xref: /netbsd-src/sys/kern/kern_malloc.c (revision 74809f044aa1c34df701bf79c7bfb9794896fdf8)
1 /*	$NetBSD: kern_malloc.c,v 1.120 2008/09/25 16:23:45 pooka Exp $	*/
2 
3 /*
4  * Copyright (c) 1987, 1991, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. Neither the name of the University nor the names of its contributors
16  *    may be used to endorse or promote products derived from this software
17  *    without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  *
31  *	@(#)kern_malloc.c	8.4 (Berkeley) 5/20/95
32  */
33 
34 /*
35  * Copyright (c) 1996 Christopher G. Demetriou.  All rights reserved.
36  *
37  * Redistribution and use in source and binary forms, with or without
38  * modification, are permitted provided that the following conditions
39  * are met:
40  * 1. Redistributions of source code must retain the above copyright
41  *    notice, this list of conditions and the following disclaimer.
42  * 2. Redistributions in binary form must reproduce the above copyright
43  *    notice, this list of conditions and the following disclaimer in the
44  *    documentation and/or other materials provided with the distribution.
45  * 3. All advertising materials mentioning features or use of this software
46  *    must display the following acknowledgement:
47  *	This product includes software developed by the University of
48  *	California, Berkeley and its contributors.
49  * 4. Neither the name of the University nor the names of its contributors
50  *    may be used to endorse or promote products derived from this software
51  *    without specific prior written permission.
52  *
53  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
54  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
55  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
56  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
57  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
58  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
59  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
60  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
61  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
62  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
63  * SUCH DAMAGE.
64  *
65  *	@(#)kern_malloc.c	8.4 (Berkeley) 5/20/95
66  */
67 
68 #include <sys/cdefs.h>
69 __KERNEL_RCSID(0, "$NetBSD: kern_malloc.c,v 1.120 2008/09/25 16:23:45 pooka Exp $");
70 
71 #include <sys/param.h>
72 #include <sys/proc.h>
73 #include <sys/kernel.h>
74 #include <sys/malloc.h>
75 #include <sys/systm.h>
76 #include <sys/debug.h>
77 #include <sys/mutex.h>
78 #include <sys/lockdebug.h>
79 
80 #include <uvm/uvm_extern.h>
81 
82 static struct vm_map_kernel kmem_map_store;
83 struct vm_map *kmem_map = NULL;
84 
85 #include "opt_kmempages.h"
86 
87 #ifdef NKMEMCLUSTERS
88 #error NKMEMCLUSTERS is obsolete; remove it from your kernel config file and use NKMEMPAGES instead or let the kernel auto-size
89 #endif
90 
91 /*
92  * Default number of pages in kmem_map.  We attempt to calculate this
93  * at run-time, but allow it to be either patched or set in the kernel
94  * config file.
95  */
96 #ifndef NKMEMPAGES
97 #define	NKMEMPAGES	0
98 #endif
99 int	nkmempages = NKMEMPAGES;
100 
101 /*
102  * Defaults for lower- and upper-bounds for the kmem_map page count.
103  * Can be overridden by kernel config options.
104  */
105 #ifndef	NKMEMPAGES_MIN
106 #define	NKMEMPAGES_MIN	NKMEMPAGES_MIN_DEFAULT
107 #endif
108 
109 #ifndef NKMEMPAGES_MAX
110 #define	NKMEMPAGES_MAX	NKMEMPAGES_MAX_DEFAULT
111 #endif
112 
113 #include "opt_kmemstats.h"
114 #include "opt_malloclog.h"
115 #include "opt_malloc_debug.h"
116 
117 #define	MINALLOCSIZE	(1 << MINBUCKET)
118 #define	BUCKETINDX(size) \
119 	((size) <= (MINALLOCSIZE * 128) \
120 		? (size) <= (MINALLOCSIZE * 8) \
121 			? (size) <= (MINALLOCSIZE * 2) \
122 				? (size) <= (MINALLOCSIZE * 1) \
123 					? (MINBUCKET + 0) \
124 					: (MINBUCKET + 1) \
125 				: (size) <= (MINALLOCSIZE * 4) \
126 					? (MINBUCKET + 2) \
127 					: (MINBUCKET + 3) \
128 			: (size) <= (MINALLOCSIZE* 32) \
129 				? (size) <= (MINALLOCSIZE * 16) \
130 					? (MINBUCKET + 4) \
131 					: (MINBUCKET + 5) \
132 				: (size) <= (MINALLOCSIZE * 64) \
133 					? (MINBUCKET + 6) \
134 					: (MINBUCKET + 7) \
135 		: (size) <= (MINALLOCSIZE * 2048) \
136 			? (size) <= (MINALLOCSIZE * 512) \
137 				? (size) <= (MINALLOCSIZE * 256) \
138 					? (MINBUCKET + 8) \
139 					: (MINBUCKET + 9) \
140 				: (size) <= (MINALLOCSIZE * 1024) \
141 					? (MINBUCKET + 10) \
142 					: (MINBUCKET + 11) \
143 			: (size) <= (MINALLOCSIZE * 8192) \
144 				? (size) <= (MINALLOCSIZE * 4096) \
145 					? (MINBUCKET + 12) \
146 					: (MINBUCKET + 13) \
147 				: (size) <= (MINALLOCSIZE * 16384) \
148 					? (MINBUCKET + 14) \
149 					: (MINBUCKET + 15))
150 
151 /*
152  * Array of descriptors that describe the contents of each page
153  */
154 struct kmemusage {
155 	short ku_indx;		/* bucket index */
156 	union {
157 		u_short freecnt;/* for small allocations, free pieces in page */
158 		u_short pagecnt;/* for large allocations, pages alloced */
159 	} ku_un;
160 };
161 #define	ku_freecnt ku_un.freecnt
162 #define	ku_pagecnt ku_un.pagecnt
163 
164 struct kmembuckets kmembuckets[MINBUCKET + 16];
165 struct kmemusage *kmemusage;
166 char *kmembase, *kmemlimit;
167 
168 #ifdef DEBUG
169 static void *malloc_freecheck;
170 #endif
171 
172 /*
173  * Turn virtual addresses into kmem map indicies
174  */
175 #define	btokup(addr)	(&kmemusage[((char *)(addr) - kmembase) >> PGSHIFT])
176 
177 struct malloc_type *kmemstatistics;
178 
179 #ifdef MALLOCLOG
180 #ifndef MALLOCLOGSIZE
181 #define	MALLOCLOGSIZE	100000
182 #endif
183 
184 struct malloclog {
185 	void *addr;
186 	long size;
187 	struct malloc_type *type;
188 	int action;
189 	const char *file;
190 	long line;
191 } malloclog[MALLOCLOGSIZE];
192 
193 long	malloclogptr;
194 
195 static void
196 domlog(void *a, long size, struct malloc_type *type, int action,
197     const char *file, long line)
198 {
199 
200 	malloclog[malloclogptr].addr = a;
201 	malloclog[malloclogptr].size = size;
202 	malloclog[malloclogptr].type = type;
203 	malloclog[malloclogptr].action = action;
204 	malloclog[malloclogptr].file = file;
205 	malloclog[malloclogptr].line = line;
206 	malloclogptr++;
207 	if (malloclogptr >= MALLOCLOGSIZE)
208 		malloclogptr = 0;
209 }
210 
211 static void
212 hitmlog(void *a)
213 {
214 	struct malloclog *lp;
215 	long l;
216 
217 #define	PRT do { \
218 	lp = &malloclog[l]; \
219 	if (lp->addr == a && lp->action) { \
220 		printf("malloc log entry %ld:\n", l); \
221 		printf("\taddr = %p\n", lp->addr); \
222 		printf("\tsize = %ld\n", lp->size); \
223 		printf("\ttype = %s\n", lp->type->ks_shortdesc); \
224 		printf("\taction = %s\n", lp->action == 1 ? "alloc" : "free"); \
225 		printf("\tfile = %s\n", lp->file); \
226 		printf("\tline = %ld\n", lp->line); \
227 	} \
228 } while (/* CONSTCOND */0)
229 
230 	for (l = malloclogptr; l < MALLOCLOGSIZE; l++)
231 		PRT;
232 
233 	for (l = 0; l < malloclogptr; l++)
234 		PRT;
235 #undef PRT
236 }
237 #endif /* MALLOCLOG */
238 
239 #ifdef DIAGNOSTIC
240 /*
241  * This structure provides a set of masks to catch unaligned frees.
242  */
243 const long addrmask[] = { 0,
244 	0x00000001, 0x00000003, 0x00000007, 0x0000000f,
245 	0x0000001f, 0x0000003f, 0x0000007f, 0x000000ff,
246 	0x000001ff, 0x000003ff, 0x000007ff, 0x00000fff,
247 	0x00001fff, 0x00003fff, 0x00007fff, 0x0000ffff,
248 };
249 
250 /*
251  * The WEIRD_ADDR is used as known text to copy into free objects so
252  * that modifications after frees can be detected.
253  */
254 #define	WEIRD_ADDR	((uint32_t) 0xdeadbeef)
255 #ifdef DEBUG
256 #define	MAX_COPY	PAGE_SIZE
257 #else
258 #define	MAX_COPY	32
259 #endif
260 
261 /*
262  * Normally the freelist structure is used only to hold the list pointer
263  * for free objects.  However, when running with diagnostics, the first
264  * 8/16 bytes of the structure is unused except for diagnostic information,
265  * and the free list pointer is at offset 8/16 in the structure.  Since the
266  * first 8 bytes is the portion of the structure most often modified, this
267  * helps to detect memory reuse problems and avoid free list corruption.
268  */
269 struct freelist {
270 	uint32_t spare0;
271 #ifdef _LP64
272 	uint32_t spare1;		/* explicit padding */
273 #endif
274 	struct malloc_type *type;
275 	void *	next;
276 };
277 #else /* !DIAGNOSTIC */
278 struct freelist {
279 	void *	next;
280 };
281 #endif /* DIAGNOSTIC */
282 
283 kmutex_t malloc_lock;
284 
285 /*
286  * Allocate a block of memory
287  */
288 #ifdef MALLOCLOG
289 void *
290 _malloc(unsigned long size, struct malloc_type *ksp, int flags,
291     const char *file, long line)
292 #else
293 void *
294 malloc(unsigned long size, struct malloc_type *ksp, int flags)
295 #endif /* MALLOCLOG */
296 {
297 	struct kmembuckets *kbp;
298 	struct kmemusage *kup;
299 	struct freelist *freep;
300 	long indx, npg, allocsize;
301 	char *va, *cp, *savedlist;
302 #ifdef DIAGNOSTIC
303 	uint32_t *end, *lp;
304 	int copysize;
305 #endif
306 
307 #ifdef LOCKDEBUG
308 	if ((flags & M_NOWAIT) == 0) {
309 		ASSERT_SLEEPABLE();
310 	}
311 #endif
312 #ifdef MALLOC_DEBUG
313 	if (debug_malloc(size, ksp, flags, (void *) &va)) {
314 		if (va != 0)
315 			FREECHECK_OUT(&malloc_freecheck, (void *)va);
316 		return ((void *) va);
317 	}
318 #endif
319 	indx = BUCKETINDX(size);
320 	kbp = &kmembuckets[indx];
321 	mutex_spin_enter(&malloc_lock);
322 #ifdef KMEMSTATS
323 	while (ksp->ks_memuse >= ksp->ks_limit) {
324 		if (flags & M_NOWAIT) {
325 			mutex_spin_exit(&malloc_lock);
326 			return ((void *) NULL);
327 		}
328 		if (ksp->ks_limblocks < 65535)
329 			ksp->ks_limblocks++;
330 		mtsleep((void *)ksp, PSWP+2, ksp->ks_shortdesc, 0,
331 			&malloc_lock);
332 	}
333 	ksp->ks_size |= 1 << indx;
334 #endif
335 #ifdef DIAGNOSTIC
336 	copysize = 1 << indx < MAX_COPY ? 1 << indx : MAX_COPY;
337 #endif
338 	if (kbp->kb_next == NULL) {
339 		int s;
340 		kbp->kb_last = NULL;
341 		if (size > MAXALLOCSAVE)
342 			allocsize = round_page(size);
343 		else
344 			allocsize = 1 << indx;
345 		npg = btoc(allocsize);
346 		mutex_spin_exit(&malloc_lock);
347 		s = splvm();
348 		va = (void *) uvm_km_alloc(kmem_map,
349 		    (vsize_t)ctob(npg), 0,
350 		    ((flags & M_NOWAIT) ? UVM_KMF_NOWAIT : 0) |
351 		    ((flags & M_CANFAIL) ? UVM_KMF_CANFAIL : 0) |
352 		    UVM_KMF_WIRED);
353 		splx(s);
354 		if (__predict_false(va == NULL)) {
355 			/*
356 			 * Kmem_malloc() can return NULL, even if it can
357 			 * wait, if there is no map space available, because
358 			 * it can't fix that problem.  Neither can we,
359 			 * right now.  (We should release pages which
360 			 * are completely free and which are in kmembuckets
361 			 * with too many free elements.)
362 			 */
363 			if ((flags & (M_NOWAIT|M_CANFAIL)) == 0)
364 				panic("malloc: out of space in kmem_map");
365 			return (NULL);
366 		}
367 		mutex_spin_enter(&malloc_lock);
368 #ifdef KMEMSTATS
369 		kbp->kb_total += kbp->kb_elmpercl;
370 #endif
371 		kup = btokup(va);
372 		kup->ku_indx = indx;
373 		if (allocsize > MAXALLOCSAVE) {
374 			if (npg > 65535)
375 				panic("malloc: allocation too large");
376 			kup->ku_pagecnt = npg;
377 #ifdef KMEMSTATS
378 			ksp->ks_memuse += allocsize;
379 #endif
380 			goto out;
381 		}
382 #ifdef KMEMSTATS
383 		kup->ku_freecnt = kbp->kb_elmpercl;
384 		kbp->kb_totalfree += kbp->kb_elmpercl;
385 #endif
386 		/*
387 		 * Just in case we blocked while allocating memory,
388 		 * and someone else also allocated memory for this
389 		 * kmembucket, don't assume the list is still empty.
390 		 */
391 		savedlist = kbp->kb_next;
392 		kbp->kb_next = cp = va + (npg << PAGE_SHIFT) - allocsize;
393 		for (;;) {
394 			freep = (struct freelist *)cp;
395 #ifdef DIAGNOSTIC
396 			/*
397 			 * Copy in known text to detect modification
398 			 * after freeing.
399 			 */
400 			end = (uint32_t *)&cp[copysize];
401 			for (lp = (uint32_t *)cp; lp < end; lp++)
402 				*lp = WEIRD_ADDR;
403 			freep->type = M_FREE;
404 #endif /* DIAGNOSTIC */
405 			if (cp <= va)
406 				break;
407 			cp -= allocsize;
408 			freep->next = cp;
409 		}
410 		freep->next = savedlist;
411 		if (savedlist == NULL)
412 			kbp->kb_last = (void *)freep;
413 	}
414 	va = kbp->kb_next;
415 	kbp->kb_next = ((struct freelist *)va)->next;
416 #ifdef DIAGNOSTIC
417 	freep = (struct freelist *)va;
418 	/* XXX potential to get garbage pointer here. */
419 	if (kbp->kb_next) {
420 		int rv;
421 		vaddr_t addr = (vaddr_t)kbp->kb_next;
422 
423 		vm_map_lock(kmem_map);
424 		rv = uvm_map_checkprot(kmem_map, addr,
425 		    addr + sizeof(struct freelist), VM_PROT_WRITE);
426 		vm_map_unlock(kmem_map);
427 
428 		if (__predict_false(rv == 0)) {
429 			printf("Data modified on freelist: "
430 			    "word %ld of object %p size %ld previous type %s "
431 			    "(invalid addr %p)\n",
432 			    (long)((int32_t *)&kbp->kb_next - (int32_t *)kbp),
433 			    va, size, "foo", kbp->kb_next);
434 #ifdef MALLOCLOG
435 			hitmlog(va);
436 #endif
437 			kbp->kb_next = NULL;
438 		}
439 	}
440 
441 	/* Fill the fields that we've used with WEIRD_ADDR */
442 #ifdef _LP64
443 	freep->type = (struct malloc_type *)
444 	    (WEIRD_ADDR | (((u_long) WEIRD_ADDR) << 32));
445 #else
446 	freep->type = (struct malloc_type *) WEIRD_ADDR;
447 #endif
448 	end = (uint32_t *)&freep->next +
449 	    (sizeof(freep->next) / sizeof(int32_t));
450 	for (lp = (uint32_t *)&freep->next; lp < end; lp++)
451 		*lp = WEIRD_ADDR;
452 
453 	/* and check that the data hasn't been modified. */
454 	end = (uint32_t *)&va[copysize];
455 	for (lp = (uint32_t *)va; lp < end; lp++) {
456 		if (__predict_true(*lp == WEIRD_ADDR))
457 			continue;
458 		printf("Data modified on freelist: "
459 		    "word %ld of object %p size %ld previous type %s "
460 		    "(0x%x != 0x%x)\n",
461 		    (long)(lp - (uint32_t *)va), va, size,
462 		    "bar", *lp, WEIRD_ADDR);
463 #ifdef MALLOCLOG
464 		hitmlog(va);
465 #endif
466 		break;
467 	}
468 
469 	freep->spare0 = 0;
470 #endif /* DIAGNOSTIC */
471 #ifdef KMEMSTATS
472 	kup = btokup(va);
473 	if (kup->ku_indx != indx)
474 		panic("malloc: wrong bucket");
475 	if (kup->ku_freecnt == 0)
476 		panic("malloc: lost data");
477 	kup->ku_freecnt--;
478 	kbp->kb_totalfree--;
479 	ksp->ks_memuse += 1 << indx;
480 out:
481 	kbp->kb_calls++;
482 	ksp->ks_inuse++;
483 	ksp->ks_calls++;
484 	if (ksp->ks_memuse > ksp->ks_maxused)
485 		ksp->ks_maxused = ksp->ks_memuse;
486 #else
487 out:
488 #endif
489 #ifdef MALLOCLOG
490 	domlog(va, size, ksp, 1, file, line);
491 #endif
492 	mutex_spin_exit(&malloc_lock);
493 	if ((flags & M_ZERO) != 0)
494 		memset(va, 0, size);
495 	FREECHECK_OUT(&malloc_freecheck, (void *)va);
496 	return ((void *) va);
497 }
498 
499 /*
500  * Free a block of memory allocated by malloc.
501  */
502 #ifdef MALLOCLOG
503 void
504 _free(void *addr, struct malloc_type *ksp, const char *file, long line)
505 #else
506 void
507 free(void *addr, struct malloc_type *ksp)
508 #endif /* MALLOCLOG */
509 {
510 	struct kmembuckets *kbp;
511 	struct kmemusage *kup;
512 	struct freelist *freep;
513 	long size;
514 #ifdef DIAGNOSTIC
515 	void *cp;
516 	int32_t *end, *lp;
517 	long alloc, copysize;
518 #endif
519 
520 	FREECHECK_IN(&malloc_freecheck, addr);
521 #ifdef MALLOC_DEBUG
522 	if (debug_free(addr, ksp))
523 		return;
524 #endif
525 
526 #ifdef DIAGNOSTIC
527 	/*
528 	 * Ensure that we're free'ing something that we could
529 	 * have allocated in the first place.  That is, check
530 	 * to see that the address is within kmem_map.
531 	 */
532 	if (__predict_false((vaddr_t)addr < vm_map_min(kmem_map) ||
533 	    (vaddr_t)addr >= vm_map_max(kmem_map)))
534 		panic("free: addr %p not within kmem_map", addr);
535 #endif
536 
537 	kup = btokup(addr);
538 	size = 1 << kup->ku_indx;
539 	kbp = &kmembuckets[kup->ku_indx];
540 
541 	LOCKDEBUG_MEM_CHECK(addr,
542 	    size <= MAXALLOCSAVE ? size : ctob(kup->ku_pagecnt));
543 
544 	mutex_spin_enter(&malloc_lock);
545 #ifdef MALLOCLOG
546 	domlog(addr, 0, ksp, 2, file, line);
547 #endif
548 #ifdef DIAGNOSTIC
549 	/*
550 	 * Check for returns of data that do not point to the
551 	 * beginning of the allocation.
552 	 */
553 	if (size > PAGE_SIZE)
554 		alloc = addrmask[BUCKETINDX(PAGE_SIZE)];
555 	else
556 		alloc = addrmask[kup->ku_indx];
557 	if (((u_long)addr & alloc) != 0)
558 		panic("free: unaligned addr %p, size %ld, type %s, mask %ld",
559 		    addr, size, ksp->ks_shortdesc, alloc);
560 #endif /* DIAGNOSTIC */
561 	if (size > MAXALLOCSAVE) {
562 		uvm_km_free(kmem_map, (vaddr_t)addr, ctob(kup->ku_pagecnt),
563 		    UVM_KMF_WIRED);
564 #ifdef KMEMSTATS
565 		size = kup->ku_pagecnt << PGSHIFT;
566 		ksp->ks_memuse -= size;
567 		kup->ku_indx = 0;
568 		kup->ku_pagecnt = 0;
569 		if (ksp->ks_memuse + size >= ksp->ks_limit &&
570 		    ksp->ks_memuse < ksp->ks_limit)
571 			wakeup((void *)ksp);
572 #ifdef DIAGNOSTIC
573 		if (ksp->ks_inuse == 0)
574 			panic("free 1: inuse 0, probable double free");
575 #endif
576 		ksp->ks_inuse--;
577 		kbp->kb_total -= 1;
578 #endif
579 		mutex_spin_exit(&malloc_lock);
580 		return;
581 	}
582 	freep = (struct freelist *)addr;
583 #ifdef DIAGNOSTIC
584 	/*
585 	 * Check for multiple frees. Use a quick check to see if
586 	 * it looks free before laboriously searching the freelist.
587 	 */
588 	if (__predict_false(freep->spare0 == WEIRD_ADDR)) {
589 		for (cp = kbp->kb_next; cp;
590 		    cp = ((struct freelist *)cp)->next) {
591 			if (addr != cp)
592 				continue;
593 			printf("multiply freed item %p\n", addr);
594 #ifdef MALLOCLOG
595 			hitmlog(addr);
596 #endif
597 			panic("free: duplicated free");
598 		}
599 	}
600 
601 	/*
602 	 * Copy in known text to detect modification after freeing
603 	 * and to make it look free. Also, save the type being freed
604 	 * so we can list likely culprit if modification is detected
605 	 * when the object is reallocated.
606 	 */
607 	copysize = size < MAX_COPY ? size : MAX_COPY;
608 	end = (int32_t *)&((char *)addr)[copysize];
609 	for (lp = (int32_t *)addr; lp < end; lp++)
610 		*lp = WEIRD_ADDR;
611 	freep->type = ksp;
612 #endif /* DIAGNOSTIC */
613 #ifdef KMEMSTATS
614 	kup->ku_freecnt++;
615 	if (kup->ku_freecnt >= kbp->kb_elmpercl) {
616 		if (kup->ku_freecnt > kbp->kb_elmpercl)
617 			panic("free: multiple frees");
618 		else if (kbp->kb_totalfree > kbp->kb_highwat)
619 			kbp->kb_couldfree++;
620 	}
621 	kbp->kb_totalfree++;
622 	ksp->ks_memuse -= size;
623 	if (ksp->ks_memuse + size >= ksp->ks_limit &&
624 	    ksp->ks_memuse < ksp->ks_limit)
625 		wakeup((void *)ksp);
626 #ifdef DIAGNOSTIC
627 	if (ksp->ks_inuse == 0)
628 		panic("free 2: inuse 0, probable double free");
629 #endif
630 	ksp->ks_inuse--;
631 #endif
632 	if (kbp->kb_next == NULL)
633 		kbp->kb_next = addr;
634 	else
635 		((struct freelist *)kbp->kb_last)->next = addr;
636 	freep->next = NULL;
637 	kbp->kb_last = addr;
638 	mutex_spin_exit(&malloc_lock);
639 }
640 
641 /*
642  * Change the size of a block of memory.
643  */
644 void *
645 realloc(void *curaddr, unsigned long newsize, struct malloc_type *ksp,
646     int flags)
647 {
648 	struct kmemusage *kup;
649 	unsigned long cursize;
650 	void *newaddr;
651 #ifdef DIAGNOSTIC
652 	long alloc;
653 #endif
654 
655 	/*
656 	 * realloc() with a NULL pointer is the same as malloc().
657 	 */
658 	if (curaddr == NULL)
659 		return (malloc(newsize, ksp, flags));
660 
661 	/*
662 	 * realloc() with zero size is the same as free().
663 	 */
664 	if (newsize == 0) {
665 		free(curaddr, ksp);
666 		return (NULL);
667 	}
668 
669 #ifdef LOCKDEBUG
670 	if ((flags & M_NOWAIT) == 0) {
671 		ASSERT_SLEEPABLE();
672 	}
673 #endif
674 
675 	/*
676 	 * Find out how large the old allocation was (and do some
677 	 * sanity checking).
678 	 */
679 	kup = btokup(curaddr);
680 	cursize = 1 << kup->ku_indx;
681 
682 #ifdef DIAGNOSTIC
683 	/*
684 	 * Check for returns of data that do not point to the
685 	 * beginning of the allocation.
686 	 */
687 	if (cursize > PAGE_SIZE)
688 		alloc = addrmask[BUCKETINDX(PAGE_SIZE)];
689 	else
690 		alloc = addrmask[kup->ku_indx];
691 	if (((u_long)curaddr & alloc) != 0)
692 		panic("realloc: "
693 		    "unaligned addr %p, size %ld, type %s, mask %ld\n",
694 		    curaddr, cursize, ksp->ks_shortdesc, alloc);
695 #endif /* DIAGNOSTIC */
696 
697 	if (cursize > MAXALLOCSAVE)
698 		cursize = ctob(kup->ku_pagecnt);
699 
700 	/*
701 	 * If we already actually have as much as they want, we're done.
702 	 */
703 	if (newsize <= cursize)
704 		return (curaddr);
705 
706 	/*
707 	 * Can't satisfy the allocation with the existing block.
708 	 * Allocate a new one and copy the data.
709 	 */
710 	newaddr = malloc(newsize, ksp, flags);
711 	if (__predict_false(newaddr == NULL)) {
712 		/*
713 		 * malloc() failed, because flags included M_NOWAIT.
714 		 * Return NULL to indicate that failure.  The old
715 		 * pointer is still valid.
716 		 */
717 		return (NULL);
718 	}
719 	memcpy(newaddr, curaddr, cursize);
720 
721 	/*
722 	 * We were successful: free the old allocation and return
723 	 * the new one.
724 	 */
725 	free(curaddr, ksp);
726 	return (newaddr);
727 }
728 
729 /*
730  * Roundup size to the actual allocation size.
731  */
732 unsigned long
733 malloc_roundup(unsigned long size)
734 {
735 
736 	if (size > MAXALLOCSAVE)
737 		return (roundup(size, PAGE_SIZE));
738 	else
739 		return (1 << BUCKETINDX(size));
740 }
741 
742 /*
743  * Add a malloc type to the system.
744  */
745 void
746 malloc_type_attach(struct malloc_type *type)
747 {
748 
749 	if (nkmempages == 0)
750 		panic("malloc_type_attach: nkmempages == 0");
751 
752 	if (type->ks_magic != M_MAGIC)
753 		panic("malloc_type_attach: bad magic");
754 
755 #ifdef DIAGNOSTIC
756 	{
757 		struct malloc_type *ksp;
758 		for (ksp = kmemstatistics; ksp != NULL; ksp = ksp->ks_next) {
759 			if (ksp == type)
760 				panic("malloc_type_attach: already on list");
761 		}
762 	}
763 #endif
764 
765 #ifdef KMEMSTATS
766 	if (type->ks_limit == 0)
767 		type->ks_limit = ((u_long)nkmempages << PAGE_SHIFT) * 6U / 10U;
768 #else
769 	type->ks_limit = 0;
770 #endif
771 
772 	type->ks_next = kmemstatistics;
773 	kmemstatistics = type;
774 }
775 
776 /*
777  * Remove a malloc type from the system..
778  */
779 void
780 malloc_type_detach(struct malloc_type *type)
781 {
782 	struct malloc_type *ksp;
783 
784 #ifdef DIAGNOSTIC
785 	if (type->ks_magic != M_MAGIC)
786 		panic("malloc_type_detach: bad magic");
787 #endif
788 
789 	if (type == kmemstatistics)
790 		kmemstatistics = type->ks_next;
791 	else {
792 		for (ksp = kmemstatistics; ksp->ks_next != NULL;
793 		     ksp = ksp->ks_next) {
794 			if (ksp->ks_next == type) {
795 				ksp->ks_next = type->ks_next;
796 				break;
797 			}
798 		}
799 #ifdef DIAGNOSTIC
800 		if (ksp->ks_next == NULL)
801 			panic("malloc_type_detach: not on list");
802 #endif
803 	}
804 	type->ks_next = NULL;
805 }
806 
807 /*
808  * Set the limit on a malloc type.
809  */
810 void
811 malloc_type_setlimit(struct malloc_type *type, u_long limit)
812 {
813 #ifdef KMEMSTATS
814 	mutex_spin_enter(&malloc_lock);
815 	type->ks_limit = limit;
816 	mutex_spin_exit(&malloc_lock);
817 #endif
818 }
819 
820 /*
821  * Compute the number of pages that kmem_map will map, that is,
822  * the size of the kernel malloc arena.
823  */
824 void
825 kmeminit_nkmempages(void)
826 {
827 	int npages;
828 
829 	if (nkmempages != 0) {
830 		/*
831 		 * It's already been set (by us being here before, or
832 		 * by patching or kernel config options), bail out now.
833 		 */
834 		return;
835 	}
836 
837 	npages = physmem;
838 
839 	if (npages > NKMEMPAGES_MAX)
840 		npages = NKMEMPAGES_MAX;
841 
842 	if (npages < NKMEMPAGES_MIN)
843 		npages = NKMEMPAGES_MIN;
844 
845 	nkmempages = npages;
846 }
847 
848 /*
849  * Initialize the kernel memory allocator
850  */
851 void
852 kmeminit(void)
853 {
854 	__link_set_decl(malloc_types, struct malloc_type);
855 	struct malloc_type * const *ksp;
856 	vaddr_t kmb, kml;
857 #ifdef KMEMSTATS
858 	long indx;
859 #endif
860 
861 #if	((MAXALLOCSAVE & (MAXALLOCSAVE - 1)) != 0)
862 		ERROR!_kmeminit:_MAXALLOCSAVE_not_power_of_2
863 #endif
864 #if	(MAXALLOCSAVE > MINALLOCSIZE * 32768)
865 		ERROR!_kmeminit:_MAXALLOCSAVE_too_big
866 #endif
867 #if	(MAXALLOCSAVE < NBPG)
868 		ERROR!_kmeminit:_MAXALLOCSAVE_too_small
869 #endif
870 
871 	if (sizeof(struct freelist) > (1 << MINBUCKET))
872 		panic("minbucket too small/struct freelist too big");
873 
874 	mutex_init(&malloc_lock, MUTEX_DEFAULT, IPL_VM);
875 
876 	/*
877 	 * Compute the number of kmem_map pages, if we have not
878 	 * done so already.
879 	 */
880 	kmeminit_nkmempages();
881 
882 	kmemusage = (struct kmemusage *) uvm_km_alloc(kernel_map,
883 	    (vsize_t)(nkmempages * sizeof(struct kmemusage)), 0,
884 	    UVM_KMF_WIRED|UVM_KMF_ZERO);
885 	kmb = 0;
886 	kmem_map = uvm_km_suballoc(kernel_map, &kmb,
887 	    &kml, ((vsize_t)nkmempages << PAGE_SHIFT),
888 	    VM_MAP_INTRSAFE, false, &kmem_map_store);
889 	uvm_km_vacache_init(kmem_map, "kvakmem", 0);
890 	kmembase = (char *)kmb;
891 	kmemlimit = (char *)kml;
892 #ifdef KMEMSTATS
893 	for (indx = 0; indx < MINBUCKET + 16; indx++) {
894 		if (1 << indx >= PAGE_SIZE)
895 			kmembuckets[indx].kb_elmpercl = 1;
896 		else
897 			kmembuckets[indx].kb_elmpercl = PAGE_SIZE / (1 << indx);
898 		kmembuckets[indx].kb_highwat =
899 			5 * kmembuckets[indx].kb_elmpercl;
900 	}
901 #endif
902 
903 	/* Attach all of the statically-linked malloc types. */
904 	__link_set_foreach(ksp, malloc_types)
905 		malloc_type_attach(*ksp);
906 }
907 
908 #ifdef DDB
909 #include <ddb/db_output.h>
910 
911 /*
912  * Dump kmem statistics from ddb.
913  *
914  * usage: call dump_kmemstats
915  */
916 void	dump_kmemstats(void);
917 
918 void
919 dump_kmemstats(void)
920 {
921 #ifdef KMEMSTATS
922 	struct malloc_type *ksp;
923 
924 	for (ksp = kmemstatistics; ksp != NULL; ksp = ksp->ks_next) {
925 		if (ksp->ks_memuse == 0)
926 			continue;
927 		db_printf("%s%.*s %ld\n", ksp->ks_shortdesc,
928 		    (int)(20 - strlen(ksp->ks_shortdesc)),
929 		    "                    ",
930 		    ksp->ks_memuse);
931 	}
932 #else
933 	db_printf("Kmem stats are not being collected.\n");
934 #endif /* KMEMSTATS */
935 }
936 #endif /* DDB */
937 
938 
939 #if 0
940 /*
941  * Diagnostic messages about "Data modified on
942  * freelist" indicate a memory corruption, but
943  * they do not help tracking it down.
944  * This function can be called at various places
945  * to sanity check malloc's freelist and discover
946  * where does the corruption take place.
947  */
948 int
949 freelist_sanitycheck(void) {
950 	int i,j;
951 	struct kmembuckets *kbp;
952 	struct freelist *freep;
953 	int rv = 0;
954 
955 	for (i = MINBUCKET; i <= MINBUCKET + 15; i++) {
956 		kbp = &kmembuckets[i];
957 		freep = (struct freelist *)kbp->kb_next;
958 		j = 0;
959 		while(freep) {
960 			vm_map_lock(kmem_map);
961 			rv = uvm_map_checkprot(kmem_map, (vaddr_t)freep,
962 			    (vaddr_t)freep + sizeof(struct freelist),
963 			    VM_PROT_WRITE);
964 			vm_map_unlock(kmem_map);
965 
966 			if ((rv == 0) || (*(int *)freep != WEIRD_ADDR)) {
967 				printf("bucket %i, chunck %d at %p modified\n",
968 				    i, j, freep);
969 				return 1;
970 			}
971 			freep = (struct freelist *)freep->next;
972 			j++;
973 		}
974 	}
975 
976 	return 0;
977 }
978 #endif
979