xref: /netbsd-src/sys/kern/kern_malloc.c (revision fff57c5525bbe431aee7bdb3983954f0627a42cb)
1 /*	$NetBSD: kern_malloc.c,v 1.119 2008/03/17 17:05:54 ad 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 1.119 2008/03/17 17:05:54 ad 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();
332 	}
333 #endif
334 #ifdef MALLOC_DEBUG
335 	if (debug_malloc(size, ksp, flags, (void *) &va)) {
336 		if (va != 0)
337 			FREECHECK_OUT(&malloc_freecheck, (void *)va);
338 		return ((void *) va);
339 	}
340 #endif
341 	indx = BUCKETINDX(size);
342 	kbp = &kmembuckets[indx];
343 	mutex_spin_enter(&malloc_lock);
344 #ifdef KMEMSTATS
345 	while (ksp->ks_memuse >= ksp->ks_limit) {
346 		if (flags & M_NOWAIT) {
347 			mutex_spin_exit(&malloc_lock);
348 			return ((void *) NULL);
349 		}
350 		if (ksp->ks_limblocks < 65535)
351 			ksp->ks_limblocks++;
352 		mtsleep((void *)ksp, PSWP+2, ksp->ks_shortdesc, 0,
353 			&malloc_lock);
354 	}
355 	ksp->ks_size |= 1 << indx;
356 #endif
357 #ifdef DIAGNOSTIC
358 	copysize = 1 << indx < MAX_COPY ? 1 << indx : MAX_COPY;
359 #endif
360 	if (kbp->kb_next == NULL) {
361 		int s;
362 		kbp->kb_last = NULL;
363 		if (size > MAXALLOCSAVE)
364 			allocsize = round_page(size);
365 		else
366 			allocsize = 1 << indx;
367 		npg = btoc(allocsize);
368 		mutex_spin_exit(&malloc_lock);
369 		s = splvm();
370 		va = (void *) uvm_km_alloc(kmem_map,
371 		    (vsize_t)ctob(npg), 0,
372 		    ((flags & M_NOWAIT) ? UVM_KMF_NOWAIT : 0) |
373 		    ((flags & M_CANFAIL) ? UVM_KMF_CANFAIL : 0) |
374 		    UVM_KMF_WIRED);
375 		splx(s);
376 		if (__predict_false(va == NULL)) {
377 			/*
378 			 * Kmem_malloc() can return NULL, even if it can
379 			 * wait, if there is no map space available, because
380 			 * it can't fix that problem.  Neither can we,
381 			 * right now.  (We should release pages which
382 			 * are completely free and which are in kmembuckets
383 			 * with too many free elements.)
384 			 */
385 			if ((flags & (M_NOWAIT|M_CANFAIL)) == 0)
386 				panic("malloc: out of space in kmem_map");
387 			return (NULL);
388 		}
389 		mutex_spin_enter(&malloc_lock);
390 #ifdef KMEMSTATS
391 		kbp->kb_total += kbp->kb_elmpercl;
392 #endif
393 		kup = btokup(va);
394 		kup->ku_indx = indx;
395 		if (allocsize > MAXALLOCSAVE) {
396 			if (npg > 65535)
397 				panic("malloc: allocation too large");
398 			kup->ku_pagecnt = npg;
399 #ifdef KMEMSTATS
400 			ksp->ks_memuse += allocsize;
401 #endif
402 			goto out;
403 		}
404 #ifdef KMEMSTATS
405 		kup->ku_freecnt = kbp->kb_elmpercl;
406 		kbp->kb_totalfree += kbp->kb_elmpercl;
407 #endif
408 		/*
409 		 * Just in case we blocked while allocating memory,
410 		 * and someone else also allocated memory for this
411 		 * kmembucket, don't assume the list is still empty.
412 		 */
413 		savedlist = kbp->kb_next;
414 		kbp->kb_next = cp = va + (npg << PAGE_SHIFT) - allocsize;
415 		for (;;) {
416 			freep = (struct freelist *)cp;
417 #ifdef DIAGNOSTIC
418 			/*
419 			 * Copy in known text to detect modification
420 			 * after freeing.
421 			 */
422 			end = (uint32_t *)&cp[copysize];
423 			for (lp = (uint32_t *)cp; lp < end; lp++)
424 				*lp = WEIRD_ADDR;
425 			freep->type = M_FREE;
426 #endif /* DIAGNOSTIC */
427 			if (cp <= va)
428 				break;
429 			cp -= allocsize;
430 			freep->next = cp;
431 		}
432 		freep->next = savedlist;
433 		if (savedlist == NULL)
434 			kbp->kb_last = (void *)freep;
435 	}
436 	va = kbp->kb_next;
437 	kbp->kb_next = ((struct freelist *)va)->next;
438 #ifdef DIAGNOSTIC
439 	freep = (struct freelist *)va;
440 	/* XXX potential to get garbage pointer here. */
441 	if (kbp->kb_next) {
442 		int rv;
443 		vaddr_t addr = (vaddr_t)kbp->kb_next;
444 
445 		vm_map_lock(kmem_map);
446 		rv = uvm_map_checkprot(kmem_map, addr,
447 		    addr + sizeof(struct freelist), VM_PROT_WRITE);
448 		vm_map_unlock(kmem_map);
449 
450 		if (__predict_false(rv == 0)) {
451 			printf("Data modified on freelist: "
452 			    "word %ld of object %p size %ld previous type %s "
453 			    "(invalid addr %p)\n",
454 			    (long)((int32_t *)&kbp->kb_next - (int32_t *)kbp),
455 			    va, size, "foo", kbp->kb_next);
456 #ifdef MALLOCLOG
457 			hitmlog(va);
458 #endif
459 			kbp->kb_next = NULL;
460 		}
461 	}
462 
463 	/* Fill the fields that we've used with WEIRD_ADDR */
464 #ifdef _LP64
465 	freep->type = (struct malloc_type *)
466 	    (WEIRD_ADDR | (((u_long) WEIRD_ADDR) << 32));
467 #else
468 	freep->type = (struct malloc_type *) WEIRD_ADDR;
469 #endif
470 	end = (uint32_t *)&freep->next +
471 	    (sizeof(freep->next) / sizeof(int32_t));
472 	for (lp = (uint32_t *)&freep->next; lp < end; lp++)
473 		*lp = WEIRD_ADDR;
474 
475 	/* and check that the data hasn't been modified. */
476 	end = (uint32_t *)&va[copysize];
477 	for (lp = (uint32_t *)va; lp < end; lp++) {
478 		if (__predict_true(*lp == WEIRD_ADDR))
479 			continue;
480 		printf("Data modified on freelist: "
481 		    "word %ld of object %p size %ld previous type %s "
482 		    "(0x%x != 0x%x)\n",
483 		    (long)(lp - (uint32_t *)va), va, size,
484 		    "bar", *lp, WEIRD_ADDR);
485 #ifdef MALLOCLOG
486 		hitmlog(va);
487 #endif
488 		break;
489 	}
490 
491 	freep->spare0 = 0;
492 #endif /* DIAGNOSTIC */
493 #ifdef KMEMSTATS
494 	kup = btokup(va);
495 	if (kup->ku_indx != indx)
496 		panic("malloc: wrong bucket");
497 	if (kup->ku_freecnt == 0)
498 		panic("malloc: lost data");
499 	kup->ku_freecnt--;
500 	kbp->kb_totalfree--;
501 	ksp->ks_memuse += 1 << indx;
502 out:
503 	kbp->kb_calls++;
504 	ksp->ks_inuse++;
505 	ksp->ks_calls++;
506 	if (ksp->ks_memuse > ksp->ks_maxused)
507 		ksp->ks_maxused = ksp->ks_memuse;
508 #else
509 out:
510 #endif
511 #ifdef MALLOCLOG
512 	domlog(va, size, ksp, 1, file, line);
513 #endif
514 	mutex_spin_exit(&malloc_lock);
515 	if ((flags & M_ZERO) != 0)
516 		memset(va, 0, size);
517 	FREECHECK_OUT(&malloc_freecheck, (void *)va);
518 	return ((void *) va);
519 }
520 
521 /*
522  * Free a block of memory allocated by malloc.
523  */
524 #ifdef MALLOCLOG
525 void
526 _free(void *addr, struct malloc_type *ksp, const char *file, long line)
527 #else
528 void
529 free(void *addr, struct malloc_type *ksp)
530 #endif /* MALLOCLOG */
531 {
532 	struct kmembuckets *kbp;
533 	struct kmemusage *kup;
534 	struct freelist *freep;
535 	long size;
536 #ifdef DIAGNOSTIC
537 	void *cp;
538 	int32_t *end, *lp;
539 	long alloc, copysize;
540 #endif
541 
542 	FREECHECK_IN(&malloc_freecheck, addr);
543 #ifdef MALLOC_DEBUG
544 	if (debug_free(addr, ksp))
545 		return;
546 #endif
547 
548 #ifdef DIAGNOSTIC
549 	/*
550 	 * Ensure that we're free'ing something that we could
551 	 * have allocated in the first place.  That is, check
552 	 * to see that the address is within kmem_map.
553 	 */
554 	if (__predict_false((vaddr_t)addr < vm_map_min(kmem_map) ||
555 	    (vaddr_t)addr >= vm_map_max(kmem_map)))
556 		panic("free: addr %p not within kmem_map", addr);
557 #endif
558 
559 	kup = btokup(addr);
560 	size = 1 << kup->ku_indx;
561 	kbp = &kmembuckets[kup->ku_indx];
562 
563 	LOCKDEBUG_MEM_CHECK(addr,
564 	    size <= MAXALLOCSAVE ? size : ctob(kup->ku_pagecnt));
565 
566 	mutex_spin_enter(&malloc_lock);
567 #ifdef MALLOCLOG
568 	domlog(addr, 0, ksp, 2, file, line);
569 #endif
570 #ifdef DIAGNOSTIC
571 	/*
572 	 * Check for returns of data that do not point to the
573 	 * beginning of the allocation.
574 	 */
575 	if (size > PAGE_SIZE)
576 		alloc = addrmask[BUCKETINDX(PAGE_SIZE)];
577 	else
578 		alloc = addrmask[kup->ku_indx];
579 	if (((u_long)addr & alloc) != 0)
580 		panic("free: unaligned addr %p, size %ld, type %s, mask %ld",
581 		    addr, size, ksp->ks_shortdesc, alloc);
582 #endif /* DIAGNOSTIC */
583 	if (size > MAXALLOCSAVE) {
584 		uvm_km_free(kmem_map, (vaddr_t)addr, ctob(kup->ku_pagecnt),
585 		    UVM_KMF_WIRED);
586 #ifdef KMEMSTATS
587 		size = kup->ku_pagecnt << PGSHIFT;
588 		ksp->ks_memuse -= size;
589 		kup->ku_indx = 0;
590 		kup->ku_pagecnt = 0;
591 		if (ksp->ks_memuse + size >= ksp->ks_limit &&
592 		    ksp->ks_memuse < ksp->ks_limit)
593 			wakeup((void *)ksp);
594 #ifdef DIAGNOSTIC
595 		if (ksp->ks_inuse == 0)
596 			panic("free 1: inuse 0, probable double free");
597 #endif
598 		ksp->ks_inuse--;
599 		kbp->kb_total -= 1;
600 #endif
601 		mutex_spin_exit(&malloc_lock);
602 		return;
603 	}
604 	freep = (struct freelist *)addr;
605 #ifdef DIAGNOSTIC
606 	/*
607 	 * Check for multiple frees. Use a quick check to see if
608 	 * it looks free before laboriously searching the freelist.
609 	 */
610 	if (__predict_false(freep->spare0 == WEIRD_ADDR)) {
611 		for (cp = kbp->kb_next; cp;
612 		    cp = ((struct freelist *)cp)->next) {
613 			if (addr != cp)
614 				continue;
615 			printf("multiply freed item %p\n", addr);
616 #ifdef MALLOCLOG
617 			hitmlog(addr);
618 #endif
619 			panic("free: duplicated free");
620 		}
621 	}
622 
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_spin_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();
694 	}
695 #endif
696 
697 	/*
698 	 * Find out how large the old allocation was (and do some
699 	 * sanity checking).
700 	 */
701 	kup = btokup(curaddr);
702 	cursize = 1 << kup->ku_indx;
703 
704 #ifdef DIAGNOSTIC
705 	/*
706 	 * Check for returns of data that do not point to the
707 	 * beginning of the allocation.
708 	 */
709 	if (cursize > PAGE_SIZE)
710 		alloc = addrmask[BUCKETINDX(PAGE_SIZE)];
711 	else
712 		alloc = addrmask[kup->ku_indx];
713 	if (((u_long)curaddr & alloc) != 0)
714 		panic("realloc: "
715 		    "unaligned addr %p, size %ld, type %s, mask %ld\n",
716 		    curaddr, cursize, ksp->ks_shortdesc, alloc);
717 #endif /* DIAGNOSTIC */
718 
719 	if (cursize > MAXALLOCSAVE)
720 		cursize = ctob(kup->ku_pagecnt);
721 
722 	/*
723 	 * If we already actually have as much as they want, we're done.
724 	 */
725 	if (newsize <= cursize)
726 		return (curaddr);
727 
728 	/*
729 	 * Can't satisfy the allocation with the existing block.
730 	 * Allocate a new one and copy the data.
731 	 */
732 	newaddr = malloc(newsize, ksp, flags);
733 	if (__predict_false(newaddr == NULL)) {
734 		/*
735 		 * malloc() failed, because flags included M_NOWAIT.
736 		 * Return NULL to indicate that failure.  The old
737 		 * pointer is still valid.
738 		 */
739 		return (NULL);
740 	}
741 	memcpy(newaddr, curaddr, cursize);
742 
743 	/*
744 	 * We were successful: free the old allocation and return
745 	 * the new one.
746 	 */
747 	free(curaddr, ksp);
748 	return (newaddr);
749 }
750 
751 /*
752  * Roundup size to the actual allocation size.
753  */
754 unsigned long
755 malloc_roundup(unsigned long size)
756 {
757 
758 	if (size > MAXALLOCSAVE)
759 		return (roundup(size, PAGE_SIZE));
760 	else
761 		return (1 << BUCKETINDX(size));
762 }
763 
764 /*
765  * Add a malloc type to the system.
766  */
767 void
768 malloc_type_attach(struct malloc_type *type)
769 {
770 
771 	if (nkmempages == 0)
772 		panic("malloc_type_attach: nkmempages == 0");
773 
774 	if (type->ks_magic != M_MAGIC)
775 		panic("malloc_type_attach: bad magic");
776 
777 #ifdef DIAGNOSTIC
778 	{
779 		struct malloc_type *ksp;
780 		for (ksp = kmemstatistics; ksp != NULL; ksp = ksp->ks_next) {
781 			if (ksp == type)
782 				panic("malloc_type_attach: already on list");
783 		}
784 	}
785 #endif
786 
787 #ifdef KMEMSTATS
788 	if (type->ks_limit == 0)
789 		type->ks_limit = ((u_long)nkmempages << PAGE_SHIFT) * 6U / 10U;
790 #else
791 	type->ks_limit = 0;
792 #endif
793 
794 	type->ks_next = kmemstatistics;
795 	kmemstatistics = type;
796 }
797 
798 /*
799  * Remove a malloc type from the system..
800  */
801 void
802 malloc_type_detach(struct malloc_type *type)
803 {
804 	struct malloc_type *ksp;
805 
806 #ifdef DIAGNOSTIC
807 	if (type->ks_magic != M_MAGIC)
808 		panic("malloc_type_detach: bad magic");
809 #endif
810 
811 	if (type == kmemstatistics)
812 		kmemstatistics = type->ks_next;
813 	else {
814 		for (ksp = kmemstatistics; ksp->ks_next != NULL;
815 		     ksp = ksp->ks_next) {
816 			if (ksp->ks_next == type) {
817 				ksp->ks_next = type->ks_next;
818 				break;
819 			}
820 		}
821 #ifdef DIAGNOSTIC
822 		if (ksp->ks_next == NULL)
823 			panic("malloc_type_detach: not on list");
824 #endif
825 	}
826 	type->ks_next = NULL;
827 }
828 
829 /*
830  * Set the limit on a malloc type.
831  */
832 void
833 malloc_type_setlimit(struct malloc_type *type, u_long limit)
834 {
835 #ifdef KMEMSTATS
836 	mutex_spin_enter(&malloc_lock);
837 	type->ks_limit = limit;
838 	mutex_spin_exit(&malloc_lock);
839 #endif
840 }
841 
842 /*
843  * Compute the number of pages that kmem_map will map, that is,
844  * the size of the kernel malloc arena.
845  */
846 void
847 kmeminit_nkmempages(void)
848 {
849 	int npages;
850 
851 	if (nkmempages != 0) {
852 		/*
853 		 * It's already been set (by us being here before, or
854 		 * by patching or kernel config options), bail out now.
855 		 */
856 		return;
857 	}
858 
859 	npages = physmem;
860 
861 	if (npages > NKMEMPAGES_MAX)
862 		npages = NKMEMPAGES_MAX;
863 
864 	if (npages < NKMEMPAGES_MIN)
865 		npages = NKMEMPAGES_MIN;
866 
867 	nkmempages = npages;
868 }
869 
870 /*
871  * Initialize the kernel memory allocator
872  */
873 void
874 kmeminit(void)
875 {
876 	__link_set_decl(malloc_types, struct malloc_type);
877 	struct malloc_type * const *ksp;
878 	vaddr_t kmb, kml;
879 #ifdef KMEMSTATS
880 	long indx;
881 #endif
882 
883 #if	((MAXALLOCSAVE & (MAXALLOCSAVE - 1)) != 0)
884 		ERROR!_kmeminit:_MAXALLOCSAVE_not_power_of_2
885 #endif
886 #if	(MAXALLOCSAVE > MINALLOCSIZE * 32768)
887 		ERROR!_kmeminit:_MAXALLOCSAVE_too_big
888 #endif
889 #if	(MAXALLOCSAVE < NBPG)
890 		ERROR!_kmeminit:_MAXALLOCSAVE_too_small
891 #endif
892 
893 	if (sizeof(struct freelist) > (1 << MINBUCKET))
894 		panic("minbucket too small/struct freelist too big");
895 
896 	mutex_init(&malloc_lock, MUTEX_DEFAULT, IPL_VM);
897 
898 	/*
899 	 * Compute the number of kmem_map pages, if we have not
900 	 * done so already.
901 	 */
902 	kmeminit_nkmempages();
903 
904 	kmemusage = (struct kmemusage *) uvm_km_alloc(kernel_map,
905 	    (vsize_t)(nkmempages * sizeof(struct kmemusage)), 0,
906 	    UVM_KMF_WIRED|UVM_KMF_ZERO);
907 	kmb = 0;
908 	kmem_map = uvm_km_suballoc(kernel_map, &kmb,
909 	    &kml, ((vsize_t)nkmempages << PAGE_SHIFT),
910 	    VM_MAP_INTRSAFE, false, &kmem_map_store);
911 	uvm_km_vacache_init(kmem_map, "kvakmem", 0);
912 	kmembase = (char *)kmb;
913 	kmemlimit = (char *)kml;
914 #ifdef KMEMSTATS
915 	for (indx = 0; indx < MINBUCKET + 16; indx++) {
916 		if (1 << indx >= PAGE_SIZE)
917 			kmembuckets[indx].kb_elmpercl = 1;
918 		else
919 			kmembuckets[indx].kb_elmpercl = PAGE_SIZE / (1 << indx);
920 		kmembuckets[indx].kb_highwat =
921 			5 * kmembuckets[indx].kb_elmpercl;
922 	}
923 #endif
924 
925 	/* Attach all of the statically-linked malloc types. */
926 	__link_set_foreach(ksp, malloc_types)
927 		malloc_type_attach(*ksp);
928 }
929 
930 #ifdef DDB
931 #include <ddb/db_output.h>
932 
933 /*
934  * Dump kmem statistics from ddb.
935  *
936  * usage: call dump_kmemstats
937  */
938 void	dump_kmemstats(void);
939 
940 void
941 dump_kmemstats(void)
942 {
943 #ifdef KMEMSTATS
944 	struct malloc_type *ksp;
945 
946 	for (ksp = kmemstatistics; ksp != NULL; ksp = ksp->ks_next) {
947 		if (ksp->ks_memuse == 0)
948 			continue;
949 		db_printf("%s%.*s %ld\n", ksp->ks_shortdesc,
950 		    (int)(20 - strlen(ksp->ks_shortdesc)),
951 		    "                    ",
952 		    ksp->ks_memuse);
953 	}
954 #else
955 	db_printf("Kmem stats are not being collected.\n");
956 #endif /* KMEMSTATS */
957 }
958 #endif /* DDB */
959 
960 
961 #if 0
962 /*
963  * Diagnostic messages about "Data modified on
964  * freelist" indicate a memory corruption, but
965  * they do not help tracking it down.
966  * This function can be called at various places
967  * to sanity check malloc's freelist and discover
968  * where does the corruption take place.
969  */
970 int
971 freelist_sanitycheck(void) {
972 	int i,j;
973 	struct kmembuckets *kbp;
974 	struct freelist *freep;
975 	int rv = 0;
976 
977 	for (i = MINBUCKET; i <= MINBUCKET + 15; i++) {
978 		kbp = &kmembuckets[i];
979 		freep = (struct freelist *)kbp->kb_next;
980 		j = 0;
981 		while(freep) {
982 			vm_map_lock(kmem_map);
983 			rv = uvm_map_checkprot(kmem_map, (vaddr_t)freep,
984 			    (vaddr_t)freep + sizeof(struct freelist),
985 			    VM_PROT_WRITE);
986 			vm_map_unlock(kmem_map);
987 
988 			if ((rv == 0) || (*(int *)freep != WEIRD_ADDR)) {
989 				printf("bucket %i, chunck %d at %p modified\n",
990 				    i, j, freep);
991 				return 1;
992 			}
993 			freep = (struct freelist *)freep->next;
994 			j++;
995 		}
996 	}
997 
998 	return 0;
999 }
1000 #endif
1001