Arena*s.  This fn converts the former to the latter.
arenaId_to_ArenaP(ArenaId arena)545 static Arena* arenaId_to_ArenaP ( ArenaId arena )
546 {
547    vg_assert(arena >= 0 && arena < VG_N_ARENAS);
548    return & vg_arena[arena];
549 }
550 
arenaP_to_ArenaId(Arena * a)551 static ArenaId arenaP_to_ArenaId ( Arena *a )
552 {
553    ArenaId arena = a -vg_arena;
554    vg_assert(arena >= 0 && arena < VG_N_ARENAS);
555    return arena;
556 }
557 
558 // Initialise an arena.  rz_szB is the (default) minimum redzone size;
559 // It might be overridden by VG_(clo_redzone_size) or VG_(clo_core_redzone_size).
560 // it might be made bigger to ensure that VG_MIN_MALLOC_SZB is observed.
561 static
arena_init(ArenaId aid,const HChar * name,SizeT rz_szB,SizeT min_sblock_szB,SizeT min_unsplittable_sblock_szB)562 void arena_init ( ArenaId aid, const HChar* name, SizeT rz_szB,
563                   SizeT min_sblock_szB, SizeT min_unsplittable_sblock_szB )
564 {
565    SizeT  i;
566    Arena* a = arenaId_to_ArenaP(aid);
567 
568    // Ensure default redzones are a reasonable size.
569    vg_assert(rz_szB <= MAX_REDZONE_SZB);
570 
571    /* Override the default redzone size if a clo value was given.
572       Note that the clo value can be significantly bigger than MAX_REDZONE_SZB
573       to allow the user to chase horrible bugs using up to 1 page
574       of protection. */
575    if (VG_AR_CLIENT == aid) {
576       if (VG_(clo_redzone_size) != -1)
577          rz_szB = VG_(clo_redzone_size);
578    } else {
579       if (VG_(clo_core_redzone_size) != rz_szB)
580          rz_szB = VG_(clo_core_redzone_size);
581    }
582 
583    // Redzones must always be at least the size of a pointer, for holding the
584    // prev/next pointer (see the layout details at the top of this file).
585    if (rz_szB < sizeof(void*)) rz_szB = sizeof(void*);
586 
587    // The size of the low and high admin sections in a block must be a
588    // multiple of VG_MIN_MALLOC_SZB.  So we round up the asked-for
589    // redzone size if necessary to achieve this.
590    a->rz_szB = rz_szB;
591    while (0 != overhead_szB_lo(a) % VG_MIN_MALLOC_SZB) a->rz_szB++;
592    vg_assert(overhead_szB_lo(a) - hp_overhead_szB() == overhead_szB_hi(a));
593 
594    // Here we have established the effective redzone size.
595 
596 
597    vg_assert((min_sblock_szB % VKI_PAGE_SIZE) == 0);
598    a->name      = name;
599    a->clientmem = ( VG_AR_CLIENT == aid ? True : False );
600 
601    a->min_sblock_szB = min_sblock_szB;
602    a->min_unsplittable_sblock_szB = min_unsplittable_sblock_szB;
603    for (i = 0; i < N_MALLOC_LISTS; i++) a->freelist[i] = NULL;
604 
605    a->sblocks                  = & a->sblocks_initial[0];
606    a->sblocks_size             = SBLOCKS_SIZE_INITIAL;
607    a->sblocks_used             = 0;
608    a->deferred_reclaimed_sb    = 0;
609    a->perm_malloc_current      = 0;
610    a->perm_malloc_limit        = 0;
611    a->stats__perm_bytes_on_loan= 0;
612    a->stats__perm_blocks       = 0;
613    a->stats__nreclaim_unsplit  = 0;
614    a->stats__nreclaim_split    = 0;
615    a->stats__bytes_on_loan     = 0;
616    a->stats__bytes_mmaped      = 0;
617    a->stats__bytes_on_loan_max = 0;
618    a->stats__bytes_mmaped_max  = 0;
619    a->stats__tot_blocks        = 0;
620    a->stats__tot_bytes         = 0;
621    a->stats__nsearches         = 0;
622    a->next_profile_at          = 25 * 1000 * 1000;
623    vg_assert(sizeof(a->sblocks_initial)
624              == SBLOCKS_SIZE_INITIAL * sizeof(Superblock*));
625 }
626 
627 /* Print vital stats for an arena. */
VG_(print_all_arena_stats)628 void VG_(print_all_arena_stats) ( void )
629 {
630    UInt i;
631    for (i = 0; i < VG_N_ARENAS; i++) {
632       Arena* a = arenaId_to_ArenaP(i);
633       VG_(message)(Vg_DebugMsg,
634                    "%-8s: %'13lu/%'13lu max/curr mmap'd, "
635                    "%llu/%llu unsplit/split sb unmmap'd,  "
636                    "%'13lu/%'13lu max/curr,  "
637                    "%10llu/%10llu totalloc-blocks/bytes,"
638                    "  %10llu searches %lu rzB\n",
639                    a->name,
640                    a->stats__bytes_mmaped_max, a->stats__bytes_mmaped,
641                    a->stats__nreclaim_unsplit, a->stats__nreclaim_split,
642                    a->stats__bytes_on_loan_max,
643                    a->stats__bytes_on_loan,
644                    a->stats__tot_blocks, a->stats__tot_bytes,
645                    a->stats__nsearches,
646                    a->rz_szB
647       );
648    }
649 }
650 
VG_(print_arena_cc_analysis)651 void VG_(print_arena_cc_analysis) ( void )
652 {
653    UInt i;
654    vg_assert( VG_(clo_profile_heap) );
655    for (i = 0; i < VG_N_ARENAS; i++) {
656       cc_analyse_alloc_arena(i);
657    }
658 }
659 
660 
661 /* This library is self-initialising, as it makes this more self-contained,
662    less coupled with the outside world.  Hence VG_(arena_malloc)() and
663    VG_(arena_free)() below always call ensure_mm_init() to ensure things are
664    correctly initialised.
665 
666    We initialise the client arena separately (and later) because the core
667    must do non-client allocation before the tool has a chance to set the
668    client arena's redzone size.
669 */
670 static Bool     client_inited = False;
671 static Bool  nonclient_inited = False;
672 
673 static
ensure_mm_init(ArenaId aid)674 void ensure_mm_init ( ArenaId aid )
675 {
676    static SizeT client_rz_szB = 8;     // default: be paranoid
677 
678    /* We use checked red zones (of various sizes) for our internal stuff,
679       and an unchecked zone of arbitrary size for the client.  Of
680       course the client's red zone can be checked by the tool, eg.
681       by using addressibility maps, but not by the mechanism implemented
682       here, which merely checks at the time of freeing that the red
683       zone bytes are unchanged.
684 
685       Nb: redzone sizes are *minimums*;  they could be made bigger to ensure
686       alignment.  Eg. with 8 byte alignment, on 32-bit machines 4 stays as
687       4, but 16 becomes 20;  but on 64-bit machines 4 becomes 8, and 16
688       stays as 16 --- the extra 4 bytes in both are accounted for by the
689       larger prev/next ptr.
690    */
691    if (VG_AR_CLIENT == aid) {
692       Int ar_client_sbszB;
693       if (client_inited) {
694          // This assertion ensures that a tool cannot try to change the client
695          // redzone size with VG_(needs_malloc_replacement)() after this module
696          // has done its first allocation from the client arena.
697          if (VG_(needs).malloc_replacement)
698             vg_assert(client_rz_szB == VG_(tdict).tool_client_redzone_szB);
699          return;
700       }
701 
702       // Check and set the client arena redzone size
703       if (VG_(needs).malloc_replacement) {
704          client_rz_szB = VG_(tdict).tool_client_redzone_szB;
705          if (client_rz_szB > MAX_REDZONE_SZB) {
706             VG_(printf)( "\nTool error:\n"
707                          "  specified redzone size is too big (%llu)\n",
708                          (ULong)client_rz_szB);
709             VG_(exit)(1);
710          }
711       }
712       // Initialise the client arena.  On all platforms,
713       // increasing the superblock size reduces the number of superblocks
714       // in the client arena, which makes findSb cheaper.
715       ar_client_sbszB = 4194304;
716       // superblocks with a size > ar_client_sbszB will be unsplittable
717       // (unless used for providing memalign-ed blocks).
718       arena_init ( VG_AR_CLIENT,    "client",   client_rz_szB,
719                    ar_client_sbszB, ar_client_sbszB+1);
720       client_inited = True;
721 
722    } else {
723       if (nonclient_inited) {
724          return;
725       }
726       set_at_init_hp_overhead_szB =
727          VG_(clo_profile_heap)  ? VG_MIN_MALLOC_SZB  : 0;
728       // Initialise the non-client arenas
729       // Similarly to client arena, big allocations will be unsplittable.
730       arena_init ( VG_AR_CORE,      "core",     CORE_REDZONE_DEFAULT_SZB,
731                    4194304, 4194304+1 );
732       arena_init ( VG_AR_DINFO,     "dinfo",    CORE_REDZONE_DEFAULT_SZB,
733                    1048576, 1048576+1 );
734       arena_init ( VG_AR_DEMANGLE,  "demangle", CORE_REDZONE_DEFAULT_SZB,
735                    65536,   65536+1 );
736       arena_init ( VG_AR_TTAUX,     "ttaux",    CORE_REDZONE_DEFAULT_SZB,
737                    65536,   65536+1 );
738       nonclient_inited = True;
739    }
740 
741 #  ifdef DEBUG_MALLOC
742    VG_(printf)("ZZZ1\n");
743    VG_(sanity_check_malloc_all)();
744    VG_(printf)("ZZZ2\n");
745 #  endif
746 }
747 
748 
749 /*------------------------------------------------------------*/
750 /*--- Superblock management                                ---*/
751 /*------------------------------------------------------------*/
752 
753 __attribute__((noreturn))
VG_(out_of_memory_NORETURN)754 void VG_(out_of_memory_NORETURN) ( const HChar* who, SizeT szB )
755 {
756    static Int outputTrial = 0;
757    // We try once to output the full memory state followed by the below message.
758    // If that fails (due to out of memory during first trial), we try to just
759    // output the below message.
760    // And then we abandon.
761 
762    ULong tot_alloc = VG_(am_get_anonsize_total)();
763    const HChar* s1 =
764       "\n"
765       "    Valgrind's memory management: out of memory:\n"
766       "       %s's request for %llu bytes failed.\n"
767       "       %'13llu bytes have already been mmap-ed ANONYMOUS.\n"
768       "    Valgrind cannot continue.  Sorry.\n\n"
769       "    There are several possible reasons for this.\n"
770       "    - You have some kind of memory limit in place.  Look at the\n"
771       "      output of 'ulimit -a'.  Is there a limit on the size of\n"
772       "      virtual memory or address space?\n"
773       "    - You have run out of swap space.\n"
774       "    - Valgrind has a bug.  If you think this is the case or you are\n"
775       "    not sure, please let us know and we'll try to fix it.\n"
776       "    Please note that programs can take substantially more memory than\n"
777       "    normal when running under Valgrind tools, eg. up to twice or\n"
778       "    more, depending on the tool.  On a 64-bit machine, Valgrind\n"
779       "    should be able to make use of up 32GB memory.  On a 32-bit\n"
780       "    machine, Valgrind should be able to use all the memory available\n"
781       "    to a single process, up to 4GB if that's how you have your\n"
782       "    kernel configured.  Most 32-bit Linux setups allow a maximum of\n"
783       "    3GB per process.\n\n"
784       "    Whatever the reason, Valgrind cannot continue.  Sorry.\n";
785 
786    if (outputTrial <= 1) {
787       if (outputTrial == 0) {
788          outputTrial++;
789          // First print the memory stats with the aspacemgr data.
790          VG_(am_show_nsegments) (0, "out_of_memory");
791          VG_(print_all_arena_stats) ();
792          if (VG_(clo_profile_heap))
793             VG_(print_arena_cc_analysis) ();
794          // And then print some other information that might help.
795          VG_(print_all_stats) (False, /* Memory stats */
796                                True /* Tool stats */);
797          VG_(show_sched_status) (True,  // host_stacktrace
798                                  True,  // valgrind_stack_usage
799                                  True); // exited_threads
800         /* In case we are an inner valgrind, asks the outer to report
801             its memory state in its log output. */
802          INNER_REQUEST(VALGRIND_MONITOR_COMMAND("v.set log_output"));
803          INNER_REQUEST(VALGRIND_MONITOR_COMMAND("v.info memory aspacemgr"));
804       }
805       outputTrial++;
806       VG_(message)(Vg_UserMsg, s1, who, (ULong)szB, tot_alloc);
807    } else {
808       VG_(debugLog)(0,"mallocfree", s1, who, (ULong)szB, tot_alloc);
809    }
810 
811    VG_(exit)(1);
812 }
813 
814 
815 // Align ptr p upwards to an align-sized boundary.
816 static
align_upwards(void * p,SizeT align)817 void* align_upwards ( void* p, SizeT align )
818 {
819    Addr a = (Addr)p;
820    if ((a % align) == 0) return (void*)a;
821    return (void*)(a - (a % align) + align);
822 }
823 
824 // Forward definition.
825 static
826 void deferred_reclaimSuperblock ( Arena* a, Superblock* sb);
827 
828 // If not enough memory available, either aborts (for non-client memory)
829 // or returns 0 (for client memory).
830 static
newSuperblock(Arena * a,SizeT cszB)831 Superblock* newSuperblock ( Arena* a, SizeT cszB )
832 {
833    Superblock* sb;
834    SysRes      sres;
835    Bool        unsplittable;
836    ArenaId     aid;
837 
838    // A new superblock is needed for arena a. We will execute the deferred
839    // reclaim in all arenas in order to minimise fragmentation and
840    // peak memory usage.
841    for (aid = 0; aid < VG_N_ARENAS; aid++) {
842       Arena* arena = arenaId_to_ArenaP(aid);
843       if (arena->deferred_reclaimed_sb != NULL)
844          deferred_reclaimSuperblock (arena, NULL);
845    }
846 
847    // Take into account admin bytes in the Superblock.
848    cszB += sizeof(Superblock);
849 
850    if (cszB < a->min_sblock_szB) cszB = a->min_sblock_szB;
851    cszB = VG_PGROUNDUP(cszB);
852 
853    if (cszB >= a->min_unsplittable_sblock_szB)
854       unsplittable = True;
855    else
856       unsplittable = False;
857 
858 
859    if (a->clientmem) {
860       // client allocation -- return 0 to client if it fails
861       sres = VG_(am_mmap_client_heap)
862          ( cszB, VKI_PROT_READ|VKI_PROT_WRITE|VKI_PROT_EXEC );
863       if (sr_isError(sres))
864          return 0;
865       sb = (Superblock*)(Addr)sr_Res(sres);
866    } else {
867       // non-client allocation -- abort if it fails
868       sres = VG_(am_mmap_anon_float_valgrind)( cszB );
869       if (sr_isError(sres)) {
870          VG_(out_of_memory_NORETURN)("newSuperblock", cszB);
871          /* NOTREACHED */
872          sb = NULL; /* keep gcc happy */
873       } else {
874          sb = (Superblock*)(Addr)sr_Res(sres);
875       }
876    }
877    vg_assert(NULL != sb);
878    INNER_REQUEST(VALGRIND_MAKE_MEM_UNDEFINED(sb, cszB));
879    vg_assert(0 == (Addr)sb % VG_MIN_MALLOC_SZB);
880    sb->n_payload_bytes = cszB - sizeof(Superblock);
881    sb->unsplittable = (unsplittable ? sb : NULL);
882    a->stats__bytes_mmaped += cszB;
883    if (a->stats__bytes_mmaped > a->stats__bytes_mmaped_max)
884       a->stats__bytes_mmaped_max = a->stats__bytes_mmaped;
885    VG_(debugLog)(1, "mallocfree",
886                     "newSuperblock at %p (pszB %7lu) %s owner %s/%s\n",
887                     sb, sb->n_payload_bytes,
888                     (unsplittable ? "unsplittable" : ""),
889                     a->clientmem ? "CLIENT" : "VALGRIND", a->name );
890    return sb;
891 }
892 
893 // Reclaims the given superblock:
894 //  * removes sb from arena sblocks list.
895 //  * munmap the superblock segment.
896 static
reclaimSuperblock(Arena * a,Superblock * sb)897 void reclaimSuperblock ( Arena* a, Superblock* sb)
898 {
899    SysRes sres;
900    SizeT  cszB;
901    UInt   i, j;
902 
903    VG_(debugLog)(1, "mallocfree",
904                     "reclaimSuperblock at %p (pszB %7lu) %s owner %s/%s\n",
905                     sb, sb->n_payload_bytes,
906                     (sb->unsplittable ? "unsplittable" : ""),
907                     a->clientmem ? "CLIENT" : "VALGRIND", a->name );
908 
909    // Take into account admin bytes in the Superblock.
910    cszB = sizeof(Superblock) + sb->n_payload_bytes;
911 
912    // removes sb from superblock list.
913    for (i = 0; i < a->sblocks_used; i++) {
914       if (a->sblocks[i] == sb)
915          break;
916    }
917    vg_assert(i >= 0 && i < a->sblocks_used);
918    for (j = i; j < a->sblocks_used; j++)
919       a->sblocks[j] = a->sblocks[j+1];
920    a->sblocks_used--;
921    a->sblocks[a->sblocks_used] = NULL;
922    // paranoia: NULLify ptr to reclaimed sb or NULLify copy of ptr to last sb.
923 
924    a->stats__bytes_mmaped -= cszB;
925    if (sb->unsplittable)
926       a->stats__nreclaim_unsplit++;
927    else
928       a->stats__nreclaim_split++;
929 
930    // Now that the sb is removed from the list, mnumap its space.
931    if (a->clientmem) {
932       // reclaimable client allocation
933       Bool need_discard = False;
934       sres = VG_(am_munmap_client)(&need_discard, (Addr) sb, cszB);
935       vg_assert2(! sr_isError(sres), "superblock client munmap failure\n");
936       /* We somewhat help the client by discarding the range.
937          Note however that if the client has JITted some code in
938          a small block that was freed, we do not provide this
939          'discard support' */
940       /* JRS 2011-Sept-26: it would be nice to move the discard
941          outwards somewhat (in terms of calls) so as to make it easier
942          to verify that there will be no nonterminating recursive set
943          of calls a result of calling VG_(discard_translations).
944          Another day, perhaps. */
945       if (need_discard)
946          VG_(discard_translations) ((Addr) sb, cszB, "reclaimSuperblock");
947    } else {
948       // reclaimable non-client allocation
949       sres = VG_(am_munmap_valgrind)((Addr) sb, cszB);
950       vg_assert2(! sr_isError(sres), "superblock valgrind munmap failure\n");
951    }
952 
953 }
954 
955 // Find the superblock containing the given chunk.
956 static
findSb(Arena * a,Block * b)957 Superblock* findSb ( Arena* a, Block* b )
958 {
959    SizeT min = 0;
960    SizeT max = a->sblocks_used;
961 
962    while (min <= max) {
963       Superblock * sb;
964       SizeT pos = min + (max - min)/2;
965 
966       vg_assert(pos >= 0 && pos < a->sblocks_used);
967       sb = a->sblocks[pos];
968       if ((Block*)&sb->payload_bytes[0] <= b
969           && b < (Block*)&sb->payload_bytes[sb->n_payload_bytes])
970       {
971          return sb;
972       } else if ((Block*)&sb->payload_bytes[0] <= b) {
973          min = pos + 1;
974       } else {
975          max = pos - 1;
976       }
977    }
978    VG_(printf)("findSb: can't find pointer %p in arena '%s'\n",
979                 b, a->name );
980    VG_(core_panic)("findSb: VG_(arena_free)() in wrong arena?");
981    return NULL; /*NOTREACHED*/
982 }
983 
984 
985 // Find the superblock containing the given address.
986 // If superblock not found, return NULL.
987 static
maybe_findSb(Arena * a,Addr ad)988 Superblock* maybe_findSb ( Arena* a, Addr ad )
989 {
990    SizeT min = 0;
991    SizeT max = a->sblocks_used;
992 
993    while (min <= max) {
994       Superblock * sb;
995       SizeT pos = min + (max - min)/2;
996       if (pos < 0 || pos >= a->sblocks_used)
997          return NULL;
998       sb = a->sblocks[pos];
999       if ((Addr)&sb->payload_bytes[0] <= ad
1000           && ad < (Addr)&sb->payload_bytes[sb->n_payload_bytes]) {
1001          return sb;
1002       } else if ((Addr)&sb->payload_bytes[0] <= ad) {
1003          min = pos + 1;
1004       } else {
1005          max = pos - 1;
1006       }
1007    }
1008    return NULL;
1009 }
1010 
1011 
1012 /*------------------------------------------------------------*/
1013 /*--- Functions for working with freelists.                ---*/
1014 /*------------------------------------------------------------*/
1015 
1016 // Nb: Determination of which freelist a block lives on is based on the
1017 // payload size, not block size.
1018 
1019 // Convert a payload size in bytes to a freelist number.
1020 static __attribute__((noinline))
pszB_to_listNo_SLOW(SizeT pszB__divided_by__VG_MIN_MALLOC_SZB)1021 UInt pszB_to_listNo_SLOW ( SizeT pszB__divided_by__VG_MIN_MALLOC_SZB )
1022 {
1023    SizeT n = pszB__divided_by__VG_MIN_MALLOC_SZB;
1024 
1025    if (n < 299) {
1026       if (n < 114) {
1027          if (n < 85) {
1028             if (n < 74) {
1029                /* -- Exponential slope up, factor 1.05 -- */
1030                if (n < 67) return 64;
1031                if (n < 70) return 65;
1032                /* else */  return 66;
1033             } else {
1034                if (n < 77) return 67;
1035                if (n < 81) return 68;
1036                /* else */  return 69;
1037             }
1038          } else {
1039             if (n < 99) {
1040                if (n < 90) return 70;
1041                if (n < 94) return 71;
1042                /* else */  return 72;
1043             } else {
1044                if (n < 104) return 73;
1045                if (n < 109) return 74;
1046                /* else */   return 75;
1047             }
1048          }
1049       } else {
1050          if (n < 169) {
1051             if (n < 133) {
1052                if (n < 120) return 76;
1053                if (n < 126) return 77;
1054                /* else */   return 78;
1055             } else {
1056                if (n < 139) return 79;
1057                /* -- Exponential slope up, factor 1.10 -- */
1058                if (n < 153) return 80;
1059                /* else */   return 81;
1060             }
1061          } else {
1062             if (n < 224) {
1063                if (n < 185) return 82;
1064                if (n < 204) return 83;
1065                /* else */   return 84;
1066             } else {
1067                if (n < 247) return 85;
1068                if (n < 272) return 86;
1069                /* else */   return 87;
1070             }
1071          }
1072       }
1073    } else {
1074       if (n < 1331) {
1075          if (n < 530) {
1076             if (n < 398) {
1077                if (n < 329) return 88;
1078                if (n < 362) return 89;
1079                /* else */   return 90;
1080             } else {
1081                if (n < 438) return 91;
1082                if (n < 482) return 92;
1083                /* else */   return 93;
1084             }
1085          } else {
1086             if (n < 770) {
1087                if (n < 583) return 94;
1088                if (n < 641) return 95;
1089                /* -- Exponential slope up, factor 1.20 -- */
1090                /* else */   return 96;
1091             } else {
1092                if (n < 924) return 97;
1093                if (n < 1109) return 98;
1094                /* else */    return 99;
1095             }
1096          }
1097       } else {
1098          if (n < 3974) {
1099             if (n < 2300) {
1100                if (n < 1597) return 100;
1101                if (n < 1916) return 101;
1102                return 102;
1103             } else {
1104                if (n < 2760) return 103;
1105                if (n < 3312) return 104;
1106                /* else */    return 105;
1107             }
1108          } else {
1109             if (n < 6868) {
1110                if (n < 4769) return 106;
1111                if (n < 5723) return 107;
1112                /* else */    return 108;
1113             } else {
1114                if (n < 8241) return 109;
1115                if (n < 9890) return 110;
1116                /* else */    return 111;
1117             }
1118          }
1119       }
1120    }
1121    /*NOTREACHED*/
1122    vg_assert(0);
1123 }
1124 
1125 static inline
pszB_to_listNo(SizeT pszB)1126 UInt pszB_to_listNo ( SizeT pszB )
1127 {
1128    SizeT n = pszB / VG_MIN_MALLOC_SZB;
1129    vg_assert(0 == (pszB % VG_MIN_MALLOC_SZB));
1130 
1131    // The first 64 lists hold blocks of size VG_MIN_MALLOC_SZB * list_num.
1132    // The final 48 hold bigger blocks and are dealt with by the _SLOW
1133    // case.
1134    if (LIKELY(n < 64)) {
1135       return (UInt)n;
1136    } else {
1137       return pszB_to_listNo_SLOW(n);
1138    }
1139 }
1140 
1141 // What is the minimum payload size for a given list?
1142 static
listNo_to_pszB_min(UInt listNo)1143 SizeT listNo_to_pszB_min ( UInt listNo )
1144 {
1145    /* Repeatedly computing this function at every request is
1146       expensive.  Hence at the first call just cache the result for
1147       every possible argument. */
1148    static SizeT cache[N_MALLOC_LISTS];
1149    static Bool  cache_valid = False;
1150    if (!cache_valid) {
1151       UInt i;
1152       for (i = 0; i < N_MALLOC_LISTS; i++) {
1153          SizeT pszB = 0;
1154          while (pszB_to_listNo(pszB) < i)
1155             pszB += VG_MIN_MALLOC_SZB;
1156          cache[i] = pszB;
1157       }
1158       cache_valid = True;
1159    }
1160    /* Returned cached answer. */
1161    vg_assert(listNo <= N_MALLOC_LISTS);
1162    return cache[listNo];
1163 }
1164 
1165 // What is the maximum payload size for a given list?
1166 static
listNo_to_pszB_max(UInt listNo)1167 SizeT listNo_to_pszB_max ( UInt listNo )
1168 {
1169    vg_assert(listNo <= N_MALLOC_LISTS);
1170    if (listNo == N_MALLOC_LISTS-1) {
1171       return MAX_PSZB;
1172    } else {
1173       return listNo_to_pszB_min(listNo+1) - 1;
1174    }
1175 }
1176 
1177 
1178 /* A nasty hack to try and reduce fragmentation.  Try and replace
1179    a->freelist[lno] with another block on the same list but with a
1180    lower address, with the idea of attempting to recycle the same
1181    blocks rather than cruise through the address space. */
1182 static
swizzle(Arena * a,UInt lno)1183 void swizzle ( Arena* a, UInt lno )
1184 {
1185    Block* p_best;
1186    Block* pp;
1187    Block* pn;
1188    UInt   i;
1189 
1190    p_best = a->freelist[lno];
1191    if (p_best == NULL) return;
1192 
1193    pn = pp = p_best;
1194 
1195    // This loop bound was 20 for a long time, but experiments showed that
1196    // reducing it to 10 gave the same result in all the tests, and 5 got the
1197    // same result in 85--100% of cases.  And it's called often enough to be
1198    // noticeable in programs that allocated a lot.
1199    for (i = 0; i < 5; i++) {
1200       pn = get_next_b(pn);
1201       pp = get_prev_b(pp);
1202       if (pn < p_best) p_best = pn;
1203       if (pp < p_best) p_best = pp;
1204    }
1205    if (p_best < a->freelist[lno]) {
1206 #     ifdef VERBOSE_MALLOC
1207       VG_(printf)("retreat by %ld\n", (Word)(a->freelist[lno] - p_best));
1208 #     endif
1209       a->freelist[lno] = p_best;
1210    }
1211 }
1212 
1213 
1214 /*------------------------------------------------------------*/
1215 /*--- Sanity-check/debugging machinery.                    ---*/
1216 /*------------------------------------------------------------*/
1217 
1218 #define REDZONE_LO_MASK    0x31
1219 #define REDZONE_HI_MASK    0x7c
1220 
1221 // Do some crude sanity checks on a Block.
1222 static
blockSane(Arena * a,Block * b)1223 Bool blockSane ( Arena* a, Block* b )
1224 {
1225 #  define BLEAT(str) VG_(printf)("blockSane: fail -- %s\n",str)
1226    UInt i;
1227    // The lo and hi size fields will be checked (indirectly) by the call
1228    // to get_rz_hi_byte().
1229    if (!a->clientmem && is_inuse_block(b)) {
1230       // In the inner, for memcheck sake, temporarily mark redzone accessible.
1231       INNER_REQUEST(mkBhdrAccess(a,b));
1232       for (i = 0; i < a->rz_szB; i++) {
1233          if (get_rz_lo_byte(b, i) !=
1234             (UByte)(((Addr)b&0xff) ^ REDZONE_LO_MASK))
1235                {BLEAT("redzone-lo");return False;}
1236          if (get_rz_hi_byte(b, i) !=
1237             (UByte)(((Addr)b&0xff) ^ REDZONE_HI_MASK))
1238                {BLEAT("redzone-hi");return False;}
1239       }
1240       INNER_REQUEST(mkBhdrNoAccess(a,b));
1241    }
1242    return True;
1243 #  undef BLEAT
1244 }
1245 
1246 // Sanity checks on a Block inside an unsplittable superblock
1247 static
unsplittableBlockSane(Arena * a,Superblock * sb,Block * b)1248 Bool unsplittableBlockSane ( Arena* a, Superblock *sb, Block* b )
1249 {
1250 #  define BLEAT(str) VG_(printf)("unsplittableBlockSane: fail -- %s\n",str)
1251    Block*      other_b;
1252    UByte* sb_start;
1253    UByte* sb_end;
1254 
1255    if (!blockSane (a, b))
1256       {BLEAT("blockSane");return False;}
1257 
1258    if (sb->unsplittable != sb)
1259       {BLEAT("unsplittable");return False;}
1260 
1261    sb_start = &sb->payload_bytes[0];
1262    sb_end   = &sb->payload_bytes[sb->n_payload_bytes - 1];
1263 
1264    // b must be first block (i.e. no unused bytes at the beginning)
1265    if ((Block*)sb_start != b)
1266       {BLEAT("sb_start");return False;}
1267 
1268    // b must be last block (i.e. no unused bytes at the end)
1269    other_b = b + get_bszB(b);
1270    if (other_b-1 != (Block*)sb_end)
1271       {BLEAT("sb_end");return False;}
1272 
1273    return True;
1274 #  undef BLEAT
1275 }
1276 
1277 // Print superblocks (only for debugging).
1278 static
ppSuperblocks(Arena * a)1279 void ppSuperblocks ( Arena* a )
1280 {
1281    UInt i, j, blockno = 1;
1282    SizeT b_bszB;
1283 
1284    for (j = 0; j < a->sblocks_used; ++j) {
1285       Superblock * sb = a->sblocks[j];
1286 
1287       VG_(printf)( "\n" );
1288       VG_(printf)( "superblock %u at %p %s, sb->n_pl_bs = %lu\n",
1289                    blockno++, sb, (sb->unsplittable ? "unsplittable" : ""),
1290                    sb->n_payload_bytes);
1291       for (i = 0; i < sb->n_payload_bytes; i += b_bszB) {
1292          Block* b = (Block*)&sb->payload_bytes[i];
1293          b_bszB   = get_bszB(b);
1294          VG_(printf)( "   block at %u, bszB %lu: ", i, b_bszB );
1295          VG_(printf)( "%s, ", is_inuse_block(b) ? "inuse" : "free");
1296          VG_(printf)( "%s\n", blockSane(a, b) ? "ok" : "BAD" );
1297       }
1298       vg_assert(i == sb->n_payload_bytes);   // no overshoot at end of Sb
1299    }
1300    VG_(printf)( "end of superblocks\n\n" );
1301 }
1302 
1303 // Sanity check both the superblocks and the chains.
sanity_check_malloc_arena(ArenaId aid)1304 static void sanity_check_malloc_arena ( ArenaId aid )
1305 {
1306    UInt        i, j, superblockctr, blockctr_sb, blockctr_li;
1307    UInt        blockctr_sb_free, listno;
1308    SizeT       b_bszB, b_pszB, list_min_pszB, list_max_pszB;
1309    Bool        thisFree, lastWasFree, sblockarrOK;
1310    Block*      b;
1311    Block*      b_prev;
1312    SizeT       arena_bytes_on_loan;
1313    Arena*      a;
1314 
1315 #  define BOMB VG_(core_panic)("sanity_check_malloc_arena")
1316 
1317    a = arenaId_to_ArenaP(aid);
1318 
1319    // Check the superblock array.
1320    sblockarrOK
1321       = a->sblocks != NULL
1322         && a->sblocks_size >= SBLOCKS_SIZE_INITIAL
1323         && a->sblocks_used <= a->sblocks_size
1324         && (a->sblocks_size == SBLOCKS_SIZE_INITIAL
1325             ? (a->sblocks == &a->sblocks_initial[0])
1326             : (a->sblocks != &a->sblocks_initial[0]));
1327    if (!sblockarrOK) {
1328       VG_(printf)("sanity_check_malloc_arena: sblock array BAD\n");
1329       BOMB;
1330    }
1331 
1332    // First, traverse all the superblocks, inspecting the Blocks in each.
1333    superblockctr = blockctr_sb = blockctr_sb_free = 0;
1334    arena_bytes_on_loan = 0;
1335    for (j = 0; j < a->sblocks_used; ++j) {
1336       Superblock * sb = a->sblocks[j];
1337       lastWasFree = False;
1338       superblockctr++;
1339       for (i = 0; i < sb->n_payload_bytes; i += mk_plain_bszB(b_bszB)) {
1340          blockctr_sb++;
1341          b     = (Block*)&sb->payload_bytes[i];
1342          b_bszB = get_bszB_as_is(b);
1343          if (!blockSane(a, b)) {
1344             VG_(printf)("sanity_check_malloc_arena: sb %p, block %u "
1345                         "(bszB %lu):  BAD\n", sb, i, b_bszB );
1346             BOMB;
1347          }
1348          thisFree = !is_inuse_block(b);
1349          if (thisFree && lastWasFree) {
1350             VG_(printf)("sanity_check_malloc_arena: sb %p, block %u "
1351                         "(bszB %lu): UNMERGED FREES\n", sb, i, b_bszB );
1352             BOMB;
1353          }
1354          if (thisFree) blockctr_sb_free++;
1355          if (!thisFree)
1356             arena_bytes_on_loan += bszB_to_pszB(a, b_bszB);
1357          lastWasFree = thisFree;
1358       }
1359       if (i > sb->n_payload_bytes) {
1360          VG_(printf)( "sanity_check_malloc_arena: sb %p: last block "
1361                       "overshoots end\n", sb);
1362          BOMB;
1363       }
1364    }
1365 
1366    arena_bytes_on_loan += a->stats__perm_bytes_on_loan;
1367 
1368    if (arena_bytes_on_loan != a->stats__bytes_on_loan) {
1369 #     ifdef VERBOSE_MALLOC
1370       VG_(printf)( "sanity_check_malloc_arena: a->bytes_on_loan %lu, "
1371                    "arena_bytes_on_loan %lu: "
1372                    "MISMATCH\n", a->stats__bytes_on_loan, arena_bytes_on_loan);
1373 #     endif
1374       ppSuperblocks(a);
1375       BOMB;
1376    }
1377 
1378    /* Second, traverse each list, checking that the back pointers make
1379       sense, counting blocks encountered, and checking that each block
1380       is an appropriate size for this list. */
1381    blockctr_li = 0;
1382    for (listno = 0; listno < N_MALLOC_LISTS; listno++) {
1383       list_min_pszB = listNo_to_pszB_min(listno);
1384       list_max_pszB = listNo_to_pszB_max(listno);
1385       b = a->freelist[listno];
1386       if (b == NULL) continue;
1387       while (True) {
1388          b_prev = b;
1389          b = get_next_b(b);
1390          if (get_prev_b(b) != b_prev) {
1391             VG_(printf)( "sanity_check_malloc_arena: list %u at %p: "
1392                          "BAD LINKAGE\n",
1393                          listno, b );
1394             BOMB;
1395          }
1396          b_pszB = get_pszB(a, b);
1397          if (b_pszB < list_min_pszB || b_pszB > list_max_pszB) {
1398             VG_(printf)(
1399                "sanity_check_malloc_arena: list %u at %p: "
1400                "WRONG CHAIN SIZE %luB (%luB, %luB)\n",
1401                listno, b, b_pszB, list_min_pszB, list_max_pszB );
1402             BOMB;
1403          }
1404          blockctr_li++;
1405          if (b == a->freelist[listno]) break;
1406       }
1407    }
1408 
1409    if (blockctr_sb_free != blockctr_li) {
1410 #     ifdef VERBOSE_MALLOC
1411       VG_(printf)( "sanity_check_malloc_arena: BLOCK COUNT MISMATCH "
1412                    "(via sbs %d, via lists %d)\n",
1413                    blockctr_sb_free, blockctr_li );
1414 #     endif
1415       ppSuperblocks(a);
1416       BOMB;
1417    }
1418 
1419    if (VG_(clo_verbosity) > 2)
1420       VG_(message)(Vg_DebugMsg,
1421                    "%-8s: %2u sbs, %5u bs, %2u/%-2u free bs, "
1422                    "%7lu mmap, %7lu loan\n",
1423                    a->name,
1424                    superblockctr,
1425                    blockctr_sb, blockctr_sb_free, blockctr_li,
1426                    a->stats__bytes_mmaped, a->stats__bytes_on_loan);
1427 #  undef BOMB
1428 }
1429 
1430 
1431 #define N_AN_CCS 1000
1432 
1433 typedef struct {
1434    ULong nBytes;
1435    ULong nBlocks;
1436    const HChar* cc;
1437 } AnCC;
1438 
1439 static AnCC anCCs[N_AN_CCS];
1440 
1441 /* Sorting by decreasing cost center nBytes, to have the biggest
1442    cost centres at the top. */
cmp_AnCC_by_vol(const void * v1,const void * v2)1443 static Int cmp_AnCC_by_vol ( const void* v1, const void* v2 ) {
1444    const AnCC* ancc1 = v1;
1445    const AnCC* ancc2 = v2;
1446    if (ancc1->nBytes < ancc2->nBytes) return 1;
1447    if (ancc1->nBytes > ancc2->nBytes) return -1;
1448    return 0;
1449 }
1450 
cc_analyse_alloc_arena(ArenaId aid)1451 static void cc_analyse_alloc_arena ( ArenaId aid )
1452 {
1453    Word i, j, k;
1454    Arena*      a;
1455    Block*      b;
1456    Bool        thisFree, lastWasFree;
1457    SizeT       b_bszB;
1458 
1459    const HChar* cc;
1460    UInt n_ccs = 0;
1461    //return;
1462    a = arenaId_to_ArenaP(aid);
1463    if (a->name == NULL) {
1464       /* arena is not in use, is not initialised and will fail the
1465          sanity check that follows. */
1466       return;
1467    }
1468 
1469    sanity_check_malloc_arena(aid);
1470 
1471    VG_(printf)(
1472       "-------- Arena \"%s\": %'lu/%'lu max/curr mmap'd, "
1473       "%llu/%llu unsplit/split sb unmmap'd, "
1474       "%'lu/%'lu max/curr on_loan %lu rzB --------\n",
1475       a->name, a->stats__bytes_mmaped_max, a->stats__bytes_mmaped,
1476       a->stats__nreclaim_unsplit, a->stats__nreclaim_split,
1477       a->stats__bytes_on_loan_max, a->stats__bytes_on_loan,
1478       a->rz_szB
1479    );
1480 
1481    for (j = 0; j < a->sblocks_used; ++j) {
1482       Superblock * sb = a->sblocks[j];
1483       lastWasFree = False;
1484       for (i = 0; i < sb->n_payload_bytes; i += mk_plain_bszB(b_bszB)) {
1485          b     = (Block*)&sb->payload_bytes[i];
1486          b_bszB = get_bszB_as_is(b);
1487          if (!blockSane(a, b)) {
1488             VG_(printf)("sanity_check_malloc_arena: sb %p, block %ld "
1489                         "(bszB %lu):  BAD\n", sb, i, b_bszB );
1490             vg_assert(0);
1491          }
1492          thisFree = !is_inuse_block(b);
1493          if (thisFree && lastWasFree) {
1494             VG_(printf)("sanity_check_malloc_arena: sb %p, block %ld "
1495                         "(bszB %lu): UNMERGED FREES\n", sb, i, b_bszB );
1496             vg_assert(0);
1497          }
1498          lastWasFree = thisFree;
1499 
1500          if (thisFree) continue;
1501 
1502          if (VG_(clo_profile_heap))
1503             cc = get_cc(b);
1504          else
1505             cc = "(--profile-heap=yes for details)";
1506          if (0)
1507          VG_(printf)("block: inUse=%d pszB=%d cc=%s\n",
1508                      (Int)(!thisFree),
1509                      (Int)bszB_to_pszB(a, b_bszB),
1510                      get_cc(b));
1511          vg_assert(cc);
1512          for (k = 0; k < n_ccs; k++) {
1513            vg_assert(anCCs[k].cc);
1514             if (0 == VG_(strcmp)(cc, anCCs[k].cc))
1515                break;
1516          }
1517          vg_assert(k >= 0 && k <= n_ccs);
1518 
1519          if (k == n_ccs) {
1520             vg_assert(n_ccs < N_AN_CCS-1);
1521             n_ccs++;
1522             anCCs[k].nBytes  = 0;
1523             anCCs[k].nBlocks = 0;
1524             anCCs[k].cc      = cc;
1525          }
1526 
1527          vg_assert(k >= 0 && k < n_ccs && k < N_AN_CCS);
1528          anCCs[k].nBytes += (ULong)bszB_to_pszB(a, b_bszB);
1529          anCCs[k].nBlocks++;
1530       }
1531       if (i > sb->n_payload_bytes) {
1532          VG_(printf)( "sanity_check_malloc_arena: sb %p: last block "
1533                       "overshoots end\n", sb);
1534          vg_assert(0);
1535       }
1536    }
1537 
1538    if (a->stats__perm_bytes_on_loan > 0) {
1539       vg_assert(n_ccs < N_AN_CCS-1);
1540       anCCs[n_ccs].nBytes  = a->stats__perm_bytes_on_loan;
1541       anCCs[n_ccs].nBlocks = a->stats__perm_blocks;
1542       anCCs[n_ccs].cc      = "perm_malloc";
1543       n_ccs++;
1544    }
1545 
1546    VG_(ssort)( &anCCs[0], n_ccs, sizeof(anCCs[0]), cmp_AnCC_by_vol );
1547 
1548    for (k = 0; k < n_ccs; k++) {
1549       VG_(printf)("%'13llu in %'9llu: %s\n",
1550                   anCCs[k].nBytes, anCCs[k].nBlocks, anCCs[k].cc );
1551    }
1552 
1553    VG_(printf)("\n");
1554 }
1555 
1556 
VG_(sanity_check_malloc_all)1557 void VG_(sanity_check_malloc_all) ( void )
1558 {
1559    UInt i;
1560    for (i = 0; i < VG_N_ARENAS; i++) {
1561       if (i == VG_AR_CLIENT && !client_inited)
1562          continue;
1563       sanity_check_malloc_arena ( i );
1564    }
1565 }
1566 
VG_(describe_arena_addr)1567 void VG_(describe_arena_addr) ( Addr a, AddrArenaInfo* aai )
1568 {
1569    UInt i;
1570    Superblock *sb;
1571    Arena      *arena;
1572 
1573    for (i = 0; i < VG_N_ARENAS; i++) {
1574       if (i == VG_AR_CLIENT && !client_inited)
1575          continue;
1576       arena = arenaId_to_ArenaP(i);
1577       sb = maybe_findSb( arena, a );
1578       if (sb != NULL) {
1579          Word   j;
1580          SizeT  b_bszB;
1581          Block *b = NULL;
1582 
1583          aai->aid = i;
1584          aai->name = arena->name;
1585          for (j = 0; j < sb->n_payload_bytes; j += mk_plain_bszB(b_bszB)) {
1586             b     = (Block*)&sb->payload_bytes[j];
1587             b_bszB = get_bszB_as_is(b);
1588             if (a < (Addr)b + mk_plain_bszB(b_bszB))
1589                break;
1590          }
1591          vg_assert (b);
1592          aai->block_szB = get_pszB(arena, b);
1593          aai->rwoffset = a - (Addr)get_block_payload(arena, b);
1594          aai->free = !is_inuse_block(b);
1595          return;
1596       }
1597    }
1598    aai->aid = 0;
1599    aai->name = NULL;
1600    aai->block_szB = 0;
1601    aai->rwoffset = 0;
1602    aai->free = False;
1603 }
1604 
1605 /*------------------------------------------------------------*/
1606 /*--- Creating and deleting blocks.                        ---*/
1607 /*------------------------------------------------------------*/
1608 
1609 // Mark the bytes at b .. b+bszB-1 as not in use, and add them to the
1610 // relevant free list.
1611 
1612 static
mkFreeBlock(Arena * a,Block * b,SizeT bszB,UInt b_lno)1613 void mkFreeBlock ( Arena* a, Block* b, SizeT bszB, UInt b_lno )
1614 {
1615    SizeT pszB = bszB_to_pszB(a, bszB);
1616    vg_assert(b_lno == pszB_to_listNo(pszB));
1617    INNER_REQUEST(VALGRIND_MAKE_MEM_UNDEFINED(b, bszB));
1618    // Set the size fields and indicate not-in-use.
1619    set_bszB(b, mk_free_bszB(bszB));
1620 
1621    // Add to the relevant list.
1622    if (a->freelist[b_lno] == NULL) {
1623       set_prev_b(b, b);
1624       set_next_b(b, b);
1625       a->freelist[b_lno] = b;
1626    } else {
1627       Block* b_prev = get_prev_b(a->freelist[b_lno]);
1628       Block* b_next = a->freelist[b_lno];
1629       set_next_b(b_prev, b);
1630       set_prev_b(b_next, b);
1631       set_next_b(b, b_next);
1632       set_prev_b(b, b_prev);
1633    }
1634 #  ifdef DEBUG_MALLOC
1635    (void)blockSane(a,b);
1636 #  endif
1637 }
1638 
1639 // Mark the bytes at b .. b+bszB-1 as in use, and set up the block
1640 // appropriately.
1641 static
mkInuseBlock(Arena * a,Block * b,SizeT bszB)1642 void mkInuseBlock ( Arena* a, Block* b, SizeT bszB )
1643 {
1644    UInt i;
1645    vg_assert(bszB >= min_useful_bszB(a));
1646    INNER_REQUEST(VALGRIND_MAKE_MEM_UNDEFINED(b, bszB));
1647    set_bszB(b, mk_inuse_bszB(bszB));
1648    set_prev_b(b, NULL);    // Take off freelist
1649    set_next_b(b, NULL);    // ditto
1650    if (!a->clientmem) {
1651       for (i = 0; i < a->rz_szB; i++) {
1652          set_rz_lo_byte(b, i, (UByte)(((Addr)b&0xff) ^ REDZONE_LO_MASK));
1653          set_rz_hi_byte(b, i, (UByte)(((Addr)b&0xff) ^ REDZONE_HI_MASK));
1654       }
1655    }
1656 #  ifdef DEBUG_MALLOC
1657    (void)blockSane(a,b);
1658 #  endif
1659 }
1660 
1661 // Mark the bytes at b .. b+bszB-1 as being part of a block that has been shrunk.
1662 static
shrinkInuseBlock(Arena * a,Block * b,SizeT bszB)1663 void shrinkInuseBlock ( Arena* a, Block* b, SizeT bszB )
1664 {
1665    UInt i;
1666 
1667    vg_assert(bszB >= min_useful_bszB(a));
1668    INNER_REQUEST(mkBhdrAccess(a,b));
1669    set_bszB(b, mk_inuse_bszB(bszB));
1670    if (!a->clientmem) {
1671       for (i = 0; i < a->rz_szB; i++) {
1672          set_rz_lo_byte(b, i, (UByte)(((Addr)b&0xff) ^ REDZONE_LO_MASK));
1673          set_rz_hi_byte(b, i, (UByte)(((Addr)b&0xff) ^ REDZONE_HI_MASK));
1674       }
1675    }
1676    INNER_REQUEST(mkBhdrNoAccess(a,b));
1677 
1678 #  ifdef DEBUG_MALLOC
1679    (void)blockSane(a,b);
1680 #  endif
1681 }
1682 
1683 // Remove a block from a given list.  Does no sanity checking.
1684 static
unlinkBlock(Arena * a,Block * b,UInt listno)1685 void unlinkBlock ( Arena* a, Block* b, UInt listno )
1686 {
1687    vg_assert(listno < N_MALLOC_LISTS);
1688    if (get_prev_b(b) == b) {
1689       // Only one element in the list; treat it specially.
1690       vg_assert(get_next_b(b) == b);
1691       a->freelist[listno] = NULL;
1692    } else {
1693       Block* b_prev = get_prev_b(b);
1694       Block* b_next = get_next_b(b);
1695       a->freelist[listno] = b_prev;
1696       set_next_b(b_prev, b_next);
1697       set_prev_b(b_next, b_prev);
1698       swizzle ( a, listno );
1699    }
1700    set_prev_b(b, NULL);
1701    set_next_b(b, NULL);
1702 }
1703 
1704 
1705 /*------------------------------------------------------------*/
1706 /*--- Core-visible functions.                              ---*/
1707 /*------------------------------------------------------------*/
1708 
1709 // Align the request size.
1710 static __inline__
align_req_pszB(SizeT req_pszB)1711 SizeT align_req_pszB ( SizeT req_pszB )
1712 {
1713    SizeT n = VG_MIN_MALLOC_SZB-1;
1714    return ((req_pszB + n) & (~n));
1715 }
1716 
1717 static
add_one_block_to_stats(Arena * a,SizeT loaned)1718 void add_one_block_to_stats (Arena* a, SizeT loaned)
1719 {
1720    a->stats__bytes_on_loan += loaned;
1721    if (a->stats__bytes_on_loan > a->stats__bytes_on_loan_max) {
1722       a->stats__bytes_on_loan_max = a->stats__bytes_on_loan;
1723       if (a->stats__bytes_on_loan_max >= a->next_profile_at) {
1724          /* next profile after 10% more growth */
1725          a->next_profile_at
1726             = (SizeT)(
1727                  (((ULong)a->stats__bytes_on_loan_max) * 105ULL) / 100ULL );
1728          if (VG_(clo_profile_heap))
1729             cc_analyse_alloc_arena(arenaP_to_ArenaId (a));
1730       }
1731    }
1732    a->stats__tot_blocks += (ULong)1;
1733    a->stats__tot_bytes  += (ULong)loaned;
1734 }
1735 
1736 /* Allocate a piece of memory of req_pszB bytes on the given arena.
1737    The function may return NULL if (and only if) aid == VG_AR_CLIENT.
1738    Otherwise, the function returns a non-NULL value. */
VG_(arena_malloc)1739 void* VG_(arena_malloc) ( ArenaId aid, const HChar* cc, SizeT req_pszB )
1740 {
1741    SizeT       req_bszB, frag_bszB, b_bszB;
1742    UInt        lno, i;
1743    Superblock* new_sb = NULL;
1744    Block*      b = NULL;
1745    Arena*      a;
1746    void*       v;
1747    UWord       stats__nsearches = 0;
1748 
1749    ensure_mm_init(aid);
1750    a = arenaId_to_ArenaP(aid);
1751 
1752    vg_assert(req_pszB < MAX_PSZB);
1753    req_pszB = align_req_pszB(req_pszB);
1754    req_bszB = pszB_to_bszB(a, req_pszB);
1755 
1756    // You must provide a cost-center name against which to charge
1757    // this allocation; it isn't optional.
1758    vg_assert(cc);
1759 
1760    // Scan through all the big-enough freelists for a block.
1761    //
1762    // Nb: this scanning might be expensive in some cases.  Eg. if you
1763    // allocate lots of small objects without freeing them, but no
1764    // medium-sized objects, it will repeatedly scanning through the whole
1765    // list, and each time not find any free blocks until the last element.
1766    //
1767    // If this becomes a noticeable problem... the loop answers the question
1768    // "where is the first nonempty list above me?"  And most of the time,
1769    // you ask the same question and get the same answer.  So it would be
1770    // good to somehow cache the results of previous searches.
1771    // One possibility is an array (with N_MALLOC_LISTS elements) of
1772    // shortcuts.  shortcut[i] would give the index number of the nearest
1773    // larger list above list i which is non-empty.  Then this loop isn't
1774    // necessary.  However, we'd have to modify some section [ .. i-1] of the
1775    // shortcut array every time a list [i] changes from empty to nonempty or
1776    // back.  This would require care to avoid pathological worst-case
1777    // behaviour.
1778    //
1779    for (lno = pszB_to_listNo(req_pszB); lno < N_MALLOC_LISTS; lno++) {
1780       UWord nsearches_this_level = 0;
1781       b = a->freelist[lno];
1782       if (NULL == b) continue;   // If this list is empty, try the next one.
1783       while (True) {
1784          stats__nsearches++;
1785          nsearches_this_level++;
1786          if (UNLIKELY(nsearches_this_level >= 100)
1787              && lno < N_MALLOC_LISTS-1) {
1788             /* Avoid excessive scanning on this freelist, and instead
1789                try the next one up.  But first, move this freelist's
1790                start pointer one element along, so as to ensure that
1791                subsequent searches of this list don't endlessly
1792                revisit only these 100 elements, but in fact slowly
1793                progress through the entire list. */
1794             b = a->freelist[lno];
1795             vg_assert(b); // this list must be nonempty!
1796             a->freelist[lno] = get_next_b(b); // step one along
1797             break;
1798          }
1799          b_bszB = get_bszB(b);
1800          if (b_bszB >= req_bszB) goto obtained_block;    // success!
1801          b = get_next_b(b);
1802          if (b == a->freelist[lno]) break;   // traversed entire freelist
1803       }
1804    }
1805 
1806    // If we reach here, no suitable block found, allocate a new superblock
1807    vg_assert(lno == N_MALLOC_LISTS);
1808    new_sb = newSuperblock(a, req_bszB);
1809    if (NULL == new_sb) {
1810       // Should only fail if for client, otherwise, should have aborted
1811       // already.
1812       vg_assert(VG_AR_CLIENT == aid);
1813       return NULL;
1814    }
1815 
1816    vg_assert(a->sblocks_used <= a->sblocks_size);
1817    if (a->sblocks_used == a->sblocks_size) {
1818       Superblock ** array;
1819       SysRes sres = VG_(am_mmap_anon_float_valgrind)(sizeof(Superblock *) *
1820                                                      a->sblocks_size * 2);
1821       if (sr_isError(sres)) {
1822          VG_(out_of_memory_NORETURN)("arena_init", sizeof(Superblock *) *
1823                                                    a->sblocks_size * 2);
1824          /* NOTREACHED */
1825       }
1826       array = (Superblock**)(Addr)sr_Res(sres);
1827       for (i = 0; i < a->sblocks_used; ++i) array[i] = a->sblocks[i];
1828 
1829       a->sblocks_size *= 2;
1830       a->sblocks = array;
1831       VG_(debugLog)(1, "mallocfree",
1832                        "sblock array for arena `%s' resized to %lu\n",
1833                        a->name, a->sblocks_size);
1834    }
1835 
1836    vg_assert(a->sblocks_used < a->sblocks_size);
1837 
1838    i = a->sblocks_used;
1839    while (i > 0) {
1840       if (a->sblocks[i-1] > new_sb) {
1841          a->sblocks[i] = a->sblocks[i-1];
1842       } else {
1843          break;
1844       }
1845       --i;
1846    }
1847    a->sblocks[i] = new_sb;
1848    a->sblocks_used++;
1849 
1850    b = (Block*)&new_sb->payload_bytes[0];
1851    lno = pszB_to_listNo(bszB_to_pszB(a, new_sb->n_payload_bytes));
1852    mkFreeBlock ( a, b, new_sb->n_payload_bytes, lno);
1853    if (VG_(clo_profile_heap))
1854       set_cc(b, "admin.free-new-sb-1");
1855    // fall through
1856 
1857   obtained_block:
1858    // Ok, we can allocate from b, which lives in list lno.
1859    vg_assert(b != NULL);
1860    vg_assert(lno < N_MALLOC_LISTS);
1861    vg_assert(a->freelist[lno] != NULL);
1862    b_bszB = get_bszB(b);
1863    // req_bszB is the size of the block we are after.  b_bszB is the
1864    // size of what we've actually got. */
1865    vg_assert(b_bszB >= req_bszB);
1866 
1867    // Could we split this block and still get a useful fragment?
1868    // A block in an unsplittable superblock can never be split.
1869    frag_bszB = b_bszB - req_bszB;
1870    if (frag_bszB >= min_useful_bszB(a)
1871        && (NULL == new_sb || ! new_sb->unsplittable)) {
1872       // Yes, split block in two, put the fragment on the appropriate free
1873       // list, and update b_bszB accordingly.
1874       // printf( "split %dB into %dB and %dB\n", b_bszB, req_bszB, frag_bszB );
1875       unlinkBlock(a, b, lno);
1876       mkInuseBlock(a, b, req_bszB);
1877       if (VG_(clo_profile_heap))
1878          set_cc(b, cc);
1879       mkFreeBlock(a, &b[req_bszB], frag_bszB,
1880                      pszB_to_listNo(bszB_to_pszB(a, frag_bszB)));
1881       if (VG_(clo_profile_heap))
1882          set_cc(&b[req_bszB], "admin.fragmentation-1");
1883       b_bszB = get_bszB(b);
1884    } else {
1885       // No, mark as in use and use as-is.
1886       unlinkBlock(a, b, lno);
1887       mkInuseBlock(a, b, b_bszB);
1888       if (VG_(clo_profile_heap))
1889          set_cc(b, cc);
1890    }
1891 
1892    // Update stats
1893    SizeT loaned = bszB_to_pszB(a, b_bszB);
1894    add_one_block_to_stats (a, loaned);
1895    a->stats__nsearches  += (ULong)stats__nsearches;
1896 
1897 #  ifdef DEBUG_MALLOC
1898    sanity_check_malloc_arena(aid);
1899 #  endif
1900 
1901    v = get_block_payload(a, b);
1902    vg_assert( (((Addr)v) & (VG_MIN_MALLOC_SZB-1)) == 0 );
1903 
1904    // Which size should we pass to VALGRIND_MALLOCLIKE_BLOCK ?
1905    // We have 2 possible options:
1906    // 1. The final resulting usable size.
1907    // 2. The initial (non-aligned) req_pszB.
1908    // Memcheck implements option 2 easily, as the initial requested size
1909    // is maintained in the mc_chunk data structure.
1910    // This is not as easy in the core, as there is no such structure.
1911    // (note: using the aligned req_pszB is not simpler than 2, as
1912    //  requesting an aligned req_pszB might still be satisfied by returning
1913    // a (slightly) bigger block than requested if the remaining part of
1914    // of a free block is not big enough to make a free block by itself).
1915    // Implement Sol 2 can be done the following way:
1916    // After having called VALGRIND_MALLOCLIKE_BLOCK, the non accessible
1917    // redzone just after the block can be used to determine the
1918    // initial requested size.
1919    // Currently, not implemented => we use Option 1.
1920    INNER_REQUEST
1921       (VALGRIND_MALLOCLIKE_BLOCK(v,
1922                                  VG_(arena_malloc_usable_size)(aid, v),
1923                                  a->rz_szB, False));
1924 
1925    /* For debugging/testing purposes, fill the newly allocated area
1926       with a definite value in an attempt to shake out any
1927       uninitialised uses of the data (by V core / V tools, not by the
1928       client).  Testing on 25 Nov 07 with the values 0x00, 0xFF, 0x55,
1929       0xAA showed no differences in the regression tests on
1930       amd64-linux.  Note, is disabled by default. */
1931    if (0 && aid != VG_AR_CLIENT)
1932       VG_(memset)(v, 0xAA, (SizeT)req_pszB);
1933 
1934    return v;
1935 }
1936 
1937 // If arena has already a deferred reclaimed superblock and
1938 // this superblock is still reclaimable, then this superblock is first
1939 // reclaimed.
1940 // sb becomes then the new arena deferred superblock.
1941 // Passing NULL as sb allows to reclaim a deferred sb without setting a new
1942 // deferred reclaim.
1943 static
deferred_reclaimSuperblock(Arena * a,Superblock * sb)1944 void deferred_reclaimSuperblock ( Arena* a, Superblock* sb)
1945 {
1946 
1947    if (sb == NULL) {
1948       if (!a->deferred_reclaimed_sb)
1949          // no deferred sb to reclaim now, nothing to do in the future =>
1950          // return directly.
1951          return;
1952 
1953       VG_(debugLog)(1, "mallocfree",
1954                     "deferred_reclaimSuperblock NULL "
1955                     "(prev %p) owner %s/%s\n",
1956                     a->deferred_reclaimed_sb,
1957                     a->clientmem ? "CLIENT" : "VALGRIND", a->name );
1958    } else
1959       VG_(debugLog)(1, "mallocfree",
1960                     "deferred_reclaimSuperblock at %p (pszB %7lu) %s "
1961                     "(prev %p) owner %s/%s\n",
1962                     sb, sb->n_payload_bytes,
1963                     (sb->unsplittable ? "unsplittable" : ""),
1964                     a->deferred_reclaimed_sb,
1965                     a->clientmem ? "CLIENT" : "VALGRIND", a->name );
1966 
1967    if (a->deferred_reclaimed_sb && a->deferred_reclaimed_sb != sb) {
1968       // If we are deferring another block that the current block deferred,
1969       // then if this block can stil be reclaimed, reclaim it now.
1970       // Note that we might have a re-deferred reclaim of the same block
1971       // with a sequence: free (causing a deferred reclaim of sb)
1972       //                  alloc (using a piece of memory of the deferred sb)
1973       //                  free of the just alloc-ed block (causing a re-defer).
1974       UByte*      def_sb_start;
1975       UByte*      def_sb_end;
1976       Superblock* def_sb;
1977       Block*      b;
1978 
1979       def_sb = a->deferred_reclaimed_sb;
1980       def_sb_start = &def_sb->payload_bytes[0];
1981       def_sb_end   = &def_sb->payload_bytes[def_sb->n_payload_bytes - 1];
1982       b = (Block *)def_sb_start;
1983       vg_assert (blockSane(a, b));
1984 
1985       // Check if the deferred_reclaimed_sb is still reclaimable.
1986       // If yes, we will execute the reclaim.
1987       if (!is_inuse_block(b)) {
1988          // b (at the beginning of def_sb) is not in use.
1989          UInt        b_listno;
1990          SizeT       b_bszB, b_pszB;
1991          b_bszB   = get_bszB(b);
1992          b_pszB   = bszB_to_pszB(a, b_bszB);
1993          if (b + b_bszB-1 == (Block*)def_sb_end) {
1994             // b (not in use) covers the full superblock.
1995             // => def_sb is still reclaimable
1996             // => execute now the reclaim of this def_sb.
1997             b_listno = pszB_to_listNo(b_pszB);
1998             unlinkBlock( a, b, b_listno );
1999             reclaimSuperblock (a, def_sb);
2000             a->deferred_reclaimed_sb = NULL;
2001          }
2002       }
2003    }
2004 
2005    // sb (possibly NULL) becomes the new deferred reclaimed superblock.
2006    a->deferred_reclaimed_sb = sb;
2007 }
2008 
2009 /* b must be a free block, of size b_bszB.
2010    If b is followed by another free block, merge them.
2011    If b is preceded by another free block, merge them.
2012    If the merge results in the superblock being fully free,
2013    deferred_reclaimSuperblock the superblock. */
mergeWithFreeNeighbours(Arena * a,Superblock * sb,Block * b,SizeT b_bszB)2014 static void mergeWithFreeNeighbours (Arena* a, Superblock* sb,
2015                                      Block* b, SizeT b_bszB)
2016 {
2017    UByte*      sb_start;
2018    UByte*      sb_end;
2019    Block*      other_b;
2020    SizeT       other_bszB;
2021    UInt        b_listno;
2022 
2023    sb_start = &sb->payload_bytes[0];
2024    sb_end   = &sb->payload_bytes[sb->n_payload_bytes - 1];
2025 
2026    b_listno = pszB_to_listNo(bszB_to_pszB(a, b_bszB));
2027 
2028    // See if this block can be merged with its successor.
2029    // First test if we're far enough before the superblock's end to possibly
2030    // have a successor.
2031    other_b = b + b_bszB;
2032    if (other_b+min_useful_bszB(a)-1 <= (Block*)sb_end) {
2033       // Ok, we have a successor, merge if it's not in use.
2034       other_bszB = get_bszB(other_b);
2035       if (!is_inuse_block(other_b)) {
2036          // VG_(printf)( "merge-successor\n");
2037 #        ifdef DEBUG_MALLOC
2038          vg_assert(blockSane(a, other_b));
2039 #        endif
2040          unlinkBlock( a, b, b_listno );
2041          unlinkBlock( a, other_b,
2042                       pszB_to_listNo(bszB_to_pszB(a,other_bszB)) );
2043          b_bszB += other_bszB;
2044          b_listno = pszB_to_listNo(bszB_to_pszB(a, b_bszB));
2045          mkFreeBlock( a, b, b_bszB, b_listno );
2046          if (VG_(clo_profile_heap))
2047             set_cc(b, "admin.free-2");
2048       }
2049    } else {
2050       // Not enough space for successor: check that b is the last block
2051       // ie. there are no unused bytes at the end of the Superblock.
2052       vg_assert(other_b-1 == (Block*)sb_end);
2053    }
2054 
2055    // Then see if this block can be merged with its predecessor.
2056    // First test if we're far enough after the superblock's start to possibly
2057    // have a predecessor.
2058    if (b >= (Block*)sb_start + min_useful_bszB(a)) {
2059       // Ok, we have a predecessor, merge if it's not in use.
2060       other_b = get_predecessor_block( b );
2061       other_bszB = get_bszB(other_b);
2062       if (!is_inuse_block(other_b)) {
2063          // VG_(printf)( "merge-predecessor\n");
2064          unlinkBlock( a, b, b_listno );
2065          unlinkBlock( a, other_b,
2066                       pszB_to_listNo(bszB_to_pszB(a, other_bszB)) );
2067          b = other_b;
2068          b_bszB += other_bszB;
2069          b_listno = pszB_to_listNo(bszB_to_pszB(a, b_bszB));
2070          mkFreeBlock( a, b, b_bszB, b_listno );
2071          if (VG_(clo_profile_heap))
2072             set_cc(b, "admin.free-3");
2073       }
2074    } else {
2075       // Not enough space for predecessor: check that b is the first block,
2076       // ie. there are no unused bytes at the start of the Superblock.
2077       vg_assert((Block*)sb_start == b);
2078    }
2079 
2080    /* If the block b just merged is the only block of the superblock sb,
2081       then we defer reclaim sb. */
2082    if ( ((Block*)sb_start == b) && (b + b_bszB-1 == (Block*)sb_end) ) {
2083       deferred_reclaimSuperblock (a, sb);
2084    }
2085 }
2086 
VG_(arena_free)2087 void VG_(arena_free) ( ArenaId aid, void* ptr )
2088 {
2089    Superblock* sb;
2090    Block*      b;
2091    SizeT       b_bszB, b_pszB;
2092    UInt        b_listno;
2093    Arena*      a;
2094 
2095    ensure_mm_init(aid);
2096    a = arenaId_to_ArenaP(aid);
2097 
2098    if (ptr == NULL) {
2099       return;
2100    }
2101 
2102    b = get_payload_block(a, ptr);
2103 
2104    /* If this is one of V's areas, check carefully the block we're
2105       getting back.  This picks up simple block-end overruns. */
2106    if (aid != VG_AR_CLIENT)
2107       vg_assert(is_inuse_block(b) && blockSane(a, b));
2108 
2109    b_bszB   = get_bszB(b);
2110    b_pszB   = bszB_to_pszB(a, b_bszB);
2111    sb       = findSb( a, b );
2112 
2113    a->stats__bytes_on_loan -= b_pszB;
2114 
2115    /* If this is one of V's areas, fill it up with junk to enhance the
2116       chances of catching any later reads of it.  Note, 0xDD is
2117       carefully chosen junk :-), in that: (1) 0xDDDDDDDD is an invalid
2118       and non-word-aligned address on most systems, and (2) 0xDD is a
2119       value which is unlikely to be generated by the new compressed
2120       Vbits representation for memcheck. */
2121    if (aid != VG_AR_CLIENT)
2122       VG_(memset)(ptr, 0xDD, (SizeT)b_pszB);
2123 
2124    if (! sb->unsplittable) {
2125       // Put this chunk back on a list somewhere.
2126       b_listno = pszB_to_listNo(b_pszB);
2127       mkFreeBlock( a, b, b_bszB, b_listno );
2128       if (VG_(clo_profile_heap))
2129          set_cc(b, "admin.free-1");
2130 
2131       /* Possibly merge b with its predecessor or successor. */
2132       mergeWithFreeNeighbours (a, sb, b, b_bszB);
2133 
2134       // Inform that ptr has been released. We give redzone size
2135       // 0 instead of a->rz_szB as proper accessibility is done just after.
2136       INNER_REQUEST(VALGRIND_FREELIKE_BLOCK(ptr, 0));
2137 
2138       // We need to (re-)establish the minimum accessibility needed
2139       // for free list management. E.g. if block ptr has been put in a free
2140       // list and a neighbour block is released afterwards, the
2141       // "lo" and "hi" portions of the block ptr will be accessed to
2142       // glue the 2 blocks together.
2143       // We could mark the whole block as not accessible, and each time
2144       // transiently mark accessible the needed lo/hi parts. Not done as this
2145       // is quite complex, for very little expected additional bug detection.
2146       // fully unaccessible. Note that the below marks the (possibly) merged
2147       // block, not the block corresponding to the ptr argument.
2148 
2149       // First mark the whole block unaccessible.
2150       INNER_REQUEST(VALGRIND_MAKE_MEM_NOACCESS(b, b_bszB));
2151       // Then mark the relevant administrative headers as defined.
2152       // No need to mark the heap profile portion as defined, this is not
2153       // used for free blocks.
2154       INNER_REQUEST(VALGRIND_MAKE_MEM_DEFINED(b + hp_overhead_szB(),
2155                                               sizeof(SizeT) + sizeof(void*)));
2156       INNER_REQUEST(VALGRIND_MAKE_MEM_DEFINED(b + b_bszB
2157                                               - sizeof(SizeT) - sizeof(void*),
2158                                               sizeof(SizeT) + sizeof(void*)));
2159    } else {
2160       vg_assert(unsplittableBlockSane(a, sb, b));
2161 
2162       // Inform that ptr has been released. Redzone size value
2163       // is not relevant (so we give  0 instead of a->rz_szB)
2164       // as it is expected that the aspacemgr munmap will be used by
2165       //  outer to mark the whole superblock as unaccessible.
2166       INNER_REQUEST(VALGRIND_FREELIKE_BLOCK(ptr, 0));
2167 
2168       // Reclaim immediately the unsplittable superblock sb.
2169       reclaimSuperblock (a, sb);
2170    }
2171 
2172 #  ifdef DEBUG_MALLOC
2173    sanity_check_malloc_arena(aid);
2174 #  endif
2175 
2176 }
2177 
2178 
2179 /*
2180    The idea for malloc_aligned() is to allocate a big block, base, and
2181    then split it into two parts: frag, which is returned to the free
2182    pool, and align, which is the bit we're really after.  Here's
2183    a picture.  L and H denote the block lower and upper overheads, in
2184    bytes.  The details are gruesome.  Note it is slightly complicated
2185    because the initial request to generate base may return a bigger
2186    block than we asked for, so it is important to distinguish the base
2187    request size and the base actual size.
2188 
2189    frag_b                   align_b
2190    |                        |
2191    |    frag_p              |    align_p
2192    |    |                   |    |
2193    v    v                   v    v
2194 
2195    +---+                +---+---+               +---+
2196    | L |----------------| H | L |---------------| H |
2197    +---+                +---+---+               +---+
2198 
2199    ^    ^                        ^
2200    |    |                        :
2201    |    base_p                   this addr must be aligned
2202    |
2203    base_b
2204 
2205    .    .               .   .   .               .   .
2206    <------ frag_bszB ------->   .               .   .
2207    .    <------------- base_pszB_act ----------->   .
2208    .    .               .   .   .               .   .
2209 
2210 */
VG_(arena_memalign)2211 void* VG_(arena_memalign) ( ArenaId aid, const HChar* cc,
2212                             SizeT req_alignB, SizeT req_pszB )
2213 {
2214    SizeT  base_pszB_req, base_pszB_act, frag_bszB;
2215    Block  *base_b, *align_b;
2216    UByte  *base_p, *align_p;
2217    SizeT  saved_bytes_on_loan;
2218    Arena* a;
2219 
2220    ensure_mm_init(aid);
2221    a = arenaId_to_ArenaP(aid);
2222 
2223    vg_assert(req_pszB < MAX_PSZB);
2224 
2225    // You must provide a cost-center name against which to charge
2226    // this allocation; it isn't optional.
2227    vg_assert(cc);
2228 
2229    // Check that the requested alignment has a plausible size.
2230    // Check that the requested alignment seems reasonable; that is, is
2231    // a power of 2.
2232    if (req_alignB < VG_MIN_MALLOC_SZB
2233        || req_alignB > 16 * 1024 * 1024
2234        || VG_(log2)( req_alignB ) == -1 /* not a power of 2 */) {
2235       VG_(printf)("VG_(arena_memalign)(%p, %lu, %lu)\n"
2236                   "bad alignment value %lu\n"
2237                   "(it is too small, too big, or not a power of two)",
2238                   a, req_alignB, req_pszB, req_alignB );
2239       VG_(core_panic)("VG_(arena_memalign)");
2240       /*NOTREACHED*/
2241    }
2242    // Paranoid
2243    vg_assert(req_alignB % VG_MIN_MALLOC_SZB == 0);
2244 
2245    /* Required payload size for the aligned chunk. */
2246    req_pszB = align_req_pszB(req_pszB);
2247 
2248    /* Payload size to request for the big block that we will split up. */
2249    base_pszB_req = req_pszB + min_useful_bszB(a) + req_alignB;
2250 
2251    /* Payload ptr for the block we are going to split.  Note this
2252       changes a->bytes_on_loan; we save and restore it ourselves. */
2253    saved_bytes_on_loan = a->stats__bytes_on_loan;
2254    {
2255       /* As we will split the block given back by VG_(arena_malloc),
2256          we have to (temporarily) disable unsplittable for this arena,
2257          as unsplittable superblocks cannot be split. */
2258       const SizeT save_min_unsplittable_sblock_szB
2259          = a->min_unsplittable_sblock_szB;
2260       a->min_unsplittable_sblock_szB = MAX_PSZB;
2261       base_p = VG_(arena_malloc) ( aid, cc, base_pszB_req );
2262       a->min_unsplittable_sblock_szB = save_min_unsplittable_sblock_szB;
2263    }
2264    a->stats__bytes_on_loan = saved_bytes_on_loan;
2265 
2266    /* Give up if we couldn't allocate enough space */
2267    if (base_p == 0)
2268       return 0;
2269    /* base_p was marked as allocated by VALGRIND_MALLOCLIKE_BLOCK
2270       inside VG_(arena_malloc). We need to indicate it is free, then
2271       we need to mark it undefined to allow the below code to access is. */
2272    INNER_REQUEST(VALGRIND_FREELIKE_BLOCK(base_p, a->rz_szB));
2273    INNER_REQUEST(VALGRIND_MAKE_MEM_UNDEFINED(base_p, base_pszB_req));
2274 
2275    /* Block ptr for the block we are going to split. */
2276    base_b = get_payload_block ( a, base_p );
2277 
2278    /* Pointer to the payload of the aligned block we are going to
2279       return.  This has to be suitably aligned. */
2280    align_p = align_upwards ( base_b + 2 * overhead_szB_lo(a)
2281                                     + overhead_szB_hi(a),
2282                              req_alignB );
2283    align_b = get_payload_block(a, align_p);
2284 
2285    /* The block size of the fragment we will create.  This must be big
2286       enough to actually create a fragment. */
2287    frag_bszB = align_b - base_b;
2288 
2289    vg_assert(frag_bszB >= min_useful_bszB(a));
2290 
2291    /* The actual payload size of the block we are going to split. */
2292    base_pszB_act = get_pszB(a, base_b);
2293 
2294    /* Create the fragment block, and put it back on the relevant free list. */
2295    mkFreeBlock ( a, base_b, frag_bszB,
2296                  pszB_to_listNo(bszB_to_pszB(a, frag_bszB)) );
2297    if (VG_(clo_profile_heap))
2298       set_cc(base_b, "admin.frag-memalign-1");
2299 
2300    /* Create the aligned block. */
2301    mkInuseBlock ( a, align_b,
2302                   base_p + base_pszB_act
2303                          + overhead_szB_hi(a) - (UByte*)align_b );
2304    if (VG_(clo_profile_heap))
2305       set_cc(align_b, cc);
2306 
2307    /* Final sanity checks. */
2308    vg_assert( is_inuse_block(get_payload_block(a, align_p)) );
2309 
2310    vg_assert(req_pszB <= get_pszB(a, get_payload_block(a, align_p)));
2311 
2312    a->stats__bytes_on_loan += get_pszB(a, get_payload_block(a, align_p));
2313    if (a->stats__bytes_on_loan > a->stats__bytes_on_loan_max) {
2314       a->stats__bytes_on_loan_max = a->stats__bytes_on_loan;
2315    }
2316    /* a->stats__tot_blocks, a->stats__tot_bytes, a->stats__nsearches
2317       are updated by the call to VG_(arena_malloc) just a few lines
2318       above.  So we don't need to update them here. */
2319 
2320 #  ifdef DEBUG_MALLOC
2321    sanity_check_malloc_arena(aid);
2322 #  endif
2323 
2324    vg_assert( (((Addr)align_p) % req_alignB) == 0 );
2325 
2326    INNER_REQUEST(VALGRIND_MALLOCLIKE_BLOCK(align_p,
2327                                            req_pszB, a->rz_szB, False));
2328 
2329    return align_p;
2330 }
2331 
2332 
VG_(arena_malloc_usable_size)2333 SizeT VG_(arena_malloc_usable_size) ( ArenaId aid, void* ptr )
2334 {
2335    Arena* a = arenaId_to_ArenaP(aid);
2336    Block* b = get_payload_block(a, ptr);
2337    return get_pszB(a, b);
2338 }
2339 
2340 
2341 // Implementation of mallinfo(). There is no recent standard that defines
2342 // the behavior of mallinfo(). The meaning of the fields in struct mallinfo
2343 // is as follows:
2344 //
2345 //     struct mallinfo  {
2346 //                int arena;     /* total space in arena            */
2347 //                int ordblks;   /* number of ordinary blocks       */
2348 //                int smblks;    /* number of small blocks          */
2349 //                int hblks;     /* number of holding blocks        */
2350 //                int hblkhd;    /* space in holding block headers  */
2351 //                int usmblks;   /* space in small blocks in use    */
2352 //                int fsmblks;   /* space in free small blocks      */
2353 //                int uordblks;  /* space in ordinary blocks in use */
2354 //                int fordblks;  /* space in free ordinary blocks   */
2355 //                int keepcost;  /* space penalty if keep option    */
2356 //                               /* is used                         */
2357 //        };
2358 //
2359 // The glibc documentation about mallinfo (which is somewhat outdated) can
2360 // be found here:
2361 // http://www.gnu.org/software/libtool/manual/libc/Statistics-of-Malloc.html
2362 //
2363 // See also http://bugs.kde.org/show_bug.cgi?id=160956.
2364 //
2365 // Regarding the implementation of VG_(mallinfo)(): we cannot return the
2366 // whole struct as the library function does, because this is called by a
2367 // client request.  So instead we use a pointer to do call by reference.
VG_(mallinfo)2368 void VG_(mallinfo) ( ThreadId tid, struct vg_mallinfo* mi )
2369 {
2370    UWord  i, free_blocks, free_blocks_size;
2371    Arena* a = arenaId_to_ArenaP(VG_AR_CLIENT);
2372 
2373    // Traverse free list and calculate free blocks statistics.
2374    // This may seem slow but glibc works the same way.
2375    free_blocks_size = free_blocks = 0;
2376    for (i = 0; i < N_MALLOC_LISTS; i++) {
2377       Block* b = a->freelist[i];
2378       if (b == NULL) continue;
2379       for (;;) {
2380          free_blocks++;
2381          free_blocks_size += (UWord)get_pszB(a, b);
2382          b = get_next_b(b);
2383          if (b == a->freelist[i]) break;
2384       }
2385    }
2386 
2387    // We don't have fastbins so smblks & fsmblks are always 0. Also we don't
2388    // have a separate mmap allocator so set hblks & hblkhd to 0.
2389    mi->arena    = a->stats__bytes_mmaped;
2390    mi->ordblks  = free_blocks + VG_(free_queue_length);
2391    mi->smblks   = 0;
2392    mi->hblks    = 0;
2393    mi->hblkhd   = 0;
2394    mi->usmblks  = 0;
2395    mi->fsmblks  = 0;
2396    mi->uordblks = a->stats__bytes_on_loan - VG_(free_queue_volume);
2397    mi->fordblks = free_blocks_size + VG_(free_queue_volume);
2398    mi->keepcost = 0; // may want some value in here
2399 }
2400 
VG_(arena_redzone_size)2401 SizeT VG_(arena_redzone_size) ( ArenaId aid )
2402 {
2403    ensure_mm_init (VG_AR_CLIENT);
2404    /*  ensure_mm_init will call arena_init if not yet done.
2405        This then ensures that the arena redzone size is properly
2406        initialised. */
2407    return arenaId_to_ArenaP(aid)->rz_szB;
2408 }
2409 
2410 /*------------------------------------------------------------*/
2411 /*--- Services layered on top of malloc/free.              ---*/
2412 /*------------------------------------------------------------*/
2413 
VG_(arena_calloc)2414 void* VG_(arena_calloc) ( ArenaId aid, const HChar* cc,
2415                           SizeT nmemb, SizeT bytes_per_memb )
2416 {
2417    SizeT  size;
2418    void*  p;
2419 
2420    size = nmemb * bytes_per_memb;
2421    vg_assert(size >= nmemb && size >= bytes_per_memb);// check against overflow
2422 
2423    p = VG_(arena_malloc) ( aid, cc, size );
2424 
2425    if (p != NULL)
2426      VG_(memset)(p, 0, size);
2427 
2428    return p;
2429 }
2430 
2431 
VG_(arena_realloc)2432 void* VG_(arena_realloc) ( ArenaId aid, const HChar* cc,
2433                            void* ptr, SizeT req_pszB )
2434 {
2435    Arena* a;
2436    SizeT  old_pszB;
2437    void*  p_new;
2438    Block* b;
2439 
2440    ensure_mm_init(aid);
2441    a = arenaId_to_ArenaP(aid);
2442 
2443    vg_assert(req_pszB < MAX_PSZB);
2444 
2445    if (NULL == ptr) {
2446       return VG_(arena_malloc)(aid, cc, req_pszB);
2447    }
2448 
2449    if (req_pszB == 0) {
2450       VG_(arena_free)(aid, ptr);
2451       return NULL;
2452    }
2453 
2454    b = get_payload_block(a, ptr);
2455    vg_assert(blockSane(a, b));
2456 
2457    vg_assert(is_inuse_block(b));
2458    old_pszB = get_pszB(a, b);
2459 
2460    if (req_pszB <= old_pszB) {
2461       return ptr;
2462    }
2463 
2464    p_new = VG_(arena_malloc) ( aid, cc, req_pszB );
2465 
2466    VG_(memcpy)(p_new, ptr, old_pszB);
2467 
2468    VG_(arena_free)(aid, ptr);
2469 
2470    return p_new;
2471 }
2472 
2473 
VG_(arena_realloc_shrink)2474 void VG_(arena_realloc_shrink) ( ArenaId aid,
2475                                  void* ptr, SizeT req_pszB )
2476 {
2477    SizeT  req_bszB, frag_bszB, b_bszB;
2478    Superblock* sb;
2479    Arena* a;
2480    SizeT  old_pszB;
2481    Block* b;
2482 
2483    ensure_mm_init(aid);
2484 
2485    a = arenaId_to_ArenaP(aid);
2486    b = get_payload_block(a, ptr);
2487    vg_assert(blockSane(a, b));
2488    vg_assert(is_inuse_block(b));
2489 
2490    old_pszB = get_pszB(a, b);
2491    req_pszB = align_req_pszB(req_pszB);
2492    vg_assert(old_pszB >= req_pszB);
2493    if (old_pszB == req_pszB)
2494       return;
2495 
2496    sb = findSb( a, b );
2497    if (sb->unsplittable) {
2498       const UByte* sb_start = &sb->payload_bytes[0];
2499       const UByte* sb_end = &sb->payload_bytes[sb->n_payload_bytes - 1];
2500       Addr  frag;
2501 
2502       vg_assert(unsplittableBlockSane(a, sb, b));
2503 
2504       frag = VG_PGROUNDUP((Addr) sb
2505                           + sizeof(Superblock) + pszB_to_bszB(a, req_pszB));
2506       frag_bszB = (Addr)sb_end - frag + 1;
2507 
2508       if (frag_bszB >= VKI_PAGE_SIZE) {
2509          SysRes sres;
2510 
2511          a->stats__bytes_on_loan -= old_pszB;
2512          b_bszB = (UByte*)frag - sb_start;
2513          shrinkInuseBlock(a, b, b_bszB);
2514          INNER_REQUEST
2515             (VALGRIND_RESIZEINPLACE_BLOCK(ptr,
2516                                           old_pszB,
2517                                           VG_(arena_malloc_usable_size)(aid, ptr),
2518                                           a->rz_szB));
2519          /* Have the minimum admin headers needed accessibility. */
2520          INNER_REQUEST(mkBhdrSzAccess(a, b));
2521          a->stats__bytes_on_loan += bszB_to_pszB(a, b_bszB);
2522 
2523          sb->n_payload_bytes -= frag_bszB;
2524          VG_(debugLog)(1, "mallocfree",
2525                        "shrink superblock %p to (pszB %7lu) "
2526                        "owner %s/%s (munmap-ing %p %7lu)\n",
2527                        sb, sb->n_payload_bytes,
2528                        a->clientmem ? "CLIENT" : "VALGRIND", a->name,
2529                        (void*) frag, frag_bszB);
2530          if (a->clientmem) {
2531             Bool need_discard = False;
2532             sres = VG_(am_munmap_client)(&need_discard,
2533                                          frag,
2534                                          frag_bszB);
2535             vg_assert (!need_discard);
2536          } else {
2537             sres = VG_(am_munmap_valgrind)(frag,
2538                                            frag_bszB);
2539          }
2540          vg_assert2(! sr_isError(sres), "shrink superblock munmap failure\n");
2541          a->stats__bytes_mmaped -= frag_bszB;
2542 
2543          vg_assert(unsplittableBlockSane(a, sb, b));
2544       }
2545    } else {
2546       req_bszB = pszB_to_bszB(a, req_pszB);
2547       b_bszB = get_bszB(b);
2548       frag_bszB = b_bszB - req_bszB;
2549       if (frag_bszB < min_useful_bszB(a))
2550          return;
2551 
2552       a->stats__bytes_on_loan -= old_pszB;
2553       shrinkInuseBlock(a, b, req_bszB);
2554       INNER_REQUEST
2555          (VALGRIND_RESIZEINPLACE_BLOCK(ptr,
2556                                        old_pszB,
2557                                        VG_(arena_malloc_usable_size)(aid, ptr),
2558                                        a->rz_szB));
2559       /* Have the minimum admin headers needed accessibility. */
2560       INNER_REQUEST(mkBhdrSzAccess(a, b));
2561 
2562       mkFreeBlock(a, &b[req_bszB], frag_bszB,
2563                   pszB_to_listNo(bszB_to_pszB(a, frag_bszB)));
2564       /* Mark the admin headers as accessible. */
2565       INNER_REQUEST(mkBhdrAccess(a, &b[req_bszB]));
2566       if (VG_(clo_profile_heap))
2567          set_cc(&b[req_bszB], "admin.fragmentation-2");
2568       /* Possibly merge &b[req_bszB] with its free neighbours. */
2569       mergeWithFreeNeighbours(a, sb, &b[req_bszB], frag_bszB);
2570 
2571       b_bszB = get_bszB(b);
2572       a->stats__bytes_on_loan += bszB_to_pszB(a, b_bszB);
2573    }
2574 
2575    vg_assert (blockSane(a, b));
2576 #  ifdef DEBUG_MALLOC
2577    sanity_check_malloc_arena(aid);
2578 #  endif
2579 }
2580 
2581 /* Inline just for the wrapper VG_(strdup) below */
VG_(arena_strdup)2582 __inline__ HChar* VG_(arena_strdup) ( ArenaId aid, const HChar* cc,
2583                                       const HChar* s )
2584 {
2585    Int   i;
2586    Int   len;
2587    HChar* res;
2588 
2589    if (s == NULL)
2590       return NULL;
2591 
2592    len = VG_(strlen)(s) + 1;
2593    res = VG_(arena_malloc) (aid, cc, len);
2594 
2595    for (i = 0; i < len; i++)
2596       res[i] = s[i];
2597    return res;
2598 }
2599 
VG_(arena_perm_malloc)2600 void* VG_(arena_perm_malloc) ( ArenaId aid, SizeT size, Int align  )
2601 {
2602    Arena*      a;
2603 
2604    ensure_mm_init(aid);
2605    a = arenaId_to_ArenaP(aid);
2606 
2607    align = align - 1;
2608    size = (size + align) & ~align;
2609 
2610    if (UNLIKELY(a->perm_malloc_current + size > a->perm_malloc_limit)) {
2611       // Get a superblock, but we will not insert it into the superblock list.
2612       // The superblock structure is not needed, so we will use the full
2613       // memory range of it. This superblock is however counted in the
2614       // mmaped statistics.
2615       Superblock* new_sb = newSuperblock (a, size);
2616       a->perm_malloc_limit = (Addr)&new_sb->payload_bytes[new_sb->n_payload_bytes - 1];
2617 
2618       // We do not mind starting allocating from the beginning of the superblock
2619       // as afterwards, we "lose" it as a superblock.
2620       a->perm_malloc_current = (Addr)new_sb;
2621    }
2622 
2623    a->stats__perm_blocks += 1;
2624    a->stats__perm_bytes_on_loan  += size;
2625    add_one_block_to_stats (a, size);
2626 
2627    a->perm_malloc_current        += size;
2628    return (void*)(a->perm_malloc_current - size);
2629 }
2630 
2631 /*------------------------------------------------------------*/
2632 /*--- Tool-visible functions.                              ---*/
2633 /*------------------------------------------------------------*/
2634 
2635 // All just wrappers to avoid exposing arenas to tools.
2636 
2637 // This function never returns NULL.
VG_(malloc)2638 void* VG_(malloc) ( const HChar* cc, SizeT nbytes )
2639 {
2640    return VG_(arena_malloc) ( VG_AR_CORE, cc, nbytes );
2641 }
2642 
VG_(free)2643 void  VG_(free) ( void* ptr )
2644 {
2645    VG_(arena_free) ( VG_AR_CORE, ptr );
2646 }
2647 
VG_(calloc)2648 void* VG_(calloc) ( const HChar* cc, SizeT nmemb, SizeT bytes_per_memb )
2649 {
2650    return VG_(arena_calloc) ( VG_AR_CORE, cc, nmemb, bytes_per_memb );
2651 }
2652 
VG_(realloc)2653 void* VG_(realloc) ( const HChar* cc, void* ptr, SizeT size )
2654 {
2655    return VG_(arena_realloc) ( VG_AR_CORE, cc, ptr, size );
2656 }
2657 
VG_(realloc_shrink)2658 void VG_(realloc_shrink) ( void* ptr, SizeT size )
2659 {
2660    VG_(arena_realloc_shrink) ( VG_AR_CORE, ptr, size );
2661 }
2662 
VG_(strdup)2663 HChar* VG_(strdup) ( const HChar* cc, const HChar* s )
2664 {
2665    return VG_(arena_strdup) ( VG_AR_CORE, cc, s );
2666 }
2667 
VG_(perm_malloc)2668 void* VG_(perm_malloc) ( SizeT size, Int align  )
2669 {
2670    return VG_(arena_perm_malloc) ( VG_AR_CORE, size, align );
2671 }
2672 
2673 
2674 /*--------------------------------------------------------------------*/
2675 /*--- end                                                          ---*/
2676 /*--------------------------------------------------------------------*/
2677