1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21
22 /*
23 * Copyright 2009 Sun Microsystems, Inc. All rights reserved.
24 * Use is subject to license terms.
25 */
26
27 #include <mtmalloc.h>
28 #include "mtmalloc_impl.h"
29 #include <unistd.h>
30 #include <synch.h>
31 #include <thread.h>
32 #include <pthread.h>
33 #include <stdio.h>
34 #include <limits.h>
35 #include <errno.h>
36 #include <string.h>
37 #include <strings.h>
38 #include <sys/param.h>
39 #include <sys/sysmacros.h>
40
41 /*
42 * To turn on the asserts just compile -DDEBUG
43 */
44
45 #ifndef DEBUG
46 #define NDEBUG
47 #endif
48
49 #include <assert.h>
50
51 /*
52 * The MT hot malloc implementation contained herein is designed to be
53 * plug-compatible with the libc version of malloc. It is not intended
54 * to replace that implementation until we decide that it is ok to break
55 * customer apps (Solaris 3.0).
56 *
57 * For requests up to 2^^16, the allocator initializes itself into NCPUS
58 * worth of chains of caches. When a memory request is made, the calling thread
59 * is vectored into one of NCPUS worth of caches. The LWP id gives us a cheap,
60 * contention-reducing index to use, eventually, this should be replaced with
61 * the actual CPU sequence number, when an interface to get it is available.
62 *
63 * Once the thread is vectored into one of the list of caches the real
64 * allocation of the memory begins. The size is determined to figure out which
65 * bucket the allocation should be satisfied from. The management of free
66 * buckets is done via a bitmask. A free bucket is represented by a 1. The
67 * first free bit represents the first free bucket. The position of the bit,
68 * represents the position of the bucket in the arena.
69 *
70 * When the memory from the arena is handed out, the address of the cache
71 * control structure is written in the word preceeding the returned memory.
72 * This cache control address is used during free() to mark the buffer free
73 * in the cache control structure.
74 *
75 * When all available memory in a cache has been depleted, a new chunk of memory
76 * is allocated via sbrk(). The new cache is allocated from this chunk of memory
77 * and initialized in the function create_cache(). New caches are installed at
78 * the front of a singly linked list of the same size memory pools. This helps
79 * to ensure that there will tend to be available memory in the beginning of the
80 * list.
81 *
82 * Long linked lists hurt performance. To decrease this effect, there is a
83 * tunable, requestsize, that bumps up the sbrk allocation size and thus
84 * increases the number of available blocks within an arena. We also keep
85 * a "hint" for each cache list, which is the last cache in the list allocated
86 * from. This lowers the cost of searching if there are a lot of fully
87 * allocated blocks at the front of the list.
88 *
89 * For requests greater than 2^^16 (oversize allocations), there are two pieces
90 * of overhead. There is the OVERHEAD used to hold the cache addr
91 * (&oversize_list), plus an oversize_t structure to further describe the block.
92 *
93 * The oversize list is kept as defragmented as possible by coalescing
94 * freed oversized allocations with adjacent neighbors.
95 *
96 * Addresses handed out are stored in a hash table, and are aligned on
97 * MTMALLOC_MIN_ALIGN-byte boundaries at both ends. Request sizes are rounded-up
98 * where necessary in order to achieve this. This eases the implementation of
99 * MTDEBUGPATTERN and MTINITPATTERN, particularly where coalescing occurs.
100 *
101 * A memalign allocation takes memalign header overhead. There's two
102 * types of memalign headers distinguished by MTMALLOC_MEMALIGN_MAGIC
103 * and MTMALLOC_MEMALIGN_MIN_MAGIC. When the size of memory taken to
104 * get to the aligned address from malloc'ed address is the minimum size
105 * OVERHEAD, we create a header taking only one OVERHEAD space with magic
106 * number MTMALLOC_MEMALIGN_MIN_MAGIC, and we know by subtracting OVERHEAD
107 * from memaligned address, we can get to the malloc'ed address. Otherwise,
108 * we create a memalign header taking two OVERHEAD space, one stores
109 * MTMALLOC_MEMALIGN_MAGIC magic number, the other one points back to the
110 * malloc'ed address.
111 */
112
113 #if defined(__i386) || defined(__amd64)
114 #include <arpa/inet.h> /* for htonl() */
115 #endif
116
117 static void * morecore(size_t);
118 static void create_cache(cache_t *, size_t bufsize, uint_t hunks);
119 static void * malloc_internal(size_t, percpu_t *);
120 static void * oversize(size_t);
121 static oversize_t *find_oversize(size_t);
122 static void add_oversize(oversize_t *);
123 static void copy_pattern(uint32_t, void *, size_t);
124 static void * verify_pattern(uint32_t, void *, size_t);
125 static void reinit_cpu_list(void);
126 static void reinit_cache(cache_t *);
127 static void free_oversize(oversize_t *);
128 static oversize_t *oversize_header_alloc(uintptr_t, size_t);
129
130 /*
131 * oversize hash table stuff
132 */
133 #define NUM_BUCKETS 67 /* must be prime */
134 #define HASH_OVERSIZE(caddr) ((uintptr_t)(caddr) % NUM_BUCKETS)
135 oversize_t *ovsz_hashtab[NUM_BUCKETS];
136
137 #define ALIGN(x, a) ((((uintptr_t)(x) + ((uintptr_t)(a) - 1)) \
138 & ~((uintptr_t)(a) - 1)))
139
140 /* need this to deal with little endianess of x86 */
141 #if defined(__i386) || defined(__amd64)
142 #define FLIP_EM(x) htonl((x))
143 #else
144 #define FLIP_EM(x) (x)
145 #endif
146
147 #define INSERT_ONLY 0
148 #define COALESCE_LEFT 0x00000001
149 #define COALESCE_RIGHT 0x00000002
150 #define COALESCE_WITH_BOTH_SIDES (COALESCE_LEFT | COALESCE_RIGHT)
151
152 #define OVERHEAD 8 /* size needed to write cache addr */
153 #define HUNKSIZE 8192 /* just a multiplier */
154
155 #define MAX_CACHED_SHIFT 16 /* 64K is the max cached size */
156 #define MAX_CACHED (1 << MAX_CACHED_SHIFT)
157 #define MIN_CACHED_SHIFT 4 /* smaller requests rounded up */
158 #define MTMALLOC_MIN_ALIGN 8 /* min guaranteed alignment */
159
160 /* maximum size before overflow */
161 #define MAX_MTMALLOC (SIZE_MAX - (SIZE_MAX % MTMALLOC_MIN_ALIGN) \
162 - OVSZ_HEADER_SIZE)
163
164 #define NUM_CACHES (MAX_CACHED_SHIFT - MIN_CACHED_SHIFT + 1)
165 #define CACHELIST_SIZE ALIGN(NUM_CACHES * sizeof (cache_head_t), \
166 CACHE_COHERENCY_UNIT)
167
168 #define MINSIZE 9 /* for requestsize, tunable */
169 #define MAXSIZE 256 /* arbitrary, big enough, for requestsize */
170
171 #define FREEPATTERN 0xdeadbeef /* debug fill pattern for free buf */
172 #define INITPATTERN 0xbaddcafe /* debug fill pattern for new buf */
173
174 #define misaligned(p) ((unsigned)(p) & (sizeof (int) - 1))
175 #define IS_OVERSIZE(x, y) (((x) < (y)) && (((x) > MAX_CACHED)? 1 : 0))
176
177 static long requestsize = MINSIZE; /* 9 pages per cache; tunable; 9 is min */
178
179 static uint_t cpu_mask;
180 static curcpu_func curcpu;
181
182 static int32_t debugopt;
183 static int32_t reinit;
184
185 static percpu_t *cpu_list;
186 static oversize_t oversize_list;
187 static mutex_t oversize_lock = DEFAULTMUTEX;
188
189 static int ncpus = 0;
190
191 #define MTMALLOC_OVERSIZE_MAGIC ((uintptr_t)&oversize_list)
192 #define MTMALLOC_MEMALIGN_MAGIC ((uintptr_t)&oversize_list + 1)
193 #define MTMALLOC_MEMALIGN_MIN_MAGIC ((uintptr_t)&oversize_list + 2)
194
195 /*
196 * We require allocations handed out to be aligned on MTMALLOC_MIN_ALIGN-byte
197 * boundaries. We round up sizeof (oversize_t) (when necessary) to ensure that
198 * this is achieved.
199 */
200 #define OVSZ_SIZE (ALIGN(sizeof (oversize_t), MTMALLOC_MIN_ALIGN))
201 #define OVSZ_HEADER_SIZE (OVSZ_SIZE + OVERHEAD)
202
203 /*
204 * memalign header takes 2 OVERHEAD space. One for memalign magic, and the
205 * other one points back to the start address of originally allocated space.
206 */
207 #define MEMALIGN_HEADER_SIZE 2 * OVERHEAD
208 #define MEMALIGN_HEADER_ALLOC(x, shift, malloc_addr)\
209 if (shift == OVERHEAD)\
210 *((uintptr_t *)((caddr_t)x - OVERHEAD)) = \
211 MTMALLOC_MEMALIGN_MIN_MAGIC; \
212 else {\
213 *((uintptr_t *)((caddr_t)x - OVERHEAD)) = \
214 MTMALLOC_MEMALIGN_MAGIC; \
215 *((uintptr_t *)((caddr_t)x - 2 * OVERHEAD)) = \
216 (uintptr_t)malloc_addr; \
217 }
218
219 /*
220 * Add big to the oversize hash table at the head of the relevant bucket.
221 */
222 static void
223 insert_hash(oversize_t *big)
224 {
225 caddr_t ret = big->addr;
226 int bucket = HASH_OVERSIZE(ret);
227
228 assert(MUTEX_HELD(&oversize_lock));
229 big->hash_next = ovsz_hashtab[bucket];
230 ovsz_hashtab[bucket] = big;
231 }
232
233 void *
234 malloc(size_t bytes)
235 {
236 percpu_t *list_rotor;
237 uint_t list_index;
238
239 if (bytes > MAX_CACHED)
240 return (oversize(bytes));
241
242 list_index = (curcpu() & cpu_mask);
243
244 list_rotor = &cpu_list[list_index];
245
246 return (malloc_internal(bytes, list_rotor));
247 }
248
249 void *
250 realloc(void * ptr, size_t bytes)
251 {
252 void *new, *data_ptr;
253 cache_t *cacheptr;
254 caddr_t mem;
255 size_t shift = 0;
256
257 if (ptr == NULL)
258 return (malloc(bytes));
259
260 if (bytes == 0) {
261 free(ptr);
262 return (NULL);
263 }
264
265 data_ptr = ptr;
266 mem = (caddr_t)ptr - OVERHEAD;
267
268 /*
269 * Optimization possibility :
270 * p = malloc(64);
271 * q = realloc(p, 64);
272 * q can be same as p.
273 * Apply this optimization for the normal
274 * sized caches for now.
275 */
276 if (*(uintptr_t *)mem < MTMALLOC_OVERSIZE_MAGIC ||
277 *(uintptr_t *)mem > MTMALLOC_MEMALIGN_MIN_MAGIC) {
278 cacheptr = (cache_t *)*(uintptr_t *)mem;
279 if (bytes <= (cacheptr->mt_size - OVERHEAD))
280 return (ptr);
281 }
282
283 new = malloc(bytes);
284
285 if (new == NULL)
286 return (NULL);
287
288 /*
289 * If new == ptr, ptr has previously been freed. Passing a freed pointer
290 * to realloc() is not allowed - unless the caller specifically states
291 * otherwise, in which case we must avoid freeing ptr (ie new) before we
292 * return new. There is (obviously) no requirement to memcpy() ptr to
293 * new before we return.
294 */
295 if (new == ptr) {
296 if (!(debugopt & MTDOUBLEFREE))
297 abort();
298 return (new);
299 }
300
301 if (*(uintptr_t *)mem == MTMALLOC_MEMALIGN_MAGIC) {
302 mem -= OVERHEAD;
303 ptr = (void *)*(uintptr_t *)mem;
304 mem = (caddr_t)ptr - OVERHEAD;
305 shift = (size_t)((uintptr_t)data_ptr - (uintptr_t)ptr);
306 } else if (*(uintptr_t *)mem == MTMALLOC_MEMALIGN_MIN_MAGIC) {
307 ptr = (void *) mem;
308 mem -= OVERHEAD;
309 shift = OVERHEAD;
310 }
311
312 if (*(uintptr_t *)mem == MTMALLOC_OVERSIZE_MAGIC) {
313 oversize_t *old;
314
315 old = (oversize_t *)(mem - OVSZ_SIZE);
316 (void) memcpy(new, data_ptr, MIN(bytes, old->size - shift));
317 free(ptr);
318 return (new);
319 }
320
321 cacheptr = (cache_t *)*(uintptr_t *)mem;
322
323 (void) memcpy(new, data_ptr,
324 MIN(cacheptr->mt_size - OVERHEAD - shift, bytes));
325 free(ptr);
326
327 return (new);
328 }
329
330 void *
331 calloc(size_t nelem, size_t bytes)
332 {
333 void * ptr;
334 size_t size = nelem * bytes;
335
336 ptr = malloc(size);
337 if (ptr == NULL)
338 return (NULL);
339 (void) memset(ptr, 0, size);
340
341 return (ptr);
342 }
343
344 void
345 free(void * ptr)
346 {
347 cache_t *cacheptr;
348 caddr_t mem;
349 int32_t i;
350 caddr_t freeblocks;
351 uintptr_t offset;
352 uchar_t mask;
353 int32_t which_bit, num_bytes;
354
355 if (ptr == NULL)
356 return;
357
358 mem = (caddr_t)ptr - OVERHEAD;
359
360 if (*(uintptr_t *)mem == MTMALLOC_MEMALIGN_MAGIC) {
361 mem -= OVERHEAD;
362 ptr = (void *)*(uintptr_t *)mem;
363 mem = (caddr_t)ptr - OVERHEAD;
364 } else if (*(uintptr_t *)mem == MTMALLOC_MEMALIGN_MIN_MAGIC) {
365 ptr = (void *) mem;
366 mem -= OVERHEAD;
367 }
368
369 if (*(uintptr_t *)mem == MTMALLOC_OVERSIZE_MAGIC) {
370 oversize_t *big, **opp;
371 int bucket;
372
373 big = (oversize_t *)(mem - OVSZ_SIZE);
374 (void) mutex_lock(&oversize_lock);
375
376 bucket = HASH_OVERSIZE(big->addr);
377 for (opp = &ovsz_hashtab[bucket]; *opp != NULL;
378 opp = &(*opp)->hash_next)
379 if (*opp == big)
380 break;
381
382 if (*opp == NULL) {
383 if (!(debugopt & MTDOUBLEFREE))
384 abort();
385 (void) mutex_unlock(&oversize_lock);
386 return;
387 }
388
389 *opp = big->hash_next; /* remove big from the hash table */
390 big->hash_next = NULL;
391
392 if (debugopt & MTDEBUGPATTERN)
393 copy_pattern(FREEPATTERN, ptr, big->size);
394 add_oversize(big);
395 (void) mutex_unlock(&oversize_lock);
396 return;
397 }
398
399 cacheptr = (cache_t *)*(uintptr_t *)mem;
400 freeblocks = cacheptr->mt_freelist;
401
402 /*
403 * This is the distance measured in bits into the arena.
404 * The value of offset is in bytes but there is a 1-1 correlation
405 * between distance into the arena and distance into the
406 * freelist bitmask.
407 */
408 offset = mem - cacheptr->mt_arena;
409
410 /*
411 * i is total number of bits to offset into freelist bitmask.
412 */
413
414 i = offset / cacheptr->mt_size;
415
416 num_bytes = i >> 3;
417
418 /*
419 * which_bit is the bit offset into the byte in the freelist.
420 * if our freelist bitmask looks like 0xf3 and we are freeing
421 * block 5 (ie: the 6th block) our mask will be 0xf7 after
422 * the free. Things go left to right that's why the mask is 0x80
423 * and not 0x01.
424 */
425 which_bit = i - (num_bytes << 3);
426
427 mask = 0x80 >> which_bit;
428
429 freeblocks += num_bytes;
430
431 if (debugopt & MTDEBUGPATTERN)
432 copy_pattern(FREEPATTERN, ptr, cacheptr->mt_size - OVERHEAD);
433
434 (void) mutex_lock(&cacheptr->mt_cache_lock);
435
436 if (*freeblocks & mask) {
437 if (!(debugopt & MTDOUBLEFREE))
438 abort();
439 } else {
440 *freeblocks |= mask;
441 cacheptr->mt_nfree++;
442 }
443
444 (void) mutex_unlock(&cacheptr->mt_cache_lock);
445 }
446
447 void *
448 memalign(size_t alignment, size_t size)
449 {
450 size_t alloc_size;
451 uintptr_t offset;
452 void *alloc_buf;
453 void *ret_buf;
454
455 if (size == 0 || alignment == 0 || misaligned(alignment) ||
456 (alignment & (alignment - 1)) != 0) {
457 errno = EINVAL;
458 return (NULL);
459 }
460
461 /* <= MTMALLOC_MIN_ALIGN, malloc can provide directly */
462 if (alignment <= MTMALLOC_MIN_ALIGN)
463 return (malloc(size));
464
465 alloc_size = size + alignment - MTMALLOC_MIN_ALIGN;
466
467 if (alloc_size < size) { /* overflow */
468 errno = ENOMEM;
469 return (NULL);
470 }
471
472 alloc_buf = malloc(alloc_size);
473
474 if (alloc_buf == NULL)
475 /* malloc sets errno */
476 return (NULL);
477
478 /*
479 * If alloc_size > MAX_CACHED, malloc() will have returned a multiple of
480 * MTMALLOC_MIN_ALIGN, having rounded-up alloc_size if necessary. Since
481 * we will use alloc_size to return the excess fragments to the free
482 * list, we also round-up alloc_size if necessary.
483 */
484 if ((alloc_size > MAX_CACHED) &&
485 (alloc_size & (MTMALLOC_MIN_ALIGN - 1)))
486 alloc_size = ALIGN(alloc_size, MTMALLOC_MIN_ALIGN);
487
488 if ((offset = (uintptr_t)alloc_buf & (alignment - 1)) == 0) {
489 /* aligned correctly */
490
491 size_t frag_size = alloc_size -
492 (size + MTMALLOC_MIN_ALIGN + OVSZ_HEADER_SIZE);
493
494 /*
495 * If the leftover piece of the memory > MAX_CACHED,
496 * split off the piece and return it back to the freelist.
497 */
498 if (IS_OVERSIZE(frag_size, alloc_size)) {
499 oversize_t *orig, *tail;
500 uintptr_t taddr;
501 size_t data_size;
502 taddr = ALIGN((uintptr_t)alloc_buf + size,
503 MTMALLOC_MIN_ALIGN);
504 data_size = taddr - (uintptr_t)alloc_buf;
505 orig = (oversize_t *)((uintptr_t)alloc_buf -
506 OVSZ_HEADER_SIZE);
507 frag_size = orig->size - data_size -
508 OVSZ_HEADER_SIZE;
509 orig->size = data_size;
510 tail = oversize_header_alloc(taddr, frag_size);
511 free_oversize(tail);
512 }
513 ret_buf = alloc_buf;
514 } else {
515 uchar_t oversize_bits = 0;
516 size_t head_sz, data_sz, tail_sz;
517 uintptr_t ret_addr, taddr, shift, tshift;
518 oversize_t *orig, *tail, *big;
519 size_t tsize;
520
521 /* needs to be aligned */
522 shift = alignment - offset;
523
524 assert(shift >= MTMALLOC_MIN_ALIGN);
525
526 ret_addr = ((uintptr_t)alloc_buf + shift);
527 ret_buf = (void *)ret_addr;
528
529 if (alloc_size <= MAX_CACHED) {
530 MEMALIGN_HEADER_ALLOC(ret_addr, shift, alloc_buf);
531 return (ret_buf);
532 }
533
534 /*
535 * Only check for the fragments when the memory is allocted
536 * from oversize_list. Split off a fragment and return it
537 * to the oversize freelist when it's > MAX_CACHED.
538 */
539
540 head_sz = shift - MAX(MEMALIGN_HEADER_SIZE, OVSZ_HEADER_SIZE);
541
542 tail_sz = alloc_size -
543 (shift + size + MTMALLOC_MIN_ALIGN + OVSZ_HEADER_SIZE);
544
545 oversize_bits |= IS_OVERSIZE(head_sz, alloc_size) |
546 IS_OVERSIZE(size, alloc_size) << DATA_SHIFT |
547 IS_OVERSIZE(tail_sz, alloc_size) << TAIL_SHIFT;
548
549 switch (oversize_bits) {
550 case NONE_OVERSIZE:
551 case DATA_OVERSIZE:
552 MEMALIGN_HEADER_ALLOC(ret_addr, shift,
553 alloc_buf);
554 break;
555 case HEAD_OVERSIZE:
556 /*
557 * If we can extend data > MAX_CACHED and have
558 * head still > MAX_CACHED, we split head-end
559 * as the case of head-end and data oversized,
560 * otherwise just create memalign header.
561 */
562 tsize = (shift + size) - (MAX_CACHED + 8 +
563 MTMALLOC_MIN_ALIGN + OVSZ_HEADER_SIZE);
564
565 if (!IS_OVERSIZE(tsize, alloc_size)) {
566 MEMALIGN_HEADER_ALLOC(ret_addr, shift,
567 alloc_buf);
568 break;
569 } else {
570 tsize += OVSZ_HEADER_SIZE;
571 taddr = ALIGN((uintptr_t)alloc_buf +
572 tsize, MTMALLOC_MIN_ALIGN);
573 tshift = ret_addr - taddr;
574 MEMALIGN_HEADER_ALLOC(ret_addr, tshift,
575 taddr);
576 ret_addr = taddr;
577 shift = ret_addr - (uintptr_t)alloc_buf;
578 }
579 /* FALLTHROUGH */
580 case HEAD_AND_DATA_OVERSIZE:
581 /*
582 * Split off the head fragment and
583 * return it back to oversize freelist.
584 * Create oversize header for the piece
585 * of (data + tail fragment).
586 */
587 orig = (oversize_t *)((uintptr_t)alloc_buf -
588 OVSZ_HEADER_SIZE);
589 big = oversize_header_alloc(ret_addr -
590 OVSZ_HEADER_SIZE, (orig->size - shift));
591 (void) mutex_lock(&oversize_lock);
592 insert_hash(big);
593 (void) mutex_unlock(&oversize_lock);
594 orig->size = shift - OVSZ_HEADER_SIZE;
595
596 /* free up the head fragment */
597 free_oversize(orig);
598 break;
599 case TAIL_OVERSIZE:
600 /*
601 * If we can extend data > MAX_CACHED and have
602 * tail-end still > MAX_CACHED, we split tail
603 * end, otherwise just create memalign header.
604 */
605 orig = (oversize_t *)((uintptr_t)alloc_buf -
606 OVSZ_HEADER_SIZE);
607 tsize = orig->size - (MAX_CACHED + 8 +
608 shift + OVSZ_HEADER_SIZE +
609 MTMALLOC_MIN_ALIGN);
610 if (!IS_OVERSIZE(tsize, alloc_size)) {
611 MEMALIGN_HEADER_ALLOC(ret_addr, shift,
612 alloc_buf);
613 break;
614 } else {
615 size = MAX_CACHED + 8;
616 }
617 /* FALLTHROUGH */
618 case DATA_AND_TAIL_OVERSIZE:
619 /*
620 * Split off the tail fragment and
621 * return it back to oversize freelist.
622 * Create memalign header and adjust
623 * the size for the piece of
624 * (head fragment + data).
625 */
626 taddr = ALIGN(ret_addr + size,
627 MTMALLOC_MIN_ALIGN);
628 data_sz = (size_t)(taddr -
629 (uintptr_t)alloc_buf);
630 orig = (oversize_t *)((uintptr_t)alloc_buf -
631 OVSZ_HEADER_SIZE);
632 tsize = orig->size - data_sz;
633 orig->size = data_sz;
634 MEMALIGN_HEADER_ALLOC(ret_buf, shift,
635 alloc_buf);
636 tsize -= OVSZ_HEADER_SIZE;
637 tail = oversize_header_alloc(taddr, tsize);
638 free_oversize(tail);
639 break;
640 case HEAD_AND_TAIL_OVERSIZE:
641 /*
642 * Split off the head fragment.
643 * We try to free up tail-end when we can
644 * extend data size to (MAX_CACHED + 8)
645 * and remain tail-end oversized.
646 * The bottom line is all split pieces
647 * should be oversize in size.
648 */
649 orig = (oversize_t *)((uintptr_t)alloc_buf -
650 OVSZ_HEADER_SIZE);
651 tsize = orig->size - (MAX_CACHED + 8 +
652 OVSZ_HEADER_SIZE + shift +
653 MTMALLOC_MIN_ALIGN);
654
655 if (!IS_OVERSIZE(tsize, alloc_size)) {
656 /*
657 * If the chunk is not big enough
658 * to make both data and tail oversize
659 * we just keep them as one piece.
660 */
661 big = oversize_header_alloc(ret_addr -
662 OVSZ_HEADER_SIZE,
663 orig->size - shift);
664 (void) mutex_lock(&oversize_lock);
665 insert_hash(big);
666 (void) mutex_unlock(&oversize_lock);
667 orig->size = shift - OVSZ_HEADER_SIZE;
668 free_oversize(orig);
669 break;
670 } else {
671 /*
672 * extend data size > MAX_CACHED
673 * and handle it as head, data, tail
674 * are all oversized.
675 */
676 size = MAX_CACHED + 8;
677 }
678 /* FALLTHROUGH */
679 case ALL_OVERSIZE:
680 /*
681 * split off the head and tail fragments,
682 * return them back to the oversize freelist.
683 * Alloc oversize header for data seg.
684 */
685 orig = (oversize_t *)((uintptr_t)alloc_buf -
686 OVSZ_HEADER_SIZE);
687 tsize = orig->size;
688 orig->size = shift - OVSZ_HEADER_SIZE;
689 free_oversize(orig);
690
691 taddr = ALIGN(ret_addr + size,
692 MTMALLOC_MIN_ALIGN);
693 data_sz = taddr - ret_addr;
694 assert(tsize > (shift + data_sz +
695 OVSZ_HEADER_SIZE));
696 tail_sz = tsize -
697 (shift + data_sz + OVSZ_HEADER_SIZE);
698
699 /* create oversize header for data seg */
700 big = oversize_header_alloc(ret_addr -
701 OVSZ_HEADER_SIZE, data_sz);
702 (void) mutex_lock(&oversize_lock);
703 insert_hash(big);
704 (void) mutex_unlock(&oversize_lock);
705
706 /* create oversize header for tail fragment */
707 tail = oversize_header_alloc(taddr, tail_sz);
708 free_oversize(tail);
709 break;
710 default:
711 /* should not reach here */
712 assert(0);
713 }
714 }
715 return (ret_buf);
716 }
717
718
719 void *
720 valloc(size_t size)
721 {
722 static unsigned pagesize;
723
724 if (size == 0)
725 return (NULL);
726
727 if (!pagesize)
728 pagesize = sysconf(_SC_PAGESIZE);
729
730 return (memalign(pagesize, size));
731 }
732
733 void
734 mallocctl(int cmd, long value)
735 {
736 switch (cmd) {
737
738 case MTDEBUGPATTERN:
739 /*
740 * Reinitialize free blocks in case malloc() is called prior
741 * to mallocctl().
742 */
743 if (value && !(debugopt & cmd)) {
744 reinit++;
745 debugopt |= cmd;
746 reinit_cpu_list();
747 }
748 /*FALLTHRU*/
749 case MTDOUBLEFREE:
750 case MTINITBUFFER:
751 if (value)
752 debugopt |= cmd;
753 else
754 debugopt &= ~cmd;
755 break;
756 case MTCHUNKSIZE:
757 if (value >= MINSIZE && value <= MAXSIZE)
758 requestsize = value;
759 break;
760 default:
761 break;
762 }
763 }
764
765 /*
766 * Initialization function, called from the init section of the library.
767 * No locking is required here because we are single-threaded during
768 * library initialization.
769 */
770 static void
771 setup_caches(void)
772 {
773 uintptr_t oldbrk;
774 uintptr_t newbrk;
775
776 size_t cache_space_needed;
777 size_t padding;
778
779 curcpu_func new_curcpu;
780 uint_t new_cpu_mask;
781 percpu_t *new_cpu_list;
782
783 uint_t i, j;
784 uintptr_t list_addr;
785
786 /*
787 * Get a decent "current cpu identifier", to be used to reduce
788 * contention. Eventually, this should be replaced by an interface
789 * to get the actual CPU sequence number in libthread/liblwp.
790 */
791 new_curcpu = (curcpu_func)thr_self;
792 if ((ncpus = 2 * sysconf(_SC_NPROCESSORS_CONF)) <= 0)
793 ncpus = 4; /* decent default value */
794
795 /* round ncpus up to a power of 2 */
796 while (ncpus & (ncpus - 1))
797 ncpus++;
798
799 new_cpu_mask = ncpus - 1; /* create the cpu mask */
800
801 /*
802 * We now do some magic with the brk. What we want to get in the
803 * end is a bunch of well-aligned stuff in a big initial allocation.
804 * Along the way, we do sanity checks to make sure no one else has
805 * touched the brk (which shouldn't happen, but it's always good to
806 * check)
807 *
808 * First, make sure sbrk is sane, and store the current brk in oldbrk.
809 */
810 oldbrk = (uintptr_t)sbrk(0);
811 if ((void *)oldbrk == (void *)-1)
812 abort(); /* sbrk is broken -- we're doomed. */
813
814 /*
815 * Now, align the brk to a multiple of CACHE_COHERENCY_UNIT, so that
816 * the percpu structures and cache lists will be properly aligned.
817 *
818 * 2. All hunks will be page-aligned, assuming HUNKSIZE >= PAGESIZE,
819 * so they can be paged out individually.
820 */
821 newbrk = ALIGN(oldbrk, CACHE_COHERENCY_UNIT);
822 if (newbrk != oldbrk && (uintptr_t)sbrk(newbrk - oldbrk) != oldbrk)
823 abort(); /* sbrk is broken -- we're doomed. */
824
825 /*
826 * For each cpu, there is one percpu_t and a list of caches
827 */
828 cache_space_needed = ncpus * (sizeof (percpu_t) + CACHELIST_SIZE);
829
830 new_cpu_list = (percpu_t *)sbrk(cache_space_needed);
831
832 if (new_cpu_list == (percpu_t *)-1 ||
833 (uintptr_t)new_cpu_list != newbrk)
834 abort(); /* sbrk is broken -- we're doomed. */
835
836 /*
837 * Finally, align the brk to HUNKSIZE so that all hunks are
838 * page-aligned, to avoid edge-effects.
839 */
840
841 newbrk = (uintptr_t)new_cpu_list + cache_space_needed;
842
843 padding = ALIGN(newbrk, HUNKSIZE) - newbrk;
844
845 if (padding > 0 && (uintptr_t)sbrk(padding) != newbrk)
846 abort(); /* sbrk is broken -- we're doomed. */
847
848 list_addr = ((uintptr_t)new_cpu_list + (sizeof (percpu_t) * ncpus));
849
850 /* initialize the percpu list */
851 for (i = 0; i < ncpus; i++) {
852 new_cpu_list[i].mt_caches = (cache_head_t *)list_addr;
853 for (j = 0; j < NUM_CACHES; j++) {
854 new_cpu_list[i].mt_caches[j].mt_cache = NULL;
855 new_cpu_list[i].mt_caches[j].mt_hint = NULL;
856 }
857
858 (void) mutex_init(&new_cpu_list[i].mt_parent_lock,
859 USYNC_THREAD, NULL);
860
861 /* get the correct cache list alignment */
862 list_addr += CACHELIST_SIZE;
863 }
864
865 /*
866 * Initialize oversize listhead
867 */
868 oversize_list.next_bysize = &oversize_list;
869 oversize_list.prev_bysize = &oversize_list;
870 oversize_list.next_byaddr = &oversize_list;
871 oversize_list.prev_byaddr = &oversize_list;
872 oversize_list.addr = NULL;
873 oversize_list.size = 0; /* sentinal */
874
875 /*
876 * Now install the global variables.
877 */
878 curcpu = new_curcpu;
879 cpu_mask = new_cpu_mask;
880 cpu_list = new_cpu_list;
881 }
882
883 static void
884 create_cache(cache_t *cp, size_t size, uint_t chunksize)
885 {
886 long nblocks;
887
888 (void) mutex_init(&cp->mt_cache_lock, USYNC_THREAD, NULL);
889 cp->mt_size = size;
890 cp->mt_freelist = ((caddr_t)cp + sizeof (cache_t));
891 cp->mt_span = chunksize * HUNKSIZE - sizeof (cache_t);
892 cp->mt_hunks = chunksize;
893 /*
894 * rough calculation. We will need to adjust later.
895 */
896 nblocks = cp->mt_span / cp->mt_size;
897 nblocks >>= 3;
898 if (nblocks == 0) { /* less than 8 free blocks in this pool */
899 int32_t numblocks = 0;
900 long i = cp->mt_span;
901 size_t sub = cp->mt_size;
902 uchar_t mask = 0;
903
904 while (i > sub) {
905 numblocks++;
906 i -= sub;
907 }
908 nblocks = numblocks;
909 cp->mt_arena = (caddr_t)ALIGN(cp->mt_freelist + 8, 8);
910 cp->mt_nfree = numblocks;
911 while (numblocks--) {
912 mask |= 0x80 >> numblocks;
913 }
914 *(cp->mt_freelist) = mask;
915 } else {
916 cp->mt_arena = (caddr_t)ALIGN((caddr_t)cp->mt_freelist +
917 nblocks, 32);
918 /* recompute nblocks */
919 nblocks = (uintptr_t)((caddr_t)cp->mt_freelist +
920 cp->mt_span - cp->mt_arena) / cp->mt_size;
921 cp->mt_nfree = ((nblocks >> 3) << 3);
922 /* Set everything to free */
923 (void) memset(cp->mt_freelist, 0xff, nblocks >> 3);
924 }
925
926 if (debugopt & MTDEBUGPATTERN)
927 copy_pattern(FREEPATTERN, cp->mt_arena, cp->mt_size * nblocks);
928
929 cp->mt_next = NULL;
930 }
931
932 static void
933 reinit_cpu_list(void)
934 {
935 oversize_t *wp = oversize_list.next_bysize;
936 percpu_t *cpuptr;
937 cache_t *thiscache;
938 cache_head_t *cachehead;
939
940 /* Reinitialize free oversize blocks. */
941 (void) mutex_lock(&oversize_lock);
942 if (debugopt & MTDEBUGPATTERN)
943 for (; wp != &oversize_list; wp = wp->next_bysize)
944 copy_pattern(FREEPATTERN, wp->addr, wp->size);
945 (void) mutex_unlock(&oversize_lock);
946
947 /* Reinitialize free blocks. */
948 for (cpuptr = &cpu_list[0]; cpuptr < &cpu_list[ncpus]; cpuptr++) {
949 (void) mutex_lock(&cpuptr->mt_parent_lock);
950 for (cachehead = &cpuptr->mt_caches[0]; cachehead <
951 &cpuptr->mt_caches[NUM_CACHES]; cachehead++) {
952 for (thiscache = cachehead->mt_cache; thiscache != NULL;
953 thiscache = thiscache->mt_next) {
954 (void) mutex_lock(&thiscache->mt_cache_lock);
955 if (thiscache->mt_nfree == 0) {
956 (void) mutex_unlock(
957 &thiscache->mt_cache_lock);
958 continue;
959 }
960 if (thiscache != NULL)
961 reinit_cache(thiscache);
962 (void) mutex_unlock(&thiscache->mt_cache_lock);
963 }
964 }
965 (void) mutex_unlock(&cpuptr->mt_parent_lock);
966 }
967 reinit = 0;
968 }
969
970 static void
971 reinit_cache(cache_t *thiscache)
972 {
973 uint32_t *freeblocks; /* not a uintptr_t on purpose */
974 int32_t i, n;
975 caddr_t ret;
976
977 freeblocks = (uint32_t *)thiscache->mt_freelist;
978 while (freeblocks < (uint32_t *)thiscache->mt_arena) {
979 if (*freeblocks & 0xffffffff) {
980 for (i = 0; i < 32; i++) {
981 if (FLIP_EM(*freeblocks) & (0x80000000 >> i)) {
982 n = (uintptr_t)(((freeblocks -
983 (uint32_t *)thiscache->mt_freelist)
984 << 5) + i) * thiscache->mt_size;
985 ret = thiscache->mt_arena + n;
986 ret += OVERHEAD;
987 copy_pattern(FREEPATTERN, ret,
988 thiscache->mt_size);
989 }
990 }
991 }
992 freeblocks++;
993 }
994 }
995
996 static void *
997 malloc_internal(size_t size, percpu_t *cpuptr)
998 {
999 cache_head_t *cachehead;
1000 cache_t *thiscache, *hintcache;
1001 int32_t i, n, logsz, bucket;
1002 uint32_t index;
1003 uint32_t *freeblocks; /* not a uintptr_t on purpose */
1004 caddr_t ret;
1005
1006 logsz = MIN_CACHED_SHIFT;
1007
1008 while (size > (1 << logsz))
1009 logsz++;
1010
1011 bucket = logsz - MIN_CACHED_SHIFT;
1012
1013 (void) mutex_lock(&cpuptr->mt_parent_lock);
1014
1015 /*
1016 * Find a cache of the appropriate size with free buffers.
1017 *
1018 * We don't need to lock each cache as we check their mt_nfree count,
1019 * since:
1020 * 1. We are only looking for caches with mt_nfree > 0. If a
1021 * free happens during our search, it will increment mt_nfree,
1022 * which will not effect the test.
1023 * 2. Allocations can decrement mt_nfree, but they can't happen
1024 * as long as we hold mt_parent_lock.
1025 */
1026
1027 cachehead = &cpuptr->mt_caches[bucket];
1028
1029 /* Search through the list, starting at the mt_hint */
1030 thiscache = cachehead->mt_hint;
1031
1032 while (thiscache != NULL && thiscache->mt_nfree == 0)
1033 thiscache = thiscache->mt_next;
1034
1035 if (thiscache == NULL) {
1036 /* wrap around -- search up to the hint */
1037 thiscache = cachehead->mt_cache;
1038 hintcache = cachehead->mt_hint;
1039
1040 while (thiscache != NULL && thiscache != hintcache &&
1041 thiscache->mt_nfree == 0)
1042 thiscache = thiscache->mt_next;
1043
1044 if (thiscache == hintcache)
1045 thiscache = NULL;
1046 }
1047
1048
1049 if (thiscache == NULL) { /* there are no free caches */
1050 int32_t thisrequest = requestsize;
1051 int32_t buffer_size = (1 << logsz) + OVERHEAD;
1052
1053 thiscache = (cache_t *)morecore(thisrequest * HUNKSIZE);
1054
1055 if (thiscache == (cache_t *)-1) {
1056 (void) mutex_unlock(&cpuptr->mt_parent_lock);
1057 errno = EAGAIN;
1058 return (NULL);
1059 }
1060 create_cache(thiscache, buffer_size, thisrequest);
1061
1062 /* link in the new block at the beginning of the list */
1063 thiscache->mt_next = cachehead->mt_cache;
1064 cachehead->mt_cache = thiscache;
1065 }
1066
1067 /* update the hint to the cache we found or created */
1068 cachehead->mt_hint = thiscache;
1069
1070 /* thiscache now points to a cache with available space */
1071 (void) mutex_lock(&thiscache->mt_cache_lock);
1072
1073 freeblocks = (uint32_t *)thiscache->mt_freelist;
1074 while (freeblocks < (uint32_t *)thiscache->mt_arena) {
1075 if (*freeblocks & 0xffffffff)
1076 break;
1077 freeblocks++;
1078 if (freeblocks < (uint32_t *)thiscache->mt_arena &&
1079 *freeblocks & 0xffffffff)
1080 break;
1081 freeblocks++;
1082 if (freeblocks < (uint32_t *)thiscache->mt_arena &&
1083 *freeblocks & 0xffffffff)
1084 break;
1085 freeblocks++;
1086 if (freeblocks < (uint32_t *)thiscache->mt_arena &&
1087 *freeblocks & 0xffffffff)
1088 break;
1089 freeblocks++;
1090 }
1091
1092 /*
1093 * the offset from mt_freelist to freeblocks is the offset into
1094 * the arena. Be sure to include the offset into freeblocks
1095 * of the bitmask. n is the offset.
1096 */
1097 for (i = 0; i < 32; ) {
1098 if (FLIP_EM(*freeblocks) & (0x80000000 >> i++))
1099 break;
1100 if (FLIP_EM(*freeblocks) & (0x80000000 >> i++))
1101 break;
1102 if (FLIP_EM(*freeblocks) & (0x80000000 >> i++))
1103 break;
1104 if (FLIP_EM(*freeblocks) & (0x80000000 >> i++))
1105 break;
1106 }
1107 index = 0x80000000 >> --i;
1108
1109
1110 *freeblocks &= FLIP_EM(~index);
1111
1112 thiscache->mt_nfree--;
1113
1114 (void) mutex_unlock(&thiscache->mt_cache_lock);
1115 (void) mutex_unlock(&cpuptr->mt_parent_lock);
1116
1117 n = (uintptr_t)(((freeblocks - (uint32_t *)thiscache->mt_freelist) << 5)
1118 + i) * thiscache->mt_size;
1119 /*
1120 * Now you have the offset in n, you've changed the free mask
1121 * in the freelist. Nothing left to do but find the block
1122 * in the arena and put the value of thiscache in the word
1123 * ahead of the handed out address and return the memory
1124 * back to the user.
1125 */
1126 ret = thiscache->mt_arena + n;
1127
1128 /* Store the cache addr for this buf. Makes free go fast. */
1129 *(uintptr_t *)ret = (uintptr_t)thiscache;
1130
1131 /*
1132 * This assert makes sure we don't hand out memory that is not
1133 * owned by this cache.
1134 */
1135 assert(ret + thiscache->mt_size <= thiscache->mt_freelist +
1136 thiscache->mt_span);
1137
1138 ret += OVERHEAD;
1139
1140 assert(((uintptr_t)ret & 7) == 0); /* are we 8 byte aligned */
1141
1142 if (reinit == 0 && (debugopt & MTDEBUGPATTERN))
1143 if (verify_pattern(FREEPATTERN, ret, size))
1144 abort(); /* reference after free */
1145
1146 if (debugopt & MTINITBUFFER)
1147 copy_pattern(INITPATTERN, ret, size);
1148 return ((void *)ret);
1149 }
1150
1151 static void *
1152 morecore(size_t bytes)
1153 {
1154 void * ret;
1155
1156 if (bytes > LONG_MAX) {
1157 intptr_t wad;
1158 /*
1159 * The request size is too big. We need to do this in
1160 * chunks. Sbrk only takes an int for an arg.
1161 */
1162 if (bytes == ULONG_MAX)
1163 return ((void *)-1);
1164
1165 ret = sbrk(0);
1166 wad = LONG_MAX;
1167 while (wad > 0) {
1168 if (sbrk(wad) == (void *)-1) {
1169 if (ret != sbrk(0))
1170 (void) sbrk(-LONG_MAX);
1171 return ((void *)-1);
1172 }
1173 bytes -= LONG_MAX;
1174 wad = bytes;
1175 }
1176 } else
1177 ret = sbrk(bytes);
1178
1179 return (ret);
1180 }
1181
1182
1183 static void *
1184 oversize(size_t size)
1185 {
1186 caddr_t ret;
1187 oversize_t *big;
1188
1189 /* make sure we will not overflow */
1190 if (size > MAX_MTMALLOC) {
1191 errno = ENOMEM;
1192 return (NULL);
1193 }
1194
1195 /*
1196 * Since we ensure every address we hand back is
1197 * MTMALLOC_MIN_ALIGN-byte aligned, ALIGNing size ensures that the
1198 * memory handed out is MTMALLOC_MIN_ALIGN-byte aligned at both ends.
1199 * This eases the implementation of MTDEBUGPATTERN and MTINITPATTERN,
1200 * particularly where coalescing occurs.
1201 */
1202 size = ALIGN(size, MTMALLOC_MIN_ALIGN);
1203
1204 /*
1205 * The idea with the global lock is that we are sure to
1206 * block in the kernel anyway since given an oversize alloc
1207 * we are sure to have to call morecore();
1208 */
1209 (void) mutex_lock(&oversize_lock);
1210
1211 if ((big = find_oversize(size)) != NULL) {
1212 if (reinit == 0 && (debugopt & MTDEBUGPATTERN))
1213 if (verify_pattern(FREEPATTERN, big->addr, size))
1214 abort(); /* reference after free */
1215 } else {
1216 /* Get more 8-byte aligned memory from heap */
1217 ret = morecore(size + OVSZ_HEADER_SIZE);
1218 if (ret == (caddr_t)-1) {
1219 (void) mutex_unlock(&oversize_lock);
1220 errno = ENOMEM;
1221 return (NULL);
1222 }
1223 big = oversize_header_alloc((uintptr_t)ret, size);
1224 }
1225 ret = big->addr;
1226
1227 insert_hash(big);
1228
1229 if (debugopt & MTINITBUFFER)
1230 copy_pattern(INITPATTERN, ret, size);
1231
1232 (void) mutex_unlock(&oversize_lock);
1233 assert(((uintptr_t)ret & 7) == 0); /* are we 8 byte aligned */
1234 return ((void *)ret);
1235 }
1236
1237 static void
1238 insert_oversize(oversize_t *op, oversize_t *nx)
1239 {
1240 oversize_t *sp;
1241
1242 /* locate correct insertion point in size-ordered list */
1243 for (sp = oversize_list.next_bysize;
1244 sp != &oversize_list && (op->size > sp->size);
1245 sp = sp->next_bysize)
1246 ;
1247
1248 /* link into size-ordered list */
1249 op->next_bysize = sp;
1250 op->prev_bysize = sp->prev_bysize;
1251 op->prev_bysize->next_bysize = op;
1252 op->next_bysize->prev_bysize = op;
1253
1254 /*
1255 * link item into address-ordered list
1256 * (caller provides insertion point as an optimization)
1257 */
1258 op->next_byaddr = nx;
1259 op->prev_byaddr = nx->prev_byaddr;
1260 op->prev_byaddr->next_byaddr = op;
1261 op->next_byaddr->prev_byaddr = op;
1262
1263 }
1264
1265 static void
1266 unlink_oversize(oversize_t *lp)
1267 {
1268 /* unlink from address list */
1269 lp->prev_byaddr->next_byaddr = lp->next_byaddr;
1270 lp->next_byaddr->prev_byaddr = lp->prev_byaddr;
1271
1272 /* unlink from size list */
1273 lp->prev_bysize->next_bysize = lp->next_bysize;
1274 lp->next_bysize->prev_bysize = lp->prev_bysize;
1275 }
1276
1277 static void
1278 position_oversize_by_size(oversize_t *op)
1279 {
1280 oversize_t *sp;
1281
1282 if (op->size > op->next_bysize->size ||
1283 op->size < op->prev_bysize->size) {
1284
1285 /* unlink from size list */
1286 op->prev_bysize->next_bysize = op->next_bysize;
1287 op->next_bysize->prev_bysize = op->prev_bysize;
1288
1289 /* locate correct insertion point in size-ordered list */
1290 for (sp = oversize_list.next_bysize;
1291 sp != &oversize_list && (op->size > sp->size);
1292 sp = sp->next_bysize)
1293 ;
1294
1295 /* link into size-ordered list */
1296 op->next_bysize = sp;
1297 op->prev_bysize = sp->prev_bysize;
1298 op->prev_bysize->next_bysize = op;
1299 op->next_bysize->prev_bysize = op;
1300 }
1301 }
1302
1303 static void
1304 add_oversize(oversize_t *lp)
1305 {
1306 int merge_flags = INSERT_ONLY;
1307 oversize_t *nx; /* ptr to item right of insertion point */
1308 oversize_t *pv; /* ptr to item left of insertion point */
1309 uint_t size_lp, size_pv, size_nx;
1310 uintptr_t endp_lp, endp_pv, endp_nx;
1311
1312 /*
1313 * Locate insertion point in address-ordered list
1314 */
1315
1316 for (nx = oversize_list.next_byaddr;
1317 nx != &oversize_list && (lp->addr > nx->addr);
1318 nx = nx->next_byaddr)
1319 ;
1320
1321 /*
1322 * Determine how to add chunk to oversize freelist
1323 */
1324
1325 size_lp = OVSZ_HEADER_SIZE + lp->size;
1326 endp_lp = ALIGN((uintptr_t)lp + size_lp, MTMALLOC_MIN_ALIGN);
1327 size_lp = endp_lp - (uintptr_t)lp;
1328
1329 pv = nx->prev_byaddr;
1330
1331 if (pv->size) {
1332
1333 size_pv = OVSZ_HEADER_SIZE + pv->size;
1334 endp_pv = ALIGN((uintptr_t)pv + size_pv,
1335 MTMALLOC_MIN_ALIGN);
1336 size_pv = endp_pv - (uintptr_t)pv;
1337
1338 /* Check for adjacency with left chunk */
1339 if ((uintptr_t)lp == endp_pv)
1340 merge_flags |= COALESCE_LEFT;
1341 }
1342
1343 if (nx->size) {
1344
1345 /* Check for adjacency with right chunk */
1346 if ((uintptr_t)nx == endp_lp) {
1347 size_nx = OVSZ_HEADER_SIZE + nx->size;
1348 endp_nx = ALIGN((uintptr_t)nx + size_nx,
1349 MTMALLOC_MIN_ALIGN);
1350 size_nx = endp_nx - (uintptr_t)nx;
1351 merge_flags |= COALESCE_RIGHT;
1352 }
1353 }
1354
1355 /*
1356 * If MTDEBUGPATTERN==1, lp->addr will have been overwritten with
1357 * FREEPATTERN for lp->size bytes. If we can merge, the oversize
1358 * header(s) that will also become part of the memory available for
1359 * reallocation (ie lp and/or nx) must also be overwritten with
1360 * FREEPATTERN or we will SIGABRT when this memory is next reallocated.
1361 */
1362 switch (merge_flags) {
1363
1364 case INSERT_ONLY: /* Coalescing not possible */
1365 insert_oversize(lp, nx);
1366 break;
1367 case COALESCE_LEFT:
1368 pv->size += size_lp;
1369 position_oversize_by_size(pv);
1370 if (debugopt & MTDEBUGPATTERN)
1371 copy_pattern(FREEPATTERN, lp, OVSZ_HEADER_SIZE);
1372 break;
1373 case COALESCE_RIGHT:
1374 unlink_oversize(nx);
1375 lp->size += size_nx;
1376 insert_oversize(lp, pv->next_byaddr);
1377 if (debugopt & MTDEBUGPATTERN)
1378 copy_pattern(FREEPATTERN, nx, OVSZ_HEADER_SIZE);
1379 break;
1380 case COALESCE_WITH_BOTH_SIDES: /* Merge (with right) to the left */
1381 pv->size += size_lp + size_nx;
1382 unlink_oversize(nx);
1383 position_oversize_by_size(pv);
1384 if (debugopt & MTDEBUGPATTERN) {
1385 copy_pattern(FREEPATTERN, lp, OVSZ_HEADER_SIZE);
1386 copy_pattern(FREEPATTERN, nx, OVSZ_HEADER_SIZE);
1387 }
1388 break;
1389 }
1390 }
1391
1392 /*
1393 * Find memory on our list that is at least size big. If we find a block that is
1394 * big enough, we break it up and return the associated oversize_t struct back
1395 * to the calling client. Any leftover piece of that block is returned to the
1396 * freelist.
1397 */
1398 static oversize_t *
1399 find_oversize(size_t size)
1400 {
1401 oversize_t *wp = oversize_list.next_bysize;
1402 while (wp != &oversize_list && size > wp->size)
1403 wp = wp->next_bysize;
1404
1405 if (wp == &oversize_list) /* empty list or nothing big enough */
1406 return (NULL);
1407 /* breaking up a chunk of memory */
1408 if ((long)((wp->size - (size + OVSZ_HEADER_SIZE + MTMALLOC_MIN_ALIGN)))
1409 > MAX_CACHED) {
1410 caddr_t off;
1411 oversize_t *np;
1412 size_t osize;
1413 off = (caddr_t)ALIGN(wp->addr + size,
1414 MTMALLOC_MIN_ALIGN);
1415 osize = wp->size;
1416 wp->size = (size_t)(off - wp->addr);
1417 np = oversize_header_alloc((uintptr_t)off,
1418 osize - (wp->size + OVSZ_HEADER_SIZE));
1419 if ((long)np->size < 0)
1420 abort();
1421 unlink_oversize(wp);
1422 add_oversize(np);
1423 } else {
1424 unlink_oversize(wp);
1425 }
1426 return (wp);
1427 }
1428
1429 static void
1430 copy_pattern(uint32_t pattern, void *buf_arg, size_t size)
1431 {
1432 uint32_t *bufend = (uint32_t *)((char *)buf_arg + size);
1433 uint32_t *buf = buf_arg;
1434
1435 while (buf < bufend - 3) {
1436 buf[3] = buf[2] = buf[1] = buf[0] = pattern;
1437 buf += 4;
1438 }
1439 while (buf < bufend)
1440 *buf++ = pattern;
1441 }
1442
1443 static void *
1444 verify_pattern(uint32_t pattern, void *buf_arg, size_t size)
1445 {
1446 uint32_t *bufend = (uint32_t *)((char *)buf_arg + size);
1447 uint32_t *buf;
1448
1449 for (buf = buf_arg; buf < bufend; buf++)
1450 if (*buf != pattern)
1451 return (buf);
1452 return (NULL);
1453 }
1454
1455 static void
1456 free_oversize(oversize_t *ovp)
1457 {
1458 assert(((uintptr_t)ovp->addr & 7) == 0); /* are we 8 byte aligned */
1459 assert(ovp->size > MAX_CACHED);
1460
1461 ovp->next_bysize = ovp->prev_bysize = NULL;
1462 ovp->next_byaddr = ovp->prev_byaddr = NULL;
1463 (void) mutex_lock(&oversize_lock);
1464 add_oversize(ovp);
1465 (void) mutex_unlock(&oversize_lock);
1466 }
1467
1468 static oversize_t *
1469 oversize_header_alloc(uintptr_t mem, size_t size)
1470 {
1471 oversize_t *ovsz_hdr;
1472
1473 assert(size > MAX_CACHED);
1474
1475 ovsz_hdr = (oversize_t *)mem;
1476 ovsz_hdr->prev_bysize = NULL;
1477 ovsz_hdr->next_bysize = NULL;
1478 ovsz_hdr->prev_byaddr = NULL;
1479 ovsz_hdr->next_byaddr = NULL;
1480 ovsz_hdr->hash_next = NULL;
1481 ovsz_hdr->size = size;
1482 mem += OVSZ_SIZE;
1483 *(uintptr_t *)mem = MTMALLOC_OVERSIZE_MAGIC;
1484 mem += OVERHEAD;
1485 assert(((uintptr_t)mem & 7) == 0); /* are we 8 byte aligned */
1486 ovsz_hdr->addr = (caddr_t)mem;
1487 return (ovsz_hdr);
1488 }
1489
1490 static void
1491 malloc_prepare()
1492 {
1493 percpu_t *cpuptr;
1494 cache_head_t *cachehead;
1495 cache_t *thiscache;
1496
1497 (void) mutex_lock(&oversize_lock);
1498 for (cpuptr = &cpu_list[0]; cpuptr < &cpu_list[ncpus]; cpuptr++) {
1499 (void) mutex_lock(&cpuptr->mt_parent_lock);
1500 for (cachehead = &cpuptr->mt_caches[0];
1501 cachehead < &cpuptr->mt_caches[NUM_CACHES];
1502 cachehead++) {
1503 for (thiscache = cachehead->mt_cache;
1504 thiscache != NULL;
1505 thiscache = thiscache->mt_next) {
1506 (void) mutex_lock(
1507 &thiscache->mt_cache_lock);
1508 }
1509 }
1510 }
1511 }
1512
1513 static void
1514 malloc_release()
1515 {
1516 percpu_t *cpuptr;
1517 cache_head_t *cachehead;
1518 cache_t *thiscache;
1519
1520 for (cpuptr = &cpu_list[ncpus - 1]; cpuptr >= &cpu_list[0]; cpuptr--) {
1521 for (cachehead = &cpuptr->mt_caches[NUM_CACHES - 1];
1522 cachehead >= &cpuptr->mt_caches[0];
1523 cachehead--) {
1524 for (thiscache = cachehead->mt_cache;
1525 thiscache != NULL;
1526 thiscache = thiscache->mt_next) {
1527 (void) mutex_unlock(
1528 &thiscache->mt_cache_lock);
1529 }
1530 }
1531 (void) mutex_unlock(&cpuptr->mt_parent_lock);
1532 }
1533 (void) mutex_unlock(&oversize_lock);
1534 }
1535
1536 #pragma init(malloc_init)
1537 static void
1538 malloc_init(void)
1539 {
1540 /*
1541 * This works in the init section for this library
1542 * because setup_caches() doesn't call anything in libc
1543 * that calls malloc(). If it did, disaster would ensue.
1544 *
1545 * For this to work properly, this library must be the first
1546 * one to have its init section called (after libc) by the
1547 * dynamic linker. If some other library's init section
1548 * ran first and called malloc(), disaster would ensue.
1549 * Because this is an interposer library for malloc(), the
1550 * dynamic linker arranges for its init section to run first.
1551 */
1552 (void) setup_caches();
1553
1554 (void) pthread_atfork(malloc_prepare, malloc_release, malloc_release);
1555 }