xref: /illumos-gate/usr/src/lib/libbsdmalloc/common/malloc.bsd43.c (revision 581cede61ac9c14d8d4ea452562a567189eead78)
1 /*
2  * Copyright 2004 Sun Microsystems, Inc.  All rights reserved.
3  * Use is subject to license terms.
4  */
5 
6 #pragma ident	"%Z%%M%	%I%	%E% SMI"
7 /*
8  * wizard:/space/4.3reno/usr/src/lib/libc/stdlib/malloc.c
9  * Copyright (c) 1983 Regents of the University of California.
10  * All rights reserved.
11  *
12  * Redistribution and use in source and binary forms are permitted
13  * provided that: (1) source distributions retain this entire copyright
14  * notice and comment, and (2) distributions including binaries display
15  * the following acknowledgement:  ``This product includes software
16  * developed by the University of California, Berkeley and its contributors''
17  * in the documentation or other materials provided with the distribution
18  * and in all advertising materials mentioning features or use of this
19  * software. Neither the name of the University nor the names of its
20  * contributors may be used to endorse or promote products derived
21  * from this software without specific prior written permission.
22  * THIS SOFTWARE IS PROVIDED ``AS IS'' AND WITHOUT ANY EXPRESS OR
23  * IMPLIED WARRANTIES, INCLUDING, WITHOUT LIMITATION, THE IMPLIED
24  * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE.
25  */
26 
27 /*
28  * malloc.c (Caltech) 2/21/82
29  * Chris Kingsley, kingsley@cit-20.
30  *
31  * This is a very fast storage allocator.  It allocates blocks of a small
32  * number of different sizes, and keeps free lists of each size.  Blocks that
33  * don't exactly fit are passed up to the next larger size.  In this
34  * implementation, the available sizes are 2^n-4 bytes long (ILP32)
35  * or 2^n-8 bytes long (LP64).
36  */
37 
38 /*LINTLIBRARY*/
39 #include <sys/types.h>
40 #include <stdlib.h>
41 #include <unistd.h>
42 #include <string.h>
43 #include <limits.h>
44 
45 /*
46  * The overhead on a block is at least 4 bytes.  When free, this space
47  * contains a pointer to the next free block, and the bottom two bits must
48  * be zero.  When in use, the first byte is set to MAGIC, and the second
49  * byte is the size index.  The remaining bytes are for alignment.
50  * The order of elements is critical: ov_magic must overlay the low order
51  * bits of ov_next, and ov_magic can not be a valid ov_next bit pattern.
52  * Overhead is 4 bytes for ILP32, 8 bytes for LP64
53  */
54 union	overhead {
55 	union	overhead *ov_next;	/* when free */
56 	struct {
57 #if defined(_LITTLE_ENDIAN)
58 		uchar_t	ovu_magic;	/* magic number */
59 		uchar_t	ovu_index;	/* bucket # */
60 		uchar_t	ovu_pad[sizeof (union overhead *) - 2];
61 #elif defined(_BIG_ENDIAN)
62 		uchar_t	ovu_pad[sizeof (union overhead *) - 2];
63 		uchar_t	ovu_index;	/* bucket # */
64 		uchar_t	ovu_magic;	/* magic number */
65 #else
66 #error "Endianness is not defined"
67 #endif
68 	} ovu;
69 };
70 
71 #define	ov_magic	ovu.ovu_magic
72 #define	ov_index	ovu.ovu_index
73 
74 #define	MAGIC		0xef		/* magic # on accounting info */
75 
76 /*
77  * nextf[i] is the pointer to the next free block of size 2^(i+EXP).
78  * The smallest allocatable block is 8 bytes (ILP32) or 16 bytes (LP64).
79  * The overhead information precedes the data area returned to the user.
80  */
81 #ifdef _LP64
82 #define	EXP	4
83 #define	NBUCKETS 60
84 #else
85 #define	EXP	3
86 #define	NBUCKETS 29
87 #endif
88 static	union overhead *nextf[NBUCKETS];
89 
90 static	int	pagesz;			/* page size */
91 static	long	sbrk_adjust;		/* in case sbrk() does alignment */
92 static	int	pagebucket;		/* page size bucket */
93 static	void	morecore(int);
94 static	int	findbucket(union overhead *, int);
95 
96 void *
97 malloc(size_t nbytes)
98 {
99 	union overhead *op;
100 	int bucket;
101 	ssize_t	n;
102 	size_t amt;
103 
104 	/*
105 	 * First time malloc is called, setup page size and
106 	 * align break pointer so all data will be page aligned.
107 	 */
108 	if (pagesz == 0) {
109 		pagesz = getpagesize();
110 		op = sbrk(0);
111 		n = pagesz - sizeof (*op) - ((uintptr_t)op & (pagesz - 1));
112 		if (n < 0)
113 			n += pagesz;
114 		if (n) {
115 			if (sbrk(n) == (void *)-1)
116 				return (NULL);
117 			/*
118 			 * We were careful to arrange that
119 			 * sbrk(0) + sizeof (union overhead)
120 			 * should end up on a page boundary.
121 			 * If the underlying sbrk() performs alignment
122 			 * then this is false.  We compute the adjustment.
123 			 */
124 			op = sbrk(0);
125 			sbrk_adjust = (uintptr_t)(op + 1) & (pagesz - 1);
126 		} else {
127 			sbrk_adjust = 0;
128 		}
129 		bucket = 0;
130 		amt = (1UL << EXP);
131 		while (pagesz > amt) {
132 			amt <<= 1;
133 			bucket++;
134 		}
135 		pagebucket = bucket;
136 	}
137 	/*
138 	 * Convert amount of memory requested into closest block size
139 	 * stored in hash buckets which satisfies request.
140 	 * Account for space used per block for accounting.
141 	 */
142 	if (nbytes <= (n = pagesz - sizeof (*op))) {
143 		amt = (1UL << EXP);	/* size of first bucket */
144 		bucket = 0;
145 		n = -(ssize_t)(sizeof (*op));
146 	} else {
147 		amt = pagesz;
148 		bucket = pagebucket;
149 	}
150 	while (nbytes > amt + n) {
151 		amt <<= 1;
152 		if (amt == 0)
153 			return (NULL);
154 		bucket++;
155 	}
156 	/*
157 	 * If nothing in hash bucket right now,
158 	 * request more memory from the system.
159 	 */
160 	if ((op = nextf[bucket]) == NULL) {
161 		morecore(bucket);
162 		if ((op = nextf[bucket]) == NULL)
163 			return (NULL);
164 	}
165 	/* remove from linked list */
166 	nextf[bucket] = op->ov_next;
167 	op->ov_magic = MAGIC;
168 	op->ov_index = (uchar_t)bucket;
169 	return (op + 1);
170 }
171 
172 /*
173  * Allocate more memory to the indicated bucket.
174  */
175 static void
176 morecore(int bucket)
177 {
178 	union overhead *op;
179 	size_t sz;			/* size of desired block */
180 	ssize_t amt;			/* amount to allocate */
181 	long nblks;			/* how many blocks we get */
182 
183 	sz = 1UL << (bucket + EXP);
184 	if (sz == 0)
185 		return;
186 	if (sz < pagesz) {
187 		amt = pagesz;
188 		nblks = amt / sz;
189 	} else {
190 		amt = sz + pagesz;
191 		nblks = 1;
192 	}
193 	if (amt <= 0)
194 		return;
195 	if (amt > LONG_MAX) {
196 		intptr_t	delta;
197 		/*
198 		 * the value required is too big for sbrk() to deal with
199 		 * in one go, so use sbrk() at most 2 times instead.
200 		 */
201 		op = sbrk(0);
202 		delta = LONG_MAX;
203 		while (delta > 0) {
204 			if (sbrk(delta) == (void *)-1) {
205 				if (op != sbrk(0))
206 					(void) sbrk(-LONG_MAX);
207 				return;
208 			}
209 			amt -= LONG_MAX;
210 			delta = amt;
211 		}
212 	}
213 	else
214 		op = sbrk(amt);
215 	/* no more room! */
216 	if (op == (union overhead *)-1)
217 		return;
218 	/* LINTED improper alignment */
219 	op = (union overhead *)((caddr_t)op - sbrk_adjust);
220 	/*
221 	 * Add new memory allocated to that on
222 	 * free list for this hash bucket.
223 	 */
224 	nextf[bucket] = op;
225 	while (--nblks > 0) {
226 		/* LINTED improper alignment */
227 		op->ov_next = (union overhead *)((caddr_t)op + sz);
228 		/* LINTED improper alignment */
229 		op = (union overhead *)((caddr_t)op + sz);
230 	}
231 }
232 
233 void
234 free(void *cp)
235 {
236 	int size;
237 	union overhead *op;
238 
239 	if (cp == NULL)
240 		return;
241 	/* LINTED improper alignment */
242 	op = (union overhead *)((caddr_t)cp - sizeof (union overhead));
243 	if (op->ov_magic != MAGIC)
244 		return;			/* previously freed? */
245 	size = op->ov_index;
246 	op->ov_next = nextf[size];	/* also clobbers ov_magic */
247 	nextf[size] = op;
248 }
249 
250 /*
251  * When a program attempts "storage compaction" as mentioned in the
252  * old malloc man page, it realloc's an already freed block.  Usually
253  * this is the last block it freed; occasionally it might be farther
254  * back.  We have to search all the free lists for the block in order
255  * to determine its bucket: 1st we make one pass thru the lists
256  * checking only the first block in each; if that fails we search
257  * ``realloc_srchlen'' blocks in each list for a match (the variable
258  * is extern so the caller can modify it).  If that fails we just copy
259  * however many bytes was given to realloc() and hope it's not huge.
260  */
261 int realloc_srchlen = 4;	/* 4 should be plenty, -1 =>'s whole list */
262 
263 void *
264 realloc(void *cp, size_t nbytes)
265 {
266 	size_t onb;
267 	int i;
268 	union overhead *op;
269 	char *res;
270 	int was_alloced = 0;
271 
272 	if (cp == NULL)
273 		return (malloc(nbytes));
274 	/* LINTED improper alignment */
275 	op = (union overhead *)((caddr_t)cp - sizeof (union overhead));
276 	if (op->ov_magic == MAGIC) {
277 		was_alloced++;
278 		i = op->ov_index;
279 	} else {
280 		/*
281 		 * Already free, doing "compaction".
282 		 *
283 		 * Search for the old block of memory on the
284 		 * free list.  First, check the most common
285 		 * case (last element free'd), then (this failing)
286 		 * the last ``realloc_srchlen'' items free'd.
287 		 * If all lookups fail, then just malloc() the
288 		 * space and copy the size of the new space.
289 		 */
290 		if ((i = findbucket(op, 1)) < 0 &&
291 		    (i = findbucket(op, realloc_srchlen)) < 0) {
292 			if ((res = malloc(nbytes)) != NULL)
293 				(void) memmove(res, cp, nbytes);
294 			return (res);
295 		}
296 	}
297 	onb = 1UL << (i + EXP);
298 	if (onb < pagesz)
299 		onb -= sizeof (*op);
300 	else
301 		onb += pagesz - sizeof (*op);
302 	/* avoid the copy if same size block */
303 	if (was_alloced) {
304 		size_t sz = 0;
305 		if (i) {
306 			sz = 1UL << (i + EXP - 1);
307 			if (sz < pagesz)
308 				sz -= sizeof (*op);
309 			else
310 				sz += pagesz - sizeof (*op);
311 		}
312 		if (nbytes <= onb && nbytes > sz) {
313 			return (cp);
314 		} else
315 			free(cp);
316 	}
317 	if ((res = malloc(nbytes)) == NULL)
318 		return (NULL);
319 	if (cp != res)		/* common optimization if "compacting" */
320 		(void) memmove(res, cp, (nbytes < onb) ? nbytes : onb);
321 	return (res);
322 }
323 
324 /*
325  * Search ``srchlen'' elements of each free list for a block whose
326  * header starts at ``freep''.  If srchlen is -1 search the whole list.
327  * Return bucket number, or -1 if not found.
328  */
329 static int
330 findbucket(union overhead *freep, int srchlen)
331 {
332 	union overhead *p;
333 	int i, j;
334 
335 	for (i = 0; i < NBUCKETS; i++) {
336 		j = 0;
337 		for (p = nextf[i]; p && j != srchlen; p = p->ov_next) {
338 			if (p == freep)
339 				return (i);
340 			j++;
341 		}
342 	}
343 	return (-1);
344 }
345