xref: /netbsd-src/sys/kern/kern_malloc.c (revision 0920b4f20b78ab1ccd9f2312fbe10deaf000cbf3)
1 /*	$NetBSD: kern_malloc.c,v 1.111 2007/04/19 11:03:44 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 1.111 2007/04/19 11:03:44 yamt 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 
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 #ifdef DEBUG
168 static void *malloc_freecheck;
169 #endif
170 
171 /*
172  * Turn virtual addresses into kmem map indicies
173  */
174 #define	btokup(addr)	(&kmemusage[((char *)(addr) - kmembase) >> PGSHIFT])
175 
176 struct malloc_type *kmemstatistics;
177 
178 #ifdef MALLOCLOG
179 #ifndef MALLOCLOGSIZE
180 #define	MALLOCLOGSIZE	100000
181 #endif
182 
183 struct malloclog {
184 	void *addr;
185 	long size;
186 	struct malloc_type *type;
187 	int action;
188 	const char *file;
189 	long line;
190 } malloclog[MALLOCLOGSIZE];
191 
192 long	malloclogptr;
193 
194 static void
195 domlog(void *a, long size, struct malloc_type *type, int action,
196     const char *file, long line)
197 {
198 
199 	malloclog[malloclogptr].addr = a;
200 	malloclog[malloclogptr].size = size;
201 	malloclog[malloclogptr].type = type;
202 	malloclog[malloclogptr].action = action;
203 	malloclog[malloclogptr].file = file;
204 	malloclog[malloclogptr].line = line;
205 	malloclogptr++;
206 	if (malloclogptr >= MALLOCLOGSIZE)
207 		malloclogptr = 0;
208 }
209 
210 static void
211 hitmlog(void *a)
212 {
213 	struct malloclog *lp;
214 	long l;
215 
216 #define	PRT do { \
217 	lp = &malloclog[l]; \
218 	if (lp->addr == a && lp->action) { \
219 		printf("malloc log entry %ld:\n", l); \
220 		printf("\taddr = %p\n", lp->addr); \
221 		printf("\tsize = %ld\n", lp->size); \
222 		printf("\ttype = %s\n", lp->type->ks_shortdesc); \
223 		printf("\taction = %s\n", lp->action == 1 ? "alloc" : "free"); \
224 		printf("\tfile = %s\n", lp->file); \
225 		printf("\tline = %ld\n", lp->line); \
226 	} \
227 } while (/* CONSTCOND */0)
228 
229 	for (l = malloclogptr; l < MALLOCLOGSIZE; l++)
230 		PRT;
231 
232 	for (l = 0; l < malloclogptr; l++)
233 		PRT;
234 #undef PRT
235 }
236 #endif /* MALLOCLOG */
237 
238 #ifdef DIAGNOSTIC
239 /*
240  * This structure provides a set of masks to catch unaligned frees.
241  */
242 const long addrmask[] = { 0,
243 	0x00000001, 0x00000003, 0x00000007, 0x0000000f,
244 	0x0000001f, 0x0000003f, 0x0000007f, 0x000000ff,
245 	0x000001ff, 0x000003ff, 0x000007ff, 0x00000fff,
246 	0x00001fff, 0x00003fff, 0x00007fff, 0x0000ffff,
247 };
248 
249 /*
250  * The WEIRD_ADDR is used as known text to copy into free objects so
251  * that modifications after frees can be detected.
252  */
253 #define	WEIRD_ADDR	((uint32_t) 0xdeadbeef)
254 #ifdef DEBUG
255 #define	MAX_COPY	PAGE_SIZE
256 #else
257 #define	MAX_COPY	32
258 #endif
259 
260 /*
261  * Normally the freelist structure is used only to hold the list pointer
262  * for free objects.  However, when running with diagnostics, the first
263  * 8/16 bytes of the structure is unused except for diagnostic information,
264  * and the free list pointer is at offset 8/16 in the structure.  Since the
265  * first 8 bytes is the portion of the structure most often modified, this
266  * helps to detect memory reuse problems and avoid free list corruption.
267  */
268 struct freelist {
269 	uint32_t spare0;
270 #ifdef _LP64
271 	uint32_t spare1;		/* explicit padding */
272 #endif
273 	struct malloc_type *type;
274 	void *	next;
275 };
276 #else /* !DIAGNOSTIC */
277 struct freelist {
278 	void *	next;
279 };
280 #endif /* DIAGNOSTIC */
281 
282 /*
283  * The following are standard, built-in malloc types and are not
284  * specific to any subsystem.
285  */
286 MALLOC_DEFINE(M_DEVBUF, "devbuf", "device driver memory");
287 MALLOC_DEFINE(M_DMAMAP, "DMA map", "bus_dma(9) structures");
288 MALLOC_DEFINE(M_FREE, "free", "should be on free list");
289 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
290 MALLOC_DEFINE(M_SOFTINTR, "softintr", "Softinterrupt structures");
291 MALLOC_DEFINE(M_TEMP, "temp", "misc. temporary data buffers");
292 
293 /* XXX These should all be elsewhere. */
294 MALLOC_DEFINE(M_RTABLE, "routetbl", "routing tables");
295 MALLOC_DEFINE(M_FTABLE, "fragtbl", "fragment reassembly header");
296 MALLOC_DEFINE(M_UFSMNT, "UFS mount", "UFS mount structure");
297 MALLOC_DEFINE(M_NETADDR, "Export Host", "Export host address structure");
298 MALLOC_DEFINE(M_IPMOPTS, "ip_moptions", "internet multicast options");
299 MALLOC_DEFINE(M_IPMADDR, "in_multi", "internet multicast address");
300 MALLOC_DEFINE(M_MRTABLE, "mrt", "multicast routing tables");
301 MALLOC_DEFINE(M_BWMETER, "bwmeter", "multicast upcall bw meters");
302 MALLOC_DEFINE(M_1394DATA, "1394data", "IEEE 1394 data buffers");
303 
304 kmutex_t malloc_lock;
305 
306 /*
307  * Allocate a block of memory
308  */
309 #ifdef MALLOCLOG
310 void *
311 _malloc(unsigned long size, struct malloc_type *ksp, int flags,
312     const char *file, long line)
313 #else
314 void *
315 malloc(unsigned long size, struct malloc_type *ksp, int flags)
316 #endif /* MALLOCLOG */
317 {
318 	struct kmembuckets *kbp;
319 	struct kmemusage *kup;
320 	struct freelist *freep;
321 	long indx, npg, allocsize;
322 	char *va, *cp, *savedlist;
323 #ifdef DIAGNOSTIC
324 	uint32_t *end, *lp;
325 	int copysize;
326 #endif
327 
328 #ifdef LOCKDEBUG
329 	if ((flags & M_NOWAIT) == 0)
330 		ASSERT_SLEEPABLE(NULL, "malloc");
331 #endif
332 #ifdef MALLOC_DEBUG
333 	if (debug_malloc(size, ksp, flags, (void *) &va)) {
334 		if (va != 0)
335 			FREECHECK_OUT(&malloc_freecheck, (void *)va);
336 		return ((void *) va);
337 	}
338 #endif
339 	indx = BUCKETINDX(size);
340 	kbp = &kmembuckets[indx];
341 	mutex_enter(&malloc_lock);
342 #ifdef KMEMSTATS
343 	while (ksp->ks_memuse >= ksp->ks_limit) {
344 		if (flags & M_NOWAIT) {
345 			mutex_exit(&malloc_lock);
346 			return ((void *) NULL);
347 		}
348 		if (ksp->ks_limblocks < 65535)
349 			ksp->ks_limblocks++;
350 		mtsleep((void *)ksp, PSWP+2, ksp->ks_shortdesc, 0,
351 			&malloc_lock);
352 	}
353 	ksp->ks_size |= 1 << indx;
354 #endif
355 #ifdef DIAGNOSTIC
356 	copysize = 1 << indx < MAX_COPY ? 1 << indx : MAX_COPY;
357 #endif
358 	if (kbp->kb_next == NULL) {
359 		int s;
360 		kbp->kb_last = NULL;
361 		if (size > MAXALLOCSAVE)
362 			allocsize = round_page(size);
363 		else
364 			allocsize = 1 << indx;
365 		npg = btoc(allocsize);
366 		mutex_exit(&malloc_lock);
367 		s = splvm();
368 		va = (void *) uvm_km_alloc(kmem_map,
369 		    (vsize_t)ctob(npg), 0,
370 		    ((flags & M_NOWAIT) ? UVM_KMF_NOWAIT : 0) |
371 		    ((flags & M_CANFAIL) ? UVM_KMF_CANFAIL : 0) |
372 		    UVM_KMF_WIRED);
373 		splx(s);
374 		if (__predict_false(va == NULL)) {
375 			/*
376 			 * Kmem_malloc() can return NULL, even if it can
377 			 * wait, if there is no map space available, because
378 			 * it can't fix that problem.  Neither can we,
379 			 * right now.  (We should release pages which
380 			 * are completely free and which are in kmembuckets
381 			 * with too many free elements.)
382 			 */
383 			if ((flags & (M_NOWAIT|M_CANFAIL)) == 0)
384 				panic("malloc: out of space in kmem_map");
385 			return (NULL);
386 		}
387 		mutex_enter(&malloc_lock);
388 #ifdef KMEMSTATS
389 		kbp->kb_total += kbp->kb_elmpercl;
390 #endif
391 		kup = btokup(va);
392 		kup->ku_indx = indx;
393 		if (allocsize > MAXALLOCSAVE) {
394 			if (npg > 65535)
395 				panic("malloc: allocation too large");
396 			kup->ku_pagecnt = npg;
397 #ifdef KMEMSTATS
398 			ksp->ks_memuse += allocsize;
399 #endif
400 			goto out;
401 		}
402 #ifdef KMEMSTATS
403 		kup->ku_freecnt = kbp->kb_elmpercl;
404 		kbp->kb_totalfree += kbp->kb_elmpercl;
405 #endif
406 		/*
407 		 * Just in case we blocked while allocating memory,
408 		 * and someone else also allocated memory for this
409 		 * kmembucket, don't assume the list is still empty.
410 		 */
411 		savedlist = kbp->kb_next;
412 		kbp->kb_next = cp = va + (npg << PAGE_SHIFT) - allocsize;
413 		for (;;) {
414 			freep = (struct freelist *)cp;
415 #ifdef DIAGNOSTIC
416 			/*
417 			 * Copy in known text to detect modification
418 			 * after freeing.
419 			 */
420 			end = (uint32_t *)&cp[copysize];
421 			for (lp = (uint32_t *)cp; lp < end; lp++)
422 				*lp = WEIRD_ADDR;
423 			freep->type = M_FREE;
424 #endif /* DIAGNOSTIC */
425 			if (cp <= va)
426 				break;
427 			cp -= allocsize;
428 			freep->next = cp;
429 		}
430 		freep->next = savedlist;
431 		if (kbp->kb_last == NULL)
432 			kbp->kb_last = (void *)freep;
433 	}
434 	va = kbp->kb_next;
435 	kbp->kb_next = ((struct freelist *)va)->next;
436 #ifdef DIAGNOSTIC
437 	freep = (struct freelist *)va;
438 	/* XXX potential to get garbage pointer here. */
439 	if (kbp->kb_next) {
440 		int rv;
441 		vaddr_t addr = (vaddr_t)kbp->kb_next;
442 
443 		vm_map_lock(kmem_map);
444 		rv = uvm_map_checkprot(kmem_map, addr,
445 		    addr + sizeof(struct freelist), VM_PROT_WRITE);
446 		vm_map_unlock(kmem_map);
447 
448 		if (__predict_false(rv == 0)) {
449 			printf("Data modified on freelist: "
450 			    "word %ld of object %p size %ld previous type %s "
451 			    "(invalid addr %p)\n",
452 			    (long)((int32_t *)&kbp->kb_next - (int32_t *)kbp),
453 			    va, size, "foo", kbp->kb_next);
454 #ifdef MALLOCLOG
455 			hitmlog(va);
456 #endif
457 			kbp->kb_next = NULL;
458 		}
459 	}
460 
461 	/* Fill the fields that we've used with WEIRD_ADDR */
462 #ifdef _LP64
463 	freep->type = (struct malloc_type *)
464 	    (WEIRD_ADDR | (((u_long) WEIRD_ADDR) << 32));
465 #else
466 	freep->type = (struct malloc_type *) WEIRD_ADDR;
467 #endif
468 	end = (uint32_t *)&freep->next +
469 	    (sizeof(freep->next) / sizeof(int32_t));
470 	for (lp = (uint32_t *)&freep->next; lp < end; lp++)
471 		*lp = WEIRD_ADDR;
472 
473 	/* and check that the data hasn't been modified. */
474 	end = (uint32_t *)&va[copysize];
475 	for (lp = (uint32_t *)va; lp < end; lp++) {
476 		if (__predict_true(*lp == WEIRD_ADDR))
477 			continue;
478 		printf("Data modified on freelist: "
479 		    "word %ld of object %p size %ld previous type %s "
480 		    "(0x%x != 0x%x)\n",
481 		    (long)(lp - (uint32_t *)va), va, size,
482 		    "bar", *lp, WEIRD_ADDR);
483 #ifdef MALLOCLOG
484 		hitmlog(va);
485 #endif
486 		break;
487 	}
488 
489 	freep->spare0 = 0;
490 #endif /* DIAGNOSTIC */
491 #ifdef KMEMSTATS
492 	kup = btokup(va);
493 	if (kup->ku_indx != indx)
494 		panic("malloc: wrong bucket");
495 	if (kup->ku_freecnt == 0)
496 		panic("malloc: lost data");
497 	kup->ku_freecnt--;
498 	kbp->kb_totalfree--;
499 	ksp->ks_memuse += 1 << indx;
500 out:
501 	kbp->kb_calls++;
502 	ksp->ks_inuse++;
503 	ksp->ks_calls++;
504 	if (ksp->ks_memuse > ksp->ks_maxused)
505 		ksp->ks_maxused = ksp->ks_memuse;
506 #else
507 out:
508 #endif
509 #ifdef MALLOCLOG
510 	domlog(va, size, ksp, 1, file, line);
511 #endif
512 	mutex_exit(&malloc_lock);
513 	if ((flags & M_ZERO) != 0)
514 		memset(va, 0, size);
515 	FREECHECK_OUT(&malloc_freecheck, (void *)va);
516 	return ((void *) va);
517 }
518 
519 /*
520  * Free a block of memory allocated by malloc.
521  */
522 #ifdef MALLOCLOG
523 void
524 _free(void *addr, struct malloc_type *ksp, const char *file, long line)
525 #else
526 void
527 free(void *addr, struct malloc_type *ksp)
528 #endif /* MALLOCLOG */
529 {
530 	struct kmembuckets *kbp;
531 	struct kmemusage *kup;
532 	struct freelist *freep;
533 	long size;
534 #ifdef DIAGNOSTIC
535 	void *cp;
536 	int32_t *end, *lp;
537 	long alloc, copysize;
538 #endif
539 
540 	FREECHECK_IN(&malloc_freecheck, addr);
541 
542 #ifdef MALLOC_DEBUG
543 	if (debug_free(addr, ksp))
544 		return;
545 #endif
546 
547 #ifdef DIAGNOSTIC
548 	/*
549 	 * Ensure that we're free'ing something that we could
550 	 * have allocated in the first place.  That is, check
551 	 * to see that the address is within kmem_map.
552 	 */
553 	if (__predict_false((vaddr_t)addr < vm_map_min(kmem_map) ||
554 	    (vaddr_t)addr >= vm_map_max(kmem_map)))
555 		panic("free: addr %p not within kmem_map", addr);
556 #endif
557 
558 	kup = btokup(addr);
559 	size = 1 << kup->ku_indx;
560 	kbp = &kmembuckets[kup->ku_indx];
561 	mutex_enter(&malloc_lock);
562 #ifdef MALLOCLOG
563 	domlog(addr, 0, ksp, 2, file, line);
564 #endif
565 #ifdef DIAGNOSTIC
566 	/*
567 	 * Check for returns of data that do not point to the
568 	 * beginning of the allocation.
569 	 */
570 	if (size > PAGE_SIZE)
571 		alloc = addrmask[BUCKETINDX(PAGE_SIZE)];
572 	else
573 		alloc = addrmask[kup->ku_indx];
574 	if (((u_long)addr & alloc) != 0)
575 		panic("free: unaligned addr %p, size %ld, type %s, mask %ld",
576 		    addr, size, ksp->ks_shortdesc, alloc);
577 #endif /* DIAGNOSTIC */
578 	if (size > MAXALLOCSAVE) {
579 		uvm_km_free(kmem_map, (vaddr_t)addr, ctob(kup->ku_pagecnt),
580 		    UVM_KMF_WIRED);
581 #ifdef KMEMSTATS
582 		size = kup->ku_pagecnt << PGSHIFT;
583 		ksp->ks_memuse -= size;
584 		kup->ku_indx = 0;
585 		kup->ku_pagecnt = 0;
586 		if (ksp->ks_memuse + size >= ksp->ks_limit &&
587 		    ksp->ks_memuse < ksp->ks_limit)
588 			wakeup((void *)ksp);
589 #ifdef DIAGNOSTIC
590 		if (ksp->ks_inuse == 0)
591 			panic("free 1: inuse 0, probable double free");
592 #endif
593 		ksp->ks_inuse--;
594 		kbp->kb_total -= 1;
595 #endif
596 		mutex_exit(&malloc_lock);
597 		return;
598 	}
599 	freep = (struct freelist *)addr;
600 #ifdef DIAGNOSTIC
601 	/*
602 	 * Check for multiple frees. Use a quick check to see if
603 	 * it looks free before laboriously searching the freelist.
604 	 */
605 	if (__predict_false(freep->spare0 == WEIRD_ADDR)) {
606 		for (cp = kbp->kb_next; cp;
607 		    cp = ((struct freelist *)cp)->next) {
608 			if (addr != cp)
609 				continue;
610 			printf("multiply freed item %p\n", addr);
611 #ifdef MALLOCLOG
612 			hitmlog(addr);
613 #endif
614 			panic("free: duplicated free");
615 		}
616 	}
617 #ifdef LOCKDEBUG
618 	/*
619 	 * Check if we're freeing a locked simple lock.
620 	 */
621 	simple_lock_freecheck(addr, (char *)addr + size);
622 #endif
623 	/*
624 	 * Copy in known text to detect modification after freeing
625 	 * and to make it look free. Also, save the type being freed
626 	 * so we can list likely culprit if modification is detected
627 	 * when the object is reallocated.
628 	 */
629 	copysize = size < MAX_COPY ? size : MAX_COPY;
630 	end = (int32_t *)&((char *)addr)[copysize];
631 	for (lp = (int32_t *)addr; lp < end; lp++)
632 		*lp = WEIRD_ADDR;
633 	freep->type = ksp;
634 #endif /* DIAGNOSTIC */
635 #ifdef KMEMSTATS
636 	kup->ku_freecnt++;
637 	if (kup->ku_freecnt >= kbp->kb_elmpercl) {
638 		if (kup->ku_freecnt > kbp->kb_elmpercl)
639 			panic("free: multiple frees");
640 		else if (kbp->kb_totalfree > kbp->kb_highwat)
641 			kbp->kb_couldfree++;
642 	}
643 	kbp->kb_totalfree++;
644 	ksp->ks_memuse -= size;
645 	if (ksp->ks_memuse + size >= ksp->ks_limit &&
646 	    ksp->ks_memuse < ksp->ks_limit)
647 		wakeup((void *)ksp);
648 #ifdef DIAGNOSTIC
649 	if (ksp->ks_inuse == 0)
650 		panic("free 2: inuse 0, probable double free");
651 #endif
652 	ksp->ks_inuse--;
653 #endif
654 	if (kbp->kb_next == NULL)
655 		kbp->kb_next = addr;
656 	else
657 		((struct freelist *)kbp->kb_last)->next = addr;
658 	freep->next = NULL;
659 	kbp->kb_last = addr;
660 	mutex_exit(&malloc_lock);
661 }
662 
663 /*
664  * Change the size of a block of memory.
665  */
666 void *
667 realloc(void *curaddr, unsigned long newsize, struct malloc_type *ksp,
668     int flags)
669 {
670 	struct kmemusage *kup;
671 	unsigned long cursize;
672 	void *newaddr;
673 #ifdef DIAGNOSTIC
674 	long alloc;
675 #endif
676 
677 	/*
678 	 * realloc() with a NULL pointer is the same as malloc().
679 	 */
680 	if (curaddr == NULL)
681 		return (malloc(newsize, ksp, flags));
682 
683 	/*
684 	 * realloc() with zero size is the same as free().
685 	 */
686 	if (newsize == 0) {
687 		free(curaddr, ksp);
688 		return (NULL);
689 	}
690 
691 #ifdef LOCKDEBUG
692 	if ((flags & M_NOWAIT) == 0)
693 		ASSERT_SLEEPABLE(NULL, "realloc");
694 #endif
695 
696 	/*
697 	 * Find out how large the old allocation was (and do some
698 	 * sanity checking).
699 	 */
700 	kup = btokup(curaddr);
701 	cursize = 1 << kup->ku_indx;
702 
703 #ifdef DIAGNOSTIC
704 	/*
705 	 * Check for returns of data that do not point to the
706 	 * beginning of the allocation.
707 	 */
708 	if (cursize > PAGE_SIZE)
709 		alloc = addrmask[BUCKETINDX(PAGE_SIZE)];
710 	else
711 		alloc = addrmask[kup->ku_indx];
712 	if (((u_long)curaddr & alloc) != 0)
713 		panic("realloc: "
714 		    "unaligned addr %p, size %ld, type %s, mask %ld\n",
715 		    curaddr, cursize, ksp->ks_shortdesc, alloc);
716 #endif /* DIAGNOSTIC */
717 
718 	if (cursize > MAXALLOCSAVE)
719 		cursize = ctob(kup->ku_pagecnt);
720 
721 	/*
722 	 * If we already actually have as much as they want, we're done.
723 	 */
724 	if (newsize <= cursize)
725 		return (curaddr);
726 
727 	/*
728 	 * Can't satisfy the allocation with the existing block.
729 	 * Allocate a new one and copy the data.
730 	 */
731 	newaddr = malloc(newsize, ksp, flags);
732 	if (__predict_false(newaddr == NULL)) {
733 		/*
734 		 * malloc() failed, because flags included M_NOWAIT.
735 		 * Return NULL to indicate that failure.  The old
736 		 * pointer is still valid.
737 		 */
738 		return (NULL);
739 	}
740 	memcpy(newaddr, curaddr, cursize);
741 
742 	/*
743 	 * We were successful: free the old allocation and return
744 	 * the new one.
745 	 */
746 	free(curaddr, ksp);
747 	return (newaddr);
748 }
749 
750 /*
751  * Roundup size to the actual allocation size.
752  */
753 unsigned long
754 malloc_roundup(unsigned long size)
755 {
756 
757 	if (size > MAXALLOCSAVE)
758 		return (roundup(size, PAGE_SIZE));
759 	else
760 		return (1 << BUCKETINDX(size));
761 }
762 
763 /*
764  * Add a malloc type to the system.
765  */
766 void
767 malloc_type_attach(struct malloc_type *type)
768 {
769 
770 	if (nkmempages == 0)
771 		panic("malloc_type_attach: nkmempages == 0");
772 
773 	if (type->ks_magic != M_MAGIC)
774 		panic("malloc_type_attach: bad magic");
775 
776 #ifdef DIAGNOSTIC
777 	{
778 		struct malloc_type *ksp;
779 		for (ksp = kmemstatistics; ksp != NULL; ksp = ksp->ks_next) {
780 			if (ksp == type)
781 				panic("malloc_type_attach: already on list");
782 		}
783 	}
784 #endif
785 
786 #ifdef KMEMSTATS
787 	if (type->ks_limit == 0)
788 		type->ks_limit = ((u_long)nkmempages << PAGE_SHIFT) * 6U / 10U;
789 #else
790 	type->ks_limit = 0;
791 #endif
792 
793 	type->ks_next = kmemstatistics;
794 	kmemstatistics = type;
795 }
796 
797 /*
798  * Remove a malloc type from the system..
799  */
800 void
801 malloc_type_detach(struct malloc_type *type)
802 {
803 	struct malloc_type *ksp;
804 
805 #ifdef DIAGNOSTIC
806 	if (type->ks_magic != M_MAGIC)
807 		panic("malloc_type_detach: bad magic");
808 #endif
809 
810 	if (type == kmemstatistics)
811 		kmemstatistics = type->ks_next;
812 	else {
813 		for (ksp = kmemstatistics; ksp->ks_next != NULL;
814 		     ksp = ksp->ks_next) {
815 			if (ksp->ks_next == type) {
816 				ksp->ks_next = type->ks_next;
817 				break;
818 			}
819 		}
820 #ifdef DIAGNOSTIC
821 		if (ksp->ks_next == NULL)
822 			panic("malloc_type_detach: not on list");
823 #endif
824 	}
825 	type->ks_next = NULL;
826 }
827 
828 /*
829  * Set the limit on a malloc type.
830  */
831 void
832 malloc_type_setlimit(struct malloc_type *type, u_long limit)
833 {
834 #ifdef KMEMSTATS
835 	mutex_enter(&malloc_lock);
836 	type->ks_limit = limit;
837 	mutex_exit(&malloc_lock);
838 #endif
839 }
840 
841 /*
842  * Compute the number of pages that kmem_map will map, that is,
843  * the size of the kernel malloc arena.
844  */
845 void
846 kmeminit_nkmempages(void)
847 {
848 	int npages;
849 
850 	if (nkmempages != 0) {
851 		/*
852 		 * It's already been set (by us being here before, or
853 		 * by patching or kernel config options), bail out now.
854 		 */
855 		return;
856 	}
857 
858 	npages = physmem;
859 
860 	if (npages > NKMEMPAGES_MAX)
861 		npages = NKMEMPAGES_MAX;
862 
863 	if (npages < NKMEMPAGES_MIN)
864 		npages = NKMEMPAGES_MIN;
865 
866 	nkmempages = npages;
867 }
868 
869 /*
870  * Initialize the kernel memory allocator
871  */
872 void
873 kmeminit(void)
874 {
875 	__link_set_decl(malloc_types, struct malloc_type);
876 	struct malloc_type * const *ksp;
877 	vaddr_t kmb, kml;
878 #ifdef KMEMSTATS
879 	long indx;
880 #endif
881 
882 #if	((MAXALLOCSAVE & (MAXALLOCSAVE - 1)) != 0)
883 		ERROR!_kmeminit:_MAXALLOCSAVE_not_power_of_2
884 #endif
885 #if	(MAXALLOCSAVE > MINALLOCSIZE * 32768)
886 		ERROR!_kmeminit:_MAXALLOCSAVE_too_big
887 #endif
888 #if	(MAXALLOCSAVE < NBPG)
889 		ERROR!_kmeminit:_MAXALLOCSAVE_too_small
890 #endif
891 
892 	if (sizeof(struct freelist) > (1 << MINBUCKET))
893 		panic("minbucket too small/struct freelist too big");
894 
895 	mutex_init(&malloc_lock, MUTEX_DRIVER, IPL_VM);
896 
897 	/*
898 	 * Compute the number of kmem_map pages, if we have not
899 	 * done so already.
900 	 */
901 	kmeminit_nkmempages();
902 
903 	kmemusage = (struct kmemusage *) uvm_km_alloc(kernel_map,
904 	    (vsize_t)(nkmempages * sizeof(struct kmemusage)), 0,
905 	    UVM_KMF_WIRED|UVM_KMF_ZERO);
906 	kmb = 0;
907 	kmem_map = uvm_km_suballoc(kernel_map, &kmb,
908 	    &kml, ((vsize_t)nkmempages << PAGE_SHIFT),
909 	    VM_MAP_INTRSAFE, false, &kmem_map_store);
910 	uvm_km_vacache_init(kmem_map, "kvakmem", 0);
911 	kmembase = (char *)kmb;
912 	kmemlimit = (char *)kml;
913 #ifdef KMEMSTATS
914 	for (indx = 0; indx < MINBUCKET + 16; indx++) {
915 		if (1 << indx >= PAGE_SIZE)
916 			kmembuckets[indx].kb_elmpercl = 1;
917 		else
918 			kmembuckets[indx].kb_elmpercl = PAGE_SIZE / (1 << indx);
919 		kmembuckets[indx].kb_highwat =
920 			5 * kmembuckets[indx].kb_elmpercl;
921 	}
922 #endif
923 
924 	/* Attach all of the statically-linked malloc types. */
925 	__link_set_foreach(ksp, malloc_types)
926 		malloc_type_attach(*ksp);
927 
928 #ifdef MALLOC_DEBUG
929 	debug_malloc_init();
930 #endif
931 }
932 
933 #ifdef DDB
934 #include <ddb/db_output.h>
935 
936 /*
937  * Dump kmem statistics from ddb.
938  *
939  * usage: call dump_kmemstats
940  */
941 void	dump_kmemstats(void);
942 
943 void
944 dump_kmemstats(void)
945 {
946 #ifdef KMEMSTATS
947 	struct malloc_type *ksp;
948 
949 	for (ksp = kmemstatistics; ksp != NULL; ksp = ksp->ks_next) {
950 		if (ksp->ks_memuse == 0)
951 			continue;
952 		db_printf("%s%.*s %ld\n", ksp->ks_shortdesc,
953 		    (int)(20 - strlen(ksp->ks_shortdesc)),
954 		    "                    ",
955 		    ksp->ks_memuse);
956 	}
957 #else
958 	db_printf("Kmem stats are not being collected.\n");
959 #endif /* KMEMSTATS */
960 }
961 #endif /* DDB */
962 
963 
964 #if 0
965 /*
966  * Diagnostic messages about "Data modified on
967  * freelist" indicate a memory corruption, but
968  * they do not help tracking it down.
969  * This function can be called at various places
970  * to sanity check malloc's freelist and discover
971  * where does the corruption take place.
972  */
973 int
974 freelist_sanitycheck(void) {
975 	int i,j;
976 	struct kmembuckets *kbp;
977 	struct freelist *freep;
978 	int rv = 0;
979 
980 	for (i = MINBUCKET; i <= MINBUCKET + 15; i++) {
981 		kbp = &kmembuckets[i];
982 		freep = (struct freelist *)kbp->kb_next;
983 		j = 0;
984 		while(freep) {
985 			vm_map_lock(kmem_map);
986 			rv = uvm_map_checkprot(kmem_map, (vaddr_t)freep,
987 			    (vaddr_t)freep + sizeof(struct freelist),
988 			    VM_PROT_WRITE);
989 			vm_map_unlock(kmem_map);
990 
991 			if ((rv == 0) || (*(int *)freep != WEIRD_ADDR)) {
992 				printf("bucket %i, chunck %d at %p modified\n",
993 				    i, j, freep);
994 				return 1;
995 			}
996 			freep = (struct freelist *)freep->next;
997 			j++;
998 		}
999 	}
1000 
1001 	return 0;
1002 }
1003 #endif
1004