xref: /freebsd/sys/kern/kern_malloc.c (revision 54ebdd631db8c0bba2baab0155f603a8b5cf014a)
1 /*-
2  * Copyright (c) 1987, 1991, 1993
3  *	The Regents of the University of California.
4  * Copyright (c) 2005-2006 Robert N. M. Watson
5  * 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  * 4. 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	8.3 (Berkeley) 1/4/94
32  */
33 
34 /*
35  * Kernel malloc(9) implementation -- general purpose kernel memory allocator
36  * based on memory types.  Back end is implemented using the UMA(9) zone
37  * allocator.  A set of fixed-size buckets are used for smaller allocations,
38  * and a special UMA allocation interface is used for larger allocations.
39  * Callers declare memory types, and statistics are maintained independently
40  * for each memory type.  Statistics are maintained per-CPU for performance
41  * reasons.  See malloc(9) and comments in malloc.h for a detailed
42  * description.
43  */
44 
45 #include <sys/cdefs.h>
46 __FBSDID("$FreeBSD$");
47 
48 #include "opt_ddb.h"
49 #include "opt_kdtrace.h"
50 #include "opt_vm.h"
51 
52 #include <sys/param.h>
53 #include <sys/systm.h>
54 #include <sys/kdb.h>
55 #include <sys/kernel.h>
56 #include <sys/lock.h>
57 #include <sys/malloc.h>
58 #include <sys/mbuf.h>
59 #include <sys/mutex.h>
60 #include <sys/vmmeter.h>
61 #include <sys/proc.h>
62 #include <sys/sbuf.h>
63 #include <sys/sysctl.h>
64 #include <sys/time.h>
65 
66 #include <vm/vm.h>
67 #include <vm/pmap.h>
68 #include <vm/vm_param.h>
69 #include <vm/vm_kern.h>
70 #include <vm/vm_extern.h>
71 #include <vm/vm_map.h>
72 #include <vm/vm_page.h>
73 #include <vm/uma.h>
74 #include <vm/uma_int.h>
75 #include <vm/uma_dbg.h>
76 
77 #ifdef DEBUG_MEMGUARD
78 #include <vm/memguard.h>
79 #endif
80 #ifdef DEBUG_REDZONE
81 #include <vm/redzone.h>
82 #endif
83 
84 #if defined(INVARIANTS) && defined(__i386__)
85 #include <machine/cpu.h>
86 #endif
87 
88 #include <ddb/ddb.h>
89 
90 #ifdef KDTRACE_HOOKS
91 #include <sys/dtrace_bsd.h>
92 
93 dtrace_malloc_probe_func_t	dtrace_malloc_probe;
94 #endif
95 
96 /*
97  * When realloc() is called, if the new size is sufficiently smaller than
98  * the old size, realloc() will allocate a new, smaller block to avoid
99  * wasting memory. 'Sufficiently smaller' is defined as: newsize <=
100  * oldsize / 2^n, where REALLOC_FRACTION defines the value of 'n'.
101  */
102 #ifndef REALLOC_FRACTION
103 #define	REALLOC_FRACTION	1	/* new block if <= half the size */
104 #endif
105 
106 /*
107  * Centrally define some common malloc types.
108  */
109 MALLOC_DEFINE(M_CACHE, "cache", "Various Dynamically allocated caches");
110 MALLOC_DEFINE(M_DEVBUF, "devbuf", "device driver memory");
111 MALLOC_DEFINE(M_TEMP, "temp", "misc temporary data buffers");
112 
113 MALLOC_DEFINE(M_IP6OPT, "ip6opt", "IPv6 options");
114 MALLOC_DEFINE(M_IP6NDP, "ip6ndp", "IPv6 Neighbor Discovery");
115 
116 static void kmeminit(void *);
117 SYSINIT(kmem, SI_SUB_KMEM, SI_ORDER_FIRST, kmeminit, NULL);
118 
119 static MALLOC_DEFINE(M_FREE, "free", "should be on free list");
120 
121 static struct malloc_type *kmemstatistics;
122 static vm_offset_t kmembase;
123 static vm_offset_t kmemlimit;
124 static int kmemcount;
125 
126 #define KMEM_ZSHIFT	4
127 #define KMEM_ZBASE	16
128 #define KMEM_ZMASK	(KMEM_ZBASE - 1)
129 
130 #define KMEM_ZMAX	PAGE_SIZE
131 #define KMEM_ZSIZE	(KMEM_ZMAX >> KMEM_ZSHIFT)
132 static u_int8_t kmemsize[KMEM_ZSIZE + 1];
133 
134 /*
135  * Small malloc(9) memory allocations are allocated from a set of UMA buckets
136  * of various sizes.
137  *
138  * XXX: The comment here used to read "These won't be powers of two for
139  * long."  It's possible that a significant amount of wasted memory could be
140  * recovered by tuning the sizes of these buckets.
141  */
142 struct {
143 	int kz_size;
144 	char *kz_name;
145 	uma_zone_t kz_zone;
146 } kmemzones[] = {
147 	{16, "16", NULL},
148 	{32, "32", NULL},
149 	{64, "64", NULL},
150 	{128, "128", NULL},
151 	{256, "256", NULL},
152 	{512, "512", NULL},
153 	{1024, "1024", NULL},
154 	{2048, "2048", NULL},
155 	{4096, "4096", NULL},
156 #if PAGE_SIZE > 4096
157 	{8192, "8192", NULL},
158 #if PAGE_SIZE > 8192
159 	{16384, "16384", NULL},
160 #if PAGE_SIZE > 16384
161 	{32768, "32768", NULL},
162 #if PAGE_SIZE > 32768
163 	{65536, "65536", NULL},
164 #if PAGE_SIZE > 65536
165 #error	"Unsupported PAGE_SIZE"
166 #endif	/* 65536 */
167 #endif	/* 32768 */
168 #endif	/* 16384 */
169 #endif	/* 8192 */
170 #endif	/* 4096 */
171 	{0, NULL},
172 };
173 
174 /*
175  * Zone to allocate malloc type descriptions from.  For ABI reasons, memory
176  * types are described by a data structure passed by the declaring code, but
177  * the malloc(9) implementation has its own data structure describing the
178  * type and statistics.  This permits the malloc(9)-internal data structures
179  * to be modified without breaking binary-compiled kernel modules that
180  * declare malloc types.
181  */
182 static uma_zone_t mt_zone;
183 
184 u_long vm_kmem_size;
185 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size, CTLFLAG_RD, &vm_kmem_size, 0,
186     "Size of kernel memory");
187 
188 static u_long vm_kmem_size_min;
189 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size_min, CTLFLAG_RD, &vm_kmem_size_min, 0,
190     "Minimum size of kernel memory");
191 
192 static u_long vm_kmem_size_max;
193 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size_max, CTLFLAG_RD, &vm_kmem_size_max, 0,
194     "Maximum size of kernel memory");
195 
196 static u_int vm_kmem_size_scale;
197 SYSCTL_UINT(_vm, OID_AUTO, kmem_size_scale, CTLFLAG_RD, &vm_kmem_size_scale, 0,
198     "Scale factor for kernel memory size");
199 
200 /*
201  * The malloc_mtx protects the kmemstatistics linked list.
202  */
203 struct mtx malloc_mtx;
204 
205 #ifdef MALLOC_PROFILE
206 uint64_t krequests[KMEM_ZSIZE + 1];
207 
208 static int sysctl_kern_mprof(SYSCTL_HANDLER_ARGS);
209 #endif
210 
211 static int sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS);
212 
213 /*
214  * time_uptime of the last malloc(9) failure (induced or real).
215  */
216 static time_t t_malloc_fail;
217 
218 /*
219  * malloc(9) fault injection -- cause malloc failures every (n) mallocs when
220  * the caller specifies M_NOWAIT.  If set to 0, no failures are caused.
221  */
222 #ifdef MALLOC_MAKE_FAILURES
223 SYSCTL_NODE(_debug, OID_AUTO, malloc, CTLFLAG_RD, 0,
224     "Kernel malloc debugging options");
225 
226 static int malloc_failure_rate;
227 static int malloc_nowait_count;
228 static int malloc_failure_count;
229 SYSCTL_INT(_debug_malloc, OID_AUTO, failure_rate, CTLFLAG_RW,
230     &malloc_failure_rate, 0, "Every (n) mallocs with M_NOWAIT will fail");
231 TUNABLE_INT("debug.malloc.failure_rate", &malloc_failure_rate);
232 SYSCTL_INT(_debug_malloc, OID_AUTO, failure_count, CTLFLAG_RD,
233     &malloc_failure_count, 0, "Number of imposed M_NOWAIT malloc failures");
234 #endif
235 
236 int
237 malloc_last_fail(void)
238 {
239 
240 	return (time_uptime - t_malloc_fail);
241 }
242 
243 /*
244  * An allocation has succeeded -- update malloc type statistics for the
245  * amount of bucket size.  Occurs within a critical section so that the
246  * thread isn't preempted and doesn't migrate while updating per-PCU
247  * statistics.
248  */
249 static void
250 malloc_type_zone_allocated(struct malloc_type *mtp, unsigned long size,
251     int zindx)
252 {
253 	struct malloc_type_internal *mtip;
254 	struct malloc_type_stats *mtsp;
255 
256 	critical_enter();
257 	mtip = mtp->ks_handle;
258 	mtsp = &mtip->mti_stats[curcpu];
259 	if (size > 0) {
260 		mtsp->mts_memalloced += size;
261 		mtsp->mts_numallocs++;
262 	}
263 	if (zindx != -1)
264 		mtsp->mts_size |= 1 << zindx;
265 
266 #ifdef KDTRACE_HOOKS
267 	if (dtrace_malloc_probe != NULL) {
268 		uint32_t probe_id = mtip->mti_probes[DTMALLOC_PROBE_MALLOC];
269 		if (probe_id != 0)
270 			(dtrace_malloc_probe)(probe_id,
271 			    (uintptr_t) mtp, (uintptr_t) mtip,
272 			    (uintptr_t) mtsp, size, zindx);
273 	}
274 #endif
275 
276 	critical_exit();
277 }
278 
279 void
280 malloc_type_allocated(struct malloc_type *mtp, unsigned long size)
281 {
282 
283 	if (size > 0)
284 		malloc_type_zone_allocated(mtp, size, -1);
285 }
286 
287 /*
288  * A free operation has occurred -- update malloc type statistics for the
289  * amount of the bucket size.  Occurs within a critical section so that the
290  * thread isn't preempted and doesn't migrate while updating per-CPU
291  * statistics.
292  */
293 void
294 malloc_type_freed(struct malloc_type *mtp, unsigned long size)
295 {
296 	struct malloc_type_internal *mtip;
297 	struct malloc_type_stats *mtsp;
298 
299 	critical_enter();
300 	mtip = mtp->ks_handle;
301 	mtsp = &mtip->mti_stats[curcpu];
302 	mtsp->mts_memfreed += size;
303 	mtsp->mts_numfrees++;
304 
305 #ifdef KDTRACE_HOOKS
306 	if (dtrace_malloc_probe != NULL) {
307 		uint32_t probe_id = mtip->mti_probes[DTMALLOC_PROBE_FREE];
308 		if (probe_id != 0)
309 			(dtrace_malloc_probe)(probe_id,
310 			    (uintptr_t) mtp, (uintptr_t) mtip,
311 			    (uintptr_t) mtsp, size, 0);
312 	}
313 #endif
314 
315 	critical_exit();
316 }
317 
318 /*
319  *	malloc:
320  *
321  *	Allocate a block of memory.
322  *
323  *	If M_NOWAIT is set, this routine will not block and return NULL if
324  *	the allocation fails.
325  */
326 void *
327 malloc(unsigned long size, struct malloc_type *mtp, int flags)
328 {
329 	int indx;
330 	caddr_t va;
331 	uma_zone_t zone;
332 	uma_keg_t keg;
333 #if defined(DIAGNOSTIC) || defined(DEBUG_REDZONE)
334 	unsigned long osize = size;
335 #endif
336 
337 #ifdef INVARIANTS
338 	/*
339 	 * Check that exactly one of M_WAITOK or M_NOWAIT is specified.
340 	 */
341 	indx = flags & (M_WAITOK | M_NOWAIT);
342 	if (indx != M_NOWAIT && indx != M_WAITOK) {
343 		static	struct timeval lasterr;
344 		static	int curerr, once;
345 		if (once == 0 && ppsratecheck(&lasterr, &curerr, 1)) {
346 			printf("Bad malloc flags: %x\n", indx);
347 			kdb_backtrace();
348 			flags |= M_WAITOK;
349 			once++;
350 		}
351 	}
352 #endif
353 #ifdef MALLOC_MAKE_FAILURES
354 	if ((flags & M_NOWAIT) && (malloc_failure_rate != 0)) {
355 		atomic_add_int(&malloc_nowait_count, 1);
356 		if ((malloc_nowait_count % malloc_failure_rate) == 0) {
357 			atomic_add_int(&malloc_failure_count, 1);
358 			t_malloc_fail = time_uptime;
359 			return (NULL);
360 		}
361 	}
362 #endif
363 	if (flags & M_WAITOK)
364 		KASSERT(curthread->td_intr_nesting_level == 0,
365 		   ("malloc(M_WAITOK) in interrupt context"));
366 
367 #ifdef DEBUG_MEMGUARD
368 	if (memguard_cmp(mtp))
369 		return memguard_alloc(size, flags);
370 #endif
371 
372 #ifdef DEBUG_REDZONE
373 	size = redzone_size_ntor(size);
374 #endif
375 
376 	if (size <= KMEM_ZMAX) {
377 		if (size & KMEM_ZMASK)
378 			size = (size & ~KMEM_ZMASK) + KMEM_ZBASE;
379 		indx = kmemsize[size >> KMEM_ZSHIFT];
380 		zone = kmemzones[indx].kz_zone;
381 		keg = zone->uz_keg;
382 #ifdef MALLOC_PROFILE
383 		krequests[size >> KMEM_ZSHIFT]++;
384 #endif
385 		va = uma_zalloc(zone, flags);
386 		if (va != NULL)
387 			size = keg->uk_size;
388 		malloc_type_zone_allocated(mtp, va == NULL ? 0 : size, indx);
389 	} else {
390 		size = roundup(size, PAGE_SIZE);
391 		zone = NULL;
392 		keg = NULL;
393 		va = uma_large_malloc(size, flags);
394 		malloc_type_allocated(mtp, va == NULL ? 0 : size);
395 	}
396 	if (flags & M_WAITOK)
397 		KASSERT(va != NULL, ("malloc(M_WAITOK) returned NULL"));
398 	else if (va == NULL)
399 		t_malloc_fail = time_uptime;
400 #ifdef DIAGNOSTIC
401 	if (va != NULL && !(flags & M_ZERO)) {
402 		memset(va, 0x70, osize);
403 	}
404 #endif
405 #ifdef DEBUG_REDZONE
406 	if (va != NULL)
407 		va = redzone_setup(va, osize);
408 #endif
409 	return ((void *) va);
410 }
411 
412 /*
413  *	free:
414  *
415  *	Free a block of memory allocated by malloc.
416  *
417  *	This routine may not block.
418  */
419 void
420 free(void *addr, struct malloc_type *mtp)
421 {
422 	uma_slab_t slab;
423 	u_long size;
424 
425 	/* free(NULL, ...) does nothing */
426 	if (addr == NULL)
427 		return;
428 
429 #ifdef DEBUG_MEMGUARD
430 	if (memguard_cmp(mtp)) {
431 		memguard_free(addr);
432 		return;
433 	}
434 #endif
435 
436 #ifdef DEBUG_REDZONE
437 	redzone_check(addr);
438 	addr = redzone_addr_ntor(addr);
439 #endif
440 
441 	size = 0;
442 
443 	slab = vtoslab((vm_offset_t)addr & (~UMA_SLAB_MASK));
444 
445 	if (slab == NULL)
446 		panic("free: address %p(%p) has not been allocated.\n",
447 		    addr, (void *)((u_long)addr & (~UMA_SLAB_MASK)));
448 
449 
450 	if (!(slab->us_flags & UMA_SLAB_MALLOC)) {
451 #ifdef INVARIANTS
452 		struct malloc_type **mtpp = addr;
453 #endif
454 		size = slab->us_keg->uk_size;
455 #ifdef INVARIANTS
456 		/*
457 		 * Cache a pointer to the malloc_type that most recently freed
458 		 * this memory here.  This way we know who is most likely to
459 		 * have stepped on it later.
460 		 *
461 		 * This code assumes that size is a multiple of 8 bytes for
462 		 * 64 bit machines
463 		 */
464 		mtpp = (struct malloc_type **)
465 		    ((unsigned long)mtpp & ~UMA_ALIGN_PTR);
466 		mtpp += (size - sizeof(struct malloc_type *)) /
467 		    sizeof(struct malloc_type *);
468 		*mtpp = mtp;
469 #endif
470 		uma_zfree_arg(LIST_FIRST(&slab->us_keg->uk_zones), addr, slab);
471 	} else {
472 		size = slab->us_size;
473 		uma_large_free(slab);
474 	}
475 	malloc_type_freed(mtp, size);
476 }
477 
478 /*
479  *	realloc: change the size of a memory block
480  */
481 void *
482 realloc(void *addr, unsigned long size, struct malloc_type *mtp, int flags)
483 {
484 	uma_slab_t slab;
485 	unsigned long alloc;
486 	void *newaddr;
487 
488 	/* realloc(NULL, ...) is equivalent to malloc(...) */
489 	if (addr == NULL)
490 		return (malloc(size, mtp, flags));
491 
492 	/*
493 	 * XXX: Should report free of old memory and alloc of new memory to
494 	 * per-CPU stats.
495 	 */
496 
497 #ifdef DEBUG_MEMGUARD
498 if (memguard_cmp(mtp)) {
499 	slab = NULL;
500 	alloc = size;
501 } else {
502 #endif
503 
504 #ifdef DEBUG_REDZONE
505 	slab = NULL;
506 	alloc = redzone_get_size(addr);
507 #else
508 	slab = vtoslab((vm_offset_t)addr & ~(UMA_SLAB_MASK));
509 
510 	/* Sanity check */
511 	KASSERT(slab != NULL,
512 	    ("realloc: address %p out of range", (void *)addr));
513 
514 	/* Get the size of the original block */
515 	if (!(slab->us_flags & UMA_SLAB_MALLOC))
516 		alloc = slab->us_keg->uk_size;
517 	else
518 		alloc = slab->us_size;
519 
520 	/* Reuse the original block if appropriate */
521 	if (size <= alloc
522 	    && (size > (alloc >> REALLOC_FRACTION) || alloc == MINALLOCSIZE))
523 		return (addr);
524 #endif /* !DEBUG_REDZONE */
525 
526 #ifdef DEBUG_MEMGUARD
527 }
528 #endif
529 
530 	/* Allocate a new, bigger (or smaller) block */
531 	if ((newaddr = malloc(size, mtp, flags)) == NULL)
532 		return (NULL);
533 
534 	/* Copy over original contents */
535 	bcopy(addr, newaddr, min(size, alloc));
536 	free(addr, mtp);
537 	return (newaddr);
538 }
539 
540 /*
541  *	reallocf: same as realloc() but free memory on failure.
542  */
543 void *
544 reallocf(void *addr, unsigned long size, struct malloc_type *mtp, int flags)
545 {
546 	void *mem;
547 
548 	if ((mem = realloc(addr, size, mtp, flags)) == NULL)
549 		free(addr, mtp);
550 	return (mem);
551 }
552 
553 /*
554  * Initialize the kernel memory allocator
555  */
556 /* ARGSUSED*/
557 static void
558 kmeminit(void *dummy)
559 {
560 	u_int8_t indx;
561 	u_long mem_size;
562 	int i;
563 
564 	mtx_init(&malloc_mtx, "malloc", NULL, MTX_DEF);
565 
566 	/*
567 	 * Try to auto-tune the kernel memory size, so that it is
568 	 * more applicable for a wider range of machine sizes.
569 	 * On an X86, a VM_KMEM_SIZE_SCALE value of 4 is good, while
570 	 * a VM_KMEM_SIZE of 12MB is a fair compromise.  The
571 	 * VM_KMEM_SIZE_MAX is dependent on the maximum KVA space
572 	 * available, and on an X86 with a total KVA space of 256MB,
573 	 * try to keep VM_KMEM_SIZE_MAX at 80MB or below.
574 	 *
575 	 * Note that the kmem_map is also used by the zone allocator,
576 	 * so make sure that there is enough space.
577 	 */
578 	vm_kmem_size = VM_KMEM_SIZE + nmbclusters * PAGE_SIZE;
579 	mem_size = cnt.v_page_count;
580 
581 #if defined(VM_KMEM_SIZE_SCALE)
582 	vm_kmem_size_scale = VM_KMEM_SIZE_SCALE;
583 #endif
584 	TUNABLE_INT_FETCH("vm.kmem_size_scale", &vm_kmem_size_scale);
585 	if (vm_kmem_size_scale > 0 &&
586 	    (mem_size / vm_kmem_size_scale) > (vm_kmem_size / PAGE_SIZE))
587 		vm_kmem_size = (mem_size / vm_kmem_size_scale) * PAGE_SIZE;
588 
589 #if defined(VM_KMEM_SIZE_MIN)
590 	vm_kmem_size_min = VM_KMEM_SIZE_MIN;
591 #endif
592 	TUNABLE_ULONG_FETCH("vm.kmem_size_min", &vm_kmem_size_min);
593 	if (vm_kmem_size_min > 0 && vm_kmem_size < vm_kmem_size_min) {
594 		vm_kmem_size = vm_kmem_size_min;
595 	}
596 
597 #if defined(VM_KMEM_SIZE_MAX)
598 	vm_kmem_size_max = VM_KMEM_SIZE_MAX;
599 #endif
600 	TUNABLE_ULONG_FETCH("vm.kmem_size_max", &vm_kmem_size_max);
601 	if (vm_kmem_size_max > 0 && vm_kmem_size >= vm_kmem_size_max)
602 		vm_kmem_size = vm_kmem_size_max;
603 
604 	/* Allow final override from the kernel environment */
605 #ifndef BURN_BRIDGES
606 	if (TUNABLE_ULONG_FETCH("kern.vm.kmem.size", &vm_kmem_size) != 0)
607 		printf("kern.vm.kmem.size is now called vm.kmem_size!\n");
608 #endif
609 	TUNABLE_ULONG_FETCH("vm.kmem_size", &vm_kmem_size);
610 
611 	/*
612 	 * Limit kmem virtual size to twice the physical memory.
613 	 * This allows for kmem map sparseness, but limits the size
614 	 * to something sane. Be careful to not overflow the 32bit
615 	 * ints while doing the check.
616 	 */
617 	if (((vm_kmem_size / 2) / PAGE_SIZE) > cnt.v_page_count)
618 		vm_kmem_size = 2 * cnt.v_page_count * PAGE_SIZE;
619 
620 	/*
621 	 * Tune settings based on the kmem map's size at this time.
622 	 */
623 	init_param3(vm_kmem_size / PAGE_SIZE);
624 
625 	kmem_map = kmem_suballoc(kernel_map, &kmembase, &kmemlimit,
626 	    vm_kmem_size, TRUE);
627 	kmem_map->system_map = 1;
628 
629 #ifdef DEBUG_MEMGUARD
630 	/*
631 	 * Initialize MemGuard if support compiled in.  MemGuard is a
632 	 * replacement allocator used for detecting tamper-after-free
633 	 * scenarios as they occur.  It is only used for debugging.
634 	 */
635 	vm_memguard_divisor = 10;
636 	TUNABLE_INT_FETCH("vm.memguard.divisor", &vm_memguard_divisor);
637 
638 	/* Pick a conservative value if provided value sucks. */
639 	if ((vm_memguard_divisor <= 0) ||
640 	    ((vm_kmem_size / vm_memguard_divisor) == 0))
641 		vm_memguard_divisor = 10;
642 	memguard_init(kmem_map, vm_kmem_size / vm_memguard_divisor);
643 #endif
644 
645 	uma_startup2();
646 
647 	mt_zone = uma_zcreate("mt_zone", sizeof(struct malloc_type_internal),
648 #ifdef INVARIANTS
649 	    mtrash_ctor, mtrash_dtor, mtrash_init, mtrash_fini,
650 #else
651 	    NULL, NULL, NULL, NULL,
652 #endif
653 	    UMA_ALIGN_PTR, UMA_ZONE_MALLOC);
654 	for (i = 0, indx = 0; kmemzones[indx].kz_size != 0; indx++) {
655 		int size = kmemzones[indx].kz_size;
656 		char *name = kmemzones[indx].kz_name;
657 
658 		kmemzones[indx].kz_zone = uma_zcreate(name, size,
659 #ifdef INVARIANTS
660 		    mtrash_ctor, mtrash_dtor, mtrash_init, mtrash_fini,
661 #else
662 		    NULL, NULL, NULL, NULL,
663 #endif
664 		    UMA_ALIGN_PTR, UMA_ZONE_MALLOC);
665 
666 		for (;i <= size; i+= KMEM_ZBASE)
667 			kmemsize[i >> KMEM_ZSHIFT] = indx;
668 
669 	}
670 }
671 
672 void
673 malloc_init(void *data)
674 {
675 	struct malloc_type_internal *mtip;
676 	struct malloc_type *mtp;
677 
678 	KASSERT(cnt.v_page_count != 0, ("malloc_register before vm_init"));
679 
680 	mtp = data;
681 	mtip = uma_zalloc(mt_zone, M_WAITOK | M_ZERO);
682 	mtp->ks_handle = mtip;
683 
684 	mtx_lock(&malloc_mtx);
685 	mtp->ks_next = kmemstatistics;
686 	kmemstatistics = mtp;
687 	kmemcount++;
688 	mtx_unlock(&malloc_mtx);
689 }
690 
691 void
692 malloc_uninit(void *data)
693 {
694 	struct malloc_type_internal *mtip;
695 	struct malloc_type_stats *mtsp;
696 	struct malloc_type *mtp, *temp;
697 	uma_slab_t slab;
698 	long temp_allocs, temp_bytes;
699 	int i;
700 
701 	mtp = data;
702 	KASSERT(mtp->ks_handle != NULL, ("malloc_deregister: cookie NULL"));
703 	mtx_lock(&malloc_mtx);
704 	mtip = mtp->ks_handle;
705 	mtp->ks_handle = NULL;
706 	if (mtp != kmemstatistics) {
707 		for (temp = kmemstatistics; temp != NULL;
708 		    temp = temp->ks_next) {
709 			if (temp->ks_next == mtp)
710 				temp->ks_next = mtp->ks_next;
711 		}
712 	} else
713 		kmemstatistics = mtp->ks_next;
714 	kmemcount--;
715 	mtx_unlock(&malloc_mtx);
716 
717 	/*
718 	 * Look for memory leaks.
719 	 */
720 	temp_allocs = temp_bytes = 0;
721 	for (i = 0; i < MAXCPU; i++) {
722 		mtsp = &mtip->mti_stats[i];
723 		temp_allocs += mtsp->mts_numallocs;
724 		temp_allocs -= mtsp->mts_numfrees;
725 		temp_bytes += mtsp->mts_memalloced;
726 		temp_bytes -= mtsp->mts_memfreed;
727 	}
728 	if (temp_allocs > 0 || temp_bytes > 0) {
729 		printf("Warning: memory type %s leaked memory on destroy "
730 		    "(%ld allocations, %ld bytes leaked).\n", mtp->ks_shortdesc,
731 		    temp_allocs, temp_bytes);
732 	}
733 
734 	slab = vtoslab((vm_offset_t) mtip & (~UMA_SLAB_MASK));
735 	uma_zfree_arg(mt_zone, mtip, slab);
736 }
737 
738 struct malloc_type *
739 malloc_desc2type(const char *desc)
740 {
741 	struct malloc_type *mtp;
742 
743 	mtx_assert(&malloc_mtx, MA_OWNED);
744 	for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
745 		if (strcmp(mtp->ks_shortdesc, desc) == 0)
746 			return (mtp);
747 	}
748 	return (NULL);
749 }
750 
751 static int
752 sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS)
753 {
754 	struct malloc_type_stream_header mtsh;
755 	struct malloc_type_internal *mtip;
756 	struct malloc_type_header mth;
757 	struct malloc_type *mtp;
758 	int buflen, count, error, i;
759 	struct sbuf sbuf;
760 	char *buffer;
761 
762 	mtx_lock(&malloc_mtx);
763 restart:
764 	mtx_assert(&malloc_mtx, MA_OWNED);
765 	count = kmemcount;
766 	mtx_unlock(&malloc_mtx);
767 	buflen = sizeof(mtsh) + count * (sizeof(mth) +
768 	    sizeof(struct malloc_type_stats) * MAXCPU) + 1;
769 	buffer = malloc(buflen, M_TEMP, M_WAITOK | M_ZERO);
770 	mtx_lock(&malloc_mtx);
771 	if (count < kmemcount) {
772 		free(buffer, M_TEMP);
773 		goto restart;
774 	}
775 
776 	sbuf_new(&sbuf, buffer, buflen, SBUF_FIXEDLEN);
777 
778 	/*
779 	 * Insert stream header.
780 	 */
781 	bzero(&mtsh, sizeof(mtsh));
782 	mtsh.mtsh_version = MALLOC_TYPE_STREAM_VERSION;
783 	mtsh.mtsh_maxcpus = MAXCPU;
784 	mtsh.mtsh_count = kmemcount;
785 	if (sbuf_bcat(&sbuf, &mtsh, sizeof(mtsh)) < 0) {
786 		mtx_unlock(&malloc_mtx);
787 		error = ENOMEM;
788 		goto out;
789 	}
790 
791 	/*
792 	 * Insert alternating sequence of type headers and type statistics.
793 	 */
794 	for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
795 		mtip = (struct malloc_type_internal *)mtp->ks_handle;
796 
797 		/*
798 		 * Insert type header.
799 		 */
800 		bzero(&mth, sizeof(mth));
801 		strlcpy(mth.mth_name, mtp->ks_shortdesc, MALLOC_MAX_NAME);
802 		if (sbuf_bcat(&sbuf, &mth, sizeof(mth)) < 0) {
803 			mtx_unlock(&malloc_mtx);
804 			error = ENOMEM;
805 			goto out;
806 		}
807 
808 		/*
809 		 * Insert type statistics for each CPU.
810 		 */
811 		for (i = 0; i < MAXCPU; i++) {
812 			if (sbuf_bcat(&sbuf, &mtip->mti_stats[i],
813 			    sizeof(mtip->mti_stats[i])) < 0) {
814 				mtx_unlock(&malloc_mtx);
815 				error = ENOMEM;
816 				goto out;
817 			}
818 		}
819 	}
820 	mtx_unlock(&malloc_mtx);
821 	sbuf_finish(&sbuf);
822 	error = SYSCTL_OUT(req, sbuf_data(&sbuf), sbuf_len(&sbuf));
823 out:
824 	sbuf_delete(&sbuf);
825 	free(buffer, M_TEMP);
826 	return (error);
827 }
828 
829 SYSCTL_PROC(_kern, OID_AUTO, malloc_stats, CTLFLAG_RD|CTLTYPE_STRUCT,
830     0, 0, sysctl_kern_malloc_stats, "s,malloc_type_ustats",
831     "Return malloc types");
832 
833 SYSCTL_INT(_kern, OID_AUTO, malloc_count, CTLFLAG_RD, &kmemcount, 0,
834     "Count of kernel malloc types");
835 
836 void
837 malloc_type_list(malloc_type_list_func_t *func, void *arg)
838 {
839 	struct malloc_type *mtp, **bufmtp;
840 	int count, i;
841 	size_t buflen;
842 
843 	mtx_lock(&malloc_mtx);
844 restart:
845 	mtx_assert(&malloc_mtx, MA_OWNED);
846 	count = kmemcount;
847 	mtx_unlock(&malloc_mtx);
848 
849 	buflen = sizeof(struct malloc_type *) * count;
850 	bufmtp = malloc(buflen, M_TEMP, M_WAITOK);
851 
852 	mtx_lock(&malloc_mtx);
853 
854 	if (count < kmemcount) {
855 		free(bufmtp, M_TEMP);
856 		goto restart;
857 	}
858 
859 	for (mtp = kmemstatistics, i = 0; mtp != NULL; mtp = mtp->ks_next, i++)
860 		bufmtp[i] = mtp;
861 
862 	mtx_unlock(&malloc_mtx);
863 
864 	for (i = 0; i < count; i++)
865 		(func)(bufmtp[i], arg);
866 
867 	free(bufmtp, M_TEMP);
868 }
869 
870 #ifdef DDB
871 DB_SHOW_COMMAND(malloc, db_show_malloc)
872 {
873 	struct malloc_type_internal *mtip;
874 	struct malloc_type *mtp;
875 	u_int64_t allocs, frees;
876 	u_int64_t alloced, freed;
877 	int i;
878 
879 	db_printf("%18s %12s  %12s %12s\n", "Type", "InUse", "MemUse",
880 	    "Requests");
881 	for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
882 		mtip = (struct malloc_type_internal *)mtp->ks_handle;
883 		allocs = 0;
884 		frees = 0;
885 		alloced = 0;
886 		freed = 0;
887 		for (i = 0; i < MAXCPU; i++) {
888 			allocs += mtip->mti_stats[i].mts_numallocs;
889 			frees += mtip->mti_stats[i].mts_numfrees;
890 			alloced += mtip->mti_stats[i].mts_memalloced;
891 			freed += mtip->mti_stats[i].mts_memfreed;
892 		}
893 		db_printf("%18s %12ju %12juK %12ju\n",
894 		    mtp->ks_shortdesc, allocs - frees,
895 		    (alloced - freed + 1023) / 1024, allocs);
896 	}
897 }
898 #endif
899 
900 #ifdef MALLOC_PROFILE
901 
902 static int
903 sysctl_kern_mprof(SYSCTL_HANDLER_ARGS)
904 {
905 	int linesize = 64;
906 	struct sbuf sbuf;
907 	uint64_t count;
908 	uint64_t waste;
909 	uint64_t mem;
910 	int bufsize;
911 	int error;
912 	char *buf;
913 	int rsize;
914 	int size;
915 	int i;
916 
917 	bufsize = linesize * (KMEM_ZSIZE + 1);
918 	bufsize += 128; 	/* For the stats line */
919 	bufsize += 128; 	/* For the banner line */
920 	waste = 0;
921 	mem = 0;
922 
923 	buf = malloc(bufsize, M_TEMP, M_WAITOK|M_ZERO);
924 	sbuf_new(&sbuf, buf, bufsize, SBUF_FIXEDLEN);
925 	sbuf_printf(&sbuf,
926 	    "\n  Size                    Requests  Real Size\n");
927 	for (i = 0; i < KMEM_ZSIZE; i++) {
928 		size = i << KMEM_ZSHIFT;
929 		rsize = kmemzones[kmemsize[i]].kz_size;
930 		count = (long long unsigned)krequests[i];
931 
932 		sbuf_printf(&sbuf, "%6d%28llu%11d\n", size,
933 		    (unsigned long long)count, rsize);
934 
935 		if ((rsize * count) > (size * count))
936 			waste += (rsize * count) - (size * count);
937 		mem += (rsize * count);
938 	}
939 	sbuf_printf(&sbuf,
940 	    "\nTotal memory used:\t%30llu\nTotal Memory wasted:\t%30llu\n",
941 	    (unsigned long long)mem, (unsigned long long)waste);
942 	sbuf_finish(&sbuf);
943 
944 	error = SYSCTL_OUT(req, sbuf_data(&sbuf), sbuf_len(&sbuf));
945 
946 	sbuf_delete(&sbuf);
947 	free(buf, M_TEMP);
948 	return (error);
949 }
950 
951 SYSCTL_OID(_kern, OID_AUTO, mprof, CTLTYPE_STRING|CTLFLAG_RD,
952     NULL, 0, sysctl_kern_mprof, "A", "Malloc Profiling");
953 #endif /* MALLOC_PROFILE */
954