xref: /netbsd-src/sys/kern/kern_malloc.c (revision c0179c282a5968435315a82f4128c61372c68fc3)
1 /*	$NetBSD: kern_malloc.c,v 1.105 2006/11/01 10:17:58 yamt 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 1.105 2006/11/01 10:17:58 yamt Exp $");
70 
71 #include "opt_lockdebug.h"
72 
73 #include <sys/param.h>
74 #include <sys/proc.h>
75 #include <sys/kernel.h>
76 #include <sys/malloc.h>
77 #include <sys/systm.h>
78 
79 #include <uvm/uvm_extern.h>
80 
81 static struct vm_map_kernel kmem_map_store;
82 struct vm_map *kmem_map = NULL;
83 
84 #include "opt_kmempages.h"
85 
86 #ifdef NKMEMCLUSTERS
87 #error NKMEMCLUSTERS is obsolete; remove it from your kernel config file and use NKMEMPAGES instead or let the kernel auto-size
88 #endif
89 
90 /*
91  * Default number of pages in kmem_map.  We attempt to calculate this
92  * at run-time, but allow it to be either patched or set in the kernel
93  * config file.
94  */
95 #ifndef NKMEMPAGES
96 #define	NKMEMPAGES	0
97 #endif
98 int	nkmempages = NKMEMPAGES;
99 
100 /*
101  * Defaults for lower- and upper-bounds for the kmem_map page count.
102  * Can be overridden by kernel config options.
103  */
104 #ifndef	NKMEMPAGES_MIN
105 #define	NKMEMPAGES_MIN	NKMEMPAGES_MIN_DEFAULT
106 #endif
107 
108 #ifndef NKMEMPAGES_MAX
109 #define	NKMEMPAGES_MAX	NKMEMPAGES_MAX_DEFAULT
110 #endif
111 
112 #include "opt_kmemstats.h"
113 #include "opt_malloclog.h"
114 #include "opt_malloc_debug.h"
115 
116 #define	MINALLOCSIZE	(1 << MINBUCKET)
117 #define	BUCKETINDX(size) \
118 	((size) <= (MINALLOCSIZE * 128) \
119 		? (size) <= (MINALLOCSIZE * 8) \
120 			? (size) <= (MINALLOCSIZE * 2) \
121 				? (size) <= (MINALLOCSIZE * 1) \
122 					? (MINBUCKET + 0) \
123 					: (MINBUCKET + 1) \
124 				: (size) <= (MINALLOCSIZE * 4) \
125 					? (MINBUCKET + 2) \
126 					: (MINBUCKET + 3) \
127 			: (size) <= (MINALLOCSIZE* 32) \
128 				? (size) <= (MINALLOCSIZE * 16) \
129 					? (MINBUCKET + 4) \
130 					: (MINBUCKET + 5) \
131 				: (size) <= (MINALLOCSIZE * 64) \
132 					? (MINBUCKET + 6) \
133 					: (MINBUCKET + 7) \
134 		: (size) <= (MINALLOCSIZE * 2048) \
135 			? (size) <= (MINALLOCSIZE * 512) \
136 				? (size) <= (MINALLOCSIZE * 256) \
137 					? (MINBUCKET + 8) \
138 					: (MINBUCKET + 9) \
139 				: (size) <= (MINALLOCSIZE * 1024) \
140 					? (MINBUCKET + 10) \
141 					: (MINBUCKET + 11) \
142 			: (size) <= (MINALLOCSIZE * 8192) \
143 				? (size) <= (MINALLOCSIZE * 4096) \
144 					? (MINBUCKET + 12) \
145 					: (MINBUCKET + 13) \
146 				: (size) <= (MINALLOCSIZE * 16384) \
147 					? (MINBUCKET + 14) \
148 					: (MINBUCKET + 15))
149 
150 /*
151  * Array of descriptors that describe the contents of each page
152  */
153 struct kmemusage {
154 	short ku_indx;		/* bucket index */
155 	union {
156 		u_short freecnt;/* for small allocations, free pieces in page */
157 		u_short pagecnt;/* for large allocations, pages alloced */
158 	} ku_un;
159 };
160 #define	ku_freecnt ku_un.freecnt
161 #define	ku_pagecnt ku_un.pagecnt
162 
163 struct kmembuckets kmembuckets[MINBUCKET + 16];
164 struct kmemusage *kmemusage;
165 char *kmembase, *kmemlimit;
166 
167 /*
168  * Turn virtual addresses into kmem map indicies
169  */
170 #define	btokup(addr)	(&kmemusage[((caddr_t)(addr) - kmembase) >> PGSHIFT])
171 
172 struct malloc_type *kmemstatistics;
173 
174 #ifdef MALLOCLOG
175 #ifndef MALLOCLOGSIZE
176 #define	MALLOCLOGSIZE	100000
177 #endif
178 
179 struct malloclog {
180 	void *addr;
181 	long size;
182 	struct malloc_type *type;
183 	int action;
184 	const char *file;
185 	long line;
186 } malloclog[MALLOCLOGSIZE];
187 
188 long	malloclogptr;
189 
190 static void
191 domlog(void *a, long size, struct malloc_type *type, int action,
192     const char *file, long line)
193 {
194 
195 	malloclog[malloclogptr].addr = a;
196 	malloclog[malloclogptr].size = size;
197 	malloclog[malloclogptr].type = type;
198 	malloclog[malloclogptr].action = action;
199 	malloclog[malloclogptr].file = file;
200 	malloclog[malloclogptr].line = line;
201 	malloclogptr++;
202 	if (malloclogptr >= MALLOCLOGSIZE)
203 		malloclogptr = 0;
204 }
205 
206 static void
207 hitmlog(void *a)
208 {
209 	struct malloclog *lp;
210 	long l;
211 
212 #define	PRT do { \
213 	lp = &malloclog[l]; \
214 	if (lp->addr == a && lp->action) { \
215 		printf("malloc log entry %ld:\n", l); \
216 		printf("\taddr = %p\n", lp->addr); \
217 		printf("\tsize = %ld\n", lp->size); \
218 		printf("\ttype = %s\n", lp->type->ks_shortdesc); \
219 		printf("\taction = %s\n", lp->action == 1 ? "alloc" : "free"); \
220 		printf("\tfile = %s\n", lp->file); \
221 		printf("\tline = %ld\n", lp->line); \
222 	} \
223 } while (/* CONSTCOND */0)
224 
225 	for (l = malloclogptr; l < MALLOCLOGSIZE; l++)
226 		PRT;
227 
228 	for (l = 0; l < malloclogptr; l++)
229 		PRT;
230 #undef PRT
231 }
232 #endif /* MALLOCLOG */
233 
234 #ifdef DIAGNOSTIC
235 /*
236  * This structure provides a set of masks to catch unaligned frees.
237  */
238 const long addrmask[] = { 0,
239 	0x00000001, 0x00000003, 0x00000007, 0x0000000f,
240 	0x0000001f, 0x0000003f, 0x0000007f, 0x000000ff,
241 	0x000001ff, 0x000003ff, 0x000007ff, 0x00000fff,
242 	0x00001fff, 0x00003fff, 0x00007fff, 0x0000ffff,
243 };
244 
245 /*
246  * The WEIRD_ADDR is used as known text to copy into free objects so
247  * that modifications after frees can be detected.
248  */
249 #define	WEIRD_ADDR	((uint32_t) 0xdeadbeef)
250 #ifdef DEBUG
251 #define	MAX_COPY	PAGE_SIZE
252 #else
253 #define	MAX_COPY	32
254 #endif
255 
256 /*
257  * Normally the freelist structure is used only to hold the list pointer
258  * for free objects.  However, when running with diagnostics, the first
259  * 8/16 bytes of the structure is unused except for diagnostic information,
260  * and the free list pointer is at offset 8/16 in the structure.  Since the
261  * first 8 bytes is the portion of the structure most often modified, this
262  * helps to detect memory reuse problems and avoid free list corruption.
263  */
264 struct freelist {
265 	uint32_t spare0;
266 #ifdef _LP64
267 	uint32_t spare1;		/* explicit padding */
268 #endif
269 	struct malloc_type *type;
270 	caddr_t	next;
271 };
272 #else /* !DIAGNOSTIC */
273 struct freelist {
274 	caddr_t	next;
275 };
276 #endif /* DIAGNOSTIC */
277 
278 /*
279  * The following are standard, built-in malloc types and are not
280  * specific to any subsystem.
281  */
282 MALLOC_DEFINE(M_DEVBUF, "devbuf", "device driver memory");
283 MALLOC_DEFINE(M_DMAMAP, "DMA map", "bus_dma(9) structures");
284 MALLOC_DEFINE(M_FREE, "free", "should be on free list");
285 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
286 MALLOC_DEFINE(M_SOFTINTR, "softintr", "Softinterrupt structures");
287 MALLOC_DEFINE(M_TEMP, "temp", "misc. temporary data buffers");
288 
289 /* XXX These should all be elsewhere. */
290 MALLOC_DEFINE(M_RTABLE, "routetbl", "routing tables");
291 MALLOC_DEFINE(M_FTABLE, "fragtbl", "fragment reassembly header");
292 MALLOC_DEFINE(M_UFSMNT, "UFS mount", "UFS mount structure");
293 MALLOC_DEFINE(M_NETADDR, "Export Host", "Export host address structure");
294 MALLOC_DEFINE(M_IPMOPTS, "ip_moptions", "internet multicast options");
295 MALLOC_DEFINE(M_IPMADDR, "in_multi", "internet multicast address");
296 MALLOC_DEFINE(M_MRTABLE, "mrt", "multicast routing tables");
297 MALLOC_DEFINE(M_BWMETER, "bwmeter", "multicast upcall bw meters");
298 MALLOC_DEFINE(M_1394DATA, "1394data", "IEEE 1394 data buffers");
299 
300 struct simplelock malloc_slock = SIMPLELOCK_INITIALIZER;
301 
302 /*
303  * Allocate a block of memory
304  */
305 #ifdef MALLOCLOG
306 void *
307 _malloc(unsigned long size, struct malloc_type *ksp, int flags,
308     const char *file, long line)
309 #else
310 void *
311 malloc(unsigned long size, struct malloc_type *ksp, int flags)
312 #endif /* MALLOCLOG */
313 {
314 	struct kmembuckets *kbp;
315 	struct kmemusage *kup;
316 	struct freelist *freep;
317 	long indx, npg, allocsize;
318 	int s;
319 	caddr_t va, cp, savedlist;
320 #ifdef DIAGNOSTIC
321 	uint32_t *end, *lp;
322 	int copysize;
323 #endif
324 
325 #ifdef LOCKDEBUG
326 	if ((flags & M_NOWAIT) == 0)
327 		ASSERT_SLEEPABLE(NULL, "malloc");
328 #endif
329 #ifdef MALLOC_DEBUG
330 	if (debug_malloc(size, ksp, flags, (void *) &va))
331 		return ((void *) va);
332 #endif
333 	indx = BUCKETINDX(size);
334 	kbp = &kmembuckets[indx];
335 	s = splvm();
336 	simple_lock(&malloc_slock);
337 #ifdef KMEMSTATS
338 	while (ksp->ks_memuse >= ksp->ks_limit) {
339 		if (flags & M_NOWAIT) {
340 			simple_unlock(&malloc_slock);
341 			splx(s);
342 			return ((void *) NULL);
343 		}
344 		if (ksp->ks_limblocks < 65535)
345 			ksp->ks_limblocks++;
346 		ltsleep((caddr_t)ksp, PSWP+2, ksp->ks_shortdesc, 0,
347 			&malloc_slock);
348 	}
349 	ksp->ks_size |= 1 << indx;
350 #endif
351 #ifdef DIAGNOSTIC
352 	copysize = 1 << indx < MAX_COPY ? 1 << indx : MAX_COPY;
353 #endif
354 	if (kbp->kb_next == NULL) {
355 		kbp->kb_last = NULL;
356 		if (size > MAXALLOCSAVE)
357 			allocsize = round_page(size);
358 		else
359 			allocsize = 1 << indx;
360 		npg = btoc(allocsize);
361 		simple_unlock(&malloc_slock);
362 		va = (caddr_t) uvm_km_alloc(kmem_map,
363 		    (vsize_t)ctob(npg), 0,
364 		    ((flags & M_NOWAIT) ? UVM_KMF_NOWAIT : 0) |
365 		    ((flags & M_CANFAIL) ? UVM_KMF_CANFAIL : 0) |
366 		    UVM_KMF_WIRED);
367 		if (__predict_false(va == NULL)) {
368 			/*
369 			 * Kmem_malloc() can return NULL, even if it can
370 			 * wait, if there is no map space available, because
371 			 * it can't fix that problem.  Neither can we,
372 			 * right now.  (We should release pages which
373 			 * are completely free and which are in kmembuckets
374 			 * with too many free elements.)
375 			 */
376 			if ((flags & (M_NOWAIT|M_CANFAIL)) == 0)
377 				panic("malloc: out of space in kmem_map");
378 			splx(s);
379 			return (NULL);
380 		}
381 		simple_lock(&malloc_slock);
382 #ifdef KMEMSTATS
383 		kbp->kb_total += kbp->kb_elmpercl;
384 #endif
385 		kup = btokup(va);
386 		kup->ku_indx = indx;
387 		if (allocsize > MAXALLOCSAVE) {
388 			if (npg > 65535)
389 				panic("malloc: allocation too large");
390 			kup->ku_pagecnt = npg;
391 #ifdef KMEMSTATS
392 			ksp->ks_memuse += allocsize;
393 #endif
394 			goto out;
395 		}
396 #ifdef KMEMSTATS
397 		kup->ku_freecnt = kbp->kb_elmpercl;
398 		kbp->kb_totalfree += kbp->kb_elmpercl;
399 #endif
400 		/*
401 		 * Just in case we blocked while allocating memory,
402 		 * and someone else also allocated memory for this
403 		 * kmembucket, don't assume the list is still empty.
404 		 */
405 		savedlist = kbp->kb_next;
406 		kbp->kb_next = cp = va + (npg << PAGE_SHIFT) - allocsize;
407 		for (;;) {
408 			freep = (struct freelist *)cp;
409 #ifdef DIAGNOSTIC
410 			/*
411 			 * Copy in known text to detect modification
412 			 * after freeing.
413 			 */
414 			end = (uint32_t *)&cp[copysize];
415 			for (lp = (uint32_t *)cp; lp < end; lp++)
416 				*lp = WEIRD_ADDR;
417 			freep->type = M_FREE;
418 #endif /* DIAGNOSTIC */
419 			if (cp <= va)
420 				break;
421 			cp -= allocsize;
422 			freep->next = cp;
423 		}
424 		freep->next = savedlist;
425 		if (kbp->kb_last == NULL)
426 			kbp->kb_last = (caddr_t)freep;
427 	}
428 	va = kbp->kb_next;
429 	kbp->kb_next = ((struct freelist *)va)->next;
430 #ifdef DIAGNOSTIC
431 	freep = (struct freelist *)va;
432 	/* XXX potential to get garbage pointer here. */
433 	if (kbp->kb_next) {
434 		int rv;
435 		vaddr_t addr = (vaddr_t)kbp->kb_next;
436 
437 		vm_map_lock(kmem_map);
438 		rv = uvm_map_checkprot(kmem_map, addr,
439 		    addr + sizeof(struct freelist), VM_PROT_WRITE);
440 		vm_map_unlock(kmem_map);
441 
442 		if (__predict_false(rv == 0)) {
443 			printf("Data modified on freelist: "
444 			    "word %ld of object %p size %ld previous type %s "
445 			    "(invalid addr %p)\n",
446 			    (long)((int32_t *)&kbp->kb_next - (int32_t *)kbp),
447 			    va, size, "foo", kbp->kb_next);
448 #ifdef MALLOCLOG
449 			hitmlog(va);
450 #endif
451 			kbp->kb_next = NULL;
452 		}
453 	}
454 
455 	/* Fill the fields that we've used with WEIRD_ADDR */
456 #ifdef _LP64
457 	freep->type = (struct malloc_type *)
458 	    (WEIRD_ADDR | (((u_long) WEIRD_ADDR) << 32));
459 #else
460 	freep->type = (struct malloc_type *) WEIRD_ADDR;
461 #endif
462 	end = (uint32_t *)&freep->next +
463 	    (sizeof(freep->next) / sizeof(int32_t));
464 	for (lp = (uint32_t *)&freep->next; lp < end; lp++)
465 		*lp = WEIRD_ADDR;
466 
467 	/* and check that the data hasn't been modified. */
468 	end = (uint32_t *)&va[copysize];
469 	for (lp = (uint32_t *)va; lp < end; lp++) {
470 		if (__predict_true(*lp == WEIRD_ADDR))
471 			continue;
472 		printf("Data modified on freelist: "
473 		    "word %ld of object %p size %ld previous type %s "
474 		    "(0x%x != 0x%x)\n",
475 		    (long)(lp - (uint32_t *)va), va, size,
476 		    "bar", *lp, WEIRD_ADDR);
477 #ifdef MALLOCLOG
478 		hitmlog(va);
479 #endif
480 		break;
481 	}
482 
483 	freep->spare0 = 0;
484 #endif /* DIAGNOSTIC */
485 #ifdef KMEMSTATS
486 	kup = btokup(va);
487 	if (kup->ku_indx != indx)
488 		panic("malloc: wrong bucket");
489 	if (kup->ku_freecnt == 0)
490 		panic("malloc: lost data");
491 	kup->ku_freecnt--;
492 	kbp->kb_totalfree--;
493 	ksp->ks_memuse += 1 << indx;
494 out:
495 	kbp->kb_calls++;
496 	ksp->ks_inuse++;
497 	ksp->ks_calls++;
498 	if (ksp->ks_memuse > ksp->ks_maxused)
499 		ksp->ks_maxused = ksp->ks_memuse;
500 #else
501 out:
502 #endif
503 #ifdef MALLOCLOG
504 	domlog(va, size, ksp, 1, file, line);
505 #endif
506 	simple_unlock(&malloc_slock);
507 	splx(s);
508 	if ((flags & M_ZERO) != 0)
509 		memset(va, 0, size);
510 	return ((void *) va);
511 }
512 
513 /*
514  * Free a block of memory allocated by malloc.
515  */
516 #ifdef MALLOCLOG
517 void
518 _free(void *addr, struct malloc_type *ksp, const char *file, long line)
519 #else
520 void
521 free(void *addr, struct malloc_type *ksp)
522 #endif /* MALLOCLOG */
523 {
524 	struct kmembuckets *kbp;
525 	struct kmemusage *kup;
526 	struct freelist *freep;
527 	long size;
528 	int s;
529 #ifdef DIAGNOSTIC
530 	caddr_t cp;
531 	int32_t *end, *lp;
532 	long alloc, copysize;
533 #endif
534 
535 #ifdef MALLOC_DEBUG
536 	if (debug_free(addr, ksp))
537 		return;
538 #endif
539 
540 #ifdef DIAGNOSTIC
541 	/*
542 	 * Ensure that we're free'ing something that we could
543 	 * have allocated in the first place.  That is, check
544 	 * to see that the address is within kmem_map.
545 	 */
546 	if (__predict_false((vaddr_t)addr < vm_map_min(kmem_map) ||
547 	    (vaddr_t)addr >= vm_map_max(kmem_map)))
548 		panic("free: addr %p not within kmem_map", addr);
549 #endif
550 
551 	kup = btokup(addr);
552 	size = 1 << kup->ku_indx;
553 	kbp = &kmembuckets[kup->ku_indx];
554 	s = splvm();
555 	simple_lock(&malloc_slock);
556 #ifdef MALLOCLOG
557 	domlog(addr, 0, ksp, 2, file, line);
558 #endif
559 #ifdef DIAGNOSTIC
560 	/*
561 	 * Check for returns of data that do not point to the
562 	 * beginning of the allocation.
563 	 */
564 	if (size > PAGE_SIZE)
565 		alloc = addrmask[BUCKETINDX(PAGE_SIZE)];
566 	else
567 		alloc = addrmask[kup->ku_indx];
568 	if (((u_long)addr & alloc) != 0)
569 		panic("free: unaligned addr %p, size %ld, type %s, mask %ld",
570 		    addr, size, ksp->ks_shortdesc, alloc);
571 #endif /* DIAGNOSTIC */
572 	if (size > MAXALLOCSAVE) {
573 		uvm_km_free(kmem_map, (vaddr_t)addr, ctob(kup->ku_pagecnt),
574 		    UVM_KMF_WIRED);
575 #ifdef KMEMSTATS
576 		size = kup->ku_pagecnt << PGSHIFT;
577 		ksp->ks_memuse -= size;
578 		kup->ku_indx = 0;
579 		kup->ku_pagecnt = 0;
580 		if (ksp->ks_memuse + size >= ksp->ks_limit &&
581 		    ksp->ks_memuse < ksp->ks_limit)
582 			wakeup((caddr_t)ksp);
583 #ifdef DIAGNOSTIC
584 		if (ksp->ks_inuse == 0)
585 			panic("free 1: inuse 0, probable double free");
586 #endif
587 		ksp->ks_inuse--;
588 		kbp->kb_total -= 1;
589 #endif
590 		simple_unlock(&malloc_slock);
591 		splx(s);
592 		return;
593 	}
594 	freep = (struct freelist *)addr;
595 #ifdef DIAGNOSTIC
596 	/*
597 	 * Check for multiple frees. Use a quick check to see if
598 	 * it looks free before laboriously searching the freelist.
599 	 */
600 	if (__predict_false(freep->spare0 == WEIRD_ADDR)) {
601 		for (cp = kbp->kb_next; cp;
602 		    cp = ((struct freelist *)cp)->next) {
603 			if (addr != cp)
604 				continue;
605 			printf("multiply freed item %p\n", addr);
606 #ifdef MALLOCLOG
607 			hitmlog(addr);
608 #endif
609 			panic("free: duplicated free");
610 		}
611 	}
612 #ifdef LOCKDEBUG
613 	/*
614 	 * Check if we're freeing a locked simple lock.
615 	 */
616 	simple_lock_freecheck(addr, (char *)addr + size);
617 #endif
618 	/*
619 	 * Copy in known text to detect modification after freeing
620 	 * and to make it look free. Also, save the type being freed
621 	 * so we can list likely culprit if modification is detected
622 	 * when the object is reallocated.
623 	 */
624 	copysize = size < MAX_COPY ? size : MAX_COPY;
625 	end = (int32_t *)&((caddr_t)addr)[copysize];
626 	for (lp = (int32_t *)addr; lp < end; lp++)
627 		*lp = WEIRD_ADDR;
628 	freep->type = ksp;
629 #endif /* DIAGNOSTIC */
630 #ifdef KMEMSTATS
631 	kup->ku_freecnt++;
632 	if (kup->ku_freecnt >= kbp->kb_elmpercl) {
633 		if (kup->ku_freecnt > kbp->kb_elmpercl)
634 			panic("free: multiple frees");
635 		else if (kbp->kb_totalfree > kbp->kb_highwat)
636 			kbp->kb_couldfree++;
637 	}
638 	kbp->kb_totalfree++;
639 	ksp->ks_memuse -= size;
640 	if (ksp->ks_memuse + size >= ksp->ks_limit &&
641 	    ksp->ks_memuse < ksp->ks_limit)
642 		wakeup((caddr_t)ksp);
643 #ifdef DIAGNOSTIC
644 	if (ksp->ks_inuse == 0)
645 		panic("free 2: inuse 0, probable double free");
646 #endif
647 	ksp->ks_inuse--;
648 #endif
649 	if (kbp->kb_next == NULL)
650 		kbp->kb_next = addr;
651 	else
652 		((struct freelist *)kbp->kb_last)->next = addr;
653 	freep->next = NULL;
654 	kbp->kb_last = addr;
655 	simple_unlock(&malloc_slock);
656 	splx(s);
657 }
658 
659 /*
660  * Change the size of a block of memory.
661  */
662 void *
663 realloc(void *curaddr, unsigned long newsize, struct malloc_type *ksp,
664     int flags)
665 {
666 	struct kmemusage *kup;
667 	unsigned long cursize;
668 	void *newaddr;
669 #ifdef DIAGNOSTIC
670 	long alloc;
671 #endif
672 
673 	/*
674 	 * realloc() with a NULL pointer is the same as malloc().
675 	 */
676 	if (curaddr == NULL)
677 		return (malloc(newsize, ksp, flags));
678 
679 	/*
680 	 * realloc() with zero size is the same as free().
681 	 */
682 	if (newsize == 0) {
683 		free(curaddr, ksp);
684 		return (NULL);
685 	}
686 
687 #ifdef LOCKDEBUG
688 	if ((flags & M_NOWAIT) == 0)
689 		ASSERT_SLEEPABLE(NULL, "realloc");
690 #endif
691 
692 	/*
693 	 * Find out how large the old allocation was (and do some
694 	 * sanity checking).
695 	 */
696 	kup = btokup(curaddr);
697 	cursize = 1 << kup->ku_indx;
698 
699 #ifdef DIAGNOSTIC
700 	/*
701 	 * Check for returns of data that do not point to the
702 	 * beginning of the allocation.
703 	 */
704 	if (cursize > PAGE_SIZE)
705 		alloc = addrmask[BUCKETINDX(PAGE_SIZE)];
706 	else
707 		alloc = addrmask[kup->ku_indx];
708 	if (((u_long)curaddr & alloc) != 0)
709 		panic("realloc: "
710 		    "unaligned addr %p, size %ld, type %s, mask %ld\n",
711 		    curaddr, cursize, ksp->ks_shortdesc, alloc);
712 #endif /* DIAGNOSTIC */
713 
714 	if (cursize > MAXALLOCSAVE)
715 		cursize = ctob(kup->ku_pagecnt);
716 
717 	/*
718 	 * If we already actually have as much as they want, we're done.
719 	 */
720 	if (newsize <= cursize)
721 		return (curaddr);
722 
723 	/*
724 	 * Can't satisfy the allocation with the existing block.
725 	 * Allocate a new one and copy the data.
726 	 */
727 	newaddr = malloc(newsize, ksp, flags);
728 	if (__predict_false(newaddr == NULL)) {
729 		/*
730 		 * malloc() failed, because flags included M_NOWAIT.
731 		 * Return NULL to indicate that failure.  The old
732 		 * pointer is still valid.
733 		 */
734 		return (NULL);
735 	}
736 	memcpy(newaddr, curaddr, cursize);
737 
738 	/*
739 	 * We were successful: free the old allocation and return
740 	 * the new one.
741 	 */
742 	free(curaddr, ksp);
743 	return (newaddr);
744 }
745 
746 /*
747  * Roundup size to the actual allocation size.
748  */
749 unsigned long
750 malloc_roundup(unsigned long size)
751 {
752 
753 	if (size > MAXALLOCSAVE)
754 		return (roundup(size, PAGE_SIZE));
755 	else
756 		return (1 << BUCKETINDX(size));
757 }
758 
759 /*
760  * Add a malloc type to the system.
761  */
762 void
763 malloc_type_attach(struct malloc_type *type)
764 {
765 
766 	if (nkmempages == 0)
767 		panic("malloc_type_attach: nkmempages == 0");
768 
769 	if (type->ks_magic != M_MAGIC)
770 		panic("malloc_type_attach: bad magic");
771 
772 #ifdef DIAGNOSTIC
773 	{
774 		struct malloc_type *ksp;
775 		for (ksp = kmemstatistics; ksp != NULL; ksp = ksp->ks_next) {
776 			if (ksp == type)
777 				panic("malloc_type_attach: already on list");
778 		}
779 	}
780 #endif
781 
782 #ifdef KMEMSTATS
783 	if (type->ks_limit == 0)
784 		type->ks_limit = ((u_long)nkmempages << PAGE_SHIFT) * 6U / 10U;
785 #else
786 	type->ks_limit = 0;
787 #endif
788 
789 	type->ks_next = kmemstatistics;
790 	kmemstatistics = type;
791 }
792 
793 /*
794  * Remove a malloc type from the system..
795  */
796 void
797 malloc_type_detach(struct malloc_type *type)
798 {
799 	struct malloc_type *ksp;
800 
801 #ifdef DIAGNOSTIC
802 	if (type->ks_magic != M_MAGIC)
803 		panic("malloc_type_detach: bad magic");
804 #endif
805 
806 	if (type == kmemstatistics)
807 		kmemstatistics = type->ks_next;
808 	else {
809 		for (ksp = kmemstatistics; ksp->ks_next != NULL;
810 		     ksp = ksp->ks_next) {
811 			if (ksp->ks_next == type) {
812 				ksp->ks_next = type->ks_next;
813 				break;
814 			}
815 		}
816 #ifdef DIAGNOSTIC
817 		if (ksp->ks_next == NULL)
818 			panic("malloc_type_detach: not on list");
819 #endif
820 	}
821 	type->ks_next = NULL;
822 }
823 
824 /*
825  * Set the limit on a malloc type.
826  */
827 void
828 malloc_type_setlimit(struct malloc_type *type, u_long limit)
829 {
830 #ifdef KMEMSTATS
831 	int s;
832 
833 	s = splvm();
834 	type->ks_limit = limit;
835 	splx(s);
836 #endif
837 }
838 
839 /*
840  * Compute the number of pages that kmem_map will map, that is,
841  * the size of the kernel malloc arena.
842  */
843 void
844 kmeminit_nkmempages(void)
845 {
846 	int npages;
847 
848 	if (nkmempages != 0) {
849 		/*
850 		 * It's already been set (by us being here before, or
851 		 * by patching or kernel config options), bail out now.
852 		 */
853 		return;
854 	}
855 
856 	npages = physmem;
857 
858 	if (npages > NKMEMPAGES_MAX)
859 		npages = NKMEMPAGES_MAX;
860 
861 	if (npages < NKMEMPAGES_MIN)
862 		npages = NKMEMPAGES_MIN;
863 
864 	nkmempages = npages;
865 }
866 
867 /*
868  * Initialize the kernel memory allocator
869  */
870 void
871 kmeminit(void)
872 {
873 	__link_set_decl(malloc_types, struct malloc_type);
874 	struct malloc_type * const *ksp;
875 	vaddr_t kmb, kml;
876 #ifdef KMEMSTATS
877 	long indx;
878 #endif
879 
880 #if	((MAXALLOCSAVE & (MAXALLOCSAVE - 1)) != 0)
881 		ERROR!_kmeminit:_MAXALLOCSAVE_not_power_of_2
882 #endif
883 #if	(MAXALLOCSAVE > MINALLOCSIZE * 32768)
884 		ERROR!_kmeminit:_MAXALLOCSAVE_too_big
885 #endif
886 #if	(MAXALLOCSAVE < NBPG)
887 		ERROR!_kmeminit:_MAXALLOCSAVE_too_small
888 #endif
889 
890 	if (sizeof(struct freelist) > (1 << MINBUCKET))
891 		panic("minbucket too small/struct freelist too big");
892 
893 	/*
894 	 * Compute the number of kmem_map pages, if we have not
895 	 * done so already.
896 	 */
897 	kmeminit_nkmempages();
898 
899 	kmemusage = (struct kmemusage *) uvm_km_alloc(kernel_map,
900 	    (vsize_t)(nkmempages * sizeof(struct kmemusage)), 0,
901 	    UVM_KMF_WIRED|UVM_KMF_ZERO);
902 	kmb = 0;
903 	kmem_map = uvm_km_suballoc(kernel_map, &kmb,
904 	    &kml, ((vsize_t)nkmempages << PAGE_SHIFT),
905 	    VM_MAP_INTRSAFE, FALSE, &kmem_map_store);
906 	uvm_km_vacache_init(kmem_map, "kvakmem", 0);
907 	kmembase = (char *)kmb;
908 	kmemlimit = (char *)kml;
909 #ifdef KMEMSTATS
910 	for (indx = 0; indx < MINBUCKET + 16; indx++) {
911 		if (1 << indx >= PAGE_SIZE)
912 			kmembuckets[indx].kb_elmpercl = 1;
913 		else
914 			kmembuckets[indx].kb_elmpercl = PAGE_SIZE / (1 << indx);
915 		kmembuckets[indx].kb_highwat =
916 			5 * kmembuckets[indx].kb_elmpercl;
917 	}
918 #endif
919 
920 	/* Attach all of the statically-linked malloc types. */
921 	__link_set_foreach(ksp, malloc_types)
922 		malloc_type_attach(*ksp);
923 
924 #ifdef MALLOC_DEBUG
925 	debug_malloc_init();
926 #endif
927 }
928 
929 #ifdef DDB
930 #include <ddb/db_output.h>
931 
932 /*
933  * Dump kmem statistics from ddb.
934  *
935  * usage: call dump_kmemstats
936  */
937 void	dump_kmemstats(void);
938 
939 void
940 dump_kmemstats(void)
941 {
942 #ifdef KMEMSTATS
943 	struct malloc_type *ksp;
944 
945 	for (ksp = kmemstatistics; ksp != NULL; ksp = ksp->ks_next) {
946 		if (ksp->ks_memuse == 0)
947 			continue;
948 		db_printf("%s%.*s %ld\n", ksp->ks_shortdesc,
949 		    (int)(20 - strlen(ksp->ks_shortdesc)),
950 		    "                    ",
951 		    ksp->ks_memuse);
952 	}
953 #else
954 	db_printf("Kmem stats are not being collected.\n");
955 #endif /* KMEMSTATS */
956 }
957 #endif /* DDB */
958 
959 
960 #if 0
961 /*
962  * Diagnostic messages about "Data modified on
963  * freelist" indicate a memory corruption, but
964  * they do not help tracking it down.
965  * This function can be called at various places
966  * to sanity check malloc's freelist and discover
967  * where does the corruption take place.
968  */
969 int
970 freelist_sanitycheck(void) {
971 	int i,j;
972 	struct kmembuckets *kbp;
973 	struct freelist *freep;
974 	int rv = 0;
975 
976 	for (i = MINBUCKET; i <= MINBUCKET + 15; i++) {
977 		kbp = &kmembuckets[i];
978 		freep = (struct freelist *)kbp->kb_next;
979 		j = 0;
980 		while(freep) {
981 			vm_map_lock(kmem_map);
982 			rv = uvm_map_checkprot(kmem_map, (vaddr_t)freep,
983 			    (vaddr_t)freep + sizeof(struct freelist),
984 			    VM_PROT_WRITE);
985 			vm_map_unlock(kmem_map);
986 
987 			if ((rv == 0) || (*(int *)freep != WEIRD_ADDR)) {
988 				printf("bucket %i, chunck %d at %p modified\n",
989 				    i, j, freep);
990 				return 1;
991 			}
992 			freep = (struct freelist *)freep->next;
993 			j++;
994 		}
995 	}
996 
997 	return 0;
998 }
999 #endif
1000