xref: /netbsd-src/sys/kern/kern_malloc.c (revision 267197ec1eebfcb9810ea27a89625b6ddf68e3e7)
1 /*	$NetBSD: kern_malloc.c,v 1.117 2008/01/03 01:21:08 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 1.117 2008/01/03 01:21:08 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 #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 /*
284  * The following are standard, built-in malloc types and are not
285  * specific to any subsystem.
286  */
287 MALLOC_DEFINE(M_DEVBUF, "devbuf", "device driver memory");
288 MALLOC_DEFINE(M_DMAMAP, "DMA map", "bus_dma(9) structures");
289 MALLOC_DEFINE(M_FREE, "free", "should be on free list");
290 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
291 MALLOC_DEFINE(M_SOFTINTR, "softintr", "Softinterrupt structures");
292 MALLOC_DEFINE(M_TEMP, "temp", "misc. temporary data buffers");
293 
294 /* XXX These should all be elsewhere. */
295 MALLOC_DEFINE(M_RTABLE, "routetbl", "routing tables");
296 MALLOC_DEFINE(M_FTABLE, "fragtbl", "fragment reassembly header");
297 MALLOC_DEFINE(M_UFSMNT, "UFS mount", "UFS mount structure");
298 MALLOC_DEFINE(M_NETADDR, "Export Host", "Export host address structure");
299 MALLOC_DEFINE(M_IPMOPTS, "ip_moptions", "internet multicast options");
300 MALLOC_DEFINE(M_IPMADDR, "in_multi", "internet multicast address");
301 MALLOC_DEFINE(M_MRTABLE, "mrt", "multicast routing tables");
302 MALLOC_DEFINE(M_BWMETER, "bwmeter", "multicast upcall bw meters");
303 MALLOC_DEFINE(M_1394DATA, "1394data", "IEEE 1394 data buffers");
304 
305 kmutex_t malloc_lock;
306 
307 /*
308  * Allocate a block of memory
309  */
310 #ifdef MALLOCLOG
311 void *
312 _malloc(unsigned long size, struct malloc_type *ksp, int flags,
313     const char *file, long line)
314 #else
315 void *
316 malloc(unsigned long size, struct malloc_type *ksp, int flags)
317 #endif /* MALLOCLOG */
318 {
319 	struct kmembuckets *kbp;
320 	struct kmemusage *kup;
321 	struct freelist *freep;
322 	long indx, npg, allocsize;
323 	char *va, *cp, *savedlist;
324 #ifdef DIAGNOSTIC
325 	uint32_t *end, *lp;
326 	int copysize;
327 #endif
328 
329 #ifdef LOCKDEBUG
330 	if ((flags & M_NOWAIT) == 0)
331 		ASSERT_SLEEPABLE(NULL, "malloc");
332 #endif
333 #ifdef MALLOC_DEBUG
334 	if (debug_malloc(size, ksp, flags, (void *) &va)) {
335 		if (va != 0)
336 			FREECHECK_OUT(&malloc_freecheck, (void *)va);
337 		return ((void *) va);
338 	}
339 #endif
340 	indx = BUCKETINDX(size);
341 	kbp = &kmembuckets[indx];
342 	mutex_spin_enter(&malloc_lock);
343 #ifdef KMEMSTATS
344 	while (ksp->ks_memuse >= ksp->ks_limit) {
345 		if (flags & M_NOWAIT) {
346 			mutex_spin_exit(&malloc_lock);
347 			return ((void *) NULL);
348 		}
349 		if (ksp->ks_limblocks < 65535)
350 			ksp->ks_limblocks++;
351 		mtsleep((void *)ksp, PSWP+2, ksp->ks_shortdesc, 0,
352 			&malloc_lock);
353 	}
354 	ksp->ks_size |= 1 << indx;
355 #endif
356 #ifdef DIAGNOSTIC
357 	copysize = 1 << indx < MAX_COPY ? 1 << indx : MAX_COPY;
358 #endif
359 	if (kbp->kb_next == NULL) {
360 		int s;
361 		kbp->kb_last = NULL;
362 		if (size > MAXALLOCSAVE)
363 			allocsize = round_page(size);
364 		else
365 			allocsize = 1 << indx;
366 		npg = btoc(allocsize);
367 		mutex_spin_exit(&malloc_lock);
368 		s = splvm();
369 		va = (void *) uvm_km_alloc(kmem_map,
370 		    (vsize_t)ctob(npg), 0,
371 		    ((flags & M_NOWAIT) ? UVM_KMF_NOWAIT : 0) |
372 		    ((flags & M_CANFAIL) ? UVM_KMF_CANFAIL : 0) |
373 		    UVM_KMF_WIRED);
374 		splx(s);
375 		if (__predict_false(va == NULL)) {
376 			/*
377 			 * Kmem_malloc() can return NULL, even if it can
378 			 * wait, if there is no map space available, because
379 			 * it can't fix that problem.  Neither can we,
380 			 * right now.  (We should release pages which
381 			 * are completely free and which are in kmembuckets
382 			 * with too many free elements.)
383 			 */
384 			if ((flags & (M_NOWAIT|M_CANFAIL)) == 0)
385 				panic("malloc: out of space in kmem_map");
386 			return (NULL);
387 		}
388 		mutex_spin_enter(&malloc_lock);
389 #ifdef KMEMSTATS
390 		kbp->kb_total += kbp->kb_elmpercl;
391 #endif
392 		kup = btokup(va);
393 		kup->ku_indx = indx;
394 		if (allocsize > MAXALLOCSAVE) {
395 			if (npg > 65535)
396 				panic("malloc: allocation too large");
397 			kup->ku_pagecnt = npg;
398 #ifdef KMEMSTATS
399 			ksp->ks_memuse += allocsize;
400 #endif
401 			goto out;
402 		}
403 #ifdef KMEMSTATS
404 		kup->ku_freecnt = kbp->kb_elmpercl;
405 		kbp->kb_totalfree += kbp->kb_elmpercl;
406 #endif
407 		/*
408 		 * Just in case we blocked while allocating memory,
409 		 * and someone else also allocated memory for this
410 		 * kmembucket, don't assume the list is still empty.
411 		 */
412 		savedlist = kbp->kb_next;
413 		kbp->kb_next = cp = va + (npg << PAGE_SHIFT) - allocsize;
414 		for (;;) {
415 			freep = (struct freelist *)cp;
416 #ifdef DIAGNOSTIC
417 			/*
418 			 * Copy in known text to detect modification
419 			 * after freeing.
420 			 */
421 			end = (uint32_t *)&cp[copysize];
422 			for (lp = (uint32_t *)cp; lp < end; lp++)
423 				*lp = WEIRD_ADDR;
424 			freep->type = M_FREE;
425 #endif /* DIAGNOSTIC */
426 			if (cp <= va)
427 				break;
428 			cp -= allocsize;
429 			freep->next = cp;
430 		}
431 		freep->next = savedlist;
432 		if (savedlist == NULL)
433 			kbp->kb_last = (void *)freep;
434 	}
435 	va = kbp->kb_next;
436 	kbp->kb_next = ((struct freelist *)va)->next;
437 #ifdef DIAGNOSTIC
438 	freep = (struct freelist *)va;
439 	/* XXX potential to get garbage pointer here. */
440 	if (kbp->kb_next) {
441 		int rv;
442 		vaddr_t addr = (vaddr_t)kbp->kb_next;
443 
444 		vm_map_lock(kmem_map);
445 		rv = uvm_map_checkprot(kmem_map, addr,
446 		    addr + sizeof(struct freelist), VM_PROT_WRITE);
447 		vm_map_unlock(kmem_map);
448 
449 		if (__predict_false(rv == 0)) {
450 			printf("Data modified on freelist: "
451 			    "word %ld of object %p size %ld previous type %s "
452 			    "(invalid addr %p)\n",
453 			    (long)((int32_t *)&kbp->kb_next - (int32_t *)kbp),
454 			    va, size, "foo", kbp->kb_next);
455 #ifdef MALLOCLOG
456 			hitmlog(va);
457 #endif
458 			kbp->kb_next = NULL;
459 		}
460 	}
461 
462 	/* Fill the fields that we've used with WEIRD_ADDR */
463 #ifdef _LP64
464 	freep->type = (struct malloc_type *)
465 	    (WEIRD_ADDR | (((u_long) WEIRD_ADDR) << 32));
466 #else
467 	freep->type = (struct malloc_type *) WEIRD_ADDR;
468 #endif
469 	end = (uint32_t *)&freep->next +
470 	    (sizeof(freep->next) / sizeof(int32_t));
471 	for (lp = (uint32_t *)&freep->next; lp < end; lp++)
472 		*lp = WEIRD_ADDR;
473 
474 	/* and check that the data hasn't been modified. */
475 	end = (uint32_t *)&va[copysize];
476 	for (lp = (uint32_t *)va; lp < end; lp++) {
477 		if (__predict_true(*lp == WEIRD_ADDR))
478 			continue;
479 		printf("Data modified on freelist: "
480 		    "word %ld of object %p size %ld previous type %s "
481 		    "(0x%x != 0x%x)\n",
482 		    (long)(lp - (uint32_t *)va), va, size,
483 		    "bar", *lp, WEIRD_ADDR);
484 #ifdef MALLOCLOG
485 		hitmlog(va);
486 #endif
487 		break;
488 	}
489 
490 	freep->spare0 = 0;
491 #endif /* DIAGNOSTIC */
492 #ifdef KMEMSTATS
493 	kup = btokup(va);
494 	if (kup->ku_indx != indx)
495 		panic("malloc: wrong bucket");
496 	if (kup->ku_freecnt == 0)
497 		panic("malloc: lost data");
498 	kup->ku_freecnt--;
499 	kbp->kb_totalfree--;
500 	ksp->ks_memuse += 1 << indx;
501 out:
502 	kbp->kb_calls++;
503 	ksp->ks_inuse++;
504 	ksp->ks_calls++;
505 	if (ksp->ks_memuse > ksp->ks_maxused)
506 		ksp->ks_maxused = ksp->ks_memuse;
507 #else
508 out:
509 #endif
510 #ifdef MALLOCLOG
511 	domlog(va, size, ksp, 1, file, line);
512 #endif
513 	mutex_spin_exit(&malloc_lock);
514 	if ((flags & M_ZERO) != 0)
515 		memset(va, 0, size);
516 	FREECHECK_OUT(&malloc_freecheck, (void *)va);
517 	return ((void *) va);
518 }
519 
520 /*
521  * Free a block of memory allocated by malloc.
522  */
523 #ifdef MALLOCLOG
524 void
525 _free(void *addr, struct malloc_type *ksp, const char *file, long line)
526 #else
527 void
528 free(void *addr, struct malloc_type *ksp)
529 #endif /* MALLOCLOG */
530 {
531 	struct kmembuckets *kbp;
532 	struct kmemusage *kup;
533 	struct freelist *freep;
534 	long size;
535 #ifdef DIAGNOSTIC
536 	void *cp;
537 	int32_t *end, *lp;
538 	long alloc, copysize;
539 #endif
540 
541 	FREECHECK_IN(&malloc_freecheck, addr);
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 
562 	LOCKDEBUG_MEM_CHECK(addr,
563 	    size <= MAXALLOCSAVE ? size : ctob(kup->ku_pagecnt));
564 
565 	mutex_spin_enter(&malloc_lock);
566 #ifdef MALLOCLOG
567 	domlog(addr, 0, ksp, 2, file, line);
568 #endif
569 #ifdef DIAGNOSTIC
570 	/*
571 	 * Check for returns of data that do not point to the
572 	 * beginning of the allocation.
573 	 */
574 	if (size > PAGE_SIZE)
575 		alloc = addrmask[BUCKETINDX(PAGE_SIZE)];
576 	else
577 		alloc = addrmask[kup->ku_indx];
578 	if (((u_long)addr & alloc) != 0)
579 		panic("free: unaligned addr %p, size %ld, type %s, mask %ld",
580 		    addr, size, ksp->ks_shortdesc, alloc);
581 #endif /* DIAGNOSTIC */
582 	if (size > MAXALLOCSAVE) {
583 		uvm_km_free(kmem_map, (vaddr_t)addr, ctob(kup->ku_pagecnt),
584 		    UVM_KMF_WIRED);
585 #ifdef KMEMSTATS
586 		size = kup->ku_pagecnt << PGSHIFT;
587 		ksp->ks_memuse -= size;
588 		kup->ku_indx = 0;
589 		kup->ku_pagecnt = 0;
590 		if (ksp->ks_memuse + size >= ksp->ks_limit &&
591 		    ksp->ks_memuse < ksp->ks_limit)
592 			wakeup((void *)ksp);
593 #ifdef DIAGNOSTIC
594 		if (ksp->ks_inuse == 0)
595 			panic("free 1: inuse 0, probable double free");
596 #endif
597 		ksp->ks_inuse--;
598 		kbp->kb_total -= 1;
599 #endif
600 		mutex_spin_exit(&malloc_lock);
601 		return;
602 	}
603 	freep = (struct freelist *)addr;
604 #ifdef DIAGNOSTIC
605 	/*
606 	 * Check for multiple frees. Use a quick check to see if
607 	 * it looks free before laboriously searching the freelist.
608 	 */
609 	if (__predict_false(freep->spare0 == WEIRD_ADDR)) {
610 		for (cp = kbp->kb_next; cp;
611 		    cp = ((struct freelist *)cp)->next) {
612 			if (addr != cp)
613 				continue;
614 			printf("multiply freed item %p\n", addr);
615 #ifdef MALLOCLOG
616 			hitmlog(addr);
617 #endif
618 			panic("free: duplicated free");
619 		}
620 	}
621 
622 	/*
623 	 * Copy in known text to detect modification after freeing
624 	 * and to make it look free. Also, save the type being freed
625 	 * so we can list likely culprit if modification is detected
626 	 * when the object is reallocated.
627 	 */
628 	copysize = size < MAX_COPY ? size : MAX_COPY;
629 	end = (int32_t *)&((char *)addr)[copysize];
630 	for (lp = (int32_t *)addr; lp < end; lp++)
631 		*lp = WEIRD_ADDR;
632 	freep->type = ksp;
633 #endif /* DIAGNOSTIC */
634 #ifdef KMEMSTATS
635 	kup->ku_freecnt++;
636 	if (kup->ku_freecnt >= kbp->kb_elmpercl) {
637 		if (kup->ku_freecnt > kbp->kb_elmpercl)
638 			panic("free: multiple frees");
639 		else if (kbp->kb_totalfree > kbp->kb_highwat)
640 			kbp->kb_couldfree++;
641 	}
642 	kbp->kb_totalfree++;
643 	ksp->ks_memuse -= size;
644 	if (ksp->ks_memuse + size >= ksp->ks_limit &&
645 	    ksp->ks_memuse < ksp->ks_limit)
646 		wakeup((void *)ksp);
647 #ifdef DIAGNOSTIC
648 	if (ksp->ks_inuse == 0)
649 		panic("free 2: inuse 0, probable double free");
650 #endif
651 	ksp->ks_inuse--;
652 #endif
653 	if (kbp->kb_next == NULL)
654 		kbp->kb_next = addr;
655 	else
656 		((struct freelist *)kbp->kb_last)->next = addr;
657 	freep->next = NULL;
658 	kbp->kb_last = addr;
659 	mutex_spin_exit(&malloc_lock);
660 }
661 
662 /*
663  * Change the size of a block of memory.
664  */
665 void *
666 realloc(void *curaddr, unsigned long newsize, struct malloc_type *ksp,
667     int flags)
668 {
669 	struct kmemusage *kup;
670 	unsigned long cursize;
671 	void *newaddr;
672 #ifdef DIAGNOSTIC
673 	long alloc;
674 #endif
675 
676 	/*
677 	 * realloc() with a NULL pointer is the same as malloc().
678 	 */
679 	if (curaddr == NULL)
680 		return (malloc(newsize, ksp, flags));
681 
682 	/*
683 	 * realloc() with zero size is the same as free().
684 	 */
685 	if (newsize == 0) {
686 		free(curaddr, ksp);
687 		return (NULL);
688 	}
689 
690 #ifdef LOCKDEBUG
691 	if ((flags & M_NOWAIT) == 0)
692 		ASSERT_SLEEPABLE(NULL, "realloc");
693 #endif
694 
695 	/*
696 	 * Find out how large the old allocation was (and do some
697 	 * sanity checking).
698 	 */
699 	kup = btokup(curaddr);
700 	cursize = 1 << kup->ku_indx;
701 
702 #ifdef DIAGNOSTIC
703 	/*
704 	 * Check for returns of data that do not point to the
705 	 * beginning of the allocation.
706 	 */
707 	if (cursize > PAGE_SIZE)
708 		alloc = addrmask[BUCKETINDX(PAGE_SIZE)];
709 	else
710 		alloc = addrmask[kup->ku_indx];
711 	if (((u_long)curaddr & alloc) != 0)
712 		panic("realloc: "
713 		    "unaligned addr %p, size %ld, type %s, mask %ld\n",
714 		    curaddr, cursize, ksp->ks_shortdesc, alloc);
715 #endif /* DIAGNOSTIC */
716 
717 	if (cursize > MAXALLOCSAVE)
718 		cursize = ctob(kup->ku_pagecnt);
719 
720 	/*
721 	 * If we already actually have as much as they want, we're done.
722 	 */
723 	if (newsize <= cursize)
724 		return (curaddr);
725 
726 	/*
727 	 * Can't satisfy the allocation with the existing block.
728 	 * Allocate a new one and copy the data.
729 	 */
730 	newaddr = malloc(newsize, ksp, flags);
731 	if (__predict_false(newaddr == NULL)) {
732 		/*
733 		 * malloc() failed, because flags included M_NOWAIT.
734 		 * Return NULL to indicate that failure.  The old
735 		 * pointer is still valid.
736 		 */
737 		return (NULL);
738 	}
739 	memcpy(newaddr, curaddr, cursize);
740 
741 	/*
742 	 * We were successful: free the old allocation and return
743 	 * the new one.
744 	 */
745 	free(curaddr, ksp);
746 	return (newaddr);
747 }
748 
749 /*
750  * Roundup size to the actual allocation size.
751  */
752 unsigned long
753 malloc_roundup(unsigned long size)
754 {
755 
756 	if (size > MAXALLOCSAVE)
757 		return (roundup(size, PAGE_SIZE));
758 	else
759 		return (1 << BUCKETINDX(size));
760 }
761 
762 /*
763  * Add a malloc type to the system.
764  */
765 void
766 malloc_type_attach(struct malloc_type *type)
767 {
768 
769 	if (nkmempages == 0)
770 		panic("malloc_type_attach: nkmempages == 0");
771 
772 	if (type->ks_magic != M_MAGIC)
773 		panic("malloc_type_attach: bad magic");
774 
775 #ifdef DIAGNOSTIC
776 	{
777 		struct malloc_type *ksp;
778 		for (ksp = kmemstatistics; ksp != NULL; ksp = ksp->ks_next) {
779 			if (ksp == type)
780 				panic("malloc_type_attach: already on list");
781 		}
782 	}
783 #endif
784 
785 #ifdef KMEMSTATS
786 	if (type->ks_limit == 0)
787 		type->ks_limit = ((u_long)nkmempages << PAGE_SHIFT) * 6U / 10U;
788 #else
789 	type->ks_limit = 0;
790 #endif
791 
792 	type->ks_next = kmemstatistics;
793 	kmemstatistics = type;
794 }
795 
796 /*
797  * Remove a malloc type from the system..
798  */
799 void
800 malloc_type_detach(struct malloc_type *type)
801 {
802 	struct malloc_type *ksp;
803 
804 #ifdef DIAGNOSTIC
805 	if (type->ks_magic != M_MAGIC)
806 		panic("malloc_type_detach: bad magic");
807 #endif
808 
809 	if (type == kmemstatistics)
810 		kmemstatistics = type->ks_next;
811 	else {
812 		for (ksp = kmemstatistics; ksp->ks_next != NULL;
813 		     ksp = ksp->ks_next) {
814 			if (ksp->ks_next == type) {
815 				ksp->ks_next = type->ks_next;
816 				break;
817 			}
818 		}
819 #ifdef DIAGNOSTIC
820 		if (ksp->ks_next == NULL)
821 			panic("malloc_type_detach: not on list");
822 #endif
823 	}
824 	type->ks_next = NULL;
825 }
826 
827 /*
828  * Set the limit on a malloc type.
829  */
830 void
831 malloc_type_setlimit(struct malloc_type *type, u_long limit)
832 {
833 #ifdef KMEMSTATS
834 	mutex_spin_enter(&malloc_lock);
835 	type->ks_limit = limit;
836 	mutex_spin_exit(&malloc_lock);
837 #endif
838 }
839 
840 /*
841  * Compute the number of pages that kmem_map will map, that is,
842  * the size of the kernel malloc arena.
843  */
844 void
845 kmeminit_nkmempages(void)
846 {
847 	int npages;
848 
849 	if (nkmempages != 0) {
850 		/*
851 		 * It's already been set (by us being here before, or
852 		 * by patching or kernel config options), bail out now.
853 		 */
854 		return;
855 	}
856 
857 	npages = physmem;
858 
859 	if (npages > NKMEMPAGES_MAX)
860 		npages = NKMEMPAGES_MAX;
861 
862 	if (npages < NKMEMPAGES_MIN)
863 		npages = NKMEMPAGES_MIN;
864 
865 	nkmempages = npages;
866 }
867 
868 /*
869  * Initialize the kernel memory allocator
870  */
871 void
872 kmeminit(void)
873 {
874 	__link_set_decl(malloc_types, struct malloc_type);
875 	struct malloc_type * const *ksp;
876 	vaddr_t kmb, kml;
877 #ifdef KMEMSTATS
878 	long indx;
879 #endif
880 
881 #if	((MAXALLOCSAVE & (MAXALLOCSAVE - 1)) != 0)
882 		ERROR!_kmeminit:_MAXALLOCSAVE_not_power_of_2
883 #endif
884 #if	(MAXALLOCSAVE > MINALLOCSIZE * 32768)
885 		ERROR!_kmeminit:_MAXALLOCSAVE_too_big
886 #endif
887 #if	(MAXALLOCSAVE < NBPG)
888 		ERROR!_kmeminit:_MAXALLOCSAVE_too_small
889 #endif
890 
891 	if (sizeof(struct freelist) > (1 << MINBUCKET))
892 		panic("minbucket too small/struct freelist too big");
893 
894 	mutex_init(&malloc_lock, MUTEX_DEFAULT, IPL_VM);
895 
896 	/*
897 	 * Compute the number of kmem_map pages, if we have not
898 	 * done so already.
899 	 */
900 	kmeminit_nkmempages();
901 
902 	kmemusage = (struct kmemusage *) uvm_km_alloc(kernel_map,
903 	    (vsize_t)(nkmempages * sizeof(struct kmemusage)), 0,
904 	    UVM_KMF_WIRED|UVM_KMF_ZERO);
905 	kmb = 0;
906 	kmem_map = uvm_km_suballoc(kernel_map, &kmb,
907 	    &kml, ((vsize_t)nkmempages << PAGE_SHIFT),
908 	    VM_MAP_INTRSAFE, false, &kmem_map_store);
909 	uvm_km_vacache_init(kmem_map, "kvakmem", 0);
910 	kmembase = (char *)kmb;
911 	kmemlimit = (char *)kml;
912 #ifdef KMEMSTATS
913 	for (indx = 0; indx < MINBUCKET + 16; indx++) {
914 		if (1 << indx >= PAGE_SIZE)
915 			kmembuckets[indx].kb_elmpercl = 1;
916 		else
917 			kmembuckets[indx].kb_elmpercl = PAGE_SIZE / (1 << indx);
918 		kmembuckets[indx].kb_highwat =
919 			5 * kmembuckets[indx].kb_elmpercl;
920 	}
921 #endif
922 
923 	/* Attach all of the statically-linked malloc types. */
924 	__link_set_foreach(ksp, malloc_types)
925 		malloc_type_attach(*ksp);
926 }
927 
928 #ifdef DDB
929 #include <ddb/db_output.h>
930 
931 /*
932  * Dump kmem statistics from ddb.
933  *
934  * usage: call dump_kmemstats
935  */
936 void	dump_kmemstats(void);
937 
938 void
939 dump_kmemstats(void)
940 {
941 #ifdef KMEMSTATS
942 	struct malloc_type *ksp;
943 
944 	for (ksp = kmemstatistics; ksp != NULL; ksp = ksp->ks_next) {
945 		if (ksp->ks_memuse == 0)
946 			continue;
947 		db_printf("%s%.*s %ld\n", ksp->ks_shortdesc,
948 		    (int)(20 - strlen(ksp->ks_shortdesc)),
949 		    "                    ",
950 		    ksp->ks_memuse);
951 	}
952 #else
953 	db_printf("Kmem stats are not being collected.\n");
954 #endif /* KMEMSTATS */
955 }
956 #endif /* DDB */
957 
958 
959 #if 0
960 /*
961  * Diagnostic messages about "Data modified on
962  * freelist" indicate a memory corruption, but
963  * they do not help tracking it down.
964  * This function can be called at various places
965  * to sanity check malloc's freelist and discover
966  * where does the corruption take place.
967  */
968 int
969 freelist_sanitycheck(void) {
970 	int i,j;
971 	struct kmembuckets *kbp;
972 	struct freelist *freep;
973 	int rv = 0;
974 
975 	for (i = MINBUCKET; i <= MINBUCKET + 15; i++) {
976 		kbp = &kmembuckets[i];
977 		freep = (struct freelist *)kbp->kb_next;
978 		j = 0;
979 		while(freep) {
980 			vm_map_lock(kmem_map);
981 			rv = uvm_map_checkprot(kmem_map, (vaddr_t)freep,
982 			    (vaddr_t)freep + sizeof(struct freelist),
983 			    VM_PROT_WRITE);
984 			vm_map_unlock(kmem_map);
985 
986 			if ((rv == 0) || (*(int *)freep != WEIRD_ADDR)) {
987 				printf("bucket %i, chunck %d at %p modified\n",
988 				    i, j, freep);
989 				return 1;
990 			}
991 			freep = (struct freelist *)freep->next;
992 			j++;
993 		}
994 	}
995 
996 	return 0;
997 }
998 #endif
999